摘要:

Introductory Lecture 1 Faraday Discuss. 1999 114 1»31 Oxide surfaces Hans-Joachim Freund Fritz-Haber-Institut der Max-Planck-Gesellschaft Department of Chemical Physics Faradayweg 4-6 D-14195 Berlin Germany Introduction The bulk properties of simple binary oxides are well understood and there are excellent reviews and text books available treating the various physical aspects.1h5 In sharp contrast to the situation encountered for the bulk properties rather little is known about the surfaces of oxides even the most simple ones. Only recently if compared with the thirty years of surface science that have passed by,6 researchers have started to study the surface science of oxides. There is a very useful book that marks a –rst milestone in this eÜort entitled ììThe surface science of oxidesœœ by V.E. Henrich and P. A. Cox.7 Since the publication of this book several reviews have appeared which have covered the –eld up to the present date.8h15 It is understood that there are classes of oxides exhibiting external and internal surfaces i.e. zeolites and meso-porous materials which are technologically very important. The present lecture will not discuss these even though some of the aspects which are dwelled upon here could be applied to those materials. We refer the reader to the paper of Thomas summarizing his Introductory Lecture of Faraday Discussion no. 105 where he discusses some aspects of this –eld as well.16 The present lecture has been organized as follows. In the –rst part we discuss several aspects of the geometric and electronic structure of clean oxide surfaces as determined by a variety of experimental methods.We show examples of surfaces terminating bulk single crystals as well as surfaces of epitaxial oxide –lms. This part is followed by examples attempting to illustrate some of the principles governing the interaction of molecules with oxide surfaces. A short comparison between the situation encountered on single crystalline surfaces with microcrystalline surfaces is included in order to demonstrate the in—uence of defects on the adsorption properties. The third part is dedicated to the interaction of metals with oxide surfaces and the study of deposited metal aggregates including adsorption and reaction of molecules on such systems. Such composites represent Received 6th September 1999 Oxides have gained increasing interest in surface science during recent years because of their important role in applications.In the –rst part of the lecture we review the current knowledge on morphology and structure of surfaces of bulk single crystals as well as oxide –lms. The interaction of oxide surfaces with molecules is thoroughly discussed and the role of defects on adsorption is highlighted. In a further part structure and morphology of deposited aggregates on clean and modi–ed substrates are discussed. Such systems may serve as models for heterogeneous catalysts. Electronic structure as a function of the size of the deposited particle is studied as well as size dependent adsorption properties and reactivities.This journal is( The Royal Society of Chemistry 2000 model systems for heterogeneous catalysts and allow us to try to bridge the materialœs gap between single crystal metal surfaces and real catalysts.8h15,17h19 When a physical chemist talks about catalysis the situation is similar to a mathematician trying to convince engineers that what he does is of use for them. G. H. Hardyœs A Mathematicianœs Apology 1940 contains many useful thoughts on this problem.20 One is It is one of the –rst duties of a professor for example in any subject to exaggerate a little both the importance of his subject and his own importance in it. A man who is always asking ì Is what I do worth while ? œ and ìAm I the right person to do it ? œ will always be ineÜective himself and a discouragement to others.He must shut his eyes a little and think a little more of his subject and himself than they deserve. Structure and adsorption on clean oxide surfaces 3 The preparation of a clean oxide surface is a rather difficult task. Several strategies have been followed.7,21,22 The most straightforward strategy is UHV in situ cleavage which however only leads to good results in certain cases such as MgO NiO ZnO SrTiO etc.13 Some very interesting materials such as Al2O3 SiO2 TiO2 etc. are hard to cleave.7 A disadvantage with respect to experimental investigations of cleaved bulk single crystal insulators is their low conductivity. An alternative way of bulk single crystal surface preparation is ex situ cutting and polishing followed by an in situ treatment by sputtering and subsequent annealing in oxygen.Through such a process a sufficient number of defects is created in the near surface region and in the bulk to support conductivity of the material. This leads to a situation where electron spectroscopies as well as STM can be applied.7 Single crystalline oxide surfaces may also be prepared via the growth of thin oxide –lms on single crystal metal supports.11,21,22 To such systems all surface science tools can be applied without further problems. If the oxide –lm is supposed to represent the bulk situation special care has to be taken in the control of –lm thickness. Also if adsorption and reactivity studies are intended the continuity of the –lm has to be guaranteed. There are several examples in the literature where this has been achieved.10,11,23 Probably the best studied clean oxide surfaces are the TiO (100) and TiO (110) surfaces.7,21,24 2 2 A STM image of the clean (1]1) TiO (110) surface taken by Diebold and her group25 is shown 2 in Fig.1. It is noteworthy that one of the –rst atomically resolved images of this surface was reported by Thornton and his group.26,27 The inset shows a ball and stick model of the surface. Fig. 1 Structure of the TiO (110) (1]1) surface as determined via STM (a reproduced from ref. 25) and grazing incidence X-ray scattering (b adapted from ref. 32). 2 Faraday Discuss. 1999 114 1»31 2 There is now accumulating evidence from theoretical modeling of the tunneling conditions but also from adsorbate studies using molecules which are assumed to bind to the exposed Ti-sites that the bright rows represent Ti atoms.Iwasawa and his group28h31 have successfully used formic acid in such a study and showed in line with the theoretical predictions and counterintuitive with respect to topological arguments that the Ti ions are imaged as bright lines and the oxygen rows as dark lines. Taking the resolvable interatomic distances within the surface layer the values correspond to the structure of the charge neutral truncation of the stoichiometric (110) surface.32 Interatomic distances normal to the surface however are substantially diÜerent from the bulk values as is revealed by X-ray scattering experiments.32 The top layer six-fold coordinated Ti atoms move outward and the –ve-fold-coordinated Ti atoms inward.This leads to a rumpling of 0.3^0.1 Aé . The rumpling repeats itself in the second layer down with an amplitude of about half of that in the top layer. Bond length variations range from 11.3% contraction to 9.3% expansion. These strong relaxations are not untypical for oxide surfaces and had been theoretically predicted for quite a while.33 The relaxations are particularly pronounced for the so-called charge-neutralized polar surfaces. 34h36 There are several experimental results,37h40 basically corroborating the theoretical predictions although the quantitative agreement is not always good.41h44 Speci–cally the (0001) surfaces of corundum-type materials such as Al2O3,41,42 Cr2O343 and Fe2O344 have been studied with X-ray diÜraction quantitative LEED as well as with STM and theoretical methods.Fig. 2 reminds the reader brie—y of the fact that a polar surface (e.g. (111) orientation for a rock-salt structure) exhibits if bulk terminated a diverging surface potential due to the missing compensation of the interlayer dipole moments as is nicely discussed in Nogueraœs book.36 Consequently polar surfaces reconstruct and/or relax substantially while non-polar surfaces often exhibit much less pronounced relaxations although as shown above for TiO the degree of relax- 2 ation is substantial. Fig. 3 shows the results of structural determinations for the three related Fig. 2 Schematic representation (side and top views) of the structure of a non-polar (a MeO(100) an unreconstructed polar (b MeO(111)) and hydroxylated polar (c MeO(111) adsorbate stabilized surface) surface of a rock-salt type crystal.The energies given refer to MgO.177,178 (V surface potential ; S area of surface unit cell ; N number of layers ; b interlayer spacing ; r charge on surface layer relative to a layer in the bulk.) 3 Faraday Discuss. 1999 114 1»31 Fig. 3 Experimental data on the structure of corundum-type depolarized (0001) surfaces (side and top views). Adapted from (b) ref. 33; (c) ref. 39 and (d) ref. 44. systems Al2O3(0001) Cr2O3(0001) Fe2O3(0001) and as addressed above. In all cases a stable structure in UHV is the metal ion terminated surface retaining only half of the number of metal ions in the surface as compared to a full buckled layer of metal ions within the bulk.The interlayer distances are very strongly relaxed down to several layers below the surface. The perturbation of the structure due to the presence of the surface in oxides is considerably more pronounced than in metals where the interlayer relaxations are typically of the order of a few percent.45 The absence of the screening charge in a dielectric material such as an oxide contributes to this eÜect considerably. It has recently been pointed out46 that oxide structures may not be as rigid as one might think judged on the relatively stiÜ phonon spectrum in the bulk. In fact at the surface the phonon spectrum may become soft so that the geometric structure becomes rather —exible and thus also very much dependent on the presence of adsorbed species.Bulk oxide stoichiometries depend strongly on oxygen pressure a fact that has been recognized for a long time.47 So do oxide surfaces structures and stoichiometries a fact that has been shown again in a recent study on the Fe2O3(0001) surface by the Scheffler and Schloé gl groups.44 In fact if a Fe2O3 single crystalline –lm is grown in low oxygen pressure the surface is metal terminated while growth under higher oxygen pressures leads to a complete oxygen termination.44 This surface would be formally unstable on the basis of the electrostatic arguments presented above. However calculations by the Scheffler group44 have shown that a strong rearrangement of the electron distribution as well as relaxation between the layers leads to stabilization of the system.STM images by Weiss and co-workers44 corroborate the coexistence of oxygen and iron terminated layers and thus indicate that stabilization must occur. Of course there is need for further structural characterization. The idea of polar and non-polar surfaces only really holds in its simplest version as presented above if the material is very highly ionic. Thus the most extreme cases to look at are perhaps the polar surfaces of the simple oxides with rock-salt structure48 such as MgO and NiO i.e. MgO(111) and NiO(111). Recently Barbier et al.49 have succeeded in preparing a single crystal NiO(111) surface and to characterize it via grazing incidence X-ray diÜraction (GIXD)! As was shown earlier for the case of thin NiO –lms of diÜerent crystallographic orientations i.e.NiO(100)50 and NiO(111),51,52 a surface prepared in air or under residual gas pressure exhibits a p(1]1) structure while the clean polar (111) surfaces are reconstructed. The p(2]2) reconstruction originally reported for the thin –lm system has also been found for the bulk single crystal surfaces.48,49 An initial structural analysis indicated that the actual structure is not the expected octopolar reconstruction shown in Fig. 4 but a more complicated one.48 However more recent investigations53 of more carefully prepared bulk single crystal surfaces reveal that a stoichiometric surface actually reconstructs according to the octopolar scheme.36,54 The small (100) terminated pyramids are oxygen terminated. Very recently NiO(111) –lms grown on Au(111) which were initially studied by Neddermeyer and his group,55 have been investigated Faraday Discuss.1999 114 1»31 4 Fig. 4 Schematic representation of the octopolar reconstruction of a polar rock-salt (111) surface in oxygen and metal termination. Adapted from ref. 48. by GIXD.56 The p(2]2) reconstruction was again corroborated but the structural analysis undertaken up to now would seem to favor a structure where oxygen as well as Ni-terminated octopols possibly arranged on adjacent terraces constitute the surface layer. Both the bulk single crystal surfaces as well as the NiO(111) –lm surfaces grown on Au(111) exhibit high degrees of surface order. This is probably one reason why these surfaces do not quickly restructure upon exposure to water while NiO(111) –lms grown on Ni(111) do reconstruct to form a hydroxy terminated NiO(111) surface.51,52 A microscopic mechanism would involve massive material transport across the surface which is the more unfavorable the better the order and may therefore be kinetically hindered on well ordered single crystals.We would like to note at this point that the interaction of water with polar oxide surfaces is a topic of general interest in geochemical and environmental issues57 as well as in catalysis. With respect to the latter Papp et al. have found indications that NiO catalysts prepared with preferential (111) crystallographic orientation by topotactical dehydration of Ni(OH) do show the highest activity towards DeNO -reactions after 2 x the last monolayer of H has been desorbed.58,59 Already in 1977 Derouane and co-workers60 2O had theoretically analyzed on the basis of energetic considerations that real crystallites must be terminated partly by polar surfaces whose charge has been reduced via OH adsorption.It is thus evident that the general study of the interaction of molecules with oxide surfaces represents an interesting and important –eld of study. In the following we are going to discuss several examples in order to bring out certain aspects of the bonding and interaction of molecules with oxide surfaces. This will be discussed in comparison with the adsorption of molecules on metal surfaces. Before we proceed to a more detailed discussion of the binding of adsorbed molecules a few remarks concerning the electronic structure of oxide surfaces are appropriate.Early on Hué fner and co-workers investigated the electronic structure of transition metal bulk samples and a great deal has been learned.61,62 There have also been attempts to investigate the surfaces of these materials with respect to electronic structure. Qualitatively it was expected that due to the high ionicity of some compounds there are pronounced surface eÜects. Photoelectron 5 Faraday Discuss. 1999 114 1»31 spectroscopy has been used to experimentally verify these expectations through the detection of chemically shifted core levels. However the shifts are not large enough to be detectable due to the relatively complex satellite structure accompanying metal core ionization.62 Eventually this was also rationalized via more sophisticated quantitative calculations which showed that there are several compensating contributions rendering the surface –elds only slightly diÜerent from the bulk.Applying techniques allowing for higher energy resolution have then clearly demonstrated surface eÜects. In the Merz group63 and our laboratory64 electron energy loss spectroscopy (EELS) in the regime of electronic excitations has been used to identify excitations in the surface layer. Fig. 5 shows a set of spectra taken on Ni(100) surfaces. The lowest trace has been taken on a clean single crystal. The broad features peaking at 4»5 eV correspond to charge-transfer excitations crossing the band gap of the insulating NiO bulk. In the gap there are narrow features due to excitations within the d-electron state manifold of the open shell Ni2` ions.As the excitation energy within this manifold increases the number of states increases so that near the chargetransfer band those states overlap and lead to the monotonous increase of intensity in this energy region. Most of the optically allowed transitions have been spectroscopically observed by transmission spectroscopy of bulk samples.65h67 Fromme and Kisker have recently performed spinpolarized EELS measurements which allow an assignment of the spin character of all states via the control of spin-polarization in the scattering conditions.68,69 The assignment and a spinpolarization measurement have been superimposed on the spectra. The important point here is Fig. 5 Electron energy loss spectra of NiO(100) surfaces.(a) Adsorbate covered NiO(100) –lms (i) defects OH saturated before NO adsorption (ii) after adsorption of NO. (b) Clean NiO(100) surfaces UHV-cleaved single crystal. The assignment of the features according to theory is given and supported by spin-polarized measurements (adapted from ref. 68 and 69). Faraday Discuss. 1999 114 1»31 6 g that there are additional spectroscopic features most pronounced at 0.6 eV excitation energy which are not due to excitations in the bulk but rather in the surface layer. This can be experimentally demonstrated by an adsorption study.64 Those excitations localized in the surface should be most strongly aÜected by adsorbed species. The experiment has been performed on a thin –lm sample because the surface has to be cooled to adsorb an appreciable amount of NO in this case.It is very obvious that the peak at 0.6 eV is in—uenced. In fact it is shifted towards the position of a feature originating from an excitation in the bulk. In passing we note that the NiO(100) –lm has been treated with water before the experiments had been performed in order to saturate the defects with hydroxy groups via dissociative adsorption of water. The vibrational losses caused by the hydroxys are clearly visible before NO adsorption took place. NO adsorption then induces yet a further vibrational loss at lower loss energy. What is the nature of the surface excitation and why is it diÜerent from the bulk excitation ? Staemmler and his group have performed ab initio cluster calculations,64,70 the result of which can be summarized as follows Due to the localization of the Ni-d-electrons it is sufficient to consider a single Ni2` ion within its octahedral coordination sphere if we consider the situation encountered in the bulk (see Fig.6). The ground state is a e 3E state with two unpaired d-electrons in the orbitals of the ligand –eld split set of d-orbitals. g The –rst excited state results from an excitation from the completely –lled t2g subset into the partly –lled e subset giving rise to an excitation located near 1 eV. There are many more higher g excited states some of which are assigned according to the work of Fromme et al.68,69 In the surface however one of the coordinated oxygen ions is missing and the symmetry of the local Ni t 2` site is reduced to C4v.e Consequently the degenerate subset is split. The subset is also 2g Fig. 6 Correlation of structural data with electronically excited states on NiO(100). Upper panel (left) coordination of a Ni ion in the bulk (middle) coordination of a Ni ion in the clean (100) surface as well as (right) in the case of adsorbed NO. Lower panel orbital diagram and total energies from cluster calculations.64,70 7 Faraday Discuss. 1999 114 1»31 split but this eÜect is not so important. The d-orbital of the former e subset pointing along the g Ni»O axis has lost part of its destabilizing interaction and consequently its energy decreases. The calculation shows that still both orbitals in the former e subset are singly occupied after reduction g of symmetry.Therefore the –rst excited state in a Ni2` surface ion is at lower energy than in the bulk as also revealed by the experiment. If an NO molecule is now coordinated to the Ni2` surface ion the energetic position of the orbital is raised again (similar to the presence of the sixth oxygen ion) eÜectively moving the excitation energy back close to the bulk position. It is clear from these results that the surface eÜect on the excitation energies is of the order of 0.4 eV and thus the above mentioned lack of evidence from other techniques can be rationalized. So far we have discussed the localized metal-ion states. Are there also surface modi–cations onto the charge-transfer states ? The answer is yes ! Unfortunately NiO is not a good example to support this experimentally.The electronic excitations of the Cr2O3(0001) surface on the other hand show the eÜect very clearly.71h74 Fig. 7 shows the EELS spectrum72h74 of this surface at low temperature. Again a thin –lm has been used. The sharp features in the band gap are excitations within the manifold of d-orbitals. A detailed discussion74 has shown that the excitations are characteristic of surface Cr ions with three d-electrons. However the Cr ions do not carry a net charge of 3] as expected (and found for the bulk Cr ions) but rather of 2] charge due to strong hybridization with the neighboring oxygen atoms. When we now perform EELS measurements after adsorption of CO2 not only the d-excitations are in—uenced but also the very intense feature near 3.8 eV.Again on the basis of cluster calculations performed in the Staemmler group,71 it has been possible to assign this intense feature to a surface charge-transfer excitation at the band gap which is shifted to lower energy as compared with the corresponding excitation in the bulk (see Fig. 7). The decrease in energy is reasonable because the Cr d-orbitals have been lowered the more open surface structure (see Fig. 3) and the coordination of the Cr ions to only three oxygen ions allows for better charge separation than in the bulk. In NiO the eÜect is less pronounced because the surface Ni ions are still –ve-fold coordinated. The analysis of the electronic structure of Cr2O3(0001) as discussed so far has been performed at 90 K. We note that upon increasing the temperature the structure of the surface changes,74 and we have speculated that these changes are connected with changes in the magnetic structure of the surface.Oxide surface magnetism is a –eld that needs to be explored in the future. At this point we return to the question raised above on the binding and interaction of molecules with oxide surfaces. Molecules bind to oxides via a bonding mechanism considerably diÜerent from metal surfaces. A CO molecule for example binds to metals via chemical bonds of varying strength involving charge exchanges.75 Fig. 8 illustrates the bonding of CO to a Ni-metal atom via the so-called r-donation/p-back-donation mechanism schematically and on the basis of a one electron orbital diagram. Fig. 7 EELS spectra of the clean and adsorbate covered Cr2O3(0001) surface.72h74 Faraday Discuss.1999 114 1»31 8 Fig. 8 Orbital diagram for the bonding of CO to Ni-metal (left) and to Ni-oxide (right). The r- and p-interactions lead to a relative shift of those r- and p-orbitals involved in the bond with respect to those orbitals not involved. The diagram re—ects this via the correlation lines. This may be contrasted by the electrostatically dominated interaction between a CO molecule and a Ni ion in nickel oxide.70,76 There is a noticeable r-repulsion between the CO carbon lone pair and the oxide leading to a similar shift of the CO 5r-orbital as in the case of the metal atom. However there is no or little p-back-donation so that the CO p-orbitals are not modi–ed.11,77 Conceptually the situation is transparent and one would expect that a detailed calculation reveals the diÜerences quantitatively.However as it turns out the description by ab initio calculations is very much involved and today a full account cannot be given.78 Theoretically (Table 1)70,78h89 the prediction is that CO as well as NO bind very weakly to NiO.78 The predicted binding energy of CO is of the order of 0.1 eV and it is expected to be similar to CO binding to MgO(100) i.e. the in—uence of the Ni d-electrons should be negligible.78 To shed light on this problem it was necessary to perform thermal desorption measurements on cleaved single crystal surfaces being the surfaces with the least number of defects.90 In Figs. 9 and 10 TDS data for CO and NO on vacuum-cleaved NiO(100) are compared with data for thin NiO(100) –lms grown by oxidation of Ni(100).At temperatures of 30 and 56 K multilayer desorption for CO and NO respectively shows up. The pronounced features at higher temperatures correspond to desorption of the respective adsorbate at (sub)monolayer coverage. In the case of the CO adsorbate at 34 K desorption of the second layer is found and the states at 45 and 145 K for CO and NO respectively are due to adsorption on defects as concluded from data obtained from ion bombarded surfaces (not shown here). It is obvious that for both adsorbates the thin –lm data and the data of the cleaved samples agree well in particular for NiO(100) the thin –lm data are comparable to those from the more perfect surfaces of the cleaved samples.The higher defect density of the thin –lm surfaces leads to small but clearly visible additional peaks in the TDS data which show up as shoulders near to the main peak for example in the NO spectra. Nevertheless the general shapes of the thin –lm spectra of both adsorbates are very similar to those of the cleaved samples. 9 Faraday Discuss. 1999 114 1»31 Table 1 Table of literature data for adsorption of CO and NO on NiO(100) and MgO(100)a Author Pacchioni and Bagus79 Klué ner and Freund80 Poé hlchen and Staemmler70 Cappus et al.81 Vesecky et al.82 Staemmler83 Poé hlchen84 Kuhlenbeck et al.85 Nygren and Pettersson78 Chen et al.86 Neyman et al.87 He et al.88 Furuyama et al.89 a BSSE basis set superposition error ; TDS thermal desorption spectroscopy ; DFT density functional theory ; IRS infrared spectroscopy ; Clausius»Clapeyron evaluation of pressure and temperature dependent IR intensities with the Clausius»Clapeyron equation; Redhead evaluation of TDS data with the Redhead equation.179 For CO the shift of the peak maximum with increasing coverage indicates that at higher coverage repulsive lateral interaction comes into play which may lead to occupation of energetically less favorable sites. This is not the case for the NO adsorbate which may be attributed to smaller lateral interactions and to the higher adsorption energy which makes adsorption more site speci –c and thus may inhibit compression of the layer involving site changes. TDS data for NO and CO on vacuum-cleaved MgO(100) for comparison are plotted in Figs.11 and 12. Multilayer desorption is found at 29 and 56 K for CO and NO respectively. The small features around 45 and 100 K are likely due to defect adsorption since they saturate at rather low coverage. Desorption from layers with small coverage is found at 57 and 84 K for CO and NO respectively. The data for CO and NO on NiO(100) have been evaluated using the leading edge method as well as a complete analysis. Details of the procedures may be found in ref. 91 and 92. Both methods determine the heat of adsorption as a function of the coverage of molecules already on the surface. The results of the evaluation are shown in Figs. 13 and 14. Both graphs exhibit a trend which is generally to be expected for laterally interacting adsorbate layers the adsorption energy decreases with increasing coverage.At coverages near to 1 monolayer the energies converge towards the multilayer values (0.09 and 0.18 eV for CO and NO respectively93). At low coverage the lateral interactions are most likely small so that the corresponding adsorption energies may be compared with theoretical results since in the calculations lateral interactions have not been considered. As indicated in Figs. 13 and 14 the low coverage adsorption energies are 0.30 and 0.57 eV for CO and NO respectively. The low coverage adsorption energies for CO and NO on NiO(100) and MgO(100) are compiled in Table 2. According to theory the interaction of the adsorbates with MgO(100) and NiO(100) are expected to be similar since the bonding should be mainly electrostatic in nature78 (the electric –elds at the surfaces of NiO(100) and MgO(100) are similar).However according to Table 2 the bonding energies are considerably diÜerent with the higher values being obtained for NiO(100). Covalent interactions involving the Ni 3d-electrons may play a role for the adsorbate» substrate interaction which does not show up in the calculations published so far. As far as it concerns the basis set superposition error (BSSE) corrected calculations listed in Table 1 which are expected to yield qualitatively better results as compared to the non-corrected B0 0.03 to 0.1 0.32 0.45 0.1 \0.23 0.52 0.08 0.28 0.11 0.43 0.46 0.15 to 0.17 Faraday Discuss.1999 114 1»31 10 System CO/NiO(100) NO/NiO(100) CO/NiO(100) CO/NiO(100)/Ni(100) CO/NiO(100)/Ni(100) NO/NiO(100) NO/NiO(100) NO/NiO(100)/Ni(100) and NO/NiO(100) CO/MgO(100) CO/MgO(100) CO/MgO(100) CO/MgO(100)/Mo(100) CO/MgO powder Method Ab initio cluster calculation Ab initio cluster calculation BSSE correction Ab initio cluster calculation BSSE correction TDS Redhead IRS Clausius»Clapeyron Ab initio cluster calculation BSSE correction Ab initio cluster calculation BSSE correction TDS Redhead Ab initio cluster calculation BSSE correction DFT DFT BSSE correction IRS Clausius»Clapeyron TDS Redhead IRS Clausius»Clapeyron Adsorption energy eV 0.24 Fig. 9 Thermal desorption spectra of CO on NiO(100) cleaved in vacuum (upper part) and CO on a thin NiO(100) –lm grown by oxidation of Ni(100) (lower part).The mass spectrometer was set to mass 28 (CO). CO doses are given relative to the dose needed to prepare a monolayer. calculations it appears that the theoretical results for adsorption on MgO(100) are in general in line with our experimental results whereas a similarly favorable comparison can not be made for NiO(100). It appears necessary to re-investigate the role of the Ni 3d-electrons in future theoretical studies. The adsorption of CO on MgO has been thoroughly investigated by Heidberg and his group using IR-spectroscopy94 and by Weiss and co-workers using helium scattering spectroscopy.95 They have clearly demonstrated that CO develops ordered phases on the cleavage planes and that order and spectroscopic properties depend on the quality of the prepared surfaces.From their experiments the in—uence of the presence of surface defects on adsorption properties is very obvious but a quantitative evaluation based on the number and the nature of the defects has not been reported. The quantitative evaluation of defects is a well de–ned but hard to tackle problem for future studies that has to be taken on by our research community. Water adsorption is an example that lends itself to a study of the in—uence of defects because at lower coverage the (100) cleavage planes of MgO and NiO do not dissociate water while the presence of defects does induce water Table 2 Compilation of low-coverage bonding energies for NO and CO on NiO(100) and MgO(100) obtained in this work MgO(100) NiO(100) CO NO 0.14 eV 0.22 eV 0.30 eV 0.57 eV 11 Faraday Discuss.1999 114 1»31 Fig. 10 Thermal desorption spectra of NO on NiO(100) cleaved in vacuum (upper part) and NO on a thin NiO(100) –lm grown by oxidation of Ni(100) (lower part). The mass spectrometer was set to mass 30 (NO). NO doses are given relative to the dose needed to prepare a monolayer. Fig. 11 Thermal desorption spectra of CO on MgO(100) cleaved in UHV. The mass spectrometer was set to mass 28 (CO). CO doses are given relative to the dose needed for the preparation of a monolayer. Faraday Discuss. 1999 114 1»31 12 Fig. 12 Thermal desorption spectra of NO on MgO(100) cleaved in UHV.The mass spectrometer was set to mass 30 (NO). NO doses are given relative to the dose needed for the preparation of a monolayer. dissociation. This can be seen in TDS spectra of H2O from (100) rock-salt type surfaces. Fig. 15 shows results for H desorption from MgO(100) and NiO(100).96 The most pronounced features 2O in the spectra are due to condensed water layers at lowest desorption temperature and the conversion of a compact layer (with c(4]2) periodicity in the case of MgO) to the monolayer which desorbs at 225 K (240 K for MgO(100).97,98 The diÜerence in desorption temperature between Fig. 13 Adsorption energy of CO on NiO(100) cleaved in vacuum as a function of coverage. The data have been determined from TDS spectra like those shown in Fig.9 (upper part) using the leading edge method and complete analysis. TDS data taken with heating rates of 0.1 0.2 1 and 2 K s~1 have been used. 13 Faraday Discuss. 1999 114 1»31 Fig. 14 Adsorption energy of NO on NiO(100) cleaved in vacuum as a function of coverage. The data have been determined from TDS spectra like those shown in Fig. 10 (upper part) using the leading edge method and complete analysis. TDS data taken with heating rates of 0.2 and 1 K s~1 have been used. MgO(100) and NiO(100) seems to be characteristic for the H2O substrate interaction. Most of that information is lost when we create defects via sputtering. Thermal desorption is now observed up to relatively high temperatures and the features are broad. Which kind of defects and how many have been created is not yet known.A combination of various techniques to characterize the defects by probe molecule adsorption together with infrared spectroscopy EPR and electron spectroscopies may in the future lead to a deeper understanding. Dissociative adsorption of water on oxide surfaces can also be used in a preparative way namely to chemically modify the surface by hydroxylation. We have used this technique for a thin alumina –lm to study the in—uence of the presence of hydroxy groups on the nucleation and growth of metallic aggregates as will be discussed later.99 At this point we show in Fig. 16 the result of such a hydroxylation as measured with vibrational spectroscopies such as high resolution electron energy loss spectroscopy (HREELS) and FTIR.100 In the case of the thin alumina –lm on NiAl(110) it was impossible to hydroxylate the oxide just by water dissociation while on a similar –lm on NiAl(100)101 formation of OH from dissociative H2O adsorption occurs.The clean oxide –lm surface was exposed to metallic aluminium and then the aluminium was hydrolyzed via water adsorption to form a hydroxy overlayer.99,100 In Fig. 16 at the bottom an HREELS spectrum showing the hydroxy vibration at 465 meV (3750 cm~1) is plotted atop a corresponding spectrum of the clean –lm. The peaks below 120 meV are due to the alumina phonons,102 which are broadened through hydroxylation in—uencing surface order. The observed hydroxy loss coincides very nicely with the FTIR absorption observed for the same system.In this case more water was adsorbed so that a broad band from water clusters is seen also. The sharp extra band at 3705 cm~1 is due to free OH groups at the surface of these water clusters,103 as they are known from the surface of ice. In fact if a thick ice –lm is grown on the alumina –lm this particular vibration is observed (see Fig. 16). In comparison with literature data104 it is now possible to assign the hydroxy loss on the alumina surface. According to a review article by Knoé zinger104 an OH-vibration at 3750 cm~1 is characteristic of hydroxys bridging aluminium ions both in octahedral or one in an octahedral and one in a tetrahedral site. We mention that on alumina –lms grown on a diÜerent NiAl substrate101 other types of OH species may be formed as was shown by Hemmingerœs group.Therefore it is conceivable that the in—uence of the nature of Faraday Discuss. 1999 114 1»31 14 Fig. 15 Thermal desorption spectra of H2O on UHV-cleaved MgO(100) and NiO(100). A schematic representation of the c (4]2) structure is included (reproduced from Heidberg Redlich and Wetter Ber. Bunsenges. Phys. Chem. 1995 99 1333). For comparison a thermal desorption spectrum from MgO(100) after creation of defects via sputtering is shown. the hydroxy species modifying the surface on the interaction with additional adsorbates i.e. metal deposits could be investigated. Before we move on to discuss the properties of metals on oxides we would like to brie—y discuss 2 the adsorption of CO on oxides as an example of a molecular adsorbate system with more 2 degrees of freedom.TDS spectra indicate105,106 that there are more weakly and less weakly bound CO species on 2O3(0001) surface. We have studied the nature of those species by various techniques includ- a Cr ing infrared spectroscopy. Fig. 17 shows several sets of IR spectra. The pair of sharp bands around 2300 cm~1 can easily be assigned to the more weakly bound CO with only a slightly distorted 2 structure as compared with the gas phase species. By a combination of isotopically labeling the adsorbed CO (shift of frequencies) as well as the oxide layer (no shift of CO bands) we have 2 2 demonstrated that the single band centered around 1400 cm~1 is due to the presence of a carboxylate species i.e. a bent anionic CO species and not as perhaps expected to a carbonate.72 The 2 bands between 1610 and 1700 cm~1 are missing because of the applicability of surface selection rules in thin –lm systems.This means all non-totally symmetric bands are suppressed in intensity. A quick comparison with CO adsorption on chromia microcrystalline material as shown in Fig. 2 17 indicates the presence of the bands between 1610 and 1700 cm~1 as expected for adsorption on a bulk dielectric material. It is remarkable how similar the thin –lm data are in comparison with the microcrystalline material. This has been discussed in detail by Zecchinaœs group.107 Also the 15 Faraday Discuss. 1999 114 1»31 Fig. 16 Fourier transform IR spectra (IRAS) and electron energy loss spectra (HREELS) of a clean and OH(]H O)-covered alumina –lm.2 response of the two systems with respect to preadsorption of oxygen is very similar. In fact as shown in Fig. 17 CO adsorption in the form of the less weakly bound CO2~ is fully suppressed 2 on the thin –lm system and very strongly attenuated for the microcrystalline system. This indicates that CO occupies the chromium sites because we know that oxygen from the gas phase 2 adsorbs on the chromium ions. As we remarked above ELS73 and XPS108 spectra of the Cr2O3(0001) surface have been used to deduce that the Cr-ions in the surface are in a low oxidation state i.e. Cr2` as opposed to chromium ions in the near surface and bulk regions. It is therefore not surprising that such a surface provides electrons to adsorbed molecules leading to electron transfer as for example documented by the formation of O2~ and CO2~.The low valence state of the Cr surface ions also has consequences in other reactions such as the polymerization of ethene which has been studied on Cr2O3(0001),109 and in connection with other more realistic model studies.110 A –eld that has not been investigated in any detail as far as well characterized single crystal oxide surfaces are concerned is connected with photoinduced chemical reactions on larger molecules. Photoinduced desorption of CO and NO from oxides has been studied extensively111h114 but the reactivity of larger molecules has not. Yates and his group have reported such studies on Faraday Discuss. 1999 114 1»31 16 2 Fig. 17 IRAS spectra of CO adsorbed on Cr2O3(0001) surfaces and on polycrystalline chromia.Left panel IRAS spectra at diÜerent surface temperatures and with isotopically labelled CO as well as Cr2O3 . Right 2 panel adsorption of CO after pre-adsorption of oxygen. 2 powder samples i.e. Rh complexes deposited on Al2O3 powder and very interesting results concerning C»H bond activation have been reported.115 We refer to the literature for details,115 and note that this should be considered as a new promising area in connection with single crystalline systems. Metals on oxides So far we have considered the clean oxide surface and its reactivity. In the following we will modify the oxide surface by deposition of metal onto the surface. This represents a route towards the preparation and characterization of more complex model systems in heterogeneous catalysis in order to bridge the so-called materials gap.Over past years several strategies have been followed along this route. Very early on small metal particles have been put onto oxide bulk single crystal surfaces particularly MgO and characterized by transmission electron microscopy (TEM). Poppa has been the pioneer in this –eld,18 and the very important contributions to the –eld have been recently reviewed by Henry who himself was involved in the early TEM measurements.14 While these eÜorts where mainly aimed at preparing small well de–ned particles another strategy has been followed by M‘ller and his group116h119 as well as Madey and co-workers19 by trying to prepare thin metal –lms on bulk oxide single crystals such as TiO (110) surfaces.As mentioned above the advent of scanning 2 tunneling microscopy has had a substantial in—uence on the understanding of the structure of clean oxide surfaces. Several groups120h122 have started to investigate metal deposition on TiO surfaces. Interesting initial results concerning metal particle migration and oxide migration 2 onto the metal particles (the so-called SMSI eÜect) have been obtained.121,122 Particularly well suited for the application of scanning tunneling microscopy are metal particles deposited onto thin –lm oxide surfaces.8,10,11,14 Goodmanœs group for example made major contributions to this –eld early on.10 In Fig. 18 we show the result of a STM study from our laboratory. The left panel shows the clean alumina surface as imaged by a scanning tunneling microscope.123 The surface is well ordered and there are several kinds of defects on the surface Firstly the re—ection domain boundaries between the two growth domains of Al2O3(0001) on the NiAl(110) surface the substrate on which the –lm is grown via a well established oxidation recipe.102 Secondly there are anti-phase domain boundaries within the re—ection domains and in addition there are point defects which are not resolved in the images.The image does not change dramatically after 17 Faraday Discuss. 1999 114 1»31 ”2 Al2 O3/NiAl(110) utip\8 Aé Fig. 18 Scanning tunneling images (3000]3000 V I\0.8 nA). (a) Clean alumina –lm (b) after deposition of 0.1 Aé of Rh at 90 K (c) after deposition of 2 of Rh at 300 K and (d) after deposition of 2 Aé of Rh at 300 K on hydroxylated substrate onto the pre-hydroxylated alumina –lm.hydroxylating the –lm a procedure we had mentioned above.99 The additional panels show STM images of rhodium deposits on the clean surface at low temperature and at room temperature, 15,124 as well as an image after deposition of Rh at room temperature on a hydroxylated substrate.125 Please note that the amount deposited onto the hydroxylated surface is equivalent to the amount deposited onto the clean alumina surface at room temperature. Upon deposition of Rh from the metal vapor onto the clean surface at low temperature small particles nucleate on the point defects of the substrate and a narrow distribution of sizes of particles is formed.If the deposition of Rh is carried out at room temperature the mobility of Rh atoms is considerably higher compared with low temperature so that nucleation at the pronounced line defects of the substrate becomes dominant. Consequently all the material nucleates on the re—ection domain and anti-phase domain boundaries. The particles have a relatively uniform size given by the amount of deposited material. If the same amount of material is deposited onto a hydroxylated surface the particles are considerably smaller and distributed across the entire surface showing Faraday Discuss. 1999 114 1»31 18 that hydroxylation leads to higher metal dispersion.15,99 The thermal behavior of the deposits is important with respect to studies of chemical reactivity because the ensemble of particles may undergo morphological changes adopting their equilibrium shape which could be diÜerent with and without the presence of a reactive gas phase.In the present case detailed studies have been undertaken on the particles deposited onto the clean substrate and less detailed studies for the deposit on the hydroxylated surface. As a result of these studies it is known that the morphology of the ensemble is not altered within a temperature window from 90 to approximately 450 K. The window is extended to even higher temperatures on the hydroxylated substrate. Above the upper temperature limit the particles tend to agglomerate and also start to diÜuse through the –lm into the metal substrate underneath.15 Studying this agglomeration process is an interesting subject in itself and research in this direction is only starting.15 A more basic aspect of course would be a study of metal atom diÜusion on oxide substrates.The obvious method to perform such a study is the STM.126 However in contrast to diÜusion studies on metal surfaces similar studies on oxide surfaces have not been reported. On the other hand –eld ion microscopy studies on metal atom diÜusion on oxide –lms are under way and a –rst estimate of activation energies for diÜusion has been reported.127 It is obvious that the area of diÜusion studies will considerably pro–t from atomic resolution once it is obtained routinely for deposited aggregates on oxide surfaces. While for TiO and very few other 2 oxide substrates atomic resolution may be obtained routinely there are very few studies on deposited metal particles where atomic resolution has been reported.128 The –rst report for an atomically resolved image of a Pd metal cluster on MoS was reported by Henry and his group.128 A 2 joint eÜort between Besenbacher and our group129 has led to atomically resolved images of Pd aggregates deposited on a thin alumina –lm.Fig. 19 shows such an image of an aggregate of about 80 Aé in width. The particle is obviously crystalline and exposes on its top facet the (111) Pd surface. Also the (111) facets on the side typical for a cuboctahedral particle can be discerned. The small (100) facets predicted via equilibrium shape considerations on the basis of the WulÜconstruction could not be atomically resolved.If we however apply the concept of the WulÜconstruction we may deduce the metal surface interaction energy.129 The basic equation is (1) Wadh\coxide]cmetal[cinterface Provided the surface energies (cmetal) of the various crystallographic planes of the metal are known,130 a relative work of adhesion (Wadh) may be de–ned.129 We –nd 2.9^0.2 J m~2 which is still rather diÜerent with respect to recent calculations by Jennison et al.131 where metal adsorption energies (1.05 J m~2) have been calculated on a defect free thin alumina –lm. It is not unlikely that this discrepancy is connected with the rather complicated nucleation and growth behavior of the aggregates involving defects in the substrate. Fig. 19 Scanning tunneling images at atomic resolution of Pd aggregates grown on an alumina –lm.129 19 Faraday Discuss.1999 114 1»31 While STM reveals the surface structure of deposited particles their internal structure in particular as a function of size is not easily accessible through STM. In this connection TEM studies on the same model systems can be of help.132 Fig. 20 shows a schematic drawing of a sample. After growing the –lm and deposition of the particles the sample is ion-milled from the back so that a small hole is –nally formed. In this way a wedge is obtained which is thin enough for the imaging process. A positive side eÜect of this procedure is the fact that also the unsupported –lm next to the edge can be studied.133 This opens the opportunity to judge whether the metal substrate has any structural eÜect on the deposits.On the basis of numerous high resolution TEM images and a subsequent analysis of the Moireç periodicities it has been possible to calculate the lattice constants as a function of particle size.132 The corresponding plot is depicted in Fig. 21 and indeed proves that the atomic distances continuously decrease to 90% of the bulk value at a cluster size of 10 Aé . On the other hand the lattice constant approaches the Pt bulk value already at a diameter of 30 Aé . This eÜect has also been detected for Ta and Pd clusters on the thin alumina –lm but it seems to be less pronounced in these cases.134h136 Fig. 20 Schematic drawing of a sample prepared for transmission electron microscopy (sample milling technique). Fig. 21 Lattice constants and interatomic distance of Pt particles grown on Al2O3/NiAl(110) their size (the ends of the horizontal bars represent the width and the length of the particular clusters respecas a function of tively while the vertical bars are error bars).Faraday Discuss. 1999 114 1»31 20 Of course the electronic structure of deposited metal aggregates re—ects to a certain extent the geometric structure and vice versa. The electronic structure which will be discussed next has been investigated using various methods including photoemission X-ray absorption and scanning tunneling microscopy. One particularly interesting aspect in connection with aggregates is the size dependence of the electronic structure in relation with adsorption and reactivity. Starting from an atomic level diagram Fig.22 shows how such a level diagram develops when more and more atoms are agglomerated to form an aggregate and –nally a solid with a periodic lattice. Upon formation of an aggregate from equivalent atoms the atomic levels are split into molecular orbitals many of which are degenerate if the symmetry of the system is high. The splittings are characteristic for the intermolecular interactions. Depending on the interaction strength the split levels derived from a given atomic orbital start to energetically overlap with levels derived from other atomic orbitals. As long as the system has molecular character there is an energy gap left between occupied and unoccupied levels in contrast to the situation encountered for an in–nite periodic metallic solid as represented on the right hand side of the –gure where there is no longer a gap between occupied and unoccupied levels.It is not hard to envision now that as we slowly enlarge the number of atoms in an agglomerate the gap between occupied and unoccupied orbitals eÜectively vanishes. It eÜectively vanishes if the gap energy decreases to a value close to kT . In this situation the changes in the electronic structure would be responsible for an insulator(molecule)»metal transition. The question arises how many atoms are necessary to induce such a transition ? There are several reports in the literature claiming numbers ranging from 20 to several hundred atoms to be necessary.135,137h149 In this connection there is one very interesting extrapolation deduced from spectroscopic measurements of the gap of inorganic carbonyl cluster compounds as a function of the metal cluster size.It is shown in Fig. 23139 and stems from the Longoni group in Bologna. The extrapolation would suggest that 70 atoms are sufficient to close the gap. Comparatively we have studied deposited clusters of varying size with a combination of photoelectron spectroscopy and X-ray absorption.141,147,149 Both the naked as well as the CO covered aggregates have been studied. Without discussing the results we only state that the extrapolation on the CO covered clusters yields a vanishing gap just below 100 metal –rst ionisation I1 Fig. 22 Diagram illustrating the transition from an atom to a metal. (EB binding energy; energy; e electron charge; / workfunction; ! X symmetry points in the Brillouin zone.) 21 Faraday Discuss.1999 114 1»31 atoms. It appears from the STM images that such a situation is reached when the diameter of the aggregate decreases down below 25 Aé diameter and a height of 15»20 Aé . The aggregate of this size contains 75»100 atoms a size which well correlates with the extrapolation on the spectroscopic data of carbonyl cluster compounds in Fig. 23 as well as with our results. We take this as a strong indication that at least for the carbon monoxide covered cluster a non-metal-to-metal transition occurs in the vicinity of such a size. Goodman and his group have used scanning tunneling microscopy to investigate the electronic structure of aggregates deposited on oxides.10,150h152 Fig.24 shows typical current»voltage curves for some aggregate sizes i.e. Au on TiO (110).150 While 2 the large particles do not exhibit a plateau near I\V \0 the smaller clusters do show the behavior expected for a system with a gap. However the discrete structures observed for other systems i.e. nanoparticles on graphite152 and related substrates153 are not found. The authors report on indications that it is particularly the second layer in the gold aggregates that is responsible for the non-metal-to-metal transition. Au is an interesting low temperature CO oxidation catalyst and the STM –ndings are important to understand the size speci–city of the reaction. Before we take a closer look at the reactivities of deposited particles we will brie—y discuss adsorption properties as observed mainly through the probe molecule CO.The technique to study CO adsorption is Fourier transform infrared spectroscopy (FTIR) because it provides the resolution to diÜerentiate between various adsorbed species. Again the thin –lm based systems are particularly well suited because the metallic support of the oxide –lms acts as a mirror at infrared frequencies. It is however also possible to perform such experiments on surfaces of bulk dielectrics as was shown by the Hayden group.154,155 Goodman and his group were active in this –eld early on152 and have published an interesting study of CO adsorption on Pd aggregates on Al2O3 –lms. The results have been interpreted as characteristic for the adsorption of CO on diÜerent facets of the small crystalline aggregates.While this interpretation does not take into account adsorption on the various defect sites of the Fig. 23 Electronic excitation of lowest energy for several cluster compounds as a function of metal atoms in the cluster (*E is the energy gap for cluster compounds). av Faraday Discuss. 1999 114 1»31 22 Fig. 24 Current»voltage (b) recorded for Au clusters of various sizes deposited onto a TiO (110) surface. A 2 typical STM picture of the system is shown in (a). (Adapted from ref. 151.) aggregates which has been pointed out in a more recent study,156 the data are indicative of the potential of the tool for the study of size dependent absorption studies. We have recently prepared metal deposits on well ordered alumina –lms at lower temperature including liquid He temperatures i.e.in the range of 50 to 90 K substrate temperature,157,158 in order to determine the IR characteristics of speci–c sites. The infrared spectrum taken from a Rh deposit prepared and saturated with CO at 90 K (average particle size nine atoms) is displayed in Fig. 25 (left top corresponding to the spectrum in the middle on the right). The most prominent feature in the stretching region of terminally Fig. 25 (Left) Infrared spectra taken after deposition of 0.028 ML Rh on Al2O3 and subsequent saturation with 12CO (top) and an approximately equimolar mixture of 12CO and 13CO (bottom) at 90 K. The isotopic compositions giving rise to the three dicarbonyl bands are indicated below the corresponding wavenumbers.Average particle size nine atoms. (Right) Infrared spectra recorded after CO saturation of Rh deposits at 90 K along with corresponding room temperature STM images (500 Aé ]500 Aé ). Top 0.057 ML Rh deposited at 300 K. Middle 0.057 ML Rh deposited at 90 K. Bottom 0.057 ML Rh deposited at 300 K followed by the same exposure at 90 K. 23 Faraday Discuss. 1999 114 1»31 bound CO molecules is a sharp intense band at 2117 cm~1. This signal has previously been shown to arise from isolated Rh bound to oxide defects.157 Both the number of adsorbed CO molecules and the nature of the defect site remained unclear. Features at lower frequencies are assigned to molecules on Rh aggregates. In order to get insight into the stoichiometry of the Rh»carbonyl species giving rise to the band at 2117 cm~1 isotopic mixing experiments have been carried out.These experiments allowed us to unambiguously assign this band to a Rh(CO)2 species. Large particles prepared at 300 K deposition temperature residing on line defects do not exhibit the Rh(CO) band (Fig. 25 right top spectrum). However if the spectra are recorded after 2 saturation of the line defects at 300 K and then further metal is deposited at 90 K (Fig. 25 right bottom spectrum) the Rh(CO) band is found indicating that this species resides in point defect 2 sites. By further reducing the size of the particles we have also identi–ed RhCO and species carrying more than three CO molecules.158 In summary however we may conclude that several diÜerent types of Rh particles are responsible for the observed infrared features.Presently density functional calculations on small Rh carbonyls are in progress.159 Calculated vibrational frequencies of such systems may help to identify the species present on the alumina –lm. These studies on small Rh particles have been extended to include neighboring elements in the periodic table. Infrared spectra recorded after deposition of comparable amounts of Pd Rh and Ir and subsequent CO saturation at 90 K are displayed in Fig. 26. We note diÜerences in the low wavenumber region where vibrational frequencies of molecules in multiple coordinated sites are located. As on single crystals the population of such sites is highest on Pd,160,161 while no such CO is observed on Ir.162,163 The diÜerences in the region of terminally bound CO however are much more pronounced.In the case of Ir several distinct features are observed. In analogy to the Rh(CO) band at 2117 2 cm~1 the sharp signal at 2107 cm~1 may be attributed to Ir(CO) species via isotopic mixture 2 experiments (not shown). Bands with similar frequencies have been assigned to the symmetric Ir`(CO) on technical Ir/Al stretch of catalysts (2107»2090 cm~1)164 and on the iridium- 2 2O3 loaded zeolite H-ZSM-5 (2104 cm~1).165 Fig. 26 Infrared spectra of Pd Ir and Rh deposited at 90 K and saturated with CO at the same temperature. Faraday Discuss. 1999 114 1»31 24 Fig. 27 CO dissociation on Rh/Al2O3/NiAl(110) tion at 90 K and heating to the indicated temperatures (data acquisition at 90 K). representative series of C 1s spectra taken after CO satura- By contrast no signs of atomically disperse Pd or of structurally well-de–ned aggregates are observed.Indeed the infrared spectrum is rather similar to that observed on much larger disordered Pd aggregates.156 At the same metal exposure the Pd particles are found to be larger than the Rh aggregates by room temperature STM. Our observations show that infrared spectra of adsorbed CO provide valuable information on the size of metal nanoparticles as has been long recognized in the catalysis related literature. In the nucleation regime the metal particle size increases from Ir across Rh to Pd implying the opposite trend in metal oxide interaction strength i.e. Pd\Rh\Ir. The literature contains several adsorption studies see for example,166 employing other probe molecules such as hydrocarbons but here also reaction comes into play which renders the situation even more complicated.In the following –nal section several simple chemical reactions of O and CO on small aggre- 2 gates are addressed. A simple reaction is the dissociative adsorption of oxygen on small particles. Pd aggregates as shown in Fig. 19 can be imaged at atomic resolution after a dosage to saturation with molecular oxygen from the gas phase.167 On the side facets the corrugation due to the presence of adsorption of oxygen can be identi–ed. A doubled periodicity corresponding to a p(2]2) structure can be identi–ed. This structure is very similar to the p(2]2) structure observed after dissociative oxygen adsorption on Pd(111).168 We therefore conclude that a similar situation is encountered in the case of the deposited aggregates.The p(2]2) structure interestingly appears on the diÜerent facets at diÜerent tunneling conditions. When the oxygen covered Pd aggregates are exposed to carbon monoxide the reactivity of the diÜerent facets appear to be diÜerent in the sense that the oxygen adsorbate structure is lost on the various facets at variable temperatures and exposures. It will be interesting in the future to study these eÜects in more detail. 25 Faraday Discuss. 1999 114 1»31 CO oxidation at low temperatures and as a function of size on TiO supported gold aggregates 2 has been studied by the Goodman group.151 They –nd a marked size eÜect of the catalytic activity which correlates with the original observations by the Haruta group169 for Au on large area titania catalysts.The aggregates near 35 Aé size show the maximum activity. In the future it will be important to perform kinetic measurements for such reactions under well de–ned conditions including UHV and ambient environments. A very good example has recently been reported by Henry and his group.170 Similar to the studies of CO oxidation on Au aggregates reported above we have undertaken a detailed study of CO dissociation as a function of Rh particle size.171h173 For various cluster sizes we have taken C 1s photoelectron spectra as a function of sample temperature. An example is shown in Fig. 27 covering a temperature range where we know that the morphology of the ensemble of aggregates does not change.At low temperature the signal typical for molecular CO is observed. Near 400 K a second signal appears indicating the dissociation of CO into carbon and oxygen atoms. At 500 K all molecular CO has been either dissociated or desorbed. The dissociation probability is then given by the relative ratio of the molecular and the atomic C 1s signal. This is plotted in Fig. 28 as a function of the aggregate size. Also included in the –gure are data for very small aggregates where it has been shown that CO dissociation is negligible.174 This is also true for closed packed single crystal surfaces175 which would apply to the far end of in–nitely large aggregate sizes. If however we look at data gained for stepped Rh surfaces then there is a probability for CO dissociation.176 At an intermediate aggregate size of 200»300 atoms the dissociation probabilities are maximal.According to ref. 171 and 172 the dissociation activity also passes a maximum for the 300 K deposits i.e. the Fig. 28 CO dissociation activity on Rh particles deposited on Al2O3/NiAl(110) as determined by XPS. activity decreases in the regime of small particle sizes re—ecting the behavior of the 90 K deposits. Faraday Discuss. 1999 114 1»31 26 Fig. 29 (a) C 1s spectra of CO adsorbed on Rh particles after saturation at 90 K. (b) Intensity changes for the components A and B as well as the intensity losses due to dissociation and desorption as a function of the annealing temperature (average particle size B104 atoms).Although electronic eÜects cannot be completely excluded as a reason for the onset of the dissociation phenomenon for small particles an explanation on the basis of structural properties of the system seems more likely. Since the Rh deposits are basically disordered it is easily imaginable that aggregates of medium size exhibit a maximum defect density in terms of steps kinks and other low coordinate surface atoms. Smaller units should contain less defects in particular if they are still two-dimensional. In addition to that spatial constraints may play a role here as well (accommodation of C and O on adjacent sites see Fig. 27). At high exposures the step density is reduced due to coalescence processes. For deposition at 300 K the observed tendency to form crystalline aggregates in the high coverage regime is another factor contributing to a lower defect density.This is consistent with the observation that the dissociation activity declines much faster in this case (see Fig. 28). An interesting detail concerning the dissociation process has been discovered by a closer inspection of the C 1s emission of the molecularly adsorbed CO.172 As demonstrated in Fig. 29 (a) the peak actually consists of two components denoted A and B. If the fraction of the total intensity found for component B after heating to 300 K is compared to the fraction of CO –nally dissociating (see Fig. 28) it turns out that the species giving rise to B can be regarded as a kind of dissociation precursor. In fact the evolution of these two quantities as a function of the particle size is identical i.e.both pass a maximum at the same point.172 At 90 K however this is not yet the case. Here the relative intensity step which causes a shift of intensity from component A to component B i.e. an increase of the B species which is most pronounced for the medium-sized particles. Interestingly this conversion is irreversible. Cooling down to 90 K does not lead to an intensity redistribution again. The conclusion that B is indeed a dissociation precursor is additionally corroborated by Fig. 29 (b) showing the intensity changes for the A and B peaks as well as the losses which result either from desorption or dissociation.173 Unambiguously the desorption curve follows the curve for component A whereas the dissociation curve mimics the development of the component B.Unfortunately the results allow no further statement as to the nature of the A and B species. It can be assumed however that the B species is connected with CO adsorbed on defects. Based on the fact that higher coordinated CO species give rise to lower C 1s binding energies (see above) it may be furthermore speculated whether B is associated with CO in a higher coordination as compared to the A species. The –eld of investigations of chemical reactivity as a function of aggregate size is in full development and there are more exciting results at the horizon. 27 Faraday Discuss. 1999 114 1»31 Concluding remarks After 30 years of surface science which have seen an enormous development of methods and instrumentation to be applied to the investigation of solid surfaces the –eld is now ready to tackle questions of rather complex natures.Naturally so far metal surfaces have been the focus of attention in surface science and this will continue to be the case. However also for such systems the complexity of problems is constantly increasing in particular if molecular adsorbates and selforganized systems are considered. Metal oxide surfaces have received some attention in the recent past and the study of such systems as well as more complex metal»metal-oxide composite systems will or perhaps has already de–ned a direction in surface science that promises to reveal interesting results of fundamental interest as well as of appeal towards applications.It is good to see that surface science is very healthy and alive and it has never been farther away from fatality. I am grateful to my co-workers present and past who have contributed to the results presented and whose names appear in the list of references. Also I am happy to thank many colleagues for stimulating discussions and collaboration. Special thanks go to Ralph Wichtendahl for his help with transparencies and –gures. Over the years many funding agencies and also the private sector have supported our work Deutsche Forschungsgemeinschaft Bundesministerium fué r Bildung und Forschung Ministerium fué r Wissenschaft und Forschung des Landes Nordrhein-Westfalen Fonds der Chemischen Industrie German»Israeli Foundation European Union NEDO International Joint Research Grant on Photon and Electron Controlled Surface Processes Hoechst Celanese and Synetix a member of the ICI group through their Strategy Research Fund.References 1 P. A. Cox T ransition Metal Oxides. An Introduction to their Electronic Structure and Properties Clarendon Press Oxford 1992. 2 D. A. Johnson Some T hermodynamic Aspects of Inorganic Chemistry Cambridge University Press Cambridge 1982. 3 Non-stoichiometric Oxides ed. O. T. Sorensen Academic Press New York 1981. 4 A. Hamnett and J. B. Goodenough in Binary transition metal oxides ed. O. Madelung Springer 1984. 5 J. F. Owen K. J. Teegarden and H. R. Shanks Phys. Rev. B 1978 18 3827. 6 Surface Science T he First T hirty Y ears ed. C. B. Duke Elsevier Amsterdam 1994. 7 V.E. Henrich and P. A. Cox T he Surface Science of Metal Oxides Cambridge University Press Cambridge 1994. 8 C. T. Campbell Surf. Sci. Rep. 1997 27 1. 9 C. T. Campbell Curr. Opin. Solid State Mater. Sci. 1998 3 439. 10 D. W. Goodman Surf. Rev. L ett. 1995 2 9. 11 H.-J. Freund Angew. Chem. Int. Ed. Engl. 1997 36 452. 12 H.-J. Freund Phys. Status Solidi B 1995 192 407. 13 H.-J. Freund H. Kuhlenbeck and V. Staemmler Rep. Prog. Phys. 1996 59 283. 14 C. R. Henry Surf. Sci. Rep. 1998 31 231. 15 M. Baé umer and H.-J. Freund Prog. Surf. Sci. 1999 61 127. 16 J. M. Thomas Faraday Discuss. 1996 105 1. 17 G. Ertl and H.-J. Freund Phys. T oday 1999 52 32. 18 H. Poppa Catal. Rev. Sci. Eng. 1993 35 359. 19 U. Diebold J.-M. Pan and T. E. Madey Surf. Sci. 1995 331ñ333 845.20 G. H. Hardy A Mathematicianœs Apology Cambridge University Press 1940. 21 Adsorption on Ordered Surfaces of Ionic Solids and T hin Films ed. H.-J. Freund and E. Umbach Springer Heidelberg 1993. 22 G. H. Vurens M. Salmeron and G. A. Somorjai Prog. Surf. Sci. 1989 32 333. 23 P. L. J. Gunter J. W. H. Niemantsverdriet F. H. Ribeiro and G. A. Somorjai Catal. Rev. Sci. Eng. 1997 39 77. 24 D. A. Bonnell Prog. Surf. Sci. 1998 57 187. 25 U. Diebold J. F. Anderson K.-O. Ng and D. Vanderbilt Phys. Rev. L ett. 1996 77 1322. 26 P. W. Murray F. M. Leibsle C. A. Muryn H. J. Fisher C. F. J. Flipse and G. Thornton Phys. Rev. L ett. 1994 72 689. 27 P. W. Murray N. G. Condon and G. Thornton Phys. Rev. B 1995 51 10989. 28 H. Onishi and Y. Iwasawa Surf.Sci. 1994 313 L783. 29 H. Onishi and Y. Iwasawa Chem. Phys. L ett. 1994 226 111. Faraday Discuss. 1999 114 1»31 28 30 H. Onishi K. Fukui and Y. Iwasawa Bull. Chem. Soc. Jpn. 1995 68 2447. 31 H. Onishi and Y. Iwasawa Phys. Rev. L ett. 1996 76 791. 32 G. Charlton P. B. Howes C. L. Nicklin P. Steadman J. S. G. Taylor C. A. Muryn S. P. Harte J. Mercer R. McGrath D. Norman T. S. Turner and G. Thornton Phys. Rev. L ett. 1997 78 495. 33 G. Renaud Surf. Sci. Rep. 1998 32 1. 34 P. W. Tasker Adv. Ceram. 1984 10 176. 35 W. C. Mackrodt R. J. Davey I. N. Black and R. Docherty J. Cryst. Growth 1987 82 441. 36 C. Noguera Physics and Chemistry at Oxide Surfaces Cambridge University Press 1996. 37 P. Gueç nard G. Renaud A. Barbier and M. Gantier-Soyer Surf. Rev. L ett.1998 5 321. 38 F. Rohr M. Baé umer H.-J. Freund J. A. Mejias V. Staemmler S. Mué ller L. Hammer and K. Heinz Surf. Sci. 1997 372 L 291. 39 R. Rohr M. Baé umer H.-J. Freund J. A. Mejias V. Staemmler S. Muller L. Hammer and K. Heinz Surf. Sci. 1997 389 391. 40 W. Weiss Surf. Sci. 1997 377ñ379 943. 41 L. Manassidis and M. J. Gillan Surf. Sci. 1993 285 L517. 42 L. Manassidis and M. J. Gillan J. Am. Ceram. Soc. 1994 77 335. 43 C. Rebhein N. M. Harrison and A. Wander Phys. Rev. B 1996 54 14066. 44 X.-G. Wang W. Weiss S. K. Shaikhutdinov M. Ritter M. Petersen F. Wagner R. Schloegl and M. Scheffler Phys. Rev. L ett. 1998 81 1038. 45 K. Heinz Surf. Sci. 1984 299 433. 46 N. M. Harrison X.-G. Wang M. Muscat and M. Scheffler Faraday Discuss. 1999 114 305.47 H. Schmalzried Chemical Kinetics of Solids VCH Weinheim 1995. 48 D. Wolf Phys. Rev. L ett. 1992 68 3315. 49 A. Barbier and G. Renauld Surf. Sci. 1997 392 L15. 50 D. Cappus C. Xu D. Ehrlich B. Dillmann C. A. Ventrice Jr. K. Al-Shamery H. Kuhlenbeck and H.-J. Freund Chem. Phys. 1993 177 533. 51 F. Rohr K. Wirth J. Libuda D. Cappus M. Baé umer and H.-J. Freund Surf. Sci. 1994 315 L977. 52 D. Cappus M. Hassel E. Neuhaus M. Heber F. Rohr and H.-J. Freund Surf. Sci. 1995 337 268. 53 A. Barbier Proc. 1st Int. W orkshop Oxide Surf. (IW OX1) Elmau Germany 1999. 54 R. Lacman Colloq. Int. CNRS 1965 152 195. 55 C. A. Ventrice Jr. T. Bertrams H. Hannemann A. Brodde and H. Neddermeyer Phys. Rev. B 1994 49 1773. 56 A. Barbier C. Mocuta H. Kuhlenbeck K. Peters B.Richter and G. Renaud in preparation. 57 W. H. Casey H. R. Westrich and G. W. Arnold Geochim. Cosmochim. Acta 1988 53 2795. 58 H. Papp and B. Egersdoé rfer personal communication. 59 B. Egersdoé rfer PhD Thesis Bochum 1993. 60 J. G. Fripiat A. A. Lucas J. M. Andreç and E. G. Derouane Chem. Phys. 1977 21 101. 61 S. Hué fner P. Steiner I. Sander M. Neumann and S. Witzel Z. Phys. B 1991 83 185. 62 S. Hué fner Photoelectron Spectroscopy Principles and Applications Springer-Verlag Berlin Heidelberg 1995. 63 A. Gorschlué ter and H. Merz Phys. Rev. B 1994 49 17293. 64 A. Freitag V. Staemmler D. Cappus C. A. Ventrice Jr. K. Al-Shamery H. Kuhlenbeck and H.-J. Freund Chem. Phys. L ett. 1993 210 10. 65 R. Newman and R. M. Chrenko Phys. Rev. 1959 114 1507.66 R. J. Powell Report no. 5220-1 (unpublished) Stanford Electronics Laboratory Stanford 1967. 67 D. Adler and J. Feinleib Phys. Rev. B 1970 2 3112. 68 B. Fromme C. Koch R. Deussen and E. Kisker Phys. Rev. L ett. 1995 75 693. 69 B. Fromme M. Moé ller T. Anschué tz C. Bethke and E. Kisker Phys. Rev. L ett. 1996 77 1548. 70 M. Poé hlchen and V. Staemmler J. Chem. Phys. 1992 97 2583. 71 J. A. Mejias V. Staemmler and H.-J. Freund J. Phys. Condens. Matter in press. 72 H. Kuhlenbeck C. Xu B. Dillmann M. Haêel B. Adam D. Ehrlich S. Wohlrab H.-J. Freund U. A. Ditzinger H. Neddermeyer M. Neuber and M. Neumann Ber. Bunsen-ges. Phys. Chem. 1992 96 15. 73 C. Xu B. Dillmann H. Kuhlenbeck and H.-J. Freund Phys. Rev. L ett. 1991 67 3551. 74 M. Bender D. Ehrlich I.N. Yakovkin F. Rohr M. Baé umer H. Kuhlenbeck H. J. Freund and V. Staemmler J. Phys. Condens. Matter 1995 7 5289. Ober—aé chenphysik 75 M. Henzler and W. Goé pel des Festkoé rpers Teubner»Verlag Stuttgart 1991. 76 K. M. Neyman G. Pacchioni and N. Roé sch in Recent Developments and Applications of Modern Density Functional T heory and Computational Chemistry ed. J. M. Seminario Elsevier Amsterdam 1996 p. 569. 77 M. Baé umer J. Libuda and H.-J. Freund in Chemisorption and Reactivity on Supported Clusters and T hin Films ed. R. M. Lambert and G. Pacchioni Kluwer Academic Press Dordrecht 1997 p. 61. 78 M. A. Nygren and L. G. M. Pettersson J. Chem. Phys. 1996 105 9339. 79 G. Pacchioni and P. S. Bagus in Adsorption on Ordered Surfaces of Ionic Solids and T hin Films ed.H.-J. Freund and E. Umbach Springer Verlag Berlin 1993 p. 180. 80 T. Klué ner personal communication. 81 D. Cappus J. Klinkmann H. Kuhlenbeck and H.-J. Freund Surf. Sci. 1995 325 L 421. 29 Faraday Discuss. 1999 114 1»31 82 S. M. Vesecky X. Xu and D. W. Goodman J. V ac. Sci. T echnol. A 1994 12 2114. 83 V. Staemmler in Adsorption on Ordered Surfaces of Ionic Solids and T hin Films ed. H.-J. Freund and E. Umbach Springer Verlag Berlin 1993 p. 169. 84 M. Poé hlchen PhD Thesis Ruhr-Universitaé t Bochum 1992. 85 H. Kuhlenbeck G. Odoé rfer R. Jaeger G. Illing M. Menges T. Mull H.-J. Freund M. Poé hlchen V. Staemmler S. Witzel C. Scharfschwerdt K. Wennemann T. Liedtke and M. Neumann Phys. Rev. B 1991 43 1969. 86 L. Chen R. Wu N. Kioussis and Q. Zhang Chem.Phys. L ett. 1998 290 255. 87 K. M. Neyman S. P. Ruzankin and N. Roé sch Chem. Phys. L ett. 1995 246 546. 88 J.-W. He C. A. Estrada J. S. Corneille M.-C. Wu and D. W. Goodman Surf. Sci. 1992 261 164. 89 S. Furuyama H. Fuji M. Kawamura and T. Morimoto J. Phys. Chem. 1978 82 1028. 90 R. Wichtendahl M. Rodriguez-Rodrigo U. Haé rtel H. Kuhlenbeck and H.-J. Freund Phys. Status Solidi A 1999 173 93. 91 H. Pfnué r P. Feulner and D. Menzel J. Chem. Phys. 1983 79 4613. 92 C. N. Chittenden E. D. Pylant A. L. Schwaner and J. M. White T hermal Desorption and Mass Spectrometry ed. A. T. Hubbard CRC Press Boca Raton FL 1995. 93 H. Schlichting and D. Menzel Rev. Sci. Instrum. 1993 64 2013. 94 J. Heidberg M. Kandel D. Meine and U. Wildt Surf. Sci. 1995 333 1467.95 R. Gerlach A. Glebov G. Lange J. P. Toennies and W. Weiss Surf. Sci. 1995 331ñ333 1490. 96 R. Wichtendahl PhD Thesis Freie Universitaé t Berlin 1999. 97 M. J. Stirniman C. Huang R. C. Smith J. A. Joyce and B. D. Kay J. Chem. Phys. 1996 105 1295. 98 C. Xu and D. W. Goodman Chem. Phys. L ett. 1997 L65 341. 99 J. Libuda M. Frank A. Sandell S. Andersson P. A. Brué hwiler M. Baé umer N. Ma- rtensson and H.-J. Freund Surf. Sci. 1997 384 106. 100 K. Wolter and M. Frank personal communication. 101 J. Hemminger personal communication. 102 R. M. Jaeger H. Kuhlenbeck H.-J. Freund M. Wuttig W. HoÜmann R. Franchy and H. Ibach Surf. Sci. 1991 259 235. 103 I. Engquist and B. Liedberg J. Phys. Chem. 1996 100 20089. 104 H. Knoé zinger and P. Ratnasaniy Catal.Rev. Sci. Eng. 1978 17 31. 105 D. Ehrlich PhD Thesis Bochum 1995. 106 O. Seiferth K. Wolter B. Dillmann G. Klivenyi H.-J. Freund D. Scarano and A. Zecchina Surf. Sci. 1999 421 176. 107 A. Zecchina D. Scarano S. Bordiga G. Ricchiardi G. Spoto and F. Geobaldo Catal. T oday 1996 27 403. 108 C. Xu M. Haêel H. Kuhlenbeck and H.-J. Freund Surf. Sci. 1991 258 23. 109 I. Hemmerich F. Rohr O. Seiferth B. Dillmann and H.-J. Freund Z. Phys. Chem. 1997 202 31. 110 P. C. Thué ne and J. W. Niemantsverdriet Isr. J. Chem. 1998 38 385. 111 M. Menges B. Baumeister K. Al-Shamery H.-J. Freund C. Fischer and P. Andresen J. Chem. Phys. M‘ller Surf. Sci. 1989 221 250. M‘ller and M. C. Wu Surf. Sci. 1989 224 265. 1994 101(4) 3318. 112 I. Beauport K. Al-Shamery and H.-J.Freund Chem. Phys. L ett. 1996 256 641. 113 T. Klué ner H.-J. Freund V. Staemmler and R. KosloÜ Phys. Rev. L ett. 1998 80 5208. 114 K. Al-Shamery Appl. Phys. A 1996 63 509. 115 E. A. Wovchko and J. Yates Jr. L angmuir 1999 15 3506. 116 M. C. Wu and P. J. 117 P. J. 118 M. C. Wu and P. J. M‘ller Surf. Sci. 1990 235 228. 119 P. J. M‘ller and J. Nerlov Surf. Sci. 1993 307»9 591. 120 P. W. Murray J. Shen N. G. Condon S. J. Peng and G. Thornton Surf. Sci. 1997 380 L455. 121 P. Stone R. A. Bennett and M. Bowker New J. Phys. 1998 1 8. 122 U. Diebold Proc. 1st Int. W orkshop Oxide Surf. (IW OX1) Elmau Germany 1999. 123 J. Libuda F. Winkelmann M. Baé umer H.-J. Freund T. Bertrams H. Neddermeyer and K. Mué ller Surf. Sci. 1994 318 61. 124 S. Stempel M.Baé umer and H.-J. Freund Surf. Sci. 1998 402»404 424. 125 M. Heemeier PhD Thesis Berlin in preparation. 126 T. R. Linderoth S. Horch E. Laensgaard I. Stensgaard and F. Besenbacher Surf. Sci. 1998 404 308. 127 N. Ernst B. Duncombe G. Bozdech M. Naschitzki and H.-J. Freund Ultramicroscopy 1999 79 231. 128 A. Piednoir E. Pernot S. Granjeand A. Humbert C. Chapon and C. R. Henry Surf. Sci. 1997 391 19. 129 K. H. Hansen T. Warren S. Stempel E. Laegsgaard M. Baé umer H.-J. Freund F. Besenbacher and I. Stensgaard Phys. Rev. L ett. in the press. 130 M. Methfessel D. Hennig and M. Scheffler Phys. Rev. B 1992 46 4816. 131 A. Bogicevic and D. R. Jennison Phys. Rev. L ett. 1999 82 4050. 132 M. Klimenkov S. Nepijko H. Kuhlenbeck M. Baé umer R. Schloé gl and H.-J. Freund Surf.Sci. 1997 391 27. 133 M. Klimenkov S. Nepijko H. Kuhlenbeck and H.-J. Freund Surf. Sci. 1997 385 66. Faraday Discuss. 1999 114 1»31 30 134 S. Nepijko M. Klimenkov H. Kuhlenbeck D. Zemylanov D. Herein R. Schloé gl and H.-J. Freund Surf. Sci. 1998 413 192. 135 Clusters and Colloids From T heory to Applications ed. G. Schmid VCH Weinheim 1994. 136 S. A. Nepijko M. Klimenkov M. Adelt H. Kuhlenbeck R. Stogé l and H.-J. Freund L angmuir 1999 15 5309. 137 Metal Clusters ed. W. Ekardt John Wiley Chichester 1999. 138 T ransition Metal Clusters ed. B. F. G. Johnson John Wiley Chichester 1980. 139 F. F. de Biani C. Femoni M. C. Iapalucci G. Longoni P. Zanello and A. Ceriotti Inorg. Chem. 1999 38(16) 3721. 140 G. K. Wertheim S. B. DiCenzo and D.N. E. Buchanan Phys. Rev. B 1986 33 5384. 141 A. Sandell J. Libuda P. Brué hwiler S. Andersson A. Maxwell M. Baé umer N. Ma- rtensson and H.-J. Freund J. Electron Spectrosc. Relat. Phenom. 1995 76 301. 142 R. Unwin and A. M. Bradshaw Chem. Phys. L ett. 1978 58 58. 143 G. K. Wertheim Z. Phys. D 1989 12 319. 144 S.-T. Lee G. Apai M. G. Mason R. Benbow and Z. Hurych Phys. Rev. B 1981 23 505. 145 V. de Gouveia B. Bellamy Y. Hadj Romdhane A. Mason and M. Che Z. Phys. D 1989 12 587. 146 C. Kuhrt and M. HarsdorÜ Surf. Sci. 1991 245 173. 147 A. Sandell J. Libuda P. A. Brué hwiler S. Andersson A. J. Maxwell M. Baé umer N. Ma- rtensson and H.-J. Freund J. V ac. Sci. T echnol. A 1996 14 1546. 148 O. D. Haé berlen S.-C. Chung M. Stener and N. Roé sch J. Chem. Phys.1997 106 5189. 149 A. Sandell J. Libuda P. A. Brué hwiler S. Andersson M. Baé umer A. J. Maxwell N. Ma- rtensson and H.-J. Freund Phys. Rev. B 1997 55 7233. 150 D. W. Goodman J. V ac. Sci. T echnol. A 1996 14 1526. 151 M. Valden and D. W. Goodman Science 1998 281 1647. 152 D. R. Rainer and D. W. Goodman in NAT O ASI ed. G. Pacchioni and R. M. Lambert Kluwer W. Berndt A. M. Bradshaw C. Baddeley A. F. Lee R. M. Lambert and D. P. WoodruÜ Surf. Sci. 1998 162 G. Kisters J. G. Chen S. Lehwald and H. Ibach Surf. Sci. 1991 245 65. 163 J. Lauterbach R. W. Boyle M. Schick W. J. Mitchell B. Meng and W. H. Weinberg Surf. Sci. 1996 164 F. Solymosi Eä . Novaç k and A. Molnaç r J. Phys. Chem. 1990 94 7250. 165 T. W. Voskobojnikov E. S. Shpiro H. Landmesser N. I. Jaeger and G.Schulz-EkloÜ J. Mol. Catal. A 166 C. DeLaCruz and N. Sheppard J. Chem. Soc. Faraday T rans. 1997 93 3569. 167 K. H‘jrup Hansen S. Stempel E. Laegsgaard M. Baé umer F. Besenbacher and I. Stensgaard personal Dordrecht 1997 p. 27. 153 M. Adelt S. Nepijko W. Drachsel and H.-J. Freund Chem. Phys. L ett. 1998 291 425. 154 J. Evans B. Hayden F. Mosselman and A. Murray Surf. Sci. 1992 279 L159. 155 J. Evans B. Hayden F. Mosselman and A. Murray Surf. Sci. 1994 301 61. 156 K. Wolter O. Seiferth H. Kuhlenbeck M. Baé umer and H.-J. Freund Surf. Sci. 1998 399 190. 157 M. Frank R. Kué hnemuth M. Baé umer and H.-J. Freund Surf. Sci. 1999 427/428 288. 158 M. Frank R. Kué hnemuth M. Baé umer and H.-J. Freund Surf. Sci. submitted. 159 T. Mineva and N. Russo personal communication.160 P. Uvdal P.-A. Karlsson C. Nyberg S. Andersson and N. V. Richardson Surf. Sci. 1988 202 167. 161 T. Giessel O. SchaÜ C. J. Hirschmugl V. Fernandez K. M. Schindler A. Theobald S. Bao R. Lindsay 406 90. 350 32. Chem. 1996 104 299. communication. 168 H. Conrad G. Ertl and J. Kué ppers Surf. Sci. 1978 76 323. 169 M. Haruta Catal. T oday 1997 36 153. 170 L. Piccolo C. Becker and C. R. Henry Euro. Phys. J. in the press. 171 M. Frank S. Andersson J. Libuda S. Stempel A. Sandell B. Brena A. Giertz P. A. Brué hwiler M. Baé umer N. Ma- rtensson and H.-J. Freund Chem. Phys. L ett. 1997 279 92. 172 S. Andersson M. Frank A. Sandell A. Giertz B. Brena P. A. Brué hwiler N. Ma- rtensson J. Libuda M. Baé umer and H.-J. Freund J. Chem. Phys. 1998 108 2967. 173 S. Andersson M. Frank A. Sandell J. Libuda B. Brena A. Giertz P. A. Brué hwiler M. Baé umer N. Ma- rtensson and H.-J. Freund V acuum 1998 49 167. 174 M. P. Irion personal communication. 175 J. T. Yates E. D. Williams and W. H. Weinberg Surf. Sci. 1980 91 562. 176 M. Rebholz R. Prins and N. Kruse Surf. Sci. 1991 259 L791. 177 K. Refson R. A. Wogelius D. G. Fraser M. C. Payne M. H. Lee and V. Milman Phys. Rev. B 1995 52 10823. 178 U. Birkenheuer J. C. Boettger and N. Roé sch J. Chem. Phys. 1994 100 6826. 179 P. A. Redhead V acuum 1962 12 203. Paper 9/07182B 31 Faraday Discuss. 1999 114 1»31

ISSN:1359-6640    

DOI:10.1039/a907182b

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Ab initio calculations on the Al2O3(0001) surface Iskander Batyrev Ali Alavi and Michael W. Finnis* Atomistic Simulation Group School of Mathematics and Physics T he Queenœs University of Belfast Belfast UK BT 7 1NN 1. Introduction The (0001) surface of a-Al2O3 (corundum or sapphire) has been studied experimentally and theoretically for many years since it is a widely used substrate for many kinds of thin –lms and an archetypal wide band-gap ionic oxide. This surface terminates a layered structure in which each oxygen plane in the bulk has an associated aluminium plane at a distance 0.838 ” above and below it forming a stoichiometric triple layer. The most convenient unit cell is a rhombohedral prism comprising six such (0001) oxygen planes separated by the associated pairs of aluminium planes.The cell contains just one atom in each Al plane and three atoms in each O plane. In the bulk structure the oxygen planes are separated by 2.166 ” and form a hexagonal lattice with ABABAB . . . stacking. Their positions are slightly laterally distorted from ideal hexagonal sites. The Al atoms occupy two-thirds of the octahedral holes in the oxygen sublattice at positions which alternate above and below the centres of these holes. The unoccupied octahedral holes are themselves stacked on a face-centred cubic lattice abcabc . . . A C3v symmetry axis passes through each Al atom and through the centres of the unoccupied octahedral sites. Fig. 1 shows an elevation of the structure. Only recently with the advent of grazing incidence X-ray diÜraction has it become possible to compare structural predictions for the (0001) surface directly with experiment.Renaud has published a comprehensive review of the experimental and theoretical results.1 The normally observed structure is (1]1) for which one can postulate three possible bulk terminations single Al double Al or O. The observed surface is believed to be terminated by a single Al plane as in Fig. 1 in which this plane is labelled Al1. This termination is stoichiometric that is it exhibits no surface excess of Al or O. The top layers relax strongly ; best estimates of the interlayer relaxations of the top four layers from experiment2 are [51% 16% [29% and ]20%. Manassidis et al.3 were the –rst to use density functional theory (DFT) within the local density approximation (LDA) for exchange and correlation to calculate the energy and relaxation of this surface.They used a pseudopotential approach and modelled the surface with a slab containing 33 Faraday Discuss. 1999 114 33»43 Received 26th April 1999 2O3 We calculate using a density functional pseudopotential method the atomic and electronic structure of the (0001) surface of a-alumina (Al2O3). The material is studied in the form of a slab with periodic boundary conditions containing up to eight layers of the stoichiometric Al units. Five diÜerent terminations of the surface are calculated representing diÜerent surface excesses of oxygen and their free energies are estimated as a function of oxygen partial pressure. Internal relaxations of the atomic positions are obtained.The aluminium terminated surface which is stoichiometric has the lowest surface energy for a wide range of oxygen pressures. This journal is( The Royal Society of Chemistry 2000 Fig. 1 View of the corundum lattice terminated at the (0001) surface by the Al plane labelled Al1. This is the stoichiometric termination (CO\0) and the main one observed experimentally. three layers of oxygen. The most striking feature of the surface is the large inward relaxation of the surface Al plane which in this calculation reduces the –rst interplanar spacing by 85%. Their results were later con–rmed with a similar calculation by Kruse et al.4 A recent DFT calculation by Verdozzi and co-workers5 again con–rmed the large surface relaxation ; they used slabs containing up to 18 layers of oxygen eliminating any uncertainty as to the eÜect of –nite slab thickness.Their use of a thicker slab also enabled them to predict the interlayer spacings of deeper layers. For the –rst four layers they found relaxations of [87% ]3% [42% ]19%. Puchin et al.6 have also calculated the atomic and electronic structure of the stoichiometric Al-terminated surface modelled as a slab containing 3 layers of oxygen. They used the Hartree» Fock (HF) method (embodied in the CRYSTAL code) with an 85-11G/8-411G basis set for each atom. The surface Al layer was found to relax inwards by 68% somewhat less than the DFT calculations (the diÜerence being 0.016 nm) but signi–cantly more than the previous HF results (48%) using a smaller basis.7,8 Puchin et al.went on to calculate theoretical UV photoelectron spectra (UPS) and metastable impact electron spectra (MIES) and compared them with experiment concluding that the model gave a good description of the surface resolved density of states (DOS) in the valence band and even that the relaxation was more consistent with these experimental data than the somewhat larger DFT relaxation. Calculations of the atomic and electronic structure of the (0001) surface have been made with a non-self-consistent tight-binding model by Godin and LaFemina.9 They also reported a very large ([90%) relaxation of the surface Al layer which they attributed to the formation of almost planar sp2 bonds between the surface Al and the O triangle on which it sits.However this large relaxation was also found in the earliest studies with classical ionic interatomic potentials,10 which of course have no representation of chemical bonding whether sp2 or sp3. We note the above discrepancy between the interpretations of the large inward relaxation of the surface Al layer. The argument in terms of bond hybridisation we –nd less persuasive than the simple electrostatic argument; because (a) ultimately the driving force for any relaxation or reconstruction is the classical electrostatic force on the relaxing ion from the self-consistent charge density (the Hellmann»Feynman theorem) and (b) in the –rst-principles calculations there is no sign of sp2 bonding in the charge density distribution which looks very ionic.There are other seemingly contradictory statements and observations concerning this surface which are worth discussing here. It was noted for example by Godin and LaFemina9 that the surface Al»O bond length is conserved (to within 10%) as the Al relaxes inwards which is eÜected by an associated expansion of the O triangle. Their tight binding model showed this eÜect and it is also apparent Faraday Discuss. 1999 114 33»43 34 from the X-ray data referred to above from which the Al»O bond length is only 4.5% (^2.5%) shorter than the bulk nearest neighbour value. However it is not the case as Godin and LaFemina speculated that the outward lateral relaxation of the O triangle is a necessary condition for the large inward relaxation of the Al (by analogy with semiconductor surfaces and their bondlength conserving reconstructions) because the earlier –rst-principles calculations of large Al relaxation3,4 did not allow lateral relaxations at all.In the present paper we report –rst-principles calculations for this surface including full relaxation of all atomic co-ordinates in the unit cell and discuss the pattern of relaxation in some detail. The nature of the electrostatic driving force for the Al relaxation has also been described as reducing the electrostatic dipole of the –rst two atomic layers in the manner of the attraction between the plates of a capacitor. This seems a satisfactory explanation as long as one bears in mind that the –rst three atomic layers (Al»O»Al . . .) of the structure form an electrically neutral object in the sense of the ionic model.Another argument which has sometimes been aired is that only the single Al terminated surface should be stable because an alternative (1]1) surface such as the one terminated by an O plane would be charged and therefore have an in–nite surface energy. This is clearly only the case within a purely ionic model in which all ions carry their bulk formal charges. However it is a poor basis for discussing the present surface especially since it is now known experimentally that there are other rather stable terminations of it. In particular on heating in UHV this surface undergoes a series of reconstructions culminating in a (J31]J31)R^9° structure at around 1350 °C; R indicates that the reconstructed layer is rotated with respect to the substrate.This structure has now been characterised11 and is believed to comprise domains of nearly two Al(111) layers formed by losing the –rst two layers of oxygen from the stoichiometric (1]1) surface discussed above. A major aim of the present paper is to describe how the relative energetics of surfaces of diÜerent stoichiometry can be calculated theoretically. There is one independent variable needed to compare the surface energies of surfaces of diÜering stoichiometry (or surface excess of one component) and that is most usefully taken to be the partial pressure of oxygen since this is what is most directly under the control of the experimentalist. We formulate the theory in Section 3 with this in mind. First in the following section we describe brie—y our method of calculation of total energies which provide input to the theory.While we have insufficient computer resources to make calculations for the (J31]J31)R^9° surface we shall illustrate how this could be done in principle by comparing the energetics of –ve alternative (1]1) surfaces of diÜering stoichiometry. This will also show that previous arguments based on classical electrostatics for the stability of the stoichiometric Al-terminated surface are of doubtful validity. We –nd for example that the O-terminated surface could in theory be stabilised by a moderately high pressure of oxygen. Our results are described in Section 4 and we conclude in Section 5. 2. Method of calculation We base our total energy calculations on a supercell with periodic boundary conditions which enables us to use a basis of plane waves.The supercell has the form of a rhombohedral prism and in the stoichiometric slab it contains 30 atoms. This slab is exactly the thickness of one bulk unit cell of the corundum structure. Our surface calculations are made on slabs of this kind repeated periodically in the z direction with a vacuum space of thickness about equal to that of the slab. This is adequate to isolate the surfaces. The stoichiometric slab has two equivalent surfaces which are terminated by an Al plane as in Fig. 1 in which our labelling of the atoms is shown. This termination de–nes the stoichiometric surface because the slab as a whole is stoichiometric and has two equivalent surfaces.We note as an aside that surfaces which are not stoichiometric are sometimes called polar. Besides the stoichiometric surface we have studied four non-stoichiometric surfaces two with an oxygen excess and two with an aluminium excess. By stripping oÜ the surface plane of Al we obtain a surface which is O terminated with an O excess of ]1.5 atoms per unit surface cell or an Al excess of [1 atoms per unit surface cell (see eqn. (5) below). We can then proceed to remove the three surface O atoms one at a time giving three more surfaces with O excesses of 0.5 [0.5 and [1.5 respectively. The –nal surface is terminated by two layers of Al. In the actual calculations we ensure that the surfaces on each side of our slab are equivalent (to avoid a dipole moment in 35 Faraday Discuss.1999 114 33»43 the supercell). Thus we actually add a seventh layer of oxygen to create the O-terminated surfaces and remove a layer of oxygen to create the most Al-rich surfaces. The total energy of the contents of a supercell is minimised with respect to the electronic coordinates (the coefficients of each plane wave in each occupied wave function) and the atomic co-ordinates as described in previous work.12,13 The ionic potentials are represented by pseudopotentials of the Troullier»Martens form.14 All the calculations were made with two k-points in the irreducible (120°) wedge of the Brillouin zone and with a plane-wave cut-oÜ of 40 Ry. Convergence tests were reported previously.15 1[; In order to obtain information about the charge redistribution at these surfaces we calculate the Mulliken populations.These are de–ned by the formalism of Mayer.16 In our case we use the o r pseudo-orbitals iaT as a basis for projection which includes O 2s O 2p Al 3s and Al 3p components (labelled a) on each site (labelled i). The ìì spillage œœ of each occupied orbital t de–ned as ia o St oriaT o2 with this atomic-like basis is less than 1.5%. (2) 3. Surface energy and oxygen partial pressure We have chosen here to discuss the regime we think is more likely to be encountered in practice in which the Al2O3 surface is in equilibrium with Al and O in the vapour phase rather than the regime in which it is in equilibrium with Al in the solid or liquid phases. Since there are two components the temperature T and pressure P are two independent variables which can be used to specify the state of the system.To be of use in experimental situations our –nal formula for the (k0) ( or partial pressure p of surface energy will be couched in terms of the chemical potential O2 ) oxygen rather than the total pressure P. We adopt the thermodynamic de–nition of surface excesses due to Gibbs most clearly described by Cahn.17 For future application to other oxides we derive here the general formula appropriate to an oxide of metal M with the stoichiometric composition MmOn . case is given by c(T P) … A\Gslab(T P)[NM kM v (T P)[NO kO(T P) M The surface energy c in this (1) N and N are the total number of atoms of kM v is the chemical potential of the metal and O in which A is the area of the surface within the supercell counting both surfaces of the slab.Gslab is the Gibbs energy of the contents of the supercell and metal and oxygen respectively within the supercell. we use the superscript v to emphasise that at the temperature of interest the metal is in the vapour phase. If this were not the case we would be dealing with an interface of the solid or liquid metal with its oxide and this would require a diÜerent treatment. In a supercell of a few tens of atoms such as we use for ab initio calculations all the atoms can be assumed to be in the solid phase and approximately half of the supercell is a true vacuum. No serious error is made in eqn. (1) by omitting the statistical occurrence within the supercell of atoms in the vapour phase.Likewise no signi–cant error is made by not explicitly including the presence of point defects within the solid part of the supercell. It is more problematic that we are ignoring the surface terminations involving a statistical distribution of defects as in a partial layer of adsorbed oxygen. The chemical potential of oxygen is given in terms of its partial pressure pO2 by the ideal gas expression kO ° \0 but the convention of our calculations means that if we use kO\kO ° ]12kT ln pO2 in which we use superscript ° to denote the standard state (STP) and the pressure is in units of atmospheres. We choose to de–ne the chemical potentials per atom rather than per mole. In order to relate c(T P) to pO2 we wish to insert eqn.(2) into eqn. (1). Two problems arise at this point. First in ab initio calculations the zero of free energy is the energy of separated free ions and electrons at rest ; this is not the usual convention which is the free energy of the elements at STP. With the latter convention eqn. (2) to evaluate eqn. (1) we still have to know the value of kO° . Secondly the free energy of oxygen in the vapour phase is a quantity we wish to avoid calculating explicitly. Because of the paramagnetic nature of oxygen even the energy of a molecule at rest would require a more accurate treatment than the local density approximation for exchange and correlation which is Faraday Discuss. 1999 114 33»43 36 adequate for solid phases. We can deal with both these problems by making use of experimental thermodynamic data as follows.First we can use the equilibrium of the two phases to eliminate the chemical potential of the metal vapour in favour of the Gibbs energy per formula unit of the metal oxide GMO (3) GMO\mkMv ]nkO . Inserting eqn. (3) into eqn. (1) gives (4) c(T P)\ 1 A (Gslab(T P)[ m 1 NMGMO(T P))[CO kO(T P) in which we have introduced the surface excess of oxygen with respect to the metal de–ned by17,18 (5) CO\ A 1 N A O[ m n NMB. Next we use the de–nition of the standard Gibbs energy of formation to obtain the oxygen chemical potential in its standard state (6) GMO ° \mkM ° ]nkO ° ]*G°. G c(T P)\ 1 A A Combining eqn. (6) with eqn. (4) and (2) gives slab(T P)[ m 1 NMGMO(T P)B (7) [COA1 n GMO ° [ m n kM ° [ 1 n *G°B[CO 1 2 kT ln pO2 .and The quantities Eqn. (7) is the result we were seeking. The –rst term is calculated at 0 K and we omit its temperature dependence. In principle one could perform a molecular dynamics simulation and by means of thermodynamic integration from 0 K to T obtain the temperature dependence of this term. Alternatively at temperatures not too close to the melting point it could be estimated from a calculation of the phonon frequencies of the slab and of a comparison piece of bulk material using the quasiharmonic approximation. The accuracy of both approaches is limited by the sample size but we feel that either would be reasonable since the result is an integrated quantity localised near the surface.The –rst term is also where a large cancellation takes place between the diÜerence of two energies which scale with the number of atoms leaving a super–cial quantity. It is therefore advantageous to use an identical cell and k-point mesh for the slab and bulk terms in it. k entering the second term are well described by 0 K quantities which GMO ° M ° we calculate. It can be veri–ed that correcting them to STP has a negligible eÜect on the surface energy. Finally *G° is taken from experimental thermodynamic data. This leaves the pO2 term as the one that describes the important variation of c with temperature and oxygen partial pressure. Notice that in the case of a stoichoimetric termination only the –rst term is non-zero and this describes the elementary case in which the energy of bulk phase is subtracted from the energy of a slab containing exactly the same number of molecules.pO2 which we denote pmin is set by the condition O2 OpO2 min the oxide would spontaneously decompose into metal and oxygen. This would O2 The minimum physically meaningful value of that at p be the case if (8) kM c (T P)OkM v (T P) where kM c (T P) denotes the chemical potential of the metal in its condensed phase. The result in terms of pO2 is fairly well known but we brie—y rederive it here for completeness. Substituting eqns. (3) and (6) into eqn. (8) and rearranging the inequality gives (9) kO[kO ° O 1 n *G°] 1 n (GAO(T P)[GAO ° )[ m n (kM c (T P)[kM ° ).37 Faraday Discuss. 1999 114 33»43 From the speci–c heat data of the solids we know that the second and third terms are about two orders of magnitude smaller than the –rst in alumina and we expect them to be negligible in general. Hence we obtain the relation (10) ln pmin\ nkT *G°. 2 O2 It is worth noting that the above formulae could be applied directly to calculate the energy of oxide/metal interfaces. All that is required is to subtract from eqn. (7) the energy of the same amount of bulk phase of the metal with which the oxide is in contact. In this case Gslab would refer to a supercell containing a slab of oxide in contact with metal and there would be no vacuum space. Note on nomenclature for surface excess In the remainder of this paper we shall characterise the surface terminations by the natural unit of oxygen surface excess CO which is atoms per surface unit cell rather than atoms per unit area of the slab.This is equivalent to setting A\2 in the above equations (since there are two surfaces in the supercell). A simple algorithm to determine the surface excess for a given surface which does not rely on having equivalent surfaces on a slab is given by Finnis.18 4. Results and discussion 4.1. Surface relaxation The interlayer relaxations are accompanied by x-y (in-plane) relaxations of the three O atoms per plane in each unit cell. The manner of these in-plane relaxations is illustrated in Fig. 2. The Al atoms do not relax laterally but remain aligned along z to preserve the threefold axes.Since the three O atoms above an empty octahedral site in each plane are symmetry related and lie at the corners of an equilateral triangle their lateral relaxation can be characterised by two parameters Fig. 2 Plan view of the (0001) surface illustrated in Fig. 1 (CO\0) showing the lateral relaxations within the topmost O plane. The relaxed positions are shaded black. The rotation and expansion of the triangle below Al1 is indicated. Faraday Discuss. 1999 114 33»43 38 which we take to be the rotation of the triangle and the linear expansion or bond length change of the triangle. This is the triangle on which the surface Al atoms labelled Al1 in Fig. 1 are centred. The magnitudes of the relaxations for three surfaces are reported in Table 1.We –nd for the CO\0 surface the only one for which a comparison with experiment is possible qualitative agreement between calculations and experimental measurements. The most marked eÜect is the well known strong inward relaxation of Al1 which we calculate to be 77% and which as in the previous published calculations of this quantity is signi–cantly larger than the experimental value of 51%. The accompanying expansion and rotation of the O1 triangle centred on an axis through Al1 are calculated to be 3.2% and 3.05° compared to the experimental values of 4.2% and 6.7°. These are quantities for which predictions have not previously been compared directly to experiment. We believe that Al1 as it relaxes is squeezing the O1 triangle open.This triangle rotates so as to allow the O1 atoms to approach Al2 rather than to move towards other O1 atoms in the same plane. There are three symmetry-equivalent O1 atoms which thereby form a more constricting triangle about each Al2 atom so the interplanar O1»Al2 distance not surprisingly increases. This is our interpretation of the 10.6% calculated O1»Al2 interplanar relaxation which is 16% experimentally. Unfortunately previous calculations give a much smaller or even negative value for this parameter. We tend to discount the tight-binding results in this comparison because they showed the opposite sign for the third layer (Al2»Al3) relaxation compared to ours and experiment. Nevertheless there remains a signi–cant discrepancy between theory and experiment and even between the –rst-principles theories for which we have no satisfactory explanation.O\1.5 The calculations predict a similar rotation»expansion of the O1 triangle for the C (oxygen terminated) surface. In this case the topmost interlayer spacing contracts by 14.6% and these relaxations can be consistently explained by the electrostatic pull of the Al2 ions. 4.2. Mulliken charges The total charges on each ion calculated from the Mulliken population analysis are shown in Table 2. Caution is needed in interpreting these data for two reasons. First because the local basis is not complete the Mulliken charges do not exactly balance the ionic charges. For example each formula unit carries an ionic charge of 2]3]3]6\24. A spillage of 1% therefore corresponds Table 1 Calculated relaxations of the O-terminated (CO\1.5) Alterminated (CO\0) and double Al-terminated (CO\[1.5) surfacesa C 0(exp) 0 O 1.5 [1.5 0.60 0.62 4.12 4.30 » » ]13.9 6.7 4.2 [51 16 [29 20 3.05 3.20 [77.0 ([70.3) ]10.6 (]13.9) [34.3 ([38.3) ]18.5 (]22.5) » [14.6 ]6.8 ]12.1 [3.6 [14.9 O-Rotation O-Expansion Al1»O1 O1»Al2 Al2»Al3 Al3»O2 O2»Al4 Al4»O5 ]7.9 ]7.7 ]1.0 ]1.0 (]9.5) [1.9 ([4.1) a The supercells contained 33 30 and 27 atoms respectively.Values in parentheses for CO\0 were calculated for a thinner slab of only 20 atoms and are shown in order to give an indication of –nite size eÜects. The third column contains experimental results.2 The –rst row shows the rotation of the topmost O triangle which is measured in degrees all other relaxations are given in percentages of bulk distances the second row gives the O»O expansion of the topmost O triangle and subsequent rows give interplanar relaxations.The topmost O triangle contains the atoms labelled O1 except for the double Al termination for which the topmost O atoms are O2. 39 Faraday Discuss. 1999 114 33»43 Table 2 Mulliken]ionic charges for atoms near the surfacea CO 0 [1.5 [0.5 ]0.5 ]1.5 » » » [0.93 1.20 1.43 [1.03 1.57 1.72 [1.03 1.54 1.57 [0.99 1.57 » [0.76 1.65 1.61 [0.99 1.57 » [0.58 1.69 1.66 [0.99 1.57 Al1 O1 Al2 Al3 O2 Al bulk O bulk 0.75 0.88 [1.01 1.57 [1.00 [1.00 [1.00 [1.00 [1.00 a The columns label the –ve surfaces by their surface excess of O in atoms per unit surface cell.The notation for atomic planes is as in Fig. 1. to 0.24 ììmissingœœ electrons per formula unit. We have not sought to allocate these missing electrons in any way which would certainly be arbitrary. Secondly because the local basis itself is not unique the charge transfers calculated are not unique and a diÜerent choice of basis would give diÜerent results. However the trends in charge transfer can be interpreted in a meaningful way which will not be diÜerent if the charges themselves are rede–ned by a diÜerent choice of local orbitals. With these caveats in mind we comment on the results in Table 2.First considering the CO\0 surface the surface ionic charges are similar to the bulk but the ionicity is slightly enhanced. The charge on Al1 for example is 1.72 compared to 1.57 in the bulk an extra charge which is mostly drawn from Al2. There is certainly no sign of increased covalence which one might have associated with the large relaxation of Al1. Secondly on the O terminated surface (CO\1.5) the surface O is carrying 0.58 electrons compared to 1.0 in the bulk. As the oxygen excess is reduced to zero the number increases from 0.58 to 1.03 slightly exceeding the bulk value. Correspondingly as the oxygen excess decreases and becomes negative the topmost Al tends to carry less charge. In the most Al-rich case (CO\[1.5) the bulk charge on Al of 1.57 is roughly shared between the two surface layers of Al (0.75 and 0.88).These large variations in surface charge highlight the well known difficulty of using an ionic model in all but stoichiometric situations. 4.3. Densities of states The stoichiometric surface is known from calculations to be insulating with a large band gap like the bulk material however in the non-stoichiometric surfaces two kinds of surface metallisation can be expected. For Al-rich surfaces the charge on Al is reduced and the extra electrons may occupy the Al 3s and 3p states to give a localised conducting band. Conversely for O-rich surfaces electrons may be missing from the O 2p states which characterise the top of the valence band in alumina and this may provide surface localised conducting states of hole character.The above picture is consistent with the trends in Mulliken charge documented above and it is con–rmed by calculations of the densities of states on the slabs shown in Fig. 3. These show the empty states at the top of the valence band in O-rich surfaces and the metallic band of electrons in Al-rich surfaces. The strong surface localisation of these surface states is illustrated by the way their charge density decays within two or three atomic layers of the surface. An example is plotted in Fig. 4 which depicts the charge density of a HOMO in the CO\1.5 slab. The stoichiometric surface in these calculations displays a localised state in the gap just below the conduction band as was found also with the tight-binding model after relaxing the atomic positions.9 4.4.Surface energies The surface energies are plotted as a function of pO2 in Fig. 5 using eqn. (7). We see that over almost all the range of pressure up to 1 atmosphere the CO\0 surface has the lowest free energy. Only in UHV could we hope to see the Al-rich surface. This con–rms the experimental picture Faraday Discuss. 1999 114 33»43 40 Fig. 3 Total density of states of slabs. (a) CO\1.5 (b) CO\0.5 (c) CO\0 (d) CO\[0.5. Fig. 4 Charge density of the HOMO for the O-terminated slab (CO\1.5). 41 Faraday Discuss. 1999 114 33»43 Fig. 5 Calculated surface energies of various 1]1 surfaces as a function of oxygen partial pressure ; po\1 atm. Two diÜerent temperatures are shown explicitly by scaling of the horizontal axis according to eqn.(7). except we have no theoretical data for the observed Al-rich (J31]J31)R^9° surface which has been observed under UHV conditions. It has an oxygen excess CO\[6 if the postulated structure1,2,11,19 is correct so it would have a very steep slope of surface energy vs. pO2 on Fig. 5. Depending on the –rst term in eqn. (7) it may then cut below the CO\0 line on Fig. 5 towards the left of the diagram but at a higher pressure than the CO\[1.5 line does. Until the calculations are done this is a matter for speculation. We certainly expect the O-terminated (CO\1.5) surface to be stable at oxygen pressures somewhat greater than 1 atm but we would not claim sufficient accuracy in our calculations which neglect the temperature dependence of the –rst terms in eqn.(7) to make a precise prediction. 5. Conclusions 1. We have presented a formalism for calculating the dependence of surface and interfacial energies in oxides on the partial pressure of oxygen and applied it to compare the energies of –ve postulated (1]1) (0001) surfaces of corundum diÜering in their surface stoichiometry. This is essentially a more explicit and somewhat extended version of the formalism used for example by Wang et al.,20 who expressed the surface energy in terms of the chemical potential of oxygen. Some subtleties associated with the de–nition of the zero of energy have been clari–ed here and by means of a thermodynamic cycle it has proved unnecessary to calculate any properties of pure oxygen which would have been problematic.2. We –nd that the observed neutral surface terminated by Al is stable up to atmospheric pressure. At some higher pressure perhaps some tens of atmospheres depending on the temperature the oxygen-terminated surface would be more stable (see Fig. 5). An Al-rich surface will probably be stable at sufficiently low oxygen pressure but the most Al-rich surface considered here (two terminating layers) is only theoretically stable at a pressure just below the very low pressure under which corundum would decompose. 3. The relaxed atomic and electronic structures of the surfaces have been obtained. There is qualitative agreement with experimental X-ray diÜraction results for the stable surface ; this Faraday Discuss. 1999 114 33»43 42 includes the sign and approximate magnitude of the –rst four interlayer relaxations and also the lateral displacements of the surface oxygen atoms which can be thought of as a rotation and expansion of the oxygen triangle beneath the surface Al.This is explained in terms of electrostatic forces. Ours and previous results both using the local density approximation (or generalised gradient approximation) and Hartree»Fock consistently overestimate the inward relaxation of the surface Al plane compared to the X-ray data. The discrepancy is unresolved. Regarding the second interlayer relaxation our result (10.6%) is closer to experiment (16%) than previous calculations some of which omitted lateral relaxations. We have veri–ed by further tests that if we do not allow lateral relaxation of oxygen the second interlayer relaxation is suppressed.However Verdozzi and co-workers5 using a Gaussian basis and Wang et al.20 using the full-potential Linear Augmented Plane Wave method have predicted smaller second interlayer relaxations than ours even though they included lateral relaxations ; this requires further investigation although the discrepancy is less than 0.1 ”. 4. Non-stoichiometric surfaces are metallic. In the case of Al-rich surfaces the Fermi energy lies in a band of surface states below and contiguous with the conduction band. Oxygen-rich surfaces are metallic because of holes at the top of the valence band which is mainly of O 2p character. Acknowledgements We thank J. Hué tter for technical help with the calculations.This work has been supported by the UK Engineering and Physical Sciences Research Council under grants GR/L08380 and GR/ M01753. The Centre for Supercomputing in Ireland is gratefully acknowledged for computer resources. This work has bene–ted from collaborations within and has been partially funded by the Training and Mobility Network on ììElectronic Structure Calculation of Materials Properties and Processes for Industry and Basic Sciencesœœ (Contract FMRX-CT98-0178). References 1 G. Renaud Surf. Sci. Rep. 1998 32 1. 2 P. Gueç nard G. Renaud A. Barbier and M. Gautier-Soyer Surf. Rev. L ett. 1998 5 321. 3 I. Manassidis A. DeVita and M. J. Gillan Surf. Sci. L ett. 1993 285 L517. 4 C. Kruse M. W. Finnis V. Y. Milman M. C. Payne A.DeVita and M. J. Gillan J. Am. Ceram. Soc. 1994 77 431. 5 C. Verdozzi D. R. Jennison P. A. Schultz and M. P. Sears Phys. Rev. L ett. 1999 82 799. 6 V. E. Puchin J. D. Gale A. L. Shluger. A. A. Kotomin J. Gunster M. Brause and V. Kempter Surf. Sci. 1997 370 190. 7 M. Causa` R. Dovesi C. Pisani and C. Roetti Surf. Sci. 1987 215 259. 8 C. Pisani M. Causa` R. Dovesi and C. Pisani Prog. Surf. Sci. 1987 25 119. 9 T. J. Godin and J. P. LaFemina Phys. Rev. B Condens. Matter 1994 49 7691. 10 W. C. Mackrodt R. J. Davey S. W. Black and R. Docherty J. Cryst. Growth 1987 80 441. 11 G. Renaud B. Vilette I. Vilfan and A. Bourret Phys. Rev. L ett. 1994 73 1825. 12 A. Alavi J.KohanoÜ M. Parrinello and D. Frenkel Phys. Rev. L ett. 1994 73 2599. 13 A. Alavi Philos. T rans. R. Soc. L ondon Ser. A 1998 356 263. 14 N. Troullier and J.-L. Martins Phys. Rev. B Condens. Matter 1991 43 1993. 15 I. G. Batyrev A. Alavi M. W. Finnis and T. Deutsch Phys. Rev. L ett. 1999 82 1510. 16 I. Mayer Chem. Phys. L ett. 1983 97 270. 17 J. W. Cahn in Interfacial Segregation ed. W. C. Johnson and J. M. Blakely American Society for Metals Metals Park OH 1977 pp. 3»23. 18 M. W. Finnis Phys. Status Solidi A 1998 166 397. 19 M. Gautier G. Renaud L. P. Van B. Villette M. Pollak N. Thromat F. Jollet and J.-P. Duraud Sci. Alumina 1993 77 323. 20 X. G. Wang W. Weiss S. K. Shaikhutdinov M. Ritter M. Petersen F. Wagner R. Schloé gl and M. Scheffler Phys. Rev. L ett. 1998 81 1038. Paper 9/03278I 43 Faraday Discuss. 1999 114 33»43

ISSN:1359-6640    

DOI:10.1039/a903278i

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Ultrathin alumina �lm Al-sublattice structure metal island nucleation at terrace point defects and how hydroxylation a� ects wetting D. R. Jennison* and A. Bogicevic Surface and Interface Sciences Department 1114 Sandia National L aboratories Albuquerque NM 87185-1421 USA. E-mail drjenni=sandia.gov Received 7th July 1999 First principles density functional slab calculations have produced the following results (1) for 5 ” (two O-layer) alumina –lms on Al(111) and Ru(0001) with larger unit cells than in recent work the lowest energy stable –lm was found to have an even mix of tetrahedral (t) and octahedral (o) Al ions arranged in alternating zig-zag rows. This most closely resembles the j-phase of bulk alumina where this pattern results in a greater average lateral separation of Al-ions than with pure t or o.A second structure with an even mix was also found consisting of alternating stripes. These patterns can exist in any of three equivalent directions on close packed substrates. (2) Because of numerical problems associated with the very large relaxations in alumina surfaces MgO(100) was used to investigate metal island nucleation. Common point defects (vacancies pairs of vacancies and water by-products) were placed in supercells and their eÜects on Pt adatom and ad-dimer binding computed. Unexpectedly single vacancies were found to destabilize metal dimers and only the mixed (Fs-Vs) divacancy increases stability. Among the water-by-products in-surface OH (produced by H` reaction with O2~) was uninteresting but ad-OH was found to both increase adatom binding and signi–cantly stabilize dimers on the surface suggesting the latter defect nucleates metal islands even at elevated temperatures.We believe these results apply to all highly ionic oxides. (3) Finally the eÜect of a substantial coverage of hydroxy on Cu deposition and growth on a-Al2O3(0001) was investigated. While Born»Haber calculations show wetting is not thermodynamically preferred on the clean surface at experimentally relevant ad-OH coverages the strength of the Cu»oxide bond is more than doubled and wetting is strongly favored. This causes a very stable B1/3 monolayer (ML) coverage of Cu`1 which then induces layer-by-layer Cu growth in agreement with recent experiments. Hydroxy coverage can thus control deposited metal morphology across a wide spectrum.1 Introduction In this paper we include for discussion three topics of current interest in metal oxide surface science. Using –rst principles density functional theory (DFT)1 calculations we have investigated (1) the atomic-scale structure of experimentally relevant ultrathin alumina –lms (2) the role of common point defects in metal island nucleation on oxide terraces and (3) the growth and morphology of metals on oxide surfaces that have high concentrations of a common impurity. 45 Faraday Discuss. 1999 114 45»52 This journal is( The Royal Society of Chemistry 2000 1.1 Ultrathin alumina –lm structure Aluminium oxide –lms have a substantial focus for a variety of reasons. First they represent structures that can be produced during the oxidation of Al metal and are thus important for understanding the ììbarrier layer œœ which inhibits corrosion.Second thin –lms enable the study of adsorption without the charging and hydrogen impurity problems which plague bulk-terminated Al2O3 . Third since alumina is an important support material when metal nanoclusters/crystals are produced by metal deposition the –lms serve as model catalysts.2 Finally there is growing interest for microelectronics applications. Considerable experimental and theoretical work on this system has occurred recently with one main topic being metal adsorption and the islands which result.3 Obviously nucleation plays a critical role here (see below) but without knowledge of the atomic scale structure one cannot address the –rst issue the basic energetics of adsorption diÜusion and dimer stability.Two O-layer 5 ” –lms are of particular interest because this thickness appears to be self-limited when produced by NiAl(110)4 or Ni Al(111)5 oxidation and more recently by the deposition and 3 subsequent oxidation of Al on Ru(0001).6 (Two-layer oxide –lms have also been recently seen in a completely diÜerent system encapsulated Pt nanocrystals on TiO2 .7) 2O3/NiAl(110) High-resolution electron energy-loss spectroscopy (HREELS) evidence from Al suggests a mixture of octahedral (o) and tetrahedral (t) Al ions is present,8 which has resulted in the –lms being called ììgamma-likeœœ. Recently transmission electron microscopy (TEM) moireç patterns have indeed indicated that the lattice constant of the –lm is consistent with a c-phase,9 but this result is also consistent with other possible structures.(Of course –lm thinness at two O-layers prevents a de–nable hcp vs. fcc stacking which diÜerentiates the a- and c-phases.) Actual structural details are in fact unknown but a signi–cant hint has arisen from new experiments by the group of Behm et al.6 on Al deposition and subsequent oxidation on Ru(0001). While islands of 5 ” thick Al2O3 were seen at various coverages using scanning tunneling microscopy (STM) low-energy electron diÜraction (LEED) evidence for ordering in the Al-sub-lattice was either not or only weakly seen in spite of annealing well above 300 K. On the theoretical side recent calculations on two and three O-layer Al2O3 –lms on Al(111) Mo(110) and Ru(0001) produced three signi–cant –ndings:10 (1) the preferred interface between the oxide –lm and the substrate metal consists in all cases of 1]1 chemisorbed oxygen (2) on top of which is a nearly coplanar layer of Al and O ions (3) with the normal bulk preference for octahedral-(o) over tetrahedral (t)-site aluminium ions energetically reversed.The last result is due to the electrostatics and layer separations induced by the interface with the underlying metal. Howeve this initial work used only small unit cells and did not allow for the possibility of greater complexity. Here we report computational results obtained by expanding the unit cell to allow a variety of t/o ratios and structures.1.2 Metal island nucleation on terraces Surface topographs from STM or atom force microscopy (AFM) observe metal island nucleation on oxide surfaces not only at line defects (such as antiphase domain boundaries4,11 and steps,12 where metal atoms presumably bind more strongly and therefore have a tendency to collect and meet) but also on terraces. For the latter nucleation could of course occur on a perfect surface depending on the temperature and density of the adatom lattice gas (which determine the stability of metal ad-dimers the probability of attaching a third metal atom etc.). In contrast surface defects might dominate nucleation in experimental conditions.2,13h16 Even though it has been speculated that the most common defect in well prepared surfaces speci–cally isolated surface oxygen vacancies,17 may act as a nucleation site,13,14 this has not been substantiated via experiment or theory.Here we report an investigation of the in—uence of surface vacancies on Pt island nucleation.18 For completeness we also examined how water dissociation products aÜect nucleation since there have been several reports that these are common low density contaminants on prepared oxide surfaces.17,19 It is very difficult numerically to study these defects in alumina –lms or with sapphire because of the extremely large surface relaxations.20,21 Therefore we have chosen a system with an order-ofmagnitude smaller relaxations,21 MgO(100). From this –rst study of dimer stability at oxide Faraday Discuss. 1999 114 45»52 46 surface defects several –ndings are completely unexpected but are really quite intuitive in retrospect and are likely to be general for highly ionic os.1.3 How hydroxylation aÜects wetting reported layer-by-layer growth23 and Cu(I) formation at cover- 2O3(0001) Cu deposition on oxides has assumed increased recent importance because of microelectronics applications. However experimental results22h27 for Cu on alumina have been inconsistent. X-Ray photoelectron spectroscopy (XPS) studies22 of Cu deposited by thermal evaporation onto bulk truncated a-Al2O3(0001) indicated ordered layer-by-layer growth for the –rst 2»3 atomic layers. The initial Cu ad-layer was observed as oxidized Cu in the form of Cu(I) ions. Other studies on polycrystalline Al2O3 ages below 0.5 monolayers.24 In contrast a study on epitaxial B20 ” Al2O3 –lms formed on refractory metal substrates25 reported the growth of 3-D clusters of metallic Cu even at submonolayer Cu coverages.In particular XPS and low energy ion scattering (LEIS)25 indicated Cu cluster formation at the lowest observable coverages at both 300 K and 80 K with no Cu(I) observed. In addition X-ray absorption near-edge structure spectroscopy (XANES)26 measurements carried out on sapphire substrates have reported no evidence of Cu oxidation and coverage-dependent shifts in Cu core level and LMM peaks have been interpreted in terms of –nal state screening,27 rather than ionization of the Cu. Meanwhile recent ion scattering experiments by Ahn and Rabalais19 have shown that cut and polished sapphire(0001) surfaces (the basil plane is not a cleavage surface) cannot be made free of hydrogen contamination in the form of hydroxy even by annealing to 1400 K.Finally experimental studies of Rh deposited on ultrathin epitaxial Al –lms28 suggest that surface hydroxy binds the Rh to the surface as a cation and serve as 2O3 nucleation sites for Rh clusters. These studies have raised the issue of the role of surface hydroxy groups in producing the apparent disagreements summarized above. The most recent experimental work by Kelber et al.,29 indicates initial layer-by-layer growth of Cu on hydroxylated a-Al at 300 K and the exclusive presence of Cu(I) during the formation of the –rst layer. Analysis of X-ray excited Cu(LMM) Auger data indicates that changes in the spectra are due to changes in the initial electronic state of the copper rather than to –nal state screening eÜects.The degree of surface hydroxylation is estimated to be high B1/3 ML (here 1 ML means one adsorbate per surface O ion) on the basis of O 1s XPS consistent with ref. 19. and on a-Al2O3(0001) with 1/3 ML of ad-OH. In collaboration with Kelber et al.,29 we have computed the adsorption energy of Cu at 1/3 and 1 ML coverages both on clean a-Al2O3(0001) Born»Haber cycles were then used to study the relative energy of isolated adatoms vs. incorporation into 2-D islands. We –nd dramatic eÜects on the binding energies due to hydroxylation and also on the growth mode suggesting that hydroxylation may explain the discrepancies in the experimental record.Following a description of the computational details we present our results and raise some issues for future work and for discussion. 2 Computational method Our electronic structure calculations were performed using the Vienna ab initio simulation package (VASP).30 This plane-wave based density functional code uses the ultra-soft pseudopotentials of Vanderbilt,31 which have good convergence for these systems with a plane wave cut-oÜ of only B270 eV. We used either the ììstandardœœ local density approximation (LDA)32 or the PW91 generalized gradient approximation (GGA),33 as indicated below. The geometric relaxation was done –rst with a quasi-Newton algorithm using computed interatomic forces. For the alumina systems where relaxation is large and problematic because of a mix of very hard and soft modes geometry was re–ned with a damped dynamics scheme built into VASP.The vacuum gap was in all cases ” [18 and k-point sampling was tested to ensure convergence to the quoted level of accuracy. 2.1 Ultrathin alumina –lm structure Our slabs had the alumina –lm on 4»7 layers of Al(111) or Ru(0001). The x»y dimensions of the supercell depended on the structure being studied. Because LDA has shown excellent accuracy in 47 Faraday Discuss. 1999 114 45»52 alumina structural predictions20 and GGA does not improve on same,34 LDA was used here. Since our study necessitated numerous computations using large supercells the following tests were performed to ensure accuracy (1) the relative energies of 2]1 supercells (Fig.1 top) with -o-o- -t-o- and -t-t- zig-zag Al rows were computed for a –lm with seven layers of Ru substrate (bottom four frozen at bulk LDA spacings) and using eight k-points. Errors produced in relative energy by reducing the Ru slab to just four layers (bottom two frozen) and/or the number of k-points to two were found to be \0.1 eV out of energy diÜerences of B2 eV per 2]1 cell. (2) Because numerical noise (arising from small inaccuracies in force computation) is seen to grow signi–cantly during prolonged geometric relaxation tests were done to examine the eÜect of freezing the entire metal substrate (these systems mix strong and soft vibrational modes and all –rstguesses had ions at the ideal positions with respect to the extended metal lattice).It was again found that errors were small here below 0.05 eV per 2]1 cell. These results indicate that the relative energies are determined almost entirely within the oxide –lm itself which because the bands are relatively —at can be described adequately by few k-points. 2.2 Metal island nucleation on terraces A supercell of –ve MgO(100) layers with 36 atoms each was used together with GGA.18 Two types of OH impurity were studied as a ììneutralœœ species it is produced by adding OH or H to the supercell while as a ììchargedœœ species it occurs when both OH and H are added to the supercell which naturally charge separate into ad-OH~ and H` the latter reacting with a surface O2~ to produce in-surface OH~. The latter was found to reduce the binding of ad-Pt atoms (due to the reduction in charge compared with the perfect surface) and are thus uninteresting for nucleation except as a means to concentrate the density of Pt adatoms in defect free regions ; we do not discuss them further.Except at the highest coverages because of its large electron affinity ad-OH would exist as a negative ion rather than as a radical. 2.3 How hydroxylation aÜects wetting The sapphire slab had nine O-layers of one Al2O3 unit each for 45 atoms per unit cell as was used in a previous study.20 In order to compare with previously published metal adsorption energies,20,21 LDA was used for these calculations. The initial Cu positions were at the most favored sites which at 1/3 ML coverage are atop the deepest lying Al-ion and at 1 ML coverage Fig.1 The preferred structure of the thin –lm has alternating rows of tetrahedral and octahedral site Al ions in a zig-zag pattern (top) rather than the striped pattern (bottom). The large spheres are O ions the small spheres Al ions in either octahedral (black) or tetrahedral (gray) sites. Faraday Discuss. 1999 114 45»52 48 are atop O. The ad-OH was placed initially at the most favored site also atop the shallowest Al ion. Relaxations were small laterally thus preserving these site descriptions for the relaxed geometry. t/o ratio (%) type 100 pure t 50 stripe 50 zig-zag 0 pure o 3 Results and discussion 3.1 Ultrathin alumina –lm structure In order to consider only the energetically lowest-lying possibilities we impose three constraints (1) we do not allow for non-stoichiometry in the Al/O ratio ; (2) we restrict coordination to what is normal i.e.each surface O has two nearly coplanar10 Al nearest neighbors; and (3) we do not consider geometries where t and o ions are in sites which are immediately adjacent. Our analysis indicates it is not possible to produce a localized o-containing ìì defect œœ starting with all t ions or vice versa without violating the above restrictions. However it is possible to produce a zig-zag row of O ions embedded in an otherwise perfect –lm of 100% t ions by displacing a row laterally (in the vertical direction in Fig. 1) so as to move all the ions in that row from t to a neighboring o site. Note that the eÜect of such a movement is to increase locally the average Al»Al interatomic spacing.Thus this electrostatic advantage maximized by an alternating mixture of t and o rows (i.e. -t-o-t-o-) competes with the t site preference reported in ref. 10. Table 1 shows the relative energies of the pure o and t structures compared with the even mix zig-zag structure on both Al(111) and Ru(0001) substrates. We –nd for both systems that the lateral electrostatic advantage of alternating o and t dominates over the t-site advantage. A second type of -t-o- evenly mixed structure has also been found consisting of alternating stripes (see Fig. 1 bottom). It is also noted that this structure can mix with the zig-zag one producing a displacement in the zig-zag axis. However we –nd for Ru(0001) that the striped structure is signi–cantly higher in energy than the zig-zag even though the Al»Al nearest neighbor spacings are the same to several neighbor shells.It is obvious that it is also possible to have any mix of the two types of rows. For example the even zig-zag mixture with -t-o- alternating rows has a 2]1 unit cell (Fig. 1 top) relative to the primitive cell of three O and two Al ions in the surface plane of the –lm. Additionally the 3 1 ratio of ion types then has a 4]1 cell (e.g. -t-t-t-o-) but the 2 1 ratio results in a 6]1 cell because of the reversal of the phase of the zig-zag after a single -t-t-o- sequence. Thus far we have been unable to converge these other –lms geometrically as —at structures. Instead computationally the energy lowers monotonically as 3-D stripes (three oxygen layers thick) are produced separated by depleted regions.In other words the additional degrees of freedom allowed by the larger unit cells permits the geometry relaxation algorithm to –nd structures which while overall energetically preferred are perhaps not relevant to the (actually metastable) —at structures produced experimentally. Work here is continuing. If it were not for the eÜects of relaxation the relative energies of various mixes could be modeled easily. If short range row»row interactions were to dominate the cost of converting an o row into a t row or a domain of the perfect 2]1 structure -o-t-o-t-o-t- into one with reversed phase on the right -o-t-t-o-t-o- could be estimated as follows (1) a reduction in energy of B0.2 eV per primitive cell for each extra t vs.o row (see Table 1 and Ref. 10; and (2) a penalty of B1.0 eV Table 1 Relative energies in eV (per Al unit) vs. the Al ion tetrahedral/octahedral site ratio for 5 ” 2O3 alumina –lms on Ru and Al substrates (the striped structure on Al(111) has not yet been studied) Unit cell Ru(0001) Al(111) 1]1 2]1 3]1 1]1 1.2 1.8 1.0 1.6 0.0 0.0 0.5 » 49 Faraday Discuss. 1999 114 45»52 Fig. 2 Pt adsorption Pt binding energy and relative stability of water products on MgO(100) and in the presence of surface defects. The open circles under each line denote O ions the –lled circles Mg ions. Negative 2 energies indicate less stable adsorption with respect to (i) isolated gas phase H2O molecule (ii) isolated gas phase Pt atom and (iii) two isolated Pt adatoms for the three panels respectively.per primitive cell (for Ru(0001) substrates Table 1) for each t-t nearest row»row interaction instead of t-o. The net cost of phase reversal or other deviations from the 2]1 structure would be lowered by relaxation and further-than-nearest row»row interactions. These make it possible that the actual energetic cost may be sufficiently low that real –lms where local structure is also in—uenced by defects and –lm growth conditions may display a loss of long-range order in the Al-sublattice by this phase-reversal mechanism. Another possibility for such a loss is presented by the striped structure in that a small inclusion of stripe between two zig-zag portions results in the zig-zag shifting laterally.Here too the large relaxations inherent in alumina would reduce the cost of such a defect below that estimated on the basis of the perfect structures (Table 1). Without calculations with very large relaxed supercells it is difficult to predict the actual energetics. 2O335 (recently structurally determined entirely by DFT and in It seems likely that the 2]1 zig-zag dominates in real –lms (as the lateral Al»Al repulsion is minimized). Domains would then be determined by surface features such as linear defects and by –lm nucleation and growth. While a loss of long range Al-sublattice order would cause an amorphous appearance in scattering experiments such as LEED the –lm is still locally dense and its adsorption properties little aÜected since the surface sites are so similar locally.Even though our calculations were done using the experimentally relevant substrate of Ru(0001) since the calculations on Al(111) show qualitatively similar results we suggest these likely apply also to –lms grown on NiAl4 and Ni Al,5 as no Ni rises into the –lm and the –lm/ 3 substrate interface which drives the –lm structure is thus similar to having chemisorbed 1]1 oxygen on Al(111). Of all known aluminium oxide bulk phases the 2]1 zig-zag (Fig. 1 top) most closely resembles the so-called A plane of j-Al close agreement with X-ray scattering from chemical vapour deposition (CVD)-grown samples). This phase has -A-B-A-C- bulk stacking of the close-packed near-hexagonal O-layers and the A plane has an even mixture of o and t arranged in the alternating zig-zag rows (Fig.1 top) while the B and C planes have all o ions. ” 35 If the 5 –lms indeed prefer the A plane structure one may speculate that this would nucleate j-like alumina –lms if grown thicker. Faraday Discuss. 1999 114 45»52 50 Table 2 The LDA adsorption energy of Cu on a per atom basis (in eV) on clean sapphire(0001) and on hydroxylated sapphire (1/3 ML of ad-OH). The Born»Haber energy *E is positive when wetting occurs Cu coverage 1 ML 1/3 ML *E 0.5 1.1 1.3 1.8 » 5.2 [4.5 ]3.8 ]3.1 Sapphire Sapphire]OH Sapphire]O]Ha a With 1 ML of Cu it was also found exothermic for Cu to dissociate the OH as shown. 3.2 Metal island nucleation on terraces In Fig.2 we –nd the results of our study. Because water contamination is an issue we initially studied water adsorption and dissociation –nding it will not dissociate on the perfect surface15,36 but when it does dissociate (either due to defects or to solvation36) it does so as separated ions (Fig. 2 left column»favored structures are always towards the top of Fig. 2). Next we see the eÜects of vacancies and water products on Pt adatom binding noting a general increase in binding due to the presence of hydroxy (Fig. 2 center column). Finally we note the eÜects of defects on Pt addimer stability (Fig. 2 right column). Of the vacancies only the mixed divacancy increases the dimer binding while both ad-OH species increase the same signi–cantly.Note at moderate ad-OH concentrations one might expect the charged species to be present which has the greatest eÜect on dimer stability. These results may be simply understood. When a Pt adatom encounters a vacancy of either type it is drawn to it and becomes ionized by the Madelung potential. It enters the vacancy to the extent allowed by its ionic radius. Because it is an ion its ability to bind to a second Pt atom is largely destroyed (and is repulsive in the LDA approximation). These results should hold for ionic oxides all having substantial Madelung potentials. With ad-OH~ however a Pt adatom and ad-dimer –nd themselves at a ììmini-stepœœ with attractive electrostatic interactions both vertically (to the underlying O2-ions) and laterally (to the OH~).3.3 How hydroxylation aÜects wetting In Table 2 we see that on perfect sapphire(0001) Cu binds less than half as strongly as on the hydroxylated surface. In particular the increase in adatom binding (represented by 1/3 ML) not only reverses the preference for wetting given by the Born»Haber cycle but also pins the Cu adatoms so they are immobile up to very high tempeatures ([1000 K) and are unable to reach and join 3-D islands presumably nucleated at defects such as steps. This pinning is caused by the large loss of binding energy that would occur were a Cu atom to move laterally away from the adjacent OH which it sits next to in the relaxed geometry. In the Born»Haber calculation the tendency to form 2-D islands vs. separated adatoms is given by *E\E(1 ML)]2E(slab)[3E(1/ 3 ML) where the last three parts refer to the total energy of the slab with 1 ML of Cu the slab alone and the slab with 1/3 ML of Cu respectively.We noted during our study that at 1 ML coverage it was exothermic for OH to dissociate placing the H well away from the remaining ad-O which coordinates locally to two Cu atoms. This observation however does not signi–- cantly aÜect the preference for wetting (Table 2). It thus appears possible not only to increase the dispersion of ad-metal particles by hydroxylation, 2,28 but also to alter the growth mode completely if sufficient OH density is present.29 It is of course possible that this surfactant eÜect extends to other impurities as well. Acknowledgements We thank JeÜ Kelber for extensive discussions on hydroxylated sapphire and for sharing unpublished results.We also thank R. Jué rgen Behm for sharing unpublished work and stimulating 51 Faraday Discuss. 1999 114 45»52 comments and Hans-Joachim Freund for valuable discussions. VASP was developed at the Institut fué r Theoretische Physik of the Technische Universitaé t Wien. Sandia is a multiprogram laboratory operated by Sandia Corporation a Lockheed Martin Company for the United States Department of Energy under Contract DE-AC04-94AL85000. This work was partially supported by a Laboratory Directed Research and Development project. Paper 9/05456A References 1 P. Hohenberg and W. Kohn Phys. Rev. 1964 136 B864; W. Kohn and L. J. Sham Phys. Rev. 1965 140 A1133. 2 G.Ertl and H.-J. Freund Phys. T oday 1999 52 32 and references cited therein. 3 H.-J. Freund H. Kuhlenbeck and V. Staemmler Rep. Prog. Phys. 1996 59 283. 4 R. M. Jaeger H. Kuhlenbeck H.-J. Freund M. Wuttig W. HoÜmann R. Franchy and H. Ibach Surf. Sci. 1991 259 235. 5 C. Becker J. Kandler H. Raaf R. Linke T. Pelster M. Drager M. Tanemura and K. Wandelt J. V ac. Sci. T echnol. A 1998 16 1000. 6 R. J. Behm unpublished work. 7 O. Dulub W. Hebenstreit and U. Diebold unpublished work. 8 J. Libuda F. Winkelmann M. Baumer H.-J. Freund T. Bertrams H. Neddermeyer and K. Muller Surf. Sci. 1994 318 61. 9 S. Nepijko M. Klimenkov H. Kuhlenbeck R. Schloé gl and H.-J. Freund unpublished work. 10 D. R. Jennison C. Verdozzi P. A. Schultz and M. P. Sears Phys. Rev. B Condens.Matter 1999 59 R15605. 11 F. Winkelmann S. Wohlrab J. Libuda M. Baumer D. Cappus M. Menges K. Al-Shamery H. Kuhlenbeck and H.-J. Freund Surf. Sci. 1994 307ñ309 1148. 12 D. A. Chen M. C. Bartelt R. Q. Hwang and K. F. McCarty Surf. Sci. submitted. 13 P. A. Thiel and T. E. Madey Surf. Sci. Rep. 1987 7 211. 14 G. Haas A. Mench H. Brune J. V. Barth J. A. Venables and K. Kern unpublished work. 15 M. J. Stirniman C. Huang R. S. Smith S. A. Joyce and B. D. Kay J. Chem. Phys. 1996 105 1295. 16 J. Gué nster J. Stultz S. Krischok D. W. Goodman P. Stracke and V. Kempter J. V ac. Sci. T echnol. 1999 A17 1657. 17 U. Diebold J. Lehman T. Mahmoud M. Kuhn G. Leonardelli W. Hebenstreit M. Schmid and P. Varga Surf. Sci. 1998 411 137. 18 A. Bogicevic and D. R.Jennison Surf. Sci. 1999 437 4741. 19 J. Ahn and J. W. Rabalais Surf. Sci. 1997 388 121. 20 C. Verdozzi D. R. Jennison P. A. Schultz and M. P. Sears Phys. Rev. L ett. 1999 82 799 and references cited therein. 21 A. Bogicevic and D. R. Jennison Phys. Rev. L ett. 1999 82 4050. 22 S. Varma G. S. Chottiner and M. Arbab J. V ac. Sci. T echnol. 1992 A10 2857. 23 J. G. Chen M. L. Colaianni W. H. Weinberg and J. T. Yates Jr. Surf. Sci. 1992 279 223. 24 F. S. Ohuchi R. H. French and R. V. Kasowski J. Appl. Phys. 1987 62 2286. 25 Y. Wu E. Garfunkel and T. E. Madey J. V ac. Sci. T echnol. 1996 A14 1662. 26 S. Gota M. Gautier L. Douillard N. Thromat J. P. Duraud and P. Le Fevre Surf. Sci. 1995 323 163; see also M. Gautier L. Pham Van and J. P. Duraud Europhys. L ett.1992 18 175. 27 M. Gautier J. P. Duraud and L. Pham Van Surf. Sci. L ett. 1991 249 L327. 28 J. Libuda M. Frank A. Sandell S. Andersson P. A. Bruhwiler M. Baumer N. Martensson and H.-J. Freund Surf. Sci. 1997 384 106. 29 J. A. Kelber C. Niu K. Shepherd D. R. Jennison and A. Bogicevic Surf. Sci. submitted. 30 G. Kresse and J. Hafner Phys. Rev. B Condens. Matter 1993 47 558; 1994 49 14251; 1996 54 11169. 31 D. Vanderbilt Phys. Rev. B Condens. Matter 1985 32 8412; 1990 41 7892. 32 J. P. Perdew and A. Zunger Phys. Rev. B Condens. Matter 1981 23 5048; D. M. Ceperley and B. J. Alder Phys. Rev. L ett. 1980 45 566. 33 J. P. Perdew J. A. Chevary S. H. Vosko K. A. Jackson M. R. Pederson D. J. Singh and C. Fiolhais Phys. Rev. B Condens. Matter 1992 46 6671. 34 Y. Yourdshahyan U. Engberg L. Bengtsson B. I. Lundqvist and B. Hammer Phys. Rev. B Condens. Matter 1997 55 8721. 35 Y. Yourdshahyan C. Ruberto M. Halvarsson L. Bengtsson V. Langer B. I. Lundqvist S. Ruppi and U. Rolander J Am. Ceram. Soc. 1999 82 1365; B. Holm R. Ahuja Y. Yourdshahyan B. Johansson and B. I. Lundqvist Phys. Rev. B Condens. Matter 1999 59 12777. 36 See for example M. A. Johnson E. V. Stefanovich T. N. Truong J. Gunster and D. W. Goodman J. Phys. Chem. 1999 103 3391 and references cited therein. Faraday Discuss. 1999 114 45»52 52

ISSN:1359-6640    

DOI:10.1039/a905456a

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Electronic properties structure and adsorption at vanadium oxide density functional theory studies K. Hermann,*a M. Witkob and R. Druzinica a Fritz-Haber-Institut der MPG Faradayweg 4-6 D-14195 Berlin Germany b Institute of Catalysis and Surface Chemistry ul. Niezapominajek 30239 Cracow Poland 1 Introduction Transition metal oxides are well known for their enormous variety of physical and chemical properties.1h3 Many of these materials undergo phase transitions with interesting structural electronic and magnetic behavior.3 Some exhibit high temperature superconductivity and exciting optical properties or high catalytic activity.2 Among these vanadium oxides represent an important class of materials which are widely studied and used in many technological applications.4,5 In particular vanadium pentoxide V2O5 or vanadia-based compounds are used as components in various catalysts for mild oxidation ammoxidation and dehydrogenation of hydrocarbons and other organic compounds.Further they are efficient in oxidation of SO to SO and for the 2 NH3 .6,7 Despite the enormous importance of V removal of NO by selective reduction with 3 x 2O5 as a catalyst many microscopic details of its catalytic behavior are still under debate,4 which makes a detailed study of V2O5 surface properties particularly attractive. The crystal structure of vanadium pentoxide V2O5 is rather complex and can be described in diÜerent ways. The orthorhombic crystal has a layer structure in which each physical layer consists of VO sub-units linked by edges and by corners8h10 with weak inter-layer coupling.The 5 basal (010) and other non-basal planes diÜer in their bond type and in the degree of coordinative saturation of vanadium and oxygen atoms. This results in rather diÜerent behavior with respect to adsorption and catalytically supported reactions. 53 Faraday Discuss. 1999 114 53»66 Received 19th April 1999 The local electronic structure at the V surface is studied by ab initio density 2O5(010) functional theory (DFT) methods where embedded clusters as large as V20O62H24 representing one or two physical layers of the substrate are used as models. Results of local binding charging and densities of states help to characterize the detailed electronic structure of the surface. In addition electronic and geometric details of surface oxygen vacancies are studied by V2O5(010) surface cluster calculations where oxygen atoms are removed from speci–c surface sites.A comparison of the data concerning vacancy energies charging and geometric relaxation shows pronounced variations between diÜerent oxygen sites which gives further insight into possible mechanisms of surface relaxation and reconstruction. Further cluster calculations of hydrogen adsorption at structurally diÜerent surface oxygen sites (leading to surface OH and H2O) are performed. A comparison of bond strengths of surface OH and H2O with that of surface oxygen gives valuable information as to which oxygen sites are involved in speci–c adsorption desorption and reaction steps. This journal is( The Royal Society of Chemistry 2000 It is generally accepted that reactions of selective hydrocarbon oxidation at the V2O5(010) surface proceed according to a nucleophilic mechanism6,7,11h13 where an important reaction step involves adsorption and binding of hydrogen at the V2O5 surface.In a possible scenario hydrogen (being abstracted from the organic molecule) adsorbs at an oxygen site forming a surface hydroxy OH species which can desorb. Alternatively the hydroxy group may combine with another hydrogen to form surface H2O which desorbs. These processes form oxygen vacancies at the surface which may migrate into the bulk with the equivalent number of metal cations being simultaneously reduced. Gaseous oxygen participates in the oxidation reaction only after adsorption in other parts of the catalyst followed by migration through the lattice to the active site.The key point for understanding these mechanisms is to identify the structurally diÜerent surface oxygen sites which take part in the reaction. This issue has been discussed rather controversially in the literature. Some authors assume that terminal vanadyl oxygen (O2V) is removed from the vanadia catalyst surface to form a lattice vacancy6 while others argue in favor of bridging oxygens (V»O»V or V»O»Me in the case of supported vanadia catalysts) and there are reports suggesting that a mixture of V2O and V»O»V(Me) type oxygens is essential for the selective oxidation process. Previous theoretical cluster studies using semi-empirical13h20 and ab initio methods (Hartree»Fock (HF)21h23 and density functional theory (DFT)24,25) have shown consistently that after hydrogen adsorption two- and three-fold coordinated bridging oxygens can be removed more easily from the V2O5(010) surface than vanadyl oxygens.This result is con–rmed by combined numerical (semi-empirical HF) and experimental (IR) studies26 on the importance of V2O5 surface oxygen for the oxidation of SO into SO3 which stress the preference of oxygen centers 2 with the highest V coordination. However the underlying mechanisms have not been veri–ed by experiments on a microscopic basis for an overview see ref. 6. In the present theoretical work we examine the local electronic structure at the V2O5(010) surface by ab initio DFT methods where embedded clusters representing one and two physical layers of the substrate are used as models.In addition electronic and geometric details of surface oxygen vacancies are studied by cluster calculations where oxygen atoms are removed from speci –c surface sites. Further calculations of hydrogen adsorption at structurally diÜerent surface oxygen sites (leading to surface OH and H2O as intermediate reaction products in the hydrocarbon oxidation) are performed using the embedded substrate clusters. In Section 2 we describe brie—y the computational details and Section 3 presents the results and a discussion. Finally we summarize our conclusions in Section 4. 2O5 ” c\3.564 ”; V forms a layer type orthorhombic lattice structure9,27,28 (lattice 4O10 unit cell) with physical layers extending with bulk structure as a starting point in the adsorbate and 2 Theoretical details Bulk vanadium pentoxide V constants a\11.519 ” b\4.373 parallel to the (010) netplane.The physical layers are characterized by periodic arrangements of edge and corner sharing VO pyramids sticking out at both sides of the layer see Fig. 1. There are 5 three structurally diÜerent layer oxygen atoms terminal (vanadyl) oxygen O(1) coordinated to one vanadium atom through a short bond (dVhO\1.58 ”) and bridging oxygen O(2)/O(3) coordinated to two or three vanadium atoms with V ” ” »O distances ranging between 1.78 and 2.01 . This gives rise to –ve diÜerent oxygen centers at the ideal V2O5(010) surface (see Fig. 1) terminal (vanadyl) oxygen O(1) located directly above vanadium centers oxygen O(2) O(2@) bridging two vanadyl groups pointing into the bulk and sticking out of the surface respectively and oxygen O(3) O(3@) connected to three vanadyl groups (one pointing up and two pointing down for O(3) two pointing up and one pointing down for O(3@)).The O(2@) and O(3@) centers (not labeled explicitly in Fig. 1) which are ììburiedœœ between vanadyl groups are of less chemical interest and will not be considered in the following. In the calculations the local environment at the V2O5(010) surface is modeled by clusters (accounting for one physical layer) and V V 20O62H24 (\2]V10O31H12 accounting 10O31H12 for two physical layers) shown in Fig. 2. All vanadium and oxygen positions are taken from the experimental bulk structure and peripheral oxygen atoms are bond saturated by hydrogen atoms.25 A full geometry optimization on V10O31H12 29 results in an equilibrium structure of the cluster that deviates only a little from that of the bulk termination (atom shifts by less than 0.18 ”).This justi–es the use of V10O31H12 Faraday Discuss. 1999 114 53»66 54 Fig. 1 Crystal structure of orthorhombic V2O5 with netplane stacking along (010). Vanadium (oxygen) centers are shown by large (small) balls. Inequivalent oxygen centers O(1) O(2) O(3) are labeled accordingly. Note that labels O(2) and O(3) point to two centers O(2) O(2@) and O(3) O(3@) respectively which are inequivalent at the (010) surface. oxygen vacancy calculations. Further comparative cluster studies30 on the single layer clusters V10O31H12 and V16O49H18 yield almost identical electronic parameters which indicates size convergence and shows that V10O31H12 can be considered a realistic representation of the extended V 2O5(010) surface.Oxygen vacancy formation at diÜerent sites O(1»3) of the V surface is considered in 10 and O30H12 (2V10O31H12»O) V20O61H24 (2V20O62H24»O) clusters. In a –rst 2O5(010) calculations on V step the respective oxygen is removed from the cluster and the electronic structure is evaluated keeping all the atom positions frozen at the bulk geometry. In addition all cluster atoms in V10O30H12 (or V20O61H24) except the terminating hydrogen atoms are allowed to rearrange according to the lowest cluster total energy. A comparison with the data of the frozen bulk geometry shows the importance of surface relaxation induced by the vacancy.Further a comparison of the V10O30H12 and V20O61H24 data can give information about the in—uence of vacancy formation on electronic inter-layer coupling. Hydrogen adsorption at diÜerent oxygen sites of the V V10O31H12]H clusters where H is approached near the O(1»3) 2O5(010) surface is modeled by sites forming a surface OH group and both the hydrogen and the adsorption site oxygen positions are optimized according to the lowest total energy of the cluster. In addition two H atoms Fig. 2 Geometric structure of the clusters V10O31H12 (a one layer) and V (b two layers). The V (O) atoms are shown as large (small) shaded balls while very small white balls refer to hydrogen atoms used to 20O62H24 saturate oxygen atoms at the cluster boundary.55 Faraday Discuss. 1999 114 53»66 species are approached near the O(1»3) sites in a V10O31H12]2H cluster forming a surface H2O which is geometrically optimized analogous to the OH optimization. The electronic structure of the clusters is determined by ab initio density functional theory (DFT) methods where the Kohn»Sham orbitals are represented by linear combinations of atomic orbitals (LCAOs) using extended all-electron basis sets of contracted Gaussians from atom optimizations. 31,32 For the calculations the program package DeMon33 is applied where electron exchange and correlation is described by the local spin density approximation (LSDA) based on the Vosko»Wilk»Nusair functional34 as well as by gradient corrected (GGA-II) functionals.35 In addition to the total energies and equilibrium geometries (based on numerical forces) detailed analyses of the electronic structure in the clusters are performed using Mulliken populations36 and Mayer bond order indices.37,38 Further the dense energetic distribution of the Kohn»Sham valence levels in the clusters allows the de–nition of a cluster total density of states (DOS) ntot(e) and atom projected partial densities of states (PDOS) nA(e) by (1) (2) ntot(e)\; g(a e[ek) k nA(e)\; qk(A)g(a e[ek) with ; nA(e)\ntot(e) A k where g(a e[ek) denotes a gaussian broadening function of width a centered at cluster level ek and the summation goes over all occupied cluster orbitals.Further q (A) gives the population of k atom A in cluster orbital r determined by a Mulliken analysis.The computed DOS and PDOS k functions can become useful in interpreting experimental photoemission spectra from the V2O5 system as will be discussed below. 2O 3 Results and discussion 3.1 Electronic structure of V2O5(010) 10O31H12 V20O62H24 is described by both ionic and sizeable covalent contributions. Table 1 lists the geometric and electronic parameters of (a) the V2O5(010) surface clusters V and of (b) the free (neutral) molecular species OH and H where all the values are calculated within the LSDA scheme. Atom charges q(A) are obtained from Mulliken population analyses and bond orders p(A»B) refer to the Mayer bond order indices where the data are given for the V and O atoms closest to the cluster center.In addition the valence energy width D of each cluster is included. An overall comparison of the calculated atom charges and bond orders reveals very close similarity between the one and two layer clusters which indicates that the electronic inter-layer coupling is rather weak and can be neglected for the electronic surface structure. In agreement with chemical intuition all the vanadium atoms are positively charged and all the oxygen atoms are negative in the clusters. Vanadium atoms are described by V`1.4» V`1.6 where the variation re—ects the location inside the cluster. Further the negative oxygen charges scale with coordination O~0.3 for singly coordinated terminal oxygen O(1) O~0.6 for doubly coordinated bridging oxygen O(2) and O~0.8 for triply coordinated bridging oxygen O(3).This indicates for the V2O5(010) surface that bridging oxygen sites are more nucleophilic than terminal vanadyl sites which becomes important in view of the reactivity of the diÜerent sites with respect to surface chemical reactions. Altogether local charging of the diÜerent cluster atoms is found to be much smaller than formal valence charges V`5 and O~2 would suggest. Obviously inter-atomic binding in V2O5 The bond order results of Table 1 give a rough estimate of the covalent contributions to the total V»O binding in the clusters. The data con–rm the general picture based on simple valence concepts. Bonds between terminal oxygen O(1) and vanadium yield bond order values close to 2 which suggests double bonds and is consistent with the single coordination of O(1).Bonds between bridging oxygen O(2) and each of their two vanadium neighbors result in bond order values close to 1 corresponding to two single bonds per oxygen which is again reasonable based on the coordination of O(2). Finally V»O bond orders involving bridging atoms O(3) coordinated to three vanadium neighbors each give meaningful values of 0.5»0.6 per bond. The electronic structure of the clusters in the valence region is determined by occupied Kohn» Sham valence orbitals which are mainly O 2sp type with some V 3d admixture. Their energy Faraday Discuss. 1999 114 53»66 56 10O31H12 and substrate clusters and (b) the free (neutral) 2O Table 1 Atom charges q from Mulliken analyses and Mayer bond orders p for (a) the V V molecular species OH H 20O62H24 obtained by LSDA calculationsa (a) V20O62H24 V10O31H12 1.58 1.78 1.88 (]2) 2.02 1.58 1.78 1.88 (]2) 2.02 dVhO(1)/” dVhO(2)/” dVhO(3)/” 1.44 1.57 [0.28 [0.63 [0.78 1.43 [0.29 [0.62 [0.79 q(V) q(O(1)) q(O(2)) q(O(3)) 2.16 0.89 0.51 2.13 0.89 0.55 p(O(1)»V) p(O(2)»V) p(O(3)»V) 5.76 5.40 D/eV OH (b) H2O 1.00 » 0.98 105.6 dOhH/” n(H»O»H)/° [0.44 0.44 [0.85 0.43 q(O) q(H) 0.78 0.80 p(O»H) a The two entries of q(V) for V20O62H24 correspond to vanadium of the –rst and second layer respectively.In addition the valence energy widths D of each cluster are given.For further de–nitions see text. range corresponds to valence energy widths D\5.40 eV and 5.76 eV respectively. The width D is expected to converge with increasing cluster size towards the total valence band width of the extended V2O5(010) surface system. Very recent FP-LAPW (full potential linearised augmented plane wave) band structure calculations39 yield D\5.35 eV for the V2O5 bulk and D\5.05 eV for V2O5(010) single layer slabs which are reasonably close to the cluster results suggesting size convergence for the present clusters as discussed before. The distribution of the valence levels and their atom character can be described by total DOS and atom projected PDOS curves following the procedure described in Section 2. Fig. 3 shows DOS and PDOS curves for the V10O31H12 cluster where the vanadium contributions as well as those from all diÜerently coordinated oxygen atoms O(1) O(2) O(3) are included and a gaussian level broadening of 0.4 eV (FWHM) is applied.The total DOS in the energy region between [13 and [7 eV shows a multi-peak structure described by mainly O 2sp derived electron states without noticeable energetic separation between O 2s and O 2p and by smaller V 3d contributions. (Note that due to the gaussian broadening the DOSs of Fig. 3 do not exhibit a sharp cut-oÜ at the HOMO energy; [7.07 eV marked by a thin line in the plot). The additional DOS peaks between [13 and [14 eV re—ect split-oÜ energy levels arising from bond saturation of peripheral cluster oxygen atoms by hydrogen terminating atoms.They have to be considered a consequence of the cluster approach and can be neglected for the present purpose. The PDOS due to vanadium given in Fig. 3 shows moderate variations with larger values near the central part of the valence region. However its size is overall smaller compared to that of the oxygen derived PDOSs. An integration over the valence region yields populations of 3.6 electrons per V atom in agreement with the atom charge (1.4) given in Table 1. As a result the V atoms in the clusters are not fully ionic. The PDOS referring to terminal oxygen O(1) are concentrated near the center of the valence region with smaller contributions above the center and they are described Faraday Discuss. 1999 114 53»66 57 Fig. 3 Total DOS and atom projected PDOS curves for the V10O31H12 a gaussian level broadening of 0.4 eV (FWHM) and the HOMO energy is marked by a thin vertical line.cluster see text. The results refer to 20O62H24 surface clusters and the V can be compared with those of the larger clusters V 10O31H12 16O49H18 30 and as well as with results from ab initio DFT band structure methods for both bulk and V2O5(010) single layer slabs.39h42 This comparison shows very good qualitative agree- 2O5 2O5 bulk/slab by an overall con–ned (D3 eV wide) distribution. In contrast the PDOSs of bridging oxygen O(2 3) yield a broad distribution covering the full energy range of the total DOS. Obviously the O(1) derived cluster levels show a dispersion width smaller than that of the bridging O(2,3) species partly because of the spatial distribution of the diÜerent oxygen atoms in the crystal and their eÜective inter-atomic binding as discussed elsewhere.30 The DOS and PDOS curves of Fig.3 obtained for V V V2O5 ment which con–rms that the diÜerent approaches V models yield basically the same electronic structure for the oxide material and can therefore be applied alternatively to model bulk and surface properties. Recent angular resolved UV photoemission experiments for the V surface30 yield a 2O5(010) spectrum with three peaks one dominant central and two smaller peripheral in the O 2sp valence region where the total valence energy width D amounts to 5.5 eV quite close to the cluster results. Further the variation of the experimental emission intensity with energy is similar to the shape of the calculated total DOS of the V2O5 surface clusters.This suggests that the origin of the peaks observed in the photoemission experiment may be identi–ed by a comparison with the calculated PDOSs. As a result the most prominent central peak in the experimental data is assigned to emission from mainly terminal oxygen O(1) while the two peripheral peaks at the top and bottom of the valence energy region are characterized as mixtures of vanadium with O(2) and O(3) induced intensity. 3.2 Oxygen vacancies at V2O5(010) Table 2 contains results from oxygen vacancy calculation using the V10O31H12 cluster which models one physical layer of the V E (O) are 2O5(010) surface. The oxygen vacancy energies D de–ned by total energy diÜerences of the corresponding clusters (3) tot(O) refers to oxygen in its neutral ground state.In the –rst step all the atoms of V10O30H12 E (O)\oE E tot(V10O31H12)[Etot(V10O30H12)[Etot(O) o D where in V10O30H12 the oxygen has been removed from one of the sites O(1) O(2) O(3) and E are kept –xed at their positions from the ideal surface yielding frozen vacancy energies D f (O). In the second step all the cluster atoms of V10O30H12 except the terminating hydrogen atoms are allowed to rearrange according to the lowest cluster total energy which leads to relaxed values ED r (O). The energetic consequence of relaxation at each oxygen site can be described by a relax- Faraday Discuss. 1999 114 53»66 58 cluster 10O31H12 Table 2 Oxygen vacancy energies E(f r)(O) with and without surface relaxation at the oxygen sites O(1»3) obtained for the V D using the LSDA schemea E Dr(V) qr(V) qf(V) Erel ED f (O) D r (O) (0.18) 0.50 0.70 0.08 (1.44) 1.45 1.27 1.21 (1.41) 1.32 1.15 1.17 (0.93) 1.45 2.17 1.26 » 6.68 7.19 7.26 » 8.13 9.36 8.52 Substrate O(1) vacancy O(2) vacancy O(3) vacancy a In addition the table contains values for the relaxation energies E as rel well as for the charges q(f r)(V) and displacements *r(V) of the central V atom closest to the vacancy.The data of the top row of the table refer to the V10O31H12 cluster without a vacancy. For de–nitions see text. All energies are given in eV and all lengths are in ”.ation energy (4) Erel\ED f (O)[ED r (O) Further Table 2 lists the atom charges q(V) of the central V atom closest to the vacancy (from Mulliken population analyses) where in each case both frozen qf(V) and relaxed values qr(V) are included. In addition the values of the displacement *r(V) of the central V atom due to relaxation are shown. For all sites the oxygen vacancy energies are rather large (7.2»9.4 eV) which suggests that it is quite difficult to remove an oxygen by itself from the V2O5(010) surface. Based on the E frozen substrate calculations the D f (O) value is largest 9.4 eV for the two-fold bridging site O(2) with that of O(3) and O(1) site being smaller by only 0.8 and 1.3 eV respectively. Relaxation due to vacancy formation decreases these energies by 1.3»2.2 eV where the eÜect is again largest for the O(2) site.Thus the strongest binding of the oxygen to the surface leads to the largest relaxation after its removal. As a result of relaxation the vacancy energies at the O(2) and O(3) sites E are rather close D r (O)\7.2 and 7.3 eV while that of the O(1) site is only slightly smaller 6.7 eV. When substrate relaxation due to vacancy formation is accounted for the computed energetic and geometric quantities are in—uenced by the fact that the V10O31H12 cluster without vacancy and with its ideal bulk structure does not correspond to the equilibrium geometry of the cluster. Therefore a part of the relaxation eÜect is due to equilibration of the initial V10O31H12 . However the corresponding contributions can be neglected for the present purpose.This has been demonstrated by test calculations where all the cluster atoms of V10O31H12 except the terminating hydrogen atoms are allowed to rearrange according to the lowest cluster total energy. The corresponding energy lowering Erel\0.9 eV as well as atom displacements *r(V)\0.2 ” turn out to be rather small and all the atom charges in the relaxed cluster are very similar to those of the cluster in its ideal bulk structure see the q values in parentheses in Table 2. All energy values of Table 2 are computed applying the LSDA scheme for exchange and correlation. The use of gradient corrected (GGA-II) functionals35 decreases the E (O) values by 0.8»1.1 eV29 depending D on the oxygen site. However this does not aÜect the energetic sequence between the diÜerent sites and leads to corrections of only 0.2 eV in the relaxation energies Erel .Therefore the present discussion will be restricted to LSDA results. As an example of the geometric eÜect due to substrate relaxation Fig. 4 shows the geometry of the relaxed V10O30H12 cluster for an O(2) vacancy. The main result is that the relaxation eÜect is found to be locally con–ned. The strongest relaxation shifts occur for the two vanadium atoms adjacent to the vacancy which move laterally by 0.63” ” (0.7 including a upwards shift) such that the opening of the vacancy is enlarged. However the lattice topology of the V2O5(010) surface is conserved which suggests that a single O(2) vacancy will not introduce major restructuring of the surface.This result has also been found in calculations of single O(1) and (3) vacancies and even for vacancy pairs.29 The atom charges of the central V atoms closest to each oxygen vacancy see Table 2 are smaller than the corresponding values of the cluster without the vacancy which suggests chemical 59 Faraday Discuss. 1999 114 53»66 Fig. 4 Relaxed geometry of the V10O31H12 cluster with an O(2) vacancy. The cluster atoms are shown by shaded balls where ball radii represent atom charges. Dark (light) shading refers to negative (positive) charge while the radius gives the amount of charge. The white balls behind the relaxed cluster describe the system without vacancy. reduction of the metal sites. This is due to the fact that the vacancy oxygen is a negatively charged species at the surface.Therefore when the oxygen is removed as a neutral species it leaves a negative excess charge behind that is distributed over the atoms close to the vacancy and compensates some of the positive metal charges. A comparison of the frozen and relaxed cluster values qf(V) and qr(V) of Table 2 shows that the reduction eÜect decreases by relaxation where for the O(1) vacancy the decrease leads to almost no reduction of the respective metal site. It is interesting to study the consequences of vacancy formation for the electronic inter-layer 2O5(010) surface. This problem is examined by vacancy calculations using sketched in Fig. 2(b). Pre- 20O61H24 V10O30H12 clusters. When atom relaxation is allowed in the two-layer cluster the frozen at the ideal surface geometry lead to charge distributions very close to those of Erel coupling near the V clusters that model two adjacent physical layers such as V20O62H24 liminary results from these calculations29 show that oxygen vacancies with the substrate cluster V the one-layer corresponding relaxation energies amount for all oxygen vacancies O(1»3) to about twice the values obtained for the single-layer clusters.Further surface binding and charging as well as geometric consequences of relaxation are somewhat diÜerent in the two- and one-layer clusters. In particular inter layer binding can be aÜected by relaxation after vacancy formation. As an illustration Fig. 5 shows the relaxed geometry of the V20O61H24 cluster with an O(1) vacancy for views normal and parallel to the surface.Obviously the vanadium surface atom (labeled by hatching) below the missing O(1) is aÜected most by relaxation. This atom shifts downwards towards the second layer by 0.94 ” while the second layer oxygen below the vanadium moves upwards by 0.12 ” . As a result the two atoms interact with each other and form a single V»O bond as can be seen from a bond order analysis.29 Thus vacancy formation at the –rst layer of the V2O5(010) surface may increase the electronic coupling with the second layer. 3.3 H adsorption at V2O5(010) Table 3 lists the geometric and electronic parameters of the surface cluster V10O31H12]H where the hydrogen species has been added near the three oxygen sites O(1) O(2) O(3).Both the hydrogen and the adsorption site oxygen are optimized in their positions according to the lowest total cluster energy. Atom charges q from Mulliken population analyses and Mayer bond orders p are given for the corresponding oxygen site and for the V atoms closest to the cluster center. The hydrogen binding energy E (H) with respect to the clean surface (adsorption energy) is de–ned by B the total energy diÜerences of the corresponding clusters (5) (H)\oE E tot(V10O31H12]H)[Etot(V10O31H12)[Etot(H) o B E where tot(V10O31H12]H) refers to the computed equilibrium geometry while the OH desorption energy E (OH) (binding with respect to the clean surface with an oxygen vacancy) is de–ned D Faraday Discuss. 1999 114 53»66 60 Fig. 5 Relaxed geometry of the V20O61H24 cluster with an O(1) vacancy for a view (a) normal and (b) parallel to the surface.Atom shading and radii are de–ned as in Fig. 4. The white balls behind the relaxed cluster describe the system without vacancy. The central V atom next to the vacancy is emphasized by hatching. by the total energy diÜerences (OH)\oE E tot(V10O31H12]H)[Etot(V10O30H12)[Etot(OH) o D tot(V10O30H12) is computed for the cluster with the appropriate oxygen vacancy. All where E results shown in Table 3 are calculated within the LSDA scheme see below. The results of Table 3 show that hydrogen can stabilize at all oxygen sites with sizable binding energies forming rather stable surface OH groups with equilibrium geometries sketched in Fig. 6. The hydroxy group involving the terminal O(1) site is bent with respect to the surface normal by 73° with the oxygen shifted by 0.37 ” relative to its position without the hydrogen.The hydroxy Table 3 Surface OH equilibrium geometries atom charges q and bond orders p for the V10O31H12 ]H clusters with H adsorbed at the oxygen sites O(1»3) obtained by LSDA calculationsa dVhO/” dOhH/” *r(O)/” inc/° q(V) q(O) q(H) Àt p(O»V) p(O»H) E (H)/eV E (OH)/eV B D a Values dVhO dOhH refer to equilibrium distances of surface OH while À denotes the inclination angle of the OH axis with respect to the surface normal inc and *r(O) gives the shift of the respective surface oxygen due to adsorption. In addition binding energies E (H) and desorption energies E (OH) are B given for each site.For further de–nitions see text. D (6) O(3) site O(2) site O(1) site 1.97 0.99 0.34 34 1.70 1.01 0.37 73 1.87 0.99 0 0.27 1.35 [0.96 0.57 1.37 [0.85 0.54 1.46 [0.63 0.54 0.27 0.62 0.51 0.66 1.25 0.61 2.50 5.79 2.76 6.89 3.05 5.96 61 Faraday Discuss. 1999 114 53»66 Fig. 6 Equilibrium geometries of surface OH and H2O at diÜerent oxygen sites O(1) O(2) O(3) obtained from optimizations of V darker shaded balls while the surface lattice is sketched by light balls. 10O31H12]H V and 10O31H12]2H respectively. The surface species are shown by group formed at the bridging O(2) site points normal to the surface for symmetry reasons while the oxygen is shifted by 0.27 ” out of the surface.Further the OH involving the bridging O(3) site is bent by 34° pointing away from the adjacent vanadyl group with the oxygen shifted by 0.34 ” relative to its position before H adsorption. At all sites the distances between oxygen and its nearest V neighbors at the V dVhO are enlarged by the adsorption which suggests a 2O5 surface weakening of V»O binding near the adsorption site. This is con–rmed by the atom charges and bond orders given in Tables 1 and 3. For all sites the oxygen accumulates negative charge due to adsorbed hydrogen and the surface OH becomes slightly negative OH~0.1»OH~0.4 where charging is smallest for the terminal O(1) site. The latter conforms with the result that the positive charge of the vanadium neighbor at the O(1) site remains the same whereas for the higher coordination sites O(2 3) the neighboring V atoms lose positive charge (are reduced) by hydrogen adsorption.The V»O bond weakening near all adsorption sites can be seen by a comparison of Tables 1 and 3 where p(O»V) values are found to decrease due to adsorption. Obviously the V2O double bond at the O(1) site is reduced to a single bond and the weaker V»O bonds at O(2 3) are further reduced. The adsorption energies E (H) vary between 2.5 and 3.0 eV depending on the surface oxygen B site which suggests rather stable surface OH groups for all sites. The present data show the strongest adsorptive binding for the O(1) site followed by the O(2) and O(3) sites. This order is in disagreement with previous results from ab initio DFT calculations for V2O5 clusters and periodic slab models,43 which –nd that E (H) is larger for the O(2) than for the O(1) site.This discrepancy B is explained by the fact that in the previous study the hydrogen adsorbate and the oxygen site are allowed to relax only perpendicular to the V2O5 surface whereas the present results are based on a full OH geometry optimization without constraints. The E (H) values listed in Table 3 refer to B cluster calculations using the LSDA scheme for exchange and correlation which is well known to overestimate binding energies. Extended tests29,43 using gradient corrected (GGA-II) functionals35 show that the E (H) values of Table 3 are decreased by 0.5»0.6 eV43 due to the improved GGA-II B scheme.However this correction is found to be independent of the adsorption site. Thus relative binding energies can be estimated using both the LSDA and the GGA-II scheme. Further equilibrium geometries turn out to be aÜected only very little by going from the LSDA to the GGA-II scheme. Therefore we restrict ourselves to a discussion of the LSDA results. Table 3 also contains results of the OH desorption energy E (OH) which is de–ned as the D energy required to remove the OH species from the V2O5 surface leaving an oxygen vacancy behind where the de–nition in eqn. (6) assumes that OH removal proceeds without an interme- Faraday Discuss. 1999 114 53»66 62 diate reaction barrier. Further in the calculations the substrate was not allowed to relax due to the oxygen vacancy formation.The desorption energies E (OH) vary between 5.8 and 6.9 eV D depending on the surface oxygen site which reveals very strong binding of OH with its V2O5(010) surface environment at all sites. The present calculations yield the strongest binding for the O(2) site followed by the O(1) and O(3) sites. While the computed E (OH) values seem to be rather D large their actual size is not unreasonable. As discussed in Section 3.2 the removal of oxygen (without pre-adsorbed hydrogen) from the V2O5(010) surface is found to require energies (cf. ED f (O) values of Table 2) larger than the corresponding ED(OH) values by 2.2»2.7 eV depending on the oxygen site. These energy diÜerences are obviously due to the hydrogen induced V»O bond weakening at the surface which makes it easier to remove an OH than an oxygen species.Table 4 lists the geometric and electronic parameters of the surface cluster V10O31H12]2H where a second hydrogen species has been added near the OH group formed at the three oxygen sites O(1) O(2) O(3) and the resulting H2O unit is optimized in its geometry according to the lowest total cluster energy. Here atom charges and bond orders are de–ned as analogous to the corresponding parameters of Table 3. The E (H) values refer to binding of the second hydrogen to B the existing surface OH group de–ned by the total energy diÜerences (H)\oE E tot(V10O31H12]2H)[Etot(V10O31H12]H)[Etot(H) o B Etot(V10O31H12]2H) and Etot(V10O31H12]H) refer to the computed equilibrium geome- D(H2O) where tries.Further the H2O desorption energy E (binding with respect to the clean surface with an oxygen vacancy) is de–ned by the total energy diÜerences 2O (OH)\oE E tot(V10O31H12]2H)[Etot(V10O30H12)[Etot(H2O) o D As mentioned before all results shown in Table 4 are calculated within the LSDA scheme. The results of Table 4 show that an additional hydrogen stabilizes at the surface OH groups formed at all the oxygen sites thereby resulting in surface H2O groups with equilibrium geometries sketched in Fig. 6. The H involving the terminal O(1) site is bent by 33° with the oxygen shifted by 0.75 ” relative to its position at the clean V2O5(010) surface. In its most stable geometry the H formed at the bridging O(2) site points with its molecular axis normal to the 2O Table 4 Surface H2O equilibrium geometries atom charges q and bond orders p for the V10O31H12]2H clusters with 2H adsorbed at the oxygen sites O(1»3) obtained by LSDA calculationsa O(1) site dVhO/” dOhH/” *r(O)/” 1.83 1.08 0.99 0.75 33 112 inc/° n(H»O»H)/° 1.48 [0.88 0.55 0.51 0.47 0.43 0.71 2.18 2.13 À q(V) q(O) q(H) p(O»V) p(O»H) E (H)/eV EB D(H2O)/eV a Values dVhO dOhH refer to equilibrium distances of surface H2O while Àinc denotes the inclination angle of the H2O axis with respect to the surface normal and *r(O) gives the shift of the respective surface oxygen due to adsorption.In addition binding energies E (H) of the second surface hydrogen and desorption energies ED(H2O) are given for each site.For B further de–nitions see text. (7) (8) O(3) site O(2) site 2.04 1.00 0 0.62 2.27 1.01 1.00 1.11 26 110 113 1.26 [0.92 0.54 0.56 1.27 [0.96 0.59 0.19 0.61 0.58 0.22 0.59 1.56 1.33 1.36 2.23 63 Faraday Discuss. 1999 114 53»66 surface and its molecular plane extends perpendicular to the V»O(2)»V plane see Fig. 6. Further the oxygen is shifted by 0.62 ” out of the surface. Finally the H2O involving the bridging O(3) site is bent by 26° with its molecular plane perpendicular to the V»O(3)»V plane and the oxygen shifted by 1.11 ”. As for the surface OH the distances between the oxygen and its nearest V neighbors at the V2O5 surface are increased by the adsorption where the eÜect is always larger for surface H2O than for surface OH.This hints at an even more pronounced weakening of V»O binding near the adsorption site. The oxygen becomes more negative due to hydrogen adsorption where the eÜect is larger for surface H2O than for surface OH. However the increased negative oxygen charge is overcompensated by the second (positively charged) hydrogen which yields a positive surface H2O H species 2O`0.2 almost independent of the oxygen site. The charge of the vanadium neighbor at the O(1) site remains almost unchanged whereas for the higher coordination sites O(2 3) the neighboring V atoms are reduced more strongly by surface H2O surface OH. Further the V»O bond weakening near all adsorption sites observed for surface OH becomes more pronounced for surface H2O bond orders p(O»V) of Tables 3 and 4.than by as evidenced by a comparison of the corresponding The adsorption energies E (H) for the second hydrogen vary between 1.4 and 2.2 eV depending D 2O Bdesorption energy ED(H2O) E (OH) discussed above. D(H2O) values vary between 1.3 and 2.2 eV depending on the surface oxygen site which ED(H2O) values are smaller than the corresponding species easier that that of OH. B on the surface oxygen site where strongest adsorptive binding is found for the O(1) site followed by the O(3) and O(2) sites. In all cases the computed binding energy E (H) of the second hydrogen is found to be smaller than the value for the –rst H species. The H2O listed in Table 4 quanti–es the removal of surface H as analogous to The E indicates rather moderate binding of H2O V2O5(010) to its surface environment at all sites.The present calculations yield the strongest binding for the O(2) site followed by the O(1) and O(3) E (OH) sites. The values by 3.8»4.7 eV D depending on the oxygen site see Tables 3 4. This can be explained by the increased V»O bond weakening at the surface for surface H2O compared to surface OH which makes the removal of an H2O 4 Conclusions 20O62H24 V are size converged and therefore the clusters can be considered realistic models of surface. The theoretical data based on ab initio DFT methods con–rm the Altogether the present theoretical study provides a clear picture of the electronic structure of the V2O5(010) surface and its consequences for oxygen vacancy formation as well as hydrogen adsorption.The electronic parameters calculated for the present cluster models V10O31H12 and V the extended 2O5(010) mixed ionic and covalent character of V»O binding and distinguish between the diÜerently coordinated oxygen sites. Both the width of the O 2sp dominated valence band region and its total as well as atom projected DOSs are consistent with angular resolved photoemission (ARUPS) data for freshly cleaved V samples.30 Based on the cluster results the three-peak structure 2O5(010) observed in the experiment can be interpreted as originating from diÜerently coordinated surface oxygen terminal (vanadyl) O(1) dominating the central peak and bridging O(2,3) characterizing the two peripheral peaks.Thus the theoretical data suggest that the diÜerent O 2sp derived peaks observed in the photoemission experiment may be taken as monitors of diÜerently coordinated surface oxygen and can be used to study details of catalytic reactions at the oxide surface where oxygen participates. Studies on diÜerent oxygen vacancies at the V2O5(010) surface yield rather large vacancy forma- E (O) 6.7»7.3 eV depending on the oxygen site which make it quite difficult to tion energies D remove oxygen by itself from the surface. Surface relaxation caused by the vacancy contributes 1.3»2.2 eV to E (O) which is not negligible. However the geometric relaxation eÜect is for all D vacancy sites O(1»3) locally con–ned with atom displacements of at most 0.7 ”.This leaves the lattice topology of the V2O5(010) surface unchanged and suggests that single oxygen vacancies will not introduce major restructuring of the surface. Vacancy formation leads in all cases to chemical reduction of the vanadium atoms near the vacancy site as shown by population analyses. Further preliminary results from two-layer clusters indicate that surface relaxation due to vacancy formation may aÜect inter-layer coupling and may lead to additional V»O bonds between the physical layers. Faraday Discuss. 1999 114 53»66 64 2O (B E (H) where the second hydrogen is always bound more weakly formation the calculations 2O (1.3»2.2 eV) surface that involve adsorption and binding of hydrogen as which is bound Hydrogen can adsorb at all oxygen sites of the V2O5(010) surface forming rather stable surface OH groups.The H adsorption energies E (H) vary between 2.5 and 3.0 eV depending on the B surface oxygen site. The approach of a second hydrogen to the surface OH group leads to the formation of surface H between 1.4 and 2.2 eV) than the –rst. For both surface OH and H2O suggest a chemical reduction of the neighboring V metal atoms connected with V»O bond weakening where the eÜect is larger for H2O than for OH. The calculations of desorption energies ED H2O (leaving a surface oxygen vacancy behind) show a required to remove surface O OH or general trend independent of the coordination of the oxygen site the removal always requires more energy for surface O than for surface OH and more for surface OH than for surface H2O.While the desorption energies for surface O are extremely large those for surface H are within the range of typical chemical reaction energies. This has implications for hydrocarbon oxidation reactions at the V2O5(010) well as oxygen desorption from the surface. Here the present calculations suggest that oxygen removal from the surface occurs preferentially by formation of surface H2O weakly enough to desorb creating oxygen vacancies at the V2O5(010) surface. Acknowledgement This work has been supported by Deutsche Forschungsgemeinschaft SFB 1760 and by Fonds der Chemischen Industrie. Further support by grant No. 3T09A 14615 of the State Committee for Scienti–c Research in Poland is acknowledged. References 1 C.N. R. Rao and B. Raven T ransition Metal Oxides VCH Press New York 1995. 2 H. K. Kung Stud. Surf. Sci. Catal. 1989 45 1. 3 V. E. Henrich and P. A. Cox T he Surface Science of Metal Oxides Cambridge University Press Cambridge 1994. 4 B. Grzybowska-Swierkosz Appl. Catal. A Gen. 1997 157 409 and references cited therein. 5 E. E. Chain Appl. Opt. 1991 30 2782 and references cited therein. 6 Appl. Catal. A Gen. 1997 157 1. 7 V anadia Catalysts for Processes of Oxidation of Aromatic Hydrocarbons ed. B. Grzybowska-Swierkosz and J. Haber PWN-Polish Scienti–c Publishers Warsaw 1984. 8 A. Bystroé m K. A. Wilhelmi and O. Brotzen Acta Chem. Scand. 1950 4 1119. 9 H. G. Bachman F. R. Ahmed and W. H. Barnes Z. Kristallogr. Kristallgeom. Kristallphys. Kristallchem.1961 115 110. 10 R. W. G. WyckoÜ Crystal Structures Interscience Publishers John Wiley & Sons Inc. New York» London»Sydney 1965. 11 A. Bielanski and J. Haber Oxygen in Catalysis Marcel Dekker New York 1990. 12 A. Bielanski J. Piwowarczyk and J. Pozniczek J. Catal. 1988 113 334. 13 J. Haber M. Witko and R. Tokarz Appl. Catal. A Gen. 1997 157 3. 14 M. Witko R. Tokarz and J. Haber J. Mol. Catal. 1991 66 205. 15 M. Witko R. Tokarz and J. Haber J. Mol. Catal. 1991 66 357. 16 M. Witko Catal. T oday 1996 32 89. 17 M. Witko R. Tokarz and J. Haber Appl. Catal. A Gen. 1997 157 23. 18 R. F. Nalewajski J. Korchowiec R. Tokarz E. Broclawik and M. Witko J. Mol. Catal. 1992 77 165. 19 R. F. Nalewajski and J. Korchowiec J. Mol. Catal. 1993 82 383. 20 M.Witko R. Tokarz and K. Hermann Pol. J. Chem. 1998 72 1565. 21 M. Witko and K. Hermann J. Mol. Catal. 1993 81 279. 22 M. Witko and K. Hermann Stud. Surf. Sci. Catal. 1994 82 94. 23 M. Witko K. Hermann and R. Tokarz J. Electron Spectrosc. Relat. Phenom. 1994 69 89. 24 K. Hermann A. Michalak and M. Witko Catal. T oday 1996 32 321. 25 A. Michalak M. Witko and K. Hermann Surf. Sci. 1997 375 385. 26 R. Ramirez B. Casal L. Utrera and E. Ruiz-Hitzky J. Phys. Chem. 1990 94 8960. 27 L. Kihlborg Ark. Kemi 1963 21 357. 28 H. Hanke R. Bunert and H. G. Jetschewitz Z. Anorg. Allg. Chem. 1975 109 414. 29 R. Druzinic PhD thesis Free University Berlin 1999. 30 K. Hermann M. Witko R. Druzinic A. Chakrabarti B. Tepper M. Elsner A. Gorschlué ter H. Kuhlenbeck and H.-J.Freund J. Electron Spectrosc. Relat. Phenom. 1999 98»99 245. 65 Faraday Discuss. 1999 114 53»66 31 N. Godbout D. R. Salahub J. Andzelm and E. Wimmer Can. J. Phys. 1992 70 560. 32 Density Functional Methods in Chemistry ed. J. K. Labanowski and J. W. Anzelm Springer»Verlag New York 1991. 33 The DFT-LCGTO program package DeMon was developed by A. St.-Amant and D. Salahub (University of Montreal). Here a modi–ed version with extensions by L. G. M. Pettersson and K. Hermann is used. 34 S. H. Vosko L. Wilk and M. Nusair Can. J. Phys. 1980 58 1200. 35 J. P. Perdew J. A. Chevary S. H. Vosko K. A. Jackson M. R. Pederson D. J. Singh and C. Fiolhais Phys. Rev. B Condens. Matter 1992 46 6671. 36 R. S. Mulliken J. Chem. Phys. 1955 23 1833; 1841; 2388; 2343. 37 I. Mayer Chem. Phys. L ett. 1983 97 270. 38 I. Mayer T HEOCHEM 1987 149 81. 39 A. Chakrabarti K. Hermann R. Druzinic M. Witko M. Petersen and F. Wagner Phys. Rev. B Condens. Matter 1999 59 10583. 40 V. Eyert in Density Functional Methods Applications in Chemistry and Materials Science ed. M. Springborg Wiley Chichester 1997 and references cited therein. 41 V. Eyert and K.-H. Hoé ck Phys. Rev. B Condens. Matter 1998 57 12727. 42 X. Yin A. Fahmi A. Endou R. Miura I. Gunji R. Yamauchi M. Kubo A. Chatterjee and A. Miyamoto Appl. Surf. Sci. 1998 130»132 539. 43 K. Hermann A. Chakrabarti R. Druzinic and M. Witko Phys. Status Solidi 1999 173 195. Paper 9/03109J Faraday Discuss. 1999 114 53»66 66

ISSN:1359-6640    

DOI:10.1039/a903109j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

The growth of vanadium oxide on alumina and titania single crystal surfaces Robert J. Madix,* Jué rgen Biener Marcus Ba é umer and Andreas Dinger Departments of Chemical Engineering and Chemistry Stanford University Stanford CA 94305 USA 1 Introduction Understanding the interaction of metals and metal oxides with oxide surfaces is important to many technological areas such as metal/ceramic interfaces microelectronics geochemistry and heterogeneous catalysis. Important unresolved questions include the strength of this interaction the wetting characteristics and the mechanical properties of such interfaces and the electronic properties and chemical reactivity of the overlayer. At submonolayer coverages metals or metal oxides which disperse themselves over the surface may possess characteristics diÜerent from their respective bulk solids since they would be expected to coordinate strongly with the guest oxide forming in a sense a mixed surface compound.1 Although a conclusive picture is still missing it has been shown in several UHV studies that the affinity of the metal with oxygen is a crucial factor in determining the strength of interaction between metal overlayers and the support,2h4 and the heat of formation of the metal oxide can serve as a guide for assessing the interaction strength.2,3 However when the oxygen affinity of both the metal atoms of the deposit and the support is high the relative heats of formation may not provide a reliable guide to wetting behavior.In an eÜort to contribute to a better understanding of such systems we have investigated the growth and the properties of vanadium and vanadia grown on alumina and titania.As an early transition metal vanadium is expected to interact strongly with oxide surfaces due to the high heats of formation of the vanadium oxides. For example titania is reduced upon deposition of vanadium metal which in turn is oxidized,5h7 even though the heat of formation of titania exceeds that of vanadia. In this respect vanadium behaves similarly to other reactive metals on TiO2 .2,8,9 Furthermore vanadium is an interesting metal for study because of its ability to adopt diÜerent oxidation states and thus form diÜerent oxides two of which are either conducting or semiconducting at room temperature. 67 Faraday Discuss.1999 114 67»84 Received 6th April 1999 Evaporation of vanadium metal onto alumina or titania surfaces at room temperature in an oxygen ambient results in the growth of V2O3 overlayers. The results of several complementary methods including STM indicate that the oxide grows in clusters 20»30 Aé in diameter eventually covering the surface with a granular thin –lm at a coverage in excess of one monolayer of vanadium atoms. The similarity of the distribution of vanadium metal and the vanadium oxide on the surfaces observed by STM on a thin crystalline alumina –lm suggests that the oxide is formed after the metal nucleates into small clusters. No surface reduction of cations occurs on either the TiO (110) or 2 Al2O3(0001) surface when the vanadium oxide is formed.On the alumina the vanadium oxide –lm appears to be conducting at room temperature as would be expected for V2O3 formation. This journal is( The Royal Society of Chemistry 2000 On alumina however the situation is more complicated. In contrast to titania alumina cannot easily be reduced so that the question arises as to which way the interaction manifests itself in this case. Several theoretical attempts have been made to elucidate the nature of the bonding between metals and alumina,1,10h12 but the picture is still unclear. The deposition of vanadium and vanadia on single crystal surfaces of both titania and alumina was studied in the submonolayer-to-multilayer coverage regime with a combination of UHV methods. Our approach was twofold. On the one hand we employed a thin alumina –lm grown on NiAl(110) and a partially bulk-reduced TiO (110) single crystal allowing us to utilize scanning 2 tunnelling microscopy (STM) and electron spectroscopic techniques without charging problems.As described in the literature A 13,14 the alumina –lm is about 5 é thick and is oxygen-terminated. Its defect structure is dominated by a network of domain boundaries separating reasonably large rotational and antiphase domains of the oxide overlayer.13 On the other hand we employed a Al single crystal.15 The growth of the vanadium oxide overlayer was studied with 2O3(0001) LEED photoemission (XPS) and near edge X-ray absorption –ne structure (NEXAFS) spectroscopy. NEXAFS was used to determine the local symmetry and to assist with determination of the oxidation state of the metal cations.Though XPS is the traditional tool to determine the chemical state of metal cations in metal and e symmetry respectively. The e orbitals (d dz2) x2~y2 point in between the oxygen g oxides based on the correlation between core-level binding energies and the oxidation state the measurement of reliable binding energies on oxide surfaces can be hampered by sample charging and –nal state eÜects. NEXAFS is well suited to investigate metal oxides since the transition energy and the line-shape can be used as a –ngerprint of the chemical state. The line-shape of the L-edges of the 3d transition metal oxides is dominated by multiplet eÜects which are sensitive to the local symmetry.16 The NEXAFS is produced by excitation of electrons into empty or partially empty states above the Fermi level using a tunable X-ray source.Band-structure calculations of the unoccupied density of states of 3d transition metal oxides predict a metal 3d-dominated conduction band and at higher energies antibonding states related to oxygen 2p and metal 4sp orbitals.17 In octahedral coordination the ligand –eld causes the metal 3d band to split into two sub-bands with t and point directly 2g g g towards the oxygen ligands whereas the t orbitals (dxy dxz and d 2g yz) band. neighbors. Consequently the e band lies higher in energy than the t2g Recently the electronic structure of vanadium deposited on titania has been studied using single crystal surfaces.5h7,18h21 These studies reveal a strong vanadium/support interaction in the monolayer regime; deposited vanadium reduces titania even at room temperature,5h7 consistent with the high oxygen affinity of vanadium.22 The V/TiO (110) interface has been characterized by 2 means of diÜraction techniques which indicate an epitaxial growth of vanadium without longrange order.19,20 The formation of vanadia is observed if the vanadium deposition is performed in an oxygen ambient5 or if metallic vanadium is exposed to oxygen following the deposition.Using the latter technique the growth of heteroepitaxial VO layers has been achieved.21 2 2 Experimental The photoemission experiments were performed at beam line 10-1 of the Stanford Synchrotron Research Laboratory in an UHV chamber with a base pressure below 2]10~10 Torr.NEXAFS and XPS spectra were produced with monochromatized synchrotron radiation and monitored with a double-pass cylindrical mirror electron energy analyzer. The spherical grating monochromator (1000 l mm~1 50 lm slit) was operated with an energy resolution of 0.2 eV at the Ti L-edge 0.3 eV at the V L-/O K-edges and 0.5 eV at 650 eV used for photoemission. The resulting overall experimental resolution was approximately 0.3 eV and 1.0 eV for NEXAFS and XPS respectively. The NEXAFS spectra were recorded either by monitoring the CMA signal in the low energy range dominated by secondary electrons (partial yield PY) or by setting the CMA on an appropriate Auger transition (Auger yield AY). The XPS and AY-NEXAFS spectra were normalized by dividing the CMA signal by the photoelectron signal of an Au-covered grid inserted in the optical path (I -signal).For the studies on the bulk alumina single crystal a low energy electron 0 gun (4 eV) was used to stabilize the surface potential during the data accumulation. In order to eliminate any possible residual in—uence of sample surface charging in the experiments on the bulk alumina crystal all photoemission spectra were referenced to the Al 2p and O Faraday Discuss. 1999 114 67»84 68 1s peaks of the clean Al surface centered at 74.7 eV and 531.6 eV respectively.23 In the 2O3 experiments with TiO (110) the photoelectron energy analyzer was repeatedly calibrated using a clean Au sample assuming the Au 4f7@2 core-level binding energy to be 84.0 eV.24 Although the 2 spectrometer calibration was performed repeatedly it was observed that for the diÜerent TiO (110) samples used in the experiments the Ti 2p core-level binding energy varied in energy 2 from 458.5 to 459.3 eV.These values of binding energy are within the range reported previously in the literature ;25 the scatter of the data may arise from the diÜering electrical conductivities of the samples. In order to eliminate the in—uence of any charging at the surface on the energy calibration all the photoemission spectra were referenced to the Ti 2p core line at 458.5 eV. For NEXAFS the photon energy was calibrated by aligning the Ti L-edge and O K-edge of the clean TiO (110) sample according to the values reported in the literature.26 In addition L-edge 2 NEXAFS spectra from a V foil were recorded to verify the energy calibration.The TiO (110) single crystals were attached to molybdenum disks which could be heated from 2 the back either by radiation or by electron bombardment. The sample temperature was measured by means of an optical pyrometer and a chromel/alumel thermocouple attached to the sample holder. To introduce n-type semiconducting bulk properties the TiO sample was bulk-reduced 2 by high temperature annealing (1 h 1100 K). During this time the crystal color changed from colorless to pale green. Calcium and carbon were removed from the surface by repeated sputter» anneal cycles (500 eV Ar` 10 lA). The surface stoichiometry was restored by annealing in oxygen (p(O2)\5]10~5 Torr 900 K) and restoration of the stoichiometric surface was judged by the absence of reduced titanium in photoemission spectroscopy and the presence of a sharp (1]1) LEED pattern.Vanadia was deposited at 300 K by evaporating vanadium from a resistively heated vanadium –lament (Goodfellows 0.008 inch diameter [99.9%) in an oxygen ambient of 2]10~6 Torr. The vanadia deposition rate and the coverage were estimated from X-ray spectra following standard procedures,27,28 making use of the attenuation of the Ti 2p XPS signal caused by the vanadia overlayer. The deposition rate was determined to be approximately 1]10~2 ML s~1 assuming that one ML corresponds to approximately 9.4]1014 atoms cm~2 the areal density of vanadium atoms in the (0001) plane of V2O3 .29 The a-Al2O3(0001) single crystals 0.8 mm thick polished wafers were mounted and heated like the titania crystals.The sample temperature was measured by means of an optical pyrometer and the error in the temperature measurement is estimated to be less than 100 K. The Al2O3 surfaces were cleaned as follows (1) prolonged degassing ([5 h at 1000 K in UHV) and (2) annealing at 1200 K (1 h UHV) followed by an oxygen annealing step at 900»1100 K (1 h 1]10~6 Torr O2). The surface cleanliness was monitored by XPS. The oxygen annealing step was repeated until a sharp (1]1) LEED pattern was observed,30h33 and surface contamination was reduced below the detection limit. The deposition of vanadium oxide on the Al2O3(0001) samples was accomplished by evaporation of metallic vanadium in an oxygen ambient of 1]10~7 Torr at 300 K sample temperature.The vanadia deposition rate and the coverage were estimated from X-ray photoelectron spectra following standard procedures 27,28 which make use of the attenuation of the Al 2p and the increase of the V 2p XPS signal intensities as a function of the deposition time. The deposition rate employed was approximately 10~3 ML s~1. STM images were taken in an UHV system equipped with a ìJohnnie Walkerœ type scanning tunneling microscope purchased from RHK Tech. The microscope stage allowed heating of the sample by radiation (W –lament) and cooling by contact to a liquid nitrogen reservoir. The sample was mounted on a transferable mount which also contained a trisectional ramp used for the approach of the STM head. This unit could be transferred between the microscope located in a side chamber and a sample preparation chamber equipped with LEED and all instruments necessary for the preparation of the surface including a gas doser and a metal evaporator.A NiCr/Ni thermocouple was used to determine the sample temperature. The sample could be heated in the preparation chamber to at least 1000 °C by electron bombardment while cooling was possible by directing liquid nitrogen through the manipulator. The clean NiAl(110) surface was prepared by several cycles of sputtering (Ar` ions 1 keV) and –lm was obtained by following procedures previously annealing to 1300 K. The ordered Al2O3 reported in the literature.13,14 After dosing about 5000 L O at a sample temperature of 533 K 2 the crystal was annealed to 1100 K for 3 min.The quality of the oxide was checked via its distinctive LEED pattern. Vanadium metal was deposited in the STM system using procedures 69 Faraday Discuss. 1999 114 67»84 similar to those employed for the alumina and titania single crystals. The evaporation rate was calibrated by a quartz microbalance. In the STM apparatus the nominal –lm thickness determined by the thickness monitor was converted into a coverage using the lattice constant of bulk vanadium (3.03 Aé 29). The deposition rates varied between 3]10~3 and 10~2 ML s~1. 2O3(0001) 3 Results and discussion 3.1 Photoemission studies of the growth of vanadium oxide on TiO (110) and Al 2 The Ti 2p photoemission spectra for the clean and vanadia-covered TiO surface are shown in 2 Fig.1. The spectrum of the clean surface is identical to those reported in the literature indicating that the surface is stoichiometric with the cations in the Ti`4 state.25,34,35 The intensity of the Ti 2p signal decreases with increasing vanadia coverage until the signal –nally vanishes indicating that the surface is completely covered with a vanadia layer of a thickness larger than the mean free path of the 200 eV photoelectrons. The lineshape of the Ti 2p peaks does not change upon vanadia deposition over the entire range of vanadia coverage examined. Since surface reduction of TiO (110) due either to sputtering or to submonolayer deposition of vanadium is easily detected 2 by XPS as a shoulder on the low binding energy side of the Ti2p peak,5 these spectra demonstrates that the vanadia does not reduce the TiO surface.2 The corresponding O 1s and V 2p regions of the XPS spectra are displayed in Fig. 2. The V 2p 2 signal increases monotonically with vanadium exposure whereas the intensity of the O 1s signal stays nearly constant consistent with the deposition of a vanadium oxide. The V 2p peaks are centered at 515.9 eV (V p 2p3@2) 2 and 523.4 eV (V over the entire range of coverage lying 1@2) between values reported for single crystals of V2O3 (515.7/523.3 eV) and VO (516.2/523.5 eV).36 2 In another investigation of the vanadia/TiO system using a very similar preparation technique 2 Zhang and Henrich5 observed the V 2p core-lines at 515.2 and 522.7 eV. For heteroepitaxial VO2 layers on TiO (110) the V 2p3@2 line was reported to be 515.7 and 516.5 eV for submonolayer and Fig.1 The Ti 2p and Ti 2p photoemission peaks as a function of vanadium exposure time at room 1@2 temperature in oxygen. From the top spectra correspond to exposures of 0 20 40 80 160 320 and 640 s. A 3@2 photon energy of 650 eV was used. Monolayer coverage occurs at exposures near 100 s. Faraday Discuss. 1999 114 67»84 70 Fig. 2 The O 1s V 2p1@2 sures and conditions were identical to Fig. 1. The inset shows the dependence of the integrated intensities on and V 2p3@2 photoemission peaks as a function of vanadium exposure time. Expovanadium exposure. multilayer coverages respectively.21 The O 1s peak for clean and vanadia-covered TiO is cen- 2 tered at 530.0 eV. For TiO2 O 1s binding energies in the range of 529.8 to 530.2 have been reported,21,35 which are similar to the values found for V2O3 (530.1 eV) VO (529.9 eV) 36 and 2 VO /TiO (530.4 eV).21 Taken together these XPS results indicate the formation of a vanadia 2 2 overlayer but alone they do not allow the determination of the oxidation state of vanadium cations because of the problems involved in determining an absolute binding energy scale as discussed above.The inset of Fig. 2 compares the normalized V and Ti 2p photoemission intensities as a function of the vanadium exposure. The exponential decay of the Ti 2p signal intensity and the corresponding increase of the V 2p signal with vanadium exposure are consistent with a simultaneous multilayer (SM) growth mode characterized by the growth of small particles which cover the surface at sufficiently high vanadia coverage.28 However although the data allow exclusion of a threedimensional growth mode (Volmer»Weber Stranski»Krastanov) the number of data points is insufficient to distinguish between a purely layer-by-layer growth (Frank»van der Merve) and a SM growth mode.For reasons discussed below we favor the simultaneous growth of isolated nuclei of vanadia which merge into a textured –lm at multilayer coverage. Consistent with these –ndings the initially sharp (1]1) LEED pattern of TiO (110) nearly vanished following the 2 deposition of a vanadia multilayer indicating that vanadia covers the surface without any longrange order. The vanadium oxide formed introduces electronic states in the band gap of the TiO (110) 2 surface.The valence band region of the X-ray spectrum for the clean TiO (110) surface shows a 6 2 eV wide O 2p-related valence band centered approximately 5 eV below EFermi . The band gap is 3 eV wide typical of the stoichiometric TiO2 .37 As expected for an n-type semiconductor the Fermi level is pinned to the bottom of the conduction band. These –ndings are in accordance with the literature.5,34 Upon vanadium exposure a new state appears in the band gap centered approximately 1 eV below EFermi . Since reduced titania can be excluded as the source of the new feature 71 Faraday Discuss. 1999 114 67»84 in the band gap this state can be assigned to photoemission from the partially occupied V 3d band of the vanadia layer thus in agreement with the XPS data eliminating V (d0) as the 2O5 dominant oxide of vanadium present on the surface.Similarly the V 2p photoemission signal from the surface of the bulk alumina single crystal Al2O3(0001) increases with vanadium exposure and the intensity of the O 1s signal decreases rapidly to a constant value suggestive of the formation of a vanadium oxide (Figs. 3 and 4). After the deposition of multilayer quantities of vanadia the V 2p peaks are centered at 515.5 eV (V 2p and 523.2 eV (Vpsuggesting the formation of a V2O3 overlayer.36 Furthermore the V 3@2) 21@2) 2p lines exhibit a linewidth of approximately 3.5 eV indicative of a narrow-band metal like V2O3 .36 With increasing vanadium exposure the O 1s peak shifts abruptly from 531.6 eV to 530.1 eV; the latter value is characteristic of vanadium oxides (V2O3 V2 O529.9 eV).36 530.1 eV A comparison of the observed binding energy shifts of the O 1s and V 2p core-lines as the 3@2 peaks shift simultaneously toward lower binding energies by 1.5 eV.The shift of the O 1s binding energy can be interpreted vanadium coverage increases is shown in Fig. 4. After an exposure time of approximately 2000 s both the O 1s and the V 2p3@2 As all spectra are referenced to the Al 2p peak this observation cannot be caused by a shift of the Fermi level within the band gap of Al2O3 . in terms of oxygen in diÜerent chemical environments i.e. oxygen bound to aluminum and vanadium respectively. The shift to lower binding energy occurs with the completion of the second pure vanadium oxide layer on top of the interface after a vanadium exposure of 2000 s.It is tempting to interpret the higher value of the vanadium binding energy at low coverage in terms of a higher oxidation state. However the problems connected with the calibration of an absolute binding energy scale for insulating oxides prevent the unequivocal determination of the oxidation state solely on the basis of the binding energy. For example for small metallic particles supported on insulating substrates –nal state eÜects in—uence the observed core hole binding energy of the metal. Depending on the details of the interaction between the metal clusters and the Fig. 3 The O 1s V 2p1@2 and V 2p photoemission peaks with increasing vanadium exposure times at room temperature in oxygen (a) clean surface (b) 250 s (c) 500 s (d) 750 s (e) 1000 s (f ) 1250 s (g) 1500 s (h) 1750 s 3@2 (i) 2000 s ( j) 2500 s (k) 3000 s (l) 3500 s (m) 4000 s (n) 6000 s and (o) 9600 s vanadium exposure.Monolayer coverage occurs at exposures near 1000 s. Faraday Discuss. 1999 114 67»84 72 Fig. 4 The dependence of the binding energies of the O 1s and V 2p photemission peaks following expoto vanadium in oxygen at room temperature at varying times. 3@2 sure of Al2O3(0001) surface a shift of the photoemission peaks toward higher binding energies may arise from the Coulomb attraction induced by the positive charge left on the particle as the result of the photoemission process.38,39 The Coulomb energy scales with De2/R where R is the cluster radius. Thus the higher value of the V 2p binding energy at the interface may simply re—ect the formation of very small metal-like clusters with a narrow size distribution during the initial deposition phase which in turn are transformed to V2O3 at the completion of a monolayer.This interpretation is consistent with the growth of vanadia on the alumina –lm observed by STM as described below. In order to assess the amount of vanadia deposited and the mode of its growth on the surface the Al 2p V 2p and O 1s photoemission spectra were –tted with Gaussian functions. In Fig. 5 the normalized signal intensities are displayed as a function of the deposition time. The Al 2p signal decreases continuously with increasing deposition time –nally reaching the detection limit ; this –nal state indicates a surface completely covered with vanadia.Simultaneously the V 2p intensity increases to a constant value. The O 1s signal decreases initially rapidly and stabilizes after an exposure of 1000 s at about 2/3 of its original intensity. These –ndings allow us to exclude threedimensional growth modes Volmer»Weber and Stranski»Krastanov respectively. The continous variation of the intensities and absence of break points suggest a simultaneous multilayer (SM) rather than a pure layer-by-layer growth mode (Frank»van der Merve).28 The attainment of the constant oxygen signal coincides with the deposition of approximately one monolayer of vanadium oxide. In the course of the study it was observed that the kinetic energy of the Al 2p photoelectrons changed systematically with the vanadia coverage even though both the analyzer work function and the photon energy were constant during the experiments (Fig.6). Since the chemical state of aluminum is not expected to change signi–cantly upon vanadia deposition the shift in the kinetic energy of the Al 2p photoelectrons must originate from surface charging due to an incomplete neutralization of the surface during the experiment. The surface charging decreases rapidly above an exposure time of approximately 1000 s indicative of the formation of a closed conducting overlayer i.e. the deposition of a metallic vanadium oxide. This behavior is consistent with the 73 Faraday Discuss. 1999 114 67»84 Fig. 5 The evolution of the integrated Al 2p O 1s and V 2p photoelectron intensities following various to vanadium in oxygen at room temperature.Monolayer coverage is estimated 3@2 exposure times of Al to occur near 1000 s.2O3(0001) shift in the V 2p binding energy as V2O3 is metallic at room temperature,22 is consistent with the calculated deposition rate and coincides with the exposure at which the O 1s signal becomes constant. The lineshape of the Al 2p signal was checked in order to verify the validity of the assumption that the chemical state of aluminum does not change upon the deposition of vanadia. For example the formation of a reduced alumina surface layer should be re—ected by the observation of a new Al 2p photoemission peak shifted D3eV to lower binding energies.40 However the analysis revealed that the lineshape of the Al 2p peak of the clean and vanadia-covered surface are identical in the submonolayer coverage regime.However for vanadium exposures exceeding 2000 s the exposure marking completion of a conducting vanadia overlayer the line width did suddenly decrease from 3.7 eV to approximately 2.7 eV. This decrease is probably a consequence of a more homogeneous surface potential produced by the sudden change in conductivity of the overlayer. 2O3(0001) 3.2 NEXAFS studies of the growth of vanadia on TiO (110) and Al 2 The NEXAFS spectra of the clean TiO (110) surface are characteristic of the rutile phase of TiO2 26 (not shown). In agreement with the information obtained from the photoemission spectra 2 of the Ti 2p region the only eÜect of vanadia deposition is an attenuation of the NEXAFS signal intensity.The lineshape remains unchanged indicating that the vanadia does not disturb the local symmetry of the titanium cations on the TiO surface. 2 Fig. 7 shows the V L-edge spectra of the vanadia overlayers obtained at progressively higher coverages. The two broad features centered at 517.6 and 524.5 eV are related to excitations from V 2p core-levels into empty or partially occupied V 3d orbitals.26 3@2 (LIII-edge) and V 2p1@2 (LII-edge) Both the transition energy 41,42 and the lineshape 43 of the vanadium L-edge are sensitive to the oxidation state of the vanadium cations. The positions of the L-edge features of the deposited Faraday Discuss. 1999 114 67»84 74 growth of the V exposure in oxygen. The energy zero is taken to be the kinetic energy measured for the photelectrons upon Fig.6 Relative kinetic energies for Al 2p photoelectrons for Al2O3 as a function of the duration of vanadium 2O3 multilayer. vanadia do not shift with the V coverage. The transition energies are in excellent agreement with values expected for V2O3 using the calibration curves published by Chen et al.,42 who reported a linear increase of 0.7 eV per oxidation state in the V LIII-edge ranging from 515.5 eV for metallic V to 519.0 eV for V2O5 . In addition to the transition energies the –ne structure of the V L-edge can be used to deter- 2O3 but it diÜers 2O5 .43 According to calculations from de Groot et al.,44 the pre-edge and are characteristic of a 3d2 initial- state multimine the vanadium oxidation state.The details of the spectral lineshape of the V L-edge can be explained only by atomic multiplet calculations.44h47 The multiplet structure generally changes with the number of d-electrons. Therefore the oxidation state can be deduced from the lineshape. The vanadia L-edge –ne structure strongly resembles that of single crystal V from that for VO and V 2 feature and a feature between the LIII- LII-edge plet. Both features are observed in the experimental spectra of the vanadia overlayer particularly at the higher coverages thus indicating the formation of V2O3 in the present study. 2 The O K-edges of the clean TiO surface before and after deposition of a vanadia multilayer are compared in Fig. 8. The O K-edge involves the excitation of an oxygen 1s electron into unoccupied states above the Fermi level with oxygen 2p character thus providing a picture of the oxygen 2p-projected density of states.In contrast to the L-edges of the transition metals the oxygen K-edge can be described within the single-particle approximation.17,48 In order to extract the positions and the intensities of the Ti and V 3d-related peaks from the clean and vanadia-covered surfaces respectively the experimental spectra were –tted with Gaussian functions. The absorption step was approximated with an error function centered at the experimentally determined oxygen 1s binding energy value of 530.0 eV with a FWHM equal to the experimental resolution (0.3 eV).49 It is obvious from the –tting results that the integrated intensity of the t2g state for V in the vanadium oxide is signi–cantly less that would be expected for a d0 oxide.As the intensity is related to the number of unoccupied states available to the excited electron this observation indicates the presence of a partially –lled t2g band in agreement 75 Faraday Discuss. 1999 114 67»84 2 Fig. 7 Transition intensities for V-near-edge absorption –ne structure following vanadium deposition at various exposure times in oxygen onto TiO (110). Exposure times were 20 40 80 160 320 and 640 s. Monolayer coverage occurs near a deposition time of 100 s. with the appearance of states in the band gap observed in the X-ray spectra. The ligand –eld splitting of the d-orbitals in the vanadium oxide formed on the surface is 2.3 eV similar to the value of 2.2 eV reported for V2O3 and VO2 .48 Thus the K-edge data are consistent with formation of V2O3 .The NEXAFS results provide evidence for the existence of a short-range order of the vanadium cations even though the sharp (1]1) LEED pattern of the clean surface disappeared completely after the deposition of a vanadia monolayer indicating the loss of long-range order. single crystal surface using the calibration curve of Chen et al.26 The –ne structure in particular for V exposures Similar results were observed for vanadia deposition on the Al2O3(0001) (Fig. 9). The V 2p3@2 (LIII-edge) and V 2p1@2 (LII-edge) core-level transition energies of 517.5 and 524.4 eV in the multilayer coverage regime are in excellent agreement with the values expected for V2O3 exceeding 1000 s also indicates the formation of V2O3 ,43 as discussed above for vanadia formation on TiO (110).In the low coverage regime these features are not as pronounced as for the 2 higher exposures perhaps suggesting a reduced symmetry in the submonolayer regime however. 3.3 Electron energy loss spectroscopy Electron energy loss spectra both in energy loss regimes for vibrational and electronic transitions are in general agreement with the above conclusions regarding growth of the oxide overlayer. The expected strong Fuchs»Kliewer vibrational excitations are seen at 53 and 94 meV in accord with previously reported results (Fig. 10).50,51 For the sake of comparison spectra were recorded for vanadium metal evaporated onto the surface both with and without oxygen in the background.Based on the attenuation of the titanium Auger features we estimate monolayer coverage to occur at about 1000 s exposure to vanadium. Conducting adlayers are known to attenuate the Fuchs» Kliewer modes rapidly. However the presence of these losses even after 3000 s exposure to vanadium in the oxygen atmosphere indicates either that a uniform monolayer coverage is not Faraday Discuss. 1999 114 67»84 76 e Fig. 8 The oxygen K-edge spectra for clean TiO (110) and V2O3-covered TiO2(110). The spectra were –tted with Gaussian peaks in order to evaluate the relative transition intensities from the core-levels into the t2g and 2 orbitals respectively. g attained (assuming the vanadia to be conducting) or that the vanadia –lm is not sufficiently metallic to screen the Fuchs»Kliewer modes.Furthermore up to a vanadium exposure of 3000 s there is no detectable shift in the frequency of any of the vibrational losses. Thus the HREELS results are consistent with a growth mode in which the surface is populated with a patchy distribution of the oxide. Electronic spectra show clear evidence for an electronic transition accessible within the band gap which arises from the formation of the overlayer of vanadium oxide. This feature which is due to d»d transitions in the vanadium oxide grows with the amount of vanadia deposited and after 1000 s exposure to vanadium in the presence of background oxygen it appears as a prominent feature at 2.0 eV (Fig. 11). Strong transitions centered at 5.2 and 10.0 eV also appear corresponding to electronic transitions from O 2p to metal 3d states.The energies for the latter transitions are slightly lower than those observed for the clean surface. Decomposition of the losses observed for the vanadia-covered surface into those for TiO and vanadia is difficult because the relative 2 loss intensities for each of the two oxides are not known. However the shift of the band centers to lower energies with vanadia deposition is consistent with the NEXAFS results which show the O 2s to metal 3d transitions at a lower energy for the vanadia overlayer. 3.4 STM and LEED STM was employed to study the manner in which vanadium deposits grow on TiO (110)»(1]2) 2 and alumina using the alumina –lm grown on NiAl(110) as a model for a bulk single crystal of alumina.52 STM of the thin alumina –lm reveals smooth terraces interrupted by steps (Fig.12). Since these results are relevant to the growth of vanadia on these surfaces they are included here with STM studies of vanadia growth on the alumina thin –lm. Within a given terrace are large domains of the alumina –lm joined by boundaries that are clearly imaged by STM. A series of 77 Faraday Discuss. 1999 114 67»84 Fig. 9 L-edge spectra for vanadia deposited on Al for vanadium exposures given in the caption of Fig. 3. 2O3(0001) Fig. 10 Electron energy loss vibrational spectra for vanadium deposition times of 0 400 s 1000 s and 3000 s in the presence of background oxygen. Monolayer coverage occurs at exposures near 1000 s. Faraday Discuss.1999 114 67»84 78 at room temperature in oxygen. Spectra a»o are Fig. 11 Electronic energy loss spectra for clean and vanadia covered TiO (110). 2 STM images taken for diÜerent exposures of the alumina –lm to metallic vanadium at room temperature shows that vanadium prefers to nucleate in small clusters on the surface. The particle diameter appears to be nearly identical at all three coverages studied (20»30 Aé ). The particle density and height however depend on the amount of metal deposited growing larger with exposure. One such image obtained following an exposure of approximately 0.10 ML is shown on the Fig. 12 An STM image of the thin alumina –lm grown on NiAl(110). The image shows terraces criss-crossed by two-dimensional single-crystal domains of the alumina –lm as well as steps (constant current image ]3.0 V bias 0.4 nA).79 Faraday Discuss. 1999 114 67»84 left-hand side of Fig. 13. These results directly con–rm the growth mode to be of the simultaneous multilayer type. The steps and the domain boundaries are not preferentially decorated by the deposits. Rather the vanadium particles decorate the terraces randomly. Exposure of the –lm to the same vanadium exposure with a background pressure of 10~6 Torr oxygen produces a similar size and distribution of islands on the surface as shown on the right-hand side of Fig. 13. There is no detectable diÜerence in the size of the islands for the metallic vanadium or the vanadium oxide; however the corrugation of the oxide particles appears less than that of the metal particles (see the bottom of Fig.13). Neither is there a preference for the growth of the small oxide islands at defects in the surface structure of the alumina –lm. The similarity of the islands formed with or without ambient oxygen suggests that metallic vanadium is oxidized after forming small clusters. Fig. 13 STM images of vanadium (left-hand side) and vanadia deposited in submonolayer quantities onto the thin alumina –lm grown on NiAl(110). The particle sizes are similar in both instances. Faraday Discuss. 1999 114 67»84 80 LEED results (not shown) indicate that the underlying structure of the thin alumina –lm is not degraded signi–cantly at low coverages of the vanadium or vanadia adlayer. Though the intensities of the diÜraction spots in the complex LEED pattern of the alumina thin –lm diminish with exposure to vanadium (both with and without an oxygen ambient) they are not rapidly extinguished showing all diÜraction spots with signi–cant intensity up to coverages of 0.12 ML.Even at 0.30 ML diÜraction features characteristic of the clean surface are readily detectable though a bright diÜuse background has emerged. Thus the surface structure in the interstitial space between the oxide (or metallic) islands seems to be relatively unperturbed by the presence of the overlayer at appreciable coverages. At a vanadium coverage of 0.5 ML the diÜraction spots are no longer visible however. STM images indicate a similar pattern in the initial stages of deposition of vanadium on (110)»(1]2).In Fig. 14 an image of the clean surface obtained after sputtering at room rows indicate the presence of undercoordinated Ti cations. TiO2 temperature and annealing near 1100 K shows a (1]1) terrace (1]2) overlayer growth and local (2]2) structures in agreement with results of others.53h54 Fig. 15 shows the STM images for the TiO (110)»(1]2) surface prepared as described above but after annealing to 1150 K (a) 2 before and (b) after a vanadium exposure of 0.05 ML. Bright rows parallel to the [001] direction with a periodicity of 13 A Aé and a corrugation of 1 é along the [110] axis cover the surface nearly completely. The rows are terminated by small protrusions 10 A Aé in diameter and 3 é high or by step edges. In agreement with the STM images LEED indicates the formation of a (1]2) structure.The bright rows in the STM images of the clean TiO (110)»(1]2) surface have been assign- 2O3 .55,56 This interpretation of the structure was chosen over a ed to the added rows of Ti 2 competing model in which the surface is proposed to be severely reduced because the (1]2) surface does not react with formic acid. This reaction would be expected for the reduced surface as even the stoichiometric surface dissociates the acid on exposed titanium cations. The protrusions terminating the added Ti2O3 After the deposition of submonolyer amounts of vanadium additional protrusions decorating the bright added»rows of the (1]2) phase appeared in the STM images (Fig. 15). The bright features are approximately 1 nm wide and 1»2 nm long depending somewhat on the tip condition with a corrugation of 0.1 nm similar to the structure at the terminus of a (1]2) row.At the lowest coverage the elongated features apparently arise from clusters of vanadium atoms. If these features are caused by single vanadium atoms the extent to which they perturb the electronic structure must be signi–cantly greater than their ionic radius. However we must admit the possibility that these circular features arise from the nucleation of several vanadium atoms since the Fig. 14 An STM image of the clean TiO2(110)»(1]2) surface showing strings of the (2]2) structure as well as a shear plane (constant current image ]3.2 V bias 0.2 nA). 81 Faraday Discuss. 1999 114 67»84 Fig. 15 An STM image of 0.1 ML of vanadium deposited onto TiO (110)»(1]2) at room temperature.The 2 vanadium atoms appear to form small clusters located preferentially on top of the (1]2) added rows. (constant current image ]1.8 V bias 0.3 nA). (a) Before and (b) after a vanadium exposure of 0.05 ML. atoms must have sufficient mobility to migrate preferentially to the top of the added rows. The density of these features increases with the vanadium exposure whereas the size remains nearly constant and similar to the hill-like structures terminating the rows. Due to the high oxygen affinity of vanadium the most likely adsorption site is on top of the oxygen covered Ti2O3 rows where vanadium could achieve the maximum coordination with oxygen. Thus the additional protrusions on top of the rows are attributed to vanadium.As discussed above we believe that even submonolayer amounts of vanadium are oxidized to V2O3 . In view of the similar structures exhibited by titanium and vanadium (Ti2O3 and V have the corundum structure22) it is not 2O3 too surprising that the features arising from the deposition of vanadium resemble the ìadded rowœ Ti2O3 structures though it is not yet clear exactly how vanadium is incorporated. The appearance of a linear arrangement of these features along the [001] direction however is further indication that they preferentially associate themselves with the underlying (1]2) structure. Faraday Discuss. 1999 114 67»84 82 As the vanadium coverage approaches one ML these bright features coalesce the surface becomes densely covered with vanadium clusters whose irregular shape seems to re—ect the surface symmetry (not shown).On average the clusters have a height of 2.5 Aé and a diameter of 20 Aé indicating that 1»2 ML thick patches of vanadium and/or vanadia cover the surface. Due to this narrow height distribution the step structure of the underlying surface is still easy to recognize even after vanadium exposures of one ML. Furthermore even at this coverage there is a clear preference for the added vanadium to align along the (1]2) rows of the underlying titania surface. The morphologies adopted by vanadium deposited on the alumina –lm and the TiO (110) 2 single crystal are quite similar. In both cases a small number of vanadium atoms appear to coalesce into small particles dispersed more or less uniformly over the surface.Since vanadium has a high affinity for oxygen surface migration of the metal atoms is limited. Three dimensional growth of large aggregates is not permitted. Acknowledgements The authors gratefully acknowledge the support of the National Science Foundation through NSF CTS-9618807 and through the MRSEC program administered by the Center for Materials Research at Stanford. J.B. and M.B. thank the Deutsche Akademischer Austauschdienst (DAAD) and the Deutsche Forschungsgemeinschaft (DFG) respectively for research fellowships. The authors are grateful to Dr. Tom Kendelewisc and Prof. Gordon Brown for assistance with the synchrotron experiments and Prof. H. Freund for loan of the NiAl single crystal. References 1 F. Ernst Mater.Sci. Eng. R 1995 14 97. 2 U. Diebold J.-M. Pan and T. E. Madey Surf. Sci. 1995 331ñ333 845. 3 T. E. Madey U. Diebold and J.-M. Pan Springer Ser. Surf. Sci. 1993 33 147. 4 M. Baé umer J. Libuda and H.-J. Freund NAT O ASI Ser. Ser. E 1997 331 61. 5 Z. Zhang and V. E. Henrich Surf. Sci. 1992 277 263. 6 D. Robba D. M. Ori P. Sangalli G. Chiarello L. E. Depero and F. Parmigiani Surf. Sci. 1997 380 311. 7 N. Price and R. J. Madix J. Electron Spectrosc. Relat. Phenom. 1999 98/99 257. 8 C. T. Campbell Surf. Sci. Rep. 1997 27 1. 9 U. Diebold and N. D. Shinn Surf. Sci. 1995 343 53. 10 M. W. Finnis J. Phys. Condens. Matter 1996 8 5811. 11 G. Pacchioni and N. Roé sch Surf. Sci. 1994 306 169. 12 C. Verdozzi D. R. Jennison P. A. Schultz and M. P. Sears Phys.Rev. L ett. in the press. 13 J. Libuda F. Winkelmann M. Baé umer H.-J. Freund T. Bertrams H. Neddermeyer and K. Mué ller Surf. Sci. 1994 318 61. 14 R. M. Jaeger H. Kuhlenbeck H.-J. Freund M. Wuttig W. HoÜmann R. Franchy and H. Ibach Surf. Sci. 1991 259 235. 15 J. Biener M. Baé umer P. Liu E. Nelson T. Kendelewisz G. Brown and R. J. Madix Surf. Sci. in the press. 16 G. van der Laan and I. W. Kirkman J. Phys. Condens. Matter 1992 4 4189. 17 F. M. F. de Groot J. Faber J. J. M. Michiels M. T. Czyz5 yk M. Abbate and J. C. Fuggle Phys. Rev. B Condens. Matter 1993 48 2074. 18 H. Poelman K. Devriendt L. Fiermans O. Dewaele G. Heynderickx and G. F. Froment Surf. Sci. 1997 377 819. 19 M. Sambi E. Pin G. Sangiovanni L. Zaratin G. Granozzi and F. Parmigiani Surf.Sci. 1996 349 L169. 20 M. Sambi G. Sangiovanni G. Granozzi and F. Parmigiani Phys. Rev. B Condens. Matter 1996 54 13464. 21 M. Sambi G. Sangiovanni G. Granozzi and F. Parmigiani Phys. Rev. B Condens. Matter 1997 55 7850. 22 V. E. Henrich and P. A. Cox in T he Surface Science of Metal Oxides Cambridge University Press Cambridge 1994. 23 C. D. Wagner W. M. Riggs L. E. Davis J. F. Moulder and G. E. Muilenberg in Handbook of X-Ray Photoelectron Spectroscopy Perkin-Elmer Eden Prairie MN 1979. 24 S. Hué fner in Photoelectron Spectroscopy Springer Verlag Berlin Heidelberg New York 1996 p. 5. 25 J. T. Mayer U. Diebold T. E. Madey and E. Garfunkel J. Electron Spectrosc. Relat. Phenom. 1995 73 1. 26 J. G. Chen Surf. Sci. Rep. 1997 30 1. 27 M. P. Seah Surf.Sci. 1972 32 703. 28 C. Argile and G. E. Rhead Surf. Sci. Rep. 1989 10 277. 83 Faraday Discuss. 1999 114 67»84 29 CRC Handbook of Chemistry and Physics ed. R. C. Weast CRC Press Boca Raton FL 76th edn. 1995»1996. 30 T. M. French and G. A. Somorjai J. Phys. Chem. 1970 74 2489. 31 E. Gillet and B. Ealet Surf. Sci. 1992 273 427. 32 M. Gautier G. Renaud L. P. Van B. Villette M. Pollak N. Thromat F. Jollet and J.-P. Duraud J. Am. Ceram. Soc. 1994 77 323. 33 J. Ahn and J. W. Rabalais Surf. Sci. 1997 388 121. 34 W. Goé pel Surf. Sci. 1984 139 333. 35 L. S. Dake and R. J. Lad Surf. Sci. 1993 289 297. 36 G. A. Sawatzky and D. Post Phys. Rev. B Condens. Matter 1979 20 1546. 37 C. Noguera in Physics and Chemistry of Oxide Surfaces Cambridge University Press Cambridge 1996.38 G. K. Wertheim S. B. DiCenzo and S. E. Youngquist Phys. Rev. L ett. 1983 51 2310. 39 G. K. Wertheim S. B. DiCenzo and D. N. E. Buchanan Phys. Rev. B Condens. Matter 1986 33 5384. 40 S. A. Flodstroé m C. W. B. Martinsson R. Z. Bachrach S. B. M. Hagstroé m and R. S. Bauer Phys. Rev. L ett. 1978 40 907. 41 C. M. Kim B. D. DeVries B. Frué hberger and J. G. Chen Surf. Sci. 1995 327 81. 42 J. G. Chen C. M. Kim B. Frué hberger B. D. DeVries and M. S. Touvelle Surf. Sci. 1994 321 145. 43 M. Abbate H. Pen M. T. Czyz5 yk F. M. F. de Groot J. C. Fuggle Y. J. Ma C. T. Chen F. Sette A. Fujimori Y. Ueda and K. Kosuge J. Electron Spectrosc. Relat. Phenom. 1993 62 185. 44 F. M. F. de Groot J. C. Fuggle B. T. Thole and G. A. Sawatzky Phys. Rev. B Condens. Matter 1990 42 5459. 45 F. M. F. de Groot J. Electron Spectrosc. Relat. Phenom. 1994 67 529. 46 F. M. F. de Groot Z. W. Hu M. F. Lopez G. Kaindl F. Guillot and M. Tronc J. Chem. Phys. 1994 101 6570. 47 F. M. F. de Groot J. C. Fuggle B. T. Thole and G. A. Sawatzky Phys. Rev. B Condens. Matter 1990 41 928. 48 F. M. F. de Groot M. Grioni J. C. Fuggle J. Ghijsen G. A. Sawatzky and H. Petersen Phys. Rev. B Condens. Matter 1989 40 5715. 49 J. Stoé hr in NEXAFS Spectroscopy Springer-Verlag Berlin Heidelberg 1996. 50 G. Rocker J. A. Schaefer and W. Goé pel Phys. Rev. B Condens. Matter 1984 30 3704. 51 P. A. Cox R. G. Egdell S. Eriksen and W. R. Flavell J. Electron Spectrosc. Relat. Phenom. 1986 39 117. 52 M. Baé umer J. Biener and R. J. Madix Surf. Sci 1999 432 189. 53 M. Sander and T. Engel Surf. Sci. 1994 302 L263. 54 H. Onishi and Y. Iwasawa Surf. Sci. 1994 313 L783. 55 H. Onishi K. Fukui and Y. Iwasawa Bull. Chem. Soc. Jpn. 1995 68 2447. 56 K.-O. Ng and D. Vanderbilt Phys. Rev. B Condens. Matter 1997 56 10544. Paper 9/02737H Faraday Discuss. 1999 114 67»84 84

ISSN:1359-6640    

DOI:10.1039/a902737h

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

General Discussion Prof. Thornton opened the discussion of the Introductory Lecture A simple explanation for the eÜect of OH on the growth of Rh on the thin alumina –lm might be that OH saturates the defects. What models do you have for the eÜect ? Prof. Freund responded STM images of Rh deposits on pre-hydroxylated alumina –lms clearly show that Rh nucleates on the entire surface uniformly. This is not compatible with a preferred site nucleation where the site is formed by hydroxylation of the point and line defects of the clean alumina. Prof. Madix asked Prof. Freund you nicely described that adsorption of molecules onto oxide surfaces may be strongly aÜected by electrostatic interactions. Is it possible or is there evidence that there may be stable adsorption states for which there is an activation barrier largely electrostatic that could be accessed at higher pressures or higher gas kinetic energy? Prof.Freund answered It is quite conceivable that electrostatic interactions between molecules and oxide surfaces establish an activation energy for adsorption which could be overcome by increasing the kinetic energy of the incoming molecule. Dr Taylor commented On the subject of the desire for experimental measurements of the dynamics of adsorption at oxide surfaces for example methane dissociation. It may be worth noting that preliminary results of such experiments are reported by my group in a poster at this meeting. An exposure of approximately 500 L of CD4 using a seeded supersonic molecular beam (T B800 K E kJ mol~1) produced no evidence of dissociation at an MgO(100) surface transB170 at 300 K.Work continues but these early results suggest that methane dissociation at MgO surfaces does not occur simply via an early transition state. Dr Egdell said A third strategy for the preparation of ordered oxide surfaces is based on oxide-on-oxide epitaxy typically involving metal deposition on an ordered oxide substrate in the presence of an oxygen atom source. This technique has been used with particular success by Chambers and co-workers to prepare for example ordered (110) surfaces of Nb-doped TiO2 1 and RuO2 .2 Of course impetus to this –eld comes from the need to prepare oxide thin –lms with novel superconducting magnetic or ferroelectric properties. 1 Y. Gao Y.Liang and S. A. Chambers Surf. Sci. 1986 348 17. 2 Y. J. Kim Y. Gao and S. A. Chambers Appl. Surf. Sci. 1997 120 250. Prof. Freund added Dr Egdell is completely right. This is an important –eld which is not considered in the Introductory Lecture. Prof. Thomas addressed Prof. Freund Your reference to hydroxylated oxide surfaces prompts me to recall the classic work of Bradley and Serratosa on single crystal kaolinite. Some 35 years ago they did elegant work using polarised infrared absorption spectroscopy and were able to deduce the precise geometry of the dangling OH bonds at the aluminium rich (and the total absence of OH bonds at the silica rich) surface. My point is simply to draw attention to the convenience and ease of using (plentiful cheap and readily available) naturally occurring minerals containing OH rich surfaces.It is also relatively easy to introduce substitutional point defects (such as Cr3` Fe2` or Fe3` Zn2` or Mg2` ions in place of Al3`) into synthetic variants of these minerals which usually have conveniently large surface areas. 85 Faraday Discuss. 1999 114 85»103 This journal is( The Royal Society of Chemistry 2000 Prof. Freund responded This comment is particularly useful and the possibility to use hydroxides has been recognized in connection with the preparations of OH-terminated surfaces. Papp and co-workers (refs. 58 59 in the Introductory Lecture) have used Ni(OH) as a template to 2 prepare NiO crystallites with preferential (111) orientation. Prof. Bowker said You have nicely described the weak adsorption of CO on oxide surfaces.Perhaps for efficient adsorption ionic-type reactions lead to stronger reactivity. For example acid» base reactions which result in OH and ultimately water formation may lead to stronger absorption and this seems to be the case for formic acid adsorption on TiO (110) where the reaction 2 may result in ionic species formation. The example you mention is 2 Prof. Freund responded Interaction of molecules with perfect oxide surfaces is not necessarily weak as was demonstrated in the case of CO and Cr2O3(0001). another example and is mentioned in refs. 28»31 in the Introductory Lecture. In general activity is increased by introducing defects. Prof. Iwasawa asked As for the optimum feature of CO adsorption quantity against the diÜerent Rh particle size you showed did you assume no restructuring of the Rh particles during CO adsorption? Prof.Freund replied We had reported earlier a restructuring of Rh particles under exposure to CO as observed by SPA-LEED (refs. 15 99 of the Introductory Lecture). The eÜects were found to be smaller using STM. However we do observe the formation of Rh(CO) by IRAS which is 2 assumed to be involved in restructuring processes (refs. 157 158 of the Introductory Lecture). Dr Kantorovich opened the discussion of Prof. Finnisœs paper (1) I understand that you consider the solid/gas interface using DFT. Do you actually include species in the gas phase into the cell in your calculation. Can you explain the trick ? (2) Does it actually mean that the gas phase does not interact with the solid one so that there is no eÜect of the gas on the structure of the surface ? Prof.Finnis responded (1) No species in the gas phase are included explicitly in the calculation. The trick is that the cell in which calculations are done is relatively small containing only a few atoms of solid. The probability of –nding a gas molecule inside this cell is negligible so in the expression for the interfacial energy eqn. (1) of our paper only molecules in the solid phase need be counted. The gas phase is included purely conceptually in the sense that I related the pressure of the gas to the chemical potential of oxygen which does enter eqn. (1). (2) In reality the gas pressure determines the excess (or de–cit) of oxygen at the surface such that the gas is in equilibrium with the surface of lowest excess free energy.In calculations by the present formalism the eÜect of the gas on the surface is indirect insofar as the value of its pressure selects a lowest free energy structure of the surface from the set of relaxed structures which we choose to simulate not from all possible structures. Of course there is no guarantee that we have not missed a lower energy structure ! In any case I donœt think it is necessary to treat the gas explicitly. Prof. Catlow said You estimate the free energy of the oxygen molecule via a thermochemical cycle using calculated energies of two solid phases. Could you con–rm that this is a more accurate procedure than using a direct calculation on the O molecule.2 2 2. A2O3 On the other hand the energies of aluminium metal andlcan be very well calcu- Prof. Finnis replied De–nitely. The local density approximation (LDA) which we use is not good for the energy of formation of an O molecule. Hence we avoid calculating any properties of O or O lated with the LDA. Dr Willock asked The reaction of an Al terminated (0001) surface with H2O to give a fully hydroxylated surface was found in our work to lead to a much more stable surface than the Al Faraday Discuss. 1999 114 85»103 86 terminated surface (GGA DFT calculation). Could you compare the exchanges of oxygen mediated by H rather than 2O O2 with your method? Prof. Finnis answered In principle yes. One would have to include the chemical potential of H2O in eqn.(1) instead of or in addition to that of oxygen. My suggestion would be then to proceed with a calculation for solid ice in order to set the zero of chemical potentials by analogy with the calculations I have described for solid Al and alumina and then use thermodynamic tables to integrate up to the chemical potential of water at the required conditions. I believe a satisfactory calculation for ice can be made within the generalised gradient approximation (GGA). Prof. Vanderbilt asked Can you clarify what sets the lower limit of oxygen partial pressure in your calculations ? Was it the nucleation of particles of aluminium metal as I would have thought? Prof. Finnis answered Thatœs right. Prof. Catlow said Your calculations (and others using both electronic structure and static lattice simulation methods) correctly predict (at least qualitatively) the large surface relaxations that have been observed for Al2O3 .There are however quantitative discrepancies between calculated and experimental relaxations. Could you comment on the possible origins of these discrepancies and also on the diÜerences between the results obtained using the electronic structure methods and the earlier shell model/static lattice techniques. Prof. Finnis replied Some discrepancies on the theoretical side between diÜerent calculations can be due to diÜerent thicknesses of slabs and earlier electronic structure calculations used inadequate bases or did not relax the oxygen atoms laterally. These problems are under control in the more recent calculations and we do not understand all the discrepancies that remain.However it turns out that these relaxations are really rather soft so that diÜerences between diÜerent pseudopotentials or LDA errors could make a diÜerence e.g. between the variations between 3 and 11% for the second layer relaxation. We have veri–ed that diÜerences of this magnitude can arise. This is seen in the paper by Harrison et al. on TiO at this meeting. We 2 cannot explain the uniformly too large predicted relaxation of the surface compared to experiment unless the experimental surface is contaminated with OH for example. Prof. Harrison suggested Might thermal vibrations be important in the comparison between calculated and experimentally determined geometries ? Prof.Finnis replied I think so and your paper has also indicated this. However for the discrepancy I have just referred to it seems doubtful that thermal vibrations are the answer because according to classical potential calculations within the quasiharmonic approximation1 thermal eÜects do not improve the agreement. We are currently calculating some ab initio vibration frequencies of an alumina slab to investigate this point. 1 N. Allan private communication. Dr Gautier-Soyer said (1) First I have a question on the driving forces of the relaxations of the Al-terminated a-Al2O3 (0001) surface. They have been calculated by classical ionic pair potential methods as well as ab initio electronic structure calculations. They have also been measured experimentally.However even though theory and experiment qualitatively agree there is no consensus about the driving forces of the relaxations. Your interpretation is in favour of an electrostatic eÜect while other papers such as ref. 9 explain them in terms of rehybridization. Using a semiempirical self-consistent Hartree»Fock method carried out on a two unit cell thick slab (collaboration with C. Noguera) we have found that delocalization eÜects were not negligible so that their contribution should be an important part in the driving force of the relaxations. Would you like to comment on this point ? (2) Secondly I would like to make a comment on the covalence/ionicity of the Al»O bond at the surface compared to the bulk. Again in the electronic structure papers where atomic charges 87 Faraday Discuss.1999 114 85»103 QAls\3[3Ds where D is the charge s are calculated there is some disagreement some authors conclude a more covalent bond others a more ionic one (as in this paper). However if one thinks in terms of charge transfer per Al»O bond it seems that the studies are in agreement. According to the recent work of Pojani et al. a simple expression can be used to derive the anion»cation charge transfer D from the formal ionic charge and the number of –rst neighbours at the surface Q transfer between surface aluminium and subsurface oxygen. In the bulk Als\3[6D where D is the bulk charge transfer. In this way D\0 in the purely ionic state and the increase of D means a more covalent bond. Taking the charges obtained by diÜerent ab initio electronic structure calculations (ref.6 of your paper ref. 2 given here as well as your paper) the charge transfer per Al»O bond at the surface is larger than in the bulk. Using the charges of Table 2 (of your paper) one gets for example Ds\0.426 electron at the surface and D\0.245 electron in the bulk. This means that the Al»O bond is more covalent at the surface than in the bulk. How would you reconcile this point of view with your conclusion that there is no sign of increased covalence at the alumina surface ? 1 A. Pojani F. Finocchi J. Goniakowski and C. Noguera Surf. Sci. 1997 387 354. 2 W. C. Mackrodt Philos. T rans. R. Soc. L ondon Ser. A 1992 341 301. Prof. Finnis responded (1) In the spirit of the Hellmann»Feynman theorem we can describe and explain exactly the relaxation as a purely electrostatic eÜect given the exact self-consistent charge density.On the other hand in a tight-binding model some of the electrostatic energy of which the force is the derivative is hidden in the band energy. The Hellmann»Feynman forces in tight-binding do not appear to be electrostatic they involve instead the gradients of some hopping integrals although they are actually modelling forces which are simply electrostatic. It is therefore difficult to tell how much of the forces in tight-binding should be interpreted as the classical electrostatic forces from classical ions. From the ab initio point of view suppose we divide up the force into the contribution from a classical charge distribution (the spherical superimposed ions) and the contribution due to the redistribution of the electronic charge.The fact that the –rst contribution explains the relaxation suggests to me that the second contribution which includes rehybridisation eÜects is small. (2) First let me repeat the caveat which we all seem to agree about that absolute numbers for charges and charge transfer in this kind of analysis are not to be interpreted literally because they are basis dependent. Your question however concerns relative values and these should be meaningful. My answer is that I think the simple formula you have used to measure the ionicity has a particularly arbitrary character at the surface because the coordination of the surface atoms is diÜerent from the bulk.For example the coordination of the surface aluminiums is 3 whereas their bulk coordination is 6. You have used these numbers in your estimate of charge transfer along the bonds. Consider a hypothetical outcome of a calculation in which all the Al atoms both bulk and surface carried the same Mulliken charge and all the O atoms likewise (e.g. ]1.5 and [1 respectively). We would interpret that by saying that the ionicity at the surface is the same as that in the bulk just because the same charge X (e.g. 1.5) has been transferred from each Al atom. The surface Al atoms might be thought of as having transferred this charge X equally through 3 bonds X/3 through each whereas the second layer Al atoms might be thought of as having transferred X/6 through each of 6 bonds.By this measure the second layer bonds are twice as covalent. But we could just as well say that the second layer Al transfer their X through only the three bonds to their nearest neighbours the three O atoms in the layer above them and none to the three more distant O atoms below them in which case all the bonds would have the same ionicity. The resulting charges on the O atoms would be the same. It is my impression that either construct is arti–cial and not particularly helpful for the purpose of de–ning ionicity since they both rely on the unphysical concept of charge transferring along a particular bond. Prof. Harrison addressed Dr Gautier-Soyer From your analysis of the covalent vs. ionic nature of the bond can we understand the reason for the discrepancy between theory and experiment in the surface geometry? Dr Gautier-Soyer replied Unfortunately we donœt.Faraday Discuss. 1999 114 85»103 88 The bonding at the surface becomes more covalent and less ionic. This represents surface structure which surface. Dr Weiss addressed Prof. Finnis It is very interesting that your calculations predict an oxygenterminated a-Al2O3(0001) surface to become stable in high oxygen pressure environments. A similar result has been obtained from recent DFT calculations performed for the a-Fe2O3(0001) surface,1 an oxide with the same corundum crystal structure as Al2O3(0001). These calculations predict the formation of strongly relaxed surfaces exposing outermost close-packed oxygen (111) layers which in a purely ionic picture are expected to be unstable because of in–nite electrostatic surface energies.However your calculations reveal a strongly changed electron density of states at the surface and a considerably reduced Mullikan ionic charge for the topmost oxygen layer atoms of Al2O3(0001). a mechanism that can stabilise the polar oxygen-terminated Al2O3(0001) also stabilises the polar Fe2O3(0001) We have recently performed a systematic study of a-Fe2O3(0001) surface structures that are formed in de–ned oxygen partial pressures ranging over 5 orders of magnitude.2 Fig. 1 here shows a large scale (a) and an atomic resolution STM image (b) of a surface prepared in 1 mbar oxygen. More than 95% of the entire surface forms an oxygen terminated structure only small dark patches exposing an Fe-terminated surface can be seen in the large scale image on the left.In oxygen pressures between 10~1 and 10~5 mbar comparable amounts of iron- and oxygenterminated surface domains coexist in 10~5 mbar only the iron-terminated surface is formed. Recent dynamical LEED intensity calculations con–rm that the two terminations observed in these STM images correspond to oxygen- and iron-terminated surfaces in analogy to your results for Al2O3(0001). These –ndings clearly demonstrate that metal oxide surface structures formed in gaseous environments with high pressures can be very diÜerent from those formed under UHV conditions. It has important consequences on the surface chemical and catalytic properties of such materials and directly addresses the problem of the ìpressure-material gapœ in catalysis research.1 X.-G. Wang W. Weiss Sh. H. Shaikhutdinov M. Ritter M. Petersen F. Wagner R. Schloé gl and M. Scheffler Phys. Rev. L ett. 1998 81 1038. Fig. 1 90]90 nm (a) and 15]15 nm (b) STM images of an a-Fe2O3(0001) hematite surface prepared by heating an epitaxially grown hematite –lm to 1120 K in 1 mbar oxygen for 15 min. Several monoatomic steps can be seen in (a). An oxygen terminated surface structure dominates small patches of iron-terminated domains cover less than 5% of the entire surface (dark patches). 89 Faraday Discuss. 1999 114 85»103 2 Sh. Shaikhutdinov and W. Weiss Surf. Sci. 1999 432 L627. Prof. Finnis responded Thank you for those observations. It would be very interesting to see if you can observe the oxygen terminated surface at around atmospheric oxygen pressure.However it might only be stable at somewhat higher pressure. Prof. Vanderbilt commented I would like to urge caution about the notion that charge transfer can lead to self-compensation of a polar surface. This is misleading. It is necessary to distinguish two cases. One possibility is that the surface can self-compensate by becoming metallic. On the other hand if the surface remains insulating i.e. the electron chemical potential lies in a gap common to both the bulk and surface it can be shown that the net excess charge cannot be changed by charge transfer among surface atoms. This follows from a theorem relating this net surface charge to the surface-normal component of the bulk polarisation which does not change.Thus if the surface remains insulating no such self-compensation can occur. Prof. Finnis said I completely agree with that. Our calculated polar surfaces are indeed metallic and I donœt believe they could conceivably be stable otherwise for the reasons you describe. Prof. Jennison addressed Dr Gautier-Soyer Concerning the degree of covalency at the Al2O3 surface using DFT slab calculations we noted only small changes in the local DOS compared with the bulk ions indicating that the high degree of ionicity found in the bulk is preserved at the surface if the surface is fully relaxed. 2O 2 3 The very large surface relaxations which penetrate to the third oxygen layer make thin slab calculations questionable quantitatively.This places Al in considerable contrast to MgO (with high ionicity but almost negligible relaxations) and TiO (with signi–cant covalency). So I would ask you how thick was the slab in your calculations ? 2O3 In fact we computed the eÜect of relaxation on the adsorption energy of an isolated Pt atom and found by LDA that it was 2.0 eV for Al 0.4 eV for TiO (110) and only 0.2 eV for 2 MgO(100). Dr Gautier-Soyer replied The slab used in our calculation was two unit cell thick (about 26 ”) relaxed on both sides. So the slab was made of 12 (»Al»O»Al») stacking units which is rather thick. We checked that in the middle of the slab the Al and O charges were equal to the bulk ones as derived from the 3D band structure calculation.surface there are no hydroxy groups present as checked Prof. Freund said On the Cr2O3(0001) with IRAS. Therefore the presence of OH cannot be used to argue for any disagreement between experiment and theory. We did observe a new phonon mode on Cr2O3(0001) and we have strong indications through CO adsorption experiments that this is a surface phonon mode.1 1 K. Wolter PhD Thesis Fritz-Haber-Institut Berlin in preparation. Prof. Harrison asked Prof. Freund Can you deduce the symmetry of the vibrational mode at the chromia surface ? Prof. Freund answered The mode is active in specular re—ection indicative of a totally symmetric mode. Prof. Harrison added In which case it seems likely to be a vertical motion of the surface Cr ion. Prof. Freund replied Isotopic labelling of the oxide indicates that it is a mode involving mainly the motion of Cr ions.Prof. Thornton addressed Dr Renaud Perhaps you could comment on the level of agreement between theory and experiment taking into account the experimental error bars. Dr Renaud responded This is a question that is also related to previously published material.1,2 The best way to answer this question is actually to simulate the crystal truncation rod data of the Faraday Discuss. 1999 114 85»103 90 surface deduced from the diÜerent theoretical calculations of the relaxations and Al2O3(0001) compare them with all the experimental data with error bars included on the data. I intend to do that in the near future. 1 P. Gueç nard G. Renaud A. Barbier and M.Gautier-Soyer Mater. Res. Soc. Symp. Proc. 1996 437 15. 2 P. Gueç nard G. Renaud A. Barbier and M. Gautier-Soyer Surf. Rev. L ett. 1997 5 321. surface modelled in the Prof. Harrison asked How are the vibrational modes of the Al2O3 interpretation of your surface X-ray diÜraction experiment? Dr Renaud answered At the time of the data analysis no enhanced or anharmonic vibration mode was expected for the top plane Al atom and hence the thermal vibrations of the top atoms were modelled the standard way by two Debye»Waller factors one parallel and one perpendicular to the surface. Since the calculations on the TiO (110) surface suggest that anharmonic vibrations 2 could be present I will check this possibility for the sapphire surface in the near future. Prof.Friend opened the discussion of Prof. Jennisonœs paper Kinetic factors are important in metal thin –lm growth as clearly seen in the work presented by Prof. Freund in Behmœs work that you cited and also in our work on Co growth on oxidized Mo metastable structures can form. In your work you obtain a thermodynamic result. How do you relate your theory to experiment since your results are thermodynamic? Prof. Jennison responded I completely agree with your comment on the importance of kinetic considerations and we are moving towards that with some studies. However for these mixed metallic and ionic systems there are no adequate potentials to perform simulations so we have started with energetic studies to indicate likely structures the basic nature of the bonding etc.Clearly with our third topic which involves water dissociation hydrogenation and possible hydrogen evolution with metal deposition I agree that kinetics is essential to make contact with experiment. I would like to add that we know we are often interested in metastable structures ; for example in some of our work on the alumina –lms we have observed that with an adequate degree of freedom (a large unit cell) and aggressive geometry relaxation algorithms 2D –lms grow into 3D structures while we were trying to study the 2D metastable structures of experimental relevance. Dr Venables asked It is well known that DFT methods though maybe the best we have available form an uncontrolled approximation in that we donœt know how accurate they are in any particular case.Can you give us some idea of what accuracy you claim for the results shown in Fig. 2 of your paper. My queries include what con–gurations have been taken into account have long range electrostatic/elastic relaxation been included particularly in the surface plane and how has (local) charge compensation been achieved in the case of defect calculations ? I ask this because recent calculations we have undertaken for the notionally similar system Pd/MgO(001) have given a considerably smaller adsorption energy and the OH-defect is thought to come in pairs linked to a Mg2` vacancy to preserve local charge neutrality.1 In the latter case we found the Pd to bind at the Mg site with a trapping energy (relative to Pd on the terrace) of 1.15 eV i.e. very much lower than the values given in your Fig.2. I am left to wonder similarly whether your calculation also overestimates the reduction in metal pair binding from gas phase values which in our calculation is a relatively small eÜect. 1 J. A. Venables and J. H. Harding J. Crys. Growth 1999 in press. Prof. Jennison answered Well I disagree with your –rst statement concerning the accuracy for DFT in ìanyœ particular case because extensive comparisons have been made for a wide variety of materials mostly metals and semiconductors and I think the error bars are usually well known. With the aluminum oxides we and others have reported that LDA obtains the experimental lattice constant of sapphire for example to 0.2% while GGA does somewhat poorer being B0.6% over experiment»this shows electrostatics ionic radii and hard-wall repulsion are properly treated.We also know that DFT does an excellent job with metal polarization (see for example Jennison et al.1). I agree with you that there is more uncertainty for the case of metals 91 Faraday Discuss. 1999 114 85»103 interacting with oxide surfaces as here little is known concerning the energetics from the experimental side. In addition the rather large diÜerences in LDA vs. GGA metal adhesion energies (see our ref. 21) are difficult to understand. We have found that metals particularly on the right of the periodic table bind without any evidence of covalency (where we would expect LDA to overbind). Rather metals bind by becoming positively charged by donating electron density to the oxide itself as may be seen by examining the local DOS changes in the neighboring oxygen ions.The factors here are polarization ionization and electrostatics all well given by DFT. Concerning relaxation around defects on oxides here again we have electrostatics and ion repulsions both accurately given because the lattice constants are well given. Concerning now the results for Pt on MgO these are very accurate DFT calculations. We used slabs of –ve MgO layers and the supercell contained 36 atoms per layer. The vacuum gap which we –nd needs to be much larger than with calculations involving only metals was ” [15 . Because this was not a cluster geometric relaxation and long-range electrostatics are accurately and naturally included (see our ref.18). Concerning the importance of relaxations for adsorbates on MgO I reported that Pt adatom binding on the perfect (100) surface is only lowered by 0.2 eV by relaxation while the number is more like 0.4 eV for TiO (110) and 2 eV ( !) for Al2O3(0001). However relaxations are critically 2 important for obtaining the correct binding at surface vacancies. This is because the relaxations determine the height of the ionized metal adatom and the Madelung potential falls exponentially into the vacuum Thus the binding energy depends exponentially on the adsorbate height which in turn depends linearly on the outward relaxation at the vacancy. Concerning your speci–c queries the question of con–gurations (as in ìcon–guration mixingœ) does not enter this problem as an obvious DFT failing as the ionized Pt atoms are similar to s0d8 s0d9 s0d10 or s1d10 and are fractionally charged as they should be.Next as I said above the long range electrostatic/elastic relaxations have been accurately included. Finally the (local) charge compensation is not an issue as these supercells are all neutral (i.e. we did not consider the charged defect). 1 D. R. Jennison P. A. Schultz and M. P. Sears Phys. Rev. L ett. 1996 77 4828. Dr Shluger asked What was the charge of anion vacancies used in your calculations of the Pt adsorption and how does it aÜect the results ? Why does Pt prefer to adsorb in an anion vacancy? Prof. Jennison responded The anion vacancy had two trapped electrons the natural state as all our calculations used neutral supercells.We cannot compare energies between neutral and charged supercells because the VASP code (our ref. 30) can only compensate a charge cell by the common practice of adding a uniform background which therefore puts the calculations on an uneven footing. Pt adsorbs at either vacancy type by becoming ionized due to the Madelung potential and entering the vacancy to the maximum extent allowed by its ionic radius and the relaxation around the vacancy. However it does not prefer the anion vacancy to the cation vacancy (see our Fig. 2) binding several eV more strongly at the latter. Dr Egdell said Can your clarify why ìadsorbed hydroxyœ and ì in surface hydroxyœ diÜer in their propensity to promote metal nucleation. It would also be interesting to those concerned with photoemission from surface hydroxy species to know details of the location of the localised energy levels associated with the two diÜerent surface hydroxy species.Prof. Jennison answered Our calculations have shown (see also our ref. 18) that the in-surface hydroxy ion weakens the binding of metals because the binding is mostly electrostatic in nature (see our refs. 20 and 21) and the surface ionic charge is reduced from 2[ (oxygen ion) to 1[ (hydroxy ion). However ad-hydroxy ions strengthen the binding because they sit above the cation sites allowing the adsorbed metal to sit immediately adjacent and bind both vertically in the normal way to surface oxygen ions but also now laterally to the OH~; i.e. at the binding site the negative electrostatic potential is deepened.We have not yet published the local DOS of the two species but appreciate your suggestion for future work. Prof. Madix asked What is the origin of the ionization of the metals deposited on the alumina? Faraday Discuss. 1999 114 85»103 92 Prof. Jennison answered Electrostatics. Prof. Madix said That is very interesting. We have seen formations of V3` on single crystal with vanadium coverages well below a monolayer. Al2O3(0001) Prof. Jennison responded The electrostatic (Madelung) potential extends further above the surface with alumina than with other oxides for two reasons (1) the potential falls oÜ exponentially above a surface with strict layer-by-layer neutrality such as MgO(100) but only as 1/Z above a non-strictly-neutral surface and (2) the dramatically large relaxations of the alumina surface increase the non-neutrality near the adatom site (nearby Al-ion movements are about 1/2 ” downwards see our ref.21). These factors serve to substantially increase the ionizing potential at the metal site and it is this potential that can even multiply ionize isolated metal atoms as we reported for Y and Nb adatoms recently (our ref. 21). Prof. Thomas said My question arises from the remark we have just heard from Prof. Madix namely that when metallic vanadium is deposited on single crystal alumina individual V3` ions are formed. One can see from what Prof. Jennison has told us that the Madelung energy at the surface of the Al2O3 accounts for the production of Cu` or Pt` (in the presence of juxtaposed OH).But why does vanadium opt for V3` (not V2` or lower or V4` or higher) ? This presents a nice opportunity for theoreticians. Prof. Jennison responded We determined the ionicity of the adatoms by integrating the local DOS of the metal atoms when projected on atomic orbitals. Naturally we did not obtain integral values nor should we expect to since DFT is inherently a mean-–eld theory and the metal charge is determined by a balance between the ionizing potential ionic radius and the ability of neighboring atoms in the oxide to accept electron density. To obtain the approximate charge we rounded oÜ the integrated LDOS to the nearest integer which we reported in our ref. 21. This rounding amounted to typically one or two tenths of an electron.Prof. Campbell said In contrast to the implications of your calculations that surface OH greatly stabilises Cu on the oxide surface (alumina) our recent microcalorimetric measurements for the heat of adsorption of Cu on MgO(100) showed that surface OH has little eÜect on the coverage-dependent adsorption energy of Cu.1 (J. Musgrove D. Starr D. Bald J. Ranney and C. T. Campbell unpublished results). 1 J. Musgrove D. Starr D. Bald J. Ranney and C. T. Campbell unpublished results. Prof. Jennison responded If as I expect your hydroxylated surface contained even mixtures of ad-OH and in-surface OH (produced by water dissociation and the subsequent reaction of H` with O2~) I welcome this result. We recently computed the interaction of Cu adatoms with a sapphire(0001) surface saturated with water dissociation products and found it to bind Cu almost identically as the clean surface for while the ad-species strengthens the binding the in-surface species weakens it.So I see potential support here for the theoretical results. If as the calculations suggest is likely the presence of metal causes hydrogen gas to evolve from the in-surface hydroxy groups while this event would increase the binding of Cu by several eV per atom it is nearly neutral energetically (i.e. the heat of desorption of H about equals the increased heat of adsorp- 2 tion of the Cu) so I do not see that this would be seen in microcalorimetry. Prof. Madey asked My question concerns the general issue when is the surface of a thin oxide –lm a good model for the surface of a bulk crystal ? Experiments on surfaces of bulk Al2O3 are generally performed on a-alumina (sapphire samples).Your calculated structure for a two-layer alumina –lm is more like j-alumina. What are the conditions (choice of substrate –lm thickness T ) for which you expect thin sapphire –lms to be stable ? In particular is it possible to stabilize sapphire in a two-layer –lm? On another matter you indicate that water dissociates readily on Al2O3(0001) at 300 K but experimental evidence indicates a very low sticking probability. As Prof. Freund indicated in his 93 Faraday Discuss. 1999 114 85»103 talk in order to form hydroxy he ììseedsœœ his alumina surface with a fractional ML of metallic Al followed by a 20 000 L water dose.Prof. Jennison answered The calculations on the two-oxygen-layer ultrathin alumina –lms clearly suggest that the lateral Al-ion interactions dominate the site preference thereby producing a 50 50 mix of tetrahedral and octahedral site Al-ion occupations making the –lm j-like. Here because the problem of structure is essentially two dimensional the various choices are limited unless defects are introduced which violate stoichiometry or coordination. I would expect quantitative but not qualitative diÜerences in metal adsorption between this –lm and its bulk analogue of sapphire(0001) and we have already seen this (to be published). I could only guess that by perhaps six or eight oxygen layers the bulk –lm energy would be sufficiently large compared with the interfacial energy that if the surface energy permits,1 sapphire might form with sufficient annealing.You raise a very interesting issue Yes our results fail to –nd an energetic impediment for water dissociation on Al at 300 K nor did Hass et al.,2 as we would expect since this was also 2O3(0001) a DFT study. There is also evidence for a lack of sapphire hydroxylation in UHV as well as for the –lm studied in Prof. Freundœs group while hydroxylation does evidently occur with a sufficient partial pressure of water e.g. ref. 3 below. At the present time we do not know whether this is due just to sticking or also to the reaction dynamics of dissociation but this is a subject of ongoing work in our group. 1 J. M. McHale A. Auroux A. J. Perrotta and A. Navrotsky Science 1997 277 788.2 K. C. Hass W. F. Schneider A. Curioni and W. Andreoni Science 1998 282 265. 3 J. W. Elam C. E. Nelson M. A. Cameron M. A. Tolbert and S. M. George J. Phys. Chem. B 1998 102 7008. Prof. Freund added We could not detect water dissociation on a clean alumina –lm as well as on a well prepared clean single crystal alumina. Prof. Diebold said I –nd the theoretical results on the in—uence of point defects on nucleation and growth extremely interesting. My comment refers to the experimental data by JeÜ Kelberœs group. It is surprising that the Al surface would be hydroxylated with a coverage of 1 monolayer. 2O3 3 It is not straightforward to extract the coverage and even the presence of OH groups from nonmonochromatized O 1s XPS spectra.In addition it is difficult to hydroxylate Al2O3 under UHV conditions.1 I think you mentioned that breaks in AES uptake curves indicated a layer-by-layer growth mode. One should note that this procedure is far from reliable and has led to erroneous results in the past especially on oxide surfaces. 1 P. Liu T. Kendelewicz G. E. Brown Jr. E. J. Nelson and S. A. Chambers Surf. Sci. 1998 417 53. Prof. Jennison replied I agree that the breaks in the uptake curves do not prove layer-by-layer growth but are only suggestive (your point is discussed for example in ref. 1 below). What does seem clear however is the Kelber Auger data (our ref. 29) showing Cu` ions at about 1/3 ML»this cannot happen unless the Cu adatoms are bound so strongly that they cannot migrate at room temperature to join metallic islands and we have shown that a surface with a high coverage of ad-OH species is capable of doing just this.This structure is again consistent with the XPS but not proven by it. However whatever the species seen in the XPS it is at the surface (as shown by tilting the sample) and is at the correct energy to be ad-OH. Further work is needed here to de–nitively prove this and also to answer the question of how this surface came about in the –rst place for while this surface was examined in UHV it apparently was not hydroxylated in UHV. 1 C. Argile and G. E. Rhead Surf. Sci. Rep. 1989 10 277. Prof. Iwasawa commented Recently we reported the stabilization of Pt4` ions at a MgO surface replacing Mg2` ions at the top layer of the surface.The Pt4`/MgO sample was very active for catalytic combustion and oxygen-isotope exchange reactions. When the Pt4` ions were reduced to the metallic state Pt atoms were not stabilized as monomers anymore. Instead it was Faraday Discuss. 1999 114 85»103 94 suggested that six-atom Pt clusters were produced and attached on the MgO surface. The Pt/MgO sample was selective for catalytic dehydrogenation of propane butane and isobutane to the corresponding alkenes. The behaviour of Pt atoms is somewhat diÜerent from your theoretical consideration. Prof. Jennison responded This is very interesting but I think your surface might be more complex than the model surface we used in our ref. 18. I say this because the Madelung potential at the cation vacancy site is perhaps 20»30 V and is therefore insufficient to ionize Pt to 4] which would take perhaps 40 V since the second ionization potential of Pt is already at 19 eV.Therefore I suggest that perhaps Pt is not just substitutional for a surface Mg ion. The remainder of your observation concerning small stable metallic clusters we have not addressed but related calculations (our ref. 21) show that just two nearest-neighbour metal atoms are sufficient to reduce metal adatoms to the metallic state. Prof. Vanderbilt opened the discussion of Prof. Hermannœs paper Previous calculations of Lindan and co-workers have shown the importance of considering spin polarizations for oxygen vacancies in TiO2. V Can you clarify whether you took spin polarization into account for the 2O5 oxygen vacancy calculations and if so what is the spin structure that you –nd? Is it a spin triplet ? How diÜerent would the results have been if spin polarization were neglected ? Prof.Hermann responded All vacancy calculations were carried out allowing for spin polarisation of the complete cluster system. Here it was found that the energetically favourable cluster states were singlet spin states. We did not determine however local projections of spin polarisation in the vicinity of the vacancies which may well have yielded triplet type character. This is a valuable suggestion for future work. Dr Noguera asked For the simulation of oxygen vacancies did you include basis functions located at the vacancy site allowing the two electrons left behind by the missing oxygen to be possibly trapped there ? Prof.Hermann answered We performed the oxygen vacancy calculations without and with additional basis functions at the vacancy site. In the latter case we used both the original oxygen basis of the system without vacancy and a basis set augmented by diÜuse functions. This did not result in localised colour-centre type states where electrons could be trapped. The bridging oxygen vacancies at V2O5 surfaces represent large openings where electron localisation is more difficult than in compact systems like MgO where colour centres have been observed. Dr Shluger asked (1) How is the electron density localised in the neutral oxygen vacancy? Does it look more like an Eœ-centre in SiO or like an F-centre ? (2) Did you try asymmetric relaxations 2 where electrons could localise on one V ion ? (3) How could long-range lattice polarisation aÜect these results ? Prof.Hermann answered (1) As mentioned before we could not –nd localised colour-centre type vacancy states in the cluster calculations. Planar sections through the vacancy regions just show a depletion of charge density. (2) Asymmetric relaxation was a natural result of the optimisation of atom positions near the vacancy and could not be ììturned onœœ in the calculations. The electrons left behind after creating the vacancy did not yield sizeable charge accumulation near selected single V centres. (3) This could be answered only by large slab calculations which were not considered in the present study.In particular do you see the development of a 2 ? Prof. Freund said Considering the local density of states near the modelled vacancies does it start to resemble the density of states in VO feature near the Fermi energy (EF) in the density of states ? 95 Faraday Discuss. 1999 114 85»103 Prof. Hermann replied A detailed analysis of changes in the atom projected partial densities of states (PDOS) of the V10O31H12 cluster due to the presence of the three diÜerent oxygen vacancies O(1»3) has been performed after and stimulated by the discussion. It is found that vacancy formation results for all oxygen vacancy types in additional PDOS contributions near E (and located about 2 eV above the O 2sp band region) which are attributed to vanadium F centres and are of V 3d type.While this may be interpreted as ìVO resemblanceœ it should be 2 noted that VO and V2O5 are substantially diÜerent in their bulk crystal structure such that DOS 2 similarities cannot be taken as indicators of structural similarities. Dr Weiss said Your calculations reveal a strong covalent bonding character in V2O5 with atom charges of ]1.5 for V instead of ]5 expected in a purely ionic picture. The valence band is formed by O 2sp and V 3d states. I do not understand why the atom projected partial DOS related to the terminal O(1) atoms has the smallest energy width if compared to the O(2) and O(3) atoms although the V»O(1) distance is the smallest O»V distance occurring in the V2O5 lattice. I would expect the valence band width to increase with decreasing distance between neighbouring O and V atoms because of increasing orbital overlap integrals.Prof. Hermann replied The valence bandwidth of the O 2sp dominated region is determined by the lateral O»O interaction rather than by V»O coupling. The bridging oxygen atoms O(2) and O(3) form a sub-lattice with inter-atomic distances smaller than those of the terminal O(1) sublattice. Therefore the O(2,3) derived valence band dispersion must be larger than that of the O(1) sub-band as discussed in ref. 1. (note that this is ref. 30 of the manuscript). 1 K. Hermann M. Witko R. Druzinic A. Chakrabarti B. Tepper M. Elsner A. Gorschlué ter H. Kuhlenbeck and H.-J. Freund J. Electron Spectrosc. Relat. Phenom. 1999 98ñ99 245. the central peak of the O Prof.Kempter asked In your comparison of the computed V2O5 DOS with the intensity curves measured from angular resolved photoemission (ARUPS) on V2O5(010) 2sp band region diÜers by 1 eV. How can this shift be explained ? Prof. Hermann answered In the comparison the computed DOS the O 2sp band region has been rigidly shifted in energy such as to coincide with the intensity region of the ARUPS experiment. This is based on the fact that the computed valence bandwidth reproduces the experimental result. However a more detailed comparison is difficult since there are many factors contributing to diÜerences between theory and experiment. First a comparison of theoretical DOS with experimental photoemission results ignores the in—uence of transition matrix elements.Second the present ARUPS data refer to –xed normal electron emission and peaks may shift as a result of emission angle. Prof. Catlow said As is commented on in your paper the magnitude of the calculated oxygen vacancy formation energies are rather high. Do you have any explanation as to why the techniques used might be overestimating these energies ? Also has it been possible to make any quantitative comparison of your results with values derived from experimental thermochemical data such as heats of reduction of the oxide? ~x ]Ofree 0 ) range between 6.5 and 7.2 eV depending on the vacancy site. Early from extrapolations of kinetic data yield Prof. Hermann responded After careful comparison with related literature values we conclude that our calculations do not grossly overestimate oxygen vacancy formation energies.The proposed values corresponding to a transition of oxygen bound at the surface to a free neutral oxygen atom (Osurface experimental vacancy formation enthalpies of bulk V2O5 1.3»1.5 eV1 for a transition of oxygen bound at the surface to a –ctitious species 12 O2 . This translates to 3.9»4.1 eV for the transition Obulk ~x ]Ofree 0 . The experimental atomisation energy of bulk V2O5 gives an estimate of average V»O binding energies of 7.92 eV.1 Finally the experimental binding energy of the 4&~ state of the VO dimer amounts to 6.4^0.2 eV2 (our calculations yield 7.2 eV at the RPBE level). Only a modern experimental determination of oxygen vacancy formation energies at the V2O5(010) surface using a well characterised substrate without major Faraday Discuss.1999 114 85»103 96 imperfections (this may have been a problem with the early extrapolation1) can give a de–nitive answer as to the reliability of the present theoretical data. 1 P. Kofstad Nonstoichiometry DiÜusion and Electrical Conductivity in Binary Metal Oxides John Wiley New York 1972 p. 57 and 180. 2 G. Balducci G. Gigli and M. Guido J. Chem. Phys. 1983 79 5616. Prof. Harrison asked How accurate are the density functional calculations for the O and OH neutral species used as reference energies in your calculations ? Prof. Hermann answered In the calculations we have used gradient corrected functionals (GGA-II RPBE) which are known to yield meaningful total energies for atoms and small molecules such as O (3P reference) OH (2% reference).Thus we expect reliable values for adsorption 2 and vacancy formation energies in the present system. This has been con–rmed by very recent work on V surface properties1 where we have applied a revised version of the gradient cor- 2O5 rected Perdew»Burke»Ernzerhof functional2 and where binding energy results similar to the present data are obtained. 1 K. Hermann M. Witko R. Druzinic and R. Tokarz T op. Catal. in press. 2 B. Hammer L. B. Hansen and J. K. Norskov Phys. Rev. B 1999 59 7413. Prof. Finnis commented In fact I donœt think you need to calculate any properties of free oxygen atoms or dimers or pathways but you can use the same trick as we did and formulate the problem in terms of the chemical potential of oxygen.Prof. Hermann replied Thank you for the suggestion. Dr Cora` asked (1) When applying cluster models to simulate solid oxides the concept of saturating the terminal oxygen with H atoms is mutuated from the –eld of silica and zeolites. There the r O»H bonds replace the r O»Si bonds that linked the cluster to its crystalline environment. In transition metal oxides such as V2O5 in which the M ion has electronic con–guration d0 the frontier crystalline orbitals are those of p symmetry along the M»O direction. In saturating the molecular fragment with Hs we replace the p M»O bonds with a r O»H. Can you please comment on which eÜect this approximation is likely to introduce in the calculated bonding properties of the –nite V2O5 fragment? (2) A related question concerns the eÜect of the –nite cluster size and saturation with Hs on the calculated O abstraction energies that you have proposed in the paper.Even though the cluster termination with Hs may reproduce with sufficient accuracy the properties of the perfect V2O5 crystal the –nite cluster size limits the possibility to delocalise the eÜect of perturbations. In my calculations (described in a paper at this meeting) I have found that an important electronic delocalisation occurs via the p M»O crystalline orbitals. In the two-layer cluster upon reduction of the V ions (which follows the O abstraction) the reduced V ion binds to the O in the sub-surface layer. I would expect that a similar process propagates to the following layers ; this would substantially reduce the calculated O abstraction energy.Prof. Hermann replied (1) The concept of using hydrogen terminators to simulate cluster embedding is actually much older than silica and zeolite modelling. For example it has been applied in adsorption studies using silicon surface clusters a long time ago.1 The statement that vanadium in V has electronic con–guration d0 is incorrect and based on a confusion of the 2O5 formal chemical valence of the atom with its actual microscopic charge state. The DOS results indicate a d1.5 charge state. Extended test calculations have shown that the detailed nature of the bonds being formed between peripheral oxygen and terminator atoms is irrelevant for the chemical behaviour in the cluster centre. This is con–rmed by recent full-potential LAPW studies on hydrogen adsorption at V2O5(010) using repeated slabs2 where both binding energies and charge distributions are very close to the present cluster results.2O5(010) Faraday Discuss. 1999 114 85»103 (2) In contrast to the rather compact MgO crystal lattice the V2O5 substrate forms an open layer structure where electronic coupling between the layers is found to be weak. Therefore perturbations induced by vacancy formation in the –rst surface layer at V and aÜecting the 97 second layer are very unlikely to penetrate into deeper layers. More detailed results require very large cluster and/or slab calculations which exceed the present computational resources. 1 K. Hermann and P. S. Bagus Phys. Rev. B 1979 20 1603.2 K. Hermann A. Chakrabarti R. Druzinic and M. Witko Phys. Stat. Solidi 1999 173 195. Prof. Friend said Hydrocarbon radicals directly add to oxygen on MoO in calculations and 3 also to oxygen on thin –lm oxides of Mo in experiments. In our DFT calculations methyl radicals add to all three types of oxygen. Methyl bound to terminal oxygen is most stable. However in the experiments there is a kinetic preference for addition to highly coordinated oxygen. Therefore oxidation of hydrocarbons does not occur via insertion of gaseous OH into gas-phase hydrocarbon species. Prof. Waugh said Following up on the previous point and on your suggestion that it would be hydroxy species that are responsible for the oxidation of propene we have looked at the reactivity of propane with vanadium pyrophosphate by temperature programmed desorption and reaction.We have shown that the propene adsorbs as a propyl species and that the hydroxys formed recombine and desorb at lower temperatures than that at which the acrolein desorbs. Prof. Pettersson addressed Prof. Hermann I like the proposed mechanism for oxygen removal in terms of successive replacement of the oxygen to crystal bonds by O»H bonds and following easy desorption of water. However in terms of V2O5 as a dehydrogenation catalyst there seems to be a problem with the energetics the computed hydrogen affinity of the O(1) oxygen although high at about 70 kcal mol~1 is still far from compensating for the C»H bond strength at 100»110 kcal mol~1. Furthermore the computed value is at the LDA level and is likely to be somewhat overestimated.Could you speculate on possible interaction mechanisms that could provide the missing energy? Prof. Hermann responded Experimental C»H bond strengths are quoted as D(C»H)\81 kcal mol~1 in ref. 1 while the computed LSDA values of the hydrogen affinity E (H) for the diÜerent B oxygen sites O(1»3) at V range between 58 and 70 kcal mol~1. Very recent calculations 2O5(010) using the RPBE functional2 yield even smaller E (H) values between 42 and 54 kcal mol~1. There- B fore the energy required to split oÜ hydrogen from a free hydrocarbon molecule cannot be fully recovered by the energy gain due to surface O»H bond formation. However in a hydrocarbon reaction near the vanadium oxide surface the relative energetics will be aÜected by the hydrocarbon reactant being in—uenced by the electrostatic potential near the surface as well as by the surface O»H bond being modi–ed due to the presence of the reactant.Both eÜects may result in eÜective E (H) and D(C»H) values which are diÜerent from those mentioned above. Further cor- B responding surface reactions may proceed according to complex concerted mechanisms involving diÜerent reactants where energy barriers of diÜerent intermediate states rather than diÜerences between adsorption and dissociation energies determine the probability of H adsorption at a surface oxygen sites. 1 Handbook of Chemistry and Physics 76th edn. CRC Press London 1996 p. 9»52. 2 K. Hermann M. Witko R. Druzinic and R. Tokarz T op.Catal. in press. Dr Kantorovich said (1) While doing geometry relaxation in order to check if there is a Jahn» Teller relaxation it is important to make sure that the symmetry of the initial con–guration would allow you to do that. (2) Local state in the DOS is an indication of the charge localisation in the vacancy created upon the removal of the O atom. The state would pull up from the VB due to lack of the atom there. Prof. Hermann responded (1) In all cases relaxation due to oxygen vacancy formation was evaluated without symmetry constraints. While the O(1) and O(3) vacancies cannot give rise to Jahn»Teller relaxation due to their missing symmetry the eÜect was studied for the bridging O(2) vacancy. Here geometry optimisations starting from asymmetric atom arrangements did not result Faraday Discuss.1999 114 85»103 98 in a Jahn»Teller relaxation. (2) So far our DOS results from the cluster levels did not indicate vacancy induced features as you mentioned but this has to be studied in greater detail. ground state derived from the ground con–guration t 3T1g excited states derived from the con–guration t Dr Egdell opened the discussion of Prof. Madixœs paper It is interesting to explore the nature of the surface V-oxide phase in your work in relation to the electronic electron energy loss spectrum in Fig. 11 of your paper. Localised d to d excitations of transition metal ions in octahedral or nearly octahedral environments are both Laporte and parity forbidden and therefore have small oscillator strengths.In HREELS they therefore give fairly weak bands.1h3 Moreover the V3` ion in an octahedral environment has a 2g2 3T 3T with 2g1eg1. The absorption spectrum and 2g 1g of V3` doped as an isolated impurity into Al2O3 has absorption peaks at 2.15 eV and 3.12 eV associated with excitation to these states as well as a third peak at 4.27 eV due to excitation to the 3A2g state associated with the doubly excited con–guration eg2 (ref. 4). Thus your observation of a strong single new loss peak at 2.0 eV appears inconsistent with an assignment in terms of localised d to d excitations of V3`. An octahedral V4` ion has a 2T ground state and a single 2E excited 2g g state. However the problem of high band intensity remains. is a metallic oxide at room temperature and the optical properties are therefore phase.Bulk V2O3 dominated by the plasmon mode associated with the 3d conduction electrons. Plasmon losses are strongly allowed and therefore give much stronger loss features in HREELS than corresponding localised d to d transitions.5 It is tempting to assign your 2.0 eV peak with the conduction electron plasmon of V2O3 although the plasmon energy found at room temperature both in thermore —ectance spectra6 and transmission EELS7 is about 0.95 eV. Incidentally Goodman and co-workers recently found an HREELS loss peak at 0.9 eV at room temperature for ordered V 2O3(0001).8 This showed a blue shift to 1.2 eV on cooling below the 2O3(0001) –lms grown on Al metal to insulator transition in agreement with the thermore—ectance measurements.All in all the simplest assignment of the 2.0 eV loss peak is to some sort of plasmon mode but the diÜerences from bulk V2O3 are telling us that the electronic properties of your vanadia overlayers do diÜer signi–cantly and interestingly from the simple bulk V2O3 1 P. A Cox and A. A. Williams Surf. Sci. 1985 152/153 791. 2 A. Freitag V. Staemmler D. Cappus Ca. Ventrice Ka. Shamery H. Kuhlenbeck and H.-J. Freund Chem. Phys. L ett. 1993 210 10. 3 M. Hassel H. Kuhlenbeck H.-J. Freund S. Shi A. Freitag V. Staemmler S. LutkehoÜ and M. Neumann Chem. Phys. L ett. 1995 240 205. 4 M. H. L. Pryce and W. A. Runciman Discuss. Faraday Soc. 1958 26 34. 5 P. A. Cox R. G. Egdell J. B. Goodenough A. Hamnett and C. C. Naish J. Phys. C 1983 16 6221. 6 S. Stizza I.Davoli R. Bernadini A. Bianconi and M. Benfatto Solid State Commun. 1983 48 471. 7 H. Abe M. Terauchi M. Tanaka and S. Shin Jpn. J. Appl. Phys. 1998 37 584. 8 Q. Guo D. Y. Kim S. C. Street and D. W. Goodman J. V ac. Sci. T echnol. 1999 17 1887. Prof. Madix responded This is a very insightful and interesting suggestion. Taking this interpretation it is interesting to note that this collective excitation is observed at 2.0 eV from 0.5 to 5 monolayers of the vanadia suggesting that the metallic character of the vanadia does not change appreciably over this coverage range. Prof. Freund asked Do you think the preferred formation of V2O3 is structurally driven when formed on Al Could you comment on the growth of vanadium oxides on TiO2 ? 2O3 ? of course both have the corundum structure so one Prof.Madix answered V2O3 and Al2O3 could easily imagine the possibility of heteroepitaxial growth. However the fact that V2O3 is also formed on TiO (110) which has the rutile structure suggests that the force behind its formation 2 may not be so simple. Prof. Kempter asked Does the vanadia formation take place also by heating the alumina surface in the absence of oxygen? We –nd that this takes place for Ti/MgO where a TiO epitaxial layer will be formed by heating the Ti-–lm,1 even in the absence of oxygen. In this case the oxygen is supplied by the MgO-substrate. 99 Faraday Discuss. 1999 114 85»103 1 T. Suzuki R. Souda W. Maus-Friedrichs and V. Kempter Phys. Rev. B submitted. Prof. Madix replied When the vanadium is deposited on the thin alumina –lm at 800 K it diÜuses to the interface between the NiAl metal and the –lm.We have not conducted such experiments on the single crystal. Prof. Diebold said You used extremely small growth rates for vanadium deposition. Vanadium is a very reactive metal and will getter residual gases from the UHV chamber with a high probability. Did you experience any problems with contamination of the deposited V –lms? Is it conceivable that unintentional oxidation of the deposited metal by the residual gas causes some of the observed similarities in the STM results of vanadium metal and vanadium oxide –lms on TiO2 ? Prof. Madix responded We have focused in this paper primarily on the growth of vanadium oxide on these surfaces and the deposition was performed in a background of oxygen.The one comparison I showed between vanadium and vanadium oxide growth is the STM images for the alumina thin –lms. We have shown by XPS that for very similar vacuum conditions and similar deposition rates we can readily form unoxidized multilayers of vanadium metal. These XPS results show a clear transition from an oxidized form of vanadium to metallic vanadium as the surface coverage of vanadium is increased. We thus believe that oxidation of the vanadium due to residual gases is not a problem in this case. that favours its formation. Prof. Thomas commented Prof. Madix has given convincing experimental proof that it is V2O3 VO2 V2O5 not or any other oxide of vanadium that is formed under the conditions employed by him.It is intriguing to enquire whether a theoretical analysis or some computational modeling would have ìpredictedœ this oxide. The trouble is that such computations cannot readily (if at all) take into consideration the factors that lead to kinetic rather than thermodynamic stability. It could well be that it is simply the case of diÜusion of the vanadium and oxygen ions in V2O3 Prof. Madix replied This is an interesting question for the theorists to consider. Dr Carley communicated In order to provide corroborative evidence for the vanadium oxide overlayer being V2O3 have you tried to quantify the vanadium oxygen XPS intensity ratio ? Although in the early stages of growth there will be interference from the oxide substrate the kinetic energy of the O 1s photoelectrons is ca.100 eV for the photon energies you use with a correspondingly small mean free path (minimum of the mfp-KE curve). The O 1s intensity from the substrate will thus fall oÜ much more rapidly with overlayer growth than for example the Al 2p intensity (Fig. 5 of your paper) and reliable vanadium oxide stoichiometry data should be obtainable for vanadium exposures greater than ca. 3000 s (Fig. 5). It is worth noting that if Sco–eldœs cross-section data are used the V 2p cross-section is signi–cantly in error and the V 2s intensity must be employed. 1 J. H. Sco–eld J. Electron Spectrosc. 1976 8 129. Prof. Madix communicated in response This is certainly a helpful suggestion but unfortunately we did not accurately record the V 2s features in these experiments.Prof. Freund said The growth of Fe and Co on alumina seem to be very similar in the case where the metal or the corresponding oxides are grown. Dr Shluger addressed Prof. Madix (1) How thick was the alumina –lm shown in Fig. 12? (2) How thick are the –lms that can be seen using STM? (3) What is the origin of enhanced contrast at steps on the STM alumina –lm image? Prof. Madix replied (1) Approximately 5 ”. (2) We have not explored the imaging of thicker –lms. (3) Presumably it arises from diÜerences in the local structure and hence electronic structure in the vicinity of step edges. Faraday Discuss. 1999 114 85»103 100 Prof. Freund said TEM measurements on thin –lms can be used to compare the growth mode of metals on thin –lms and more bulk like materials.In the Introductory Lecture the experimental procedure was explained (see also ref. 132 of the Introductory Lecture). For alumina details of such a comparison are described in ref. 133 of the Introductory Lecture. Prof. Madey asked Prof. Madix (1) Your XPS and NEXAFS measurements of V and V2O3 growth on TiO (110) were performed on the (1]1) surface whereas the STM measurements you 2 report were carried out on the (1]2) reconstruction. Do you have STM data for V (V2O3) on the (1]1) surface ? Are there any diÜerences in growth on the two surfaces ? 2O3) (2) Also we reported evidence for two-dimensional growth of Cr on TiO (110) (1]1) up to 2 B0.8 ML;1 are your V (V islands relatively —at two-dimensional islands or are they threedimensional clusters ? 1 J.-M.Pan U. Diebold L. Zhang and T. E. Madey Surf. Sci. 1993 295 411. Prof. Madix responded (1) We do have some images for the growth of V on the (1]1) surface. On this surface exposure to V creates bright circular features most of which appear centered on the rows of coordinatively unsaturated titanium cations on the undisturbed surface. The underlying (1]1) structure of the (110) surface is disrupted. There is a notable disturbance of the structure in the vicinity of step edges. At higher vanadium exposures the vanadium-induced features increase slightly in size but the most noticeable increase is in their density. The general appearance of the STM images is similar on the (1]2) surface at low vanadium exposures. At low coverages the V is centered on the added row features of the (1]2).At higher coverages vanadium forms needle-like growths which orient their long axis along the (001) direction in registry with the underlying (1]2) structure. A preference for a double row spacing of these needles (25 ”) is observed. (2) The metallic vanadium particles appear to be hemispherical in STM on both the titania and alumina surfaces. STM suggests some —attening of the particles on the alumina surface due to reaction with background oxygen. Since there is great interest in the community regarding the surface Prof. Goodman addressed Prof. Madix You mentioned that you were able to synthesize an ìanatase-likeœ phase of TiO2 . could you elaborate on the synthetic procedure and its characteristic chemistry of anatase TiO2 structural diagnostics.Prof. Madix responded My statement was not meant to be so de–nitive but we have observed two diÜerent surfaces of TiO (110) by NEXAFS after the same nominal treatment to produce the 2 (1]1) stoichiometric surface. These two surfaces show diÜerent intensity ratios of the Ti L(III)- edge doublet which correspond closely to the diÜerences in –ne structure known for bulk rutile and anatase. This structural eÜect appears to be con–ned to the surface. Prof. Jennison said Vanadium and other reactive metals are added to brazing compounds supposedly to cause the metal to wet a ceramic (oxide) surface. When the brazed joints are then cut and examined microscopically an interfacial compound is often noted which is poorly characterized.My question is then have you seen any indication of reactions which mix the oxide substrate with the vanadium/vanadium oxide you are depositing ? Prof. Madix responded We have con–ned most of our work to room temperature and we have little evidence of the formation of complex oxides at the surface. Our STM results suggest that on the alumina thin –lm there is some incorporation of vanadium metal into the surface. However when vanadium metal deposited on either titania or the alumina thin –lm is heated it ultimately diÜuses away from the surface. The exact nature of the oxides formed in either case is however unknown. Prof. Campbell asked Did the deposition of V onto the (1]1) surface cause it to transform to the (1]2) structure (which might indicate some reduction of the Ti and formation of a mixed surface V»Ti oxide) ? 101 Faraday Discuss.1999 114 85»103 Prof. Madix answered No at a vanadium concentration of 0.03 ML large ìvacancy clusters œ appear in the terrace structure of the (1]1) surface and at higher coverages the underlying structure of the TiO (110) surface becomes indistinct. The random pattern of coverage by the 2 circular metallic features and their high density may make it difficult for the surface to rearrange into the (1]2) structure. Prof. Bowker asked Is it possible that the V2O3 formation is limited by a low oxygen dissociation probability on the oxide whereas it is likely to be very high on the initially deposited metal would be produced 2O5 atoms. Is it also possible that if the O pressure were much higher that V 2 and did you try such an experiment? Prof.Madix responded Yes we think this conversion is kinetically limited. We tried further oxidation of the vanadia layer in 5]10~5 Torr oxygen at 700 K for 100 min and observed no change. Of course this is not a very high oxygen pressure and further oxidation might be possible at higher pressures. Prof. Asscher addressed Prof. Madix The question/comment was that in order for metallic vanadium clusters to form atoms should diÜuse across multiple lattice sites. This suggests that upon initial impact from the gas phase vanadium atoms cannot be oxidised to the V3` state»it would prevent any surface mobility and thus clusters cannot be formed. It is therefore suggested that initially the vanadium atoms adsorb as neutrals diÜuse to form clusters and only then at a certain point they get oxidised by oxygen molecules which dissociatively adsorb on the metallic cluster.Prof. Madix answered The dynamics of the formation of these small particles is not clear and I am hesitant to speculate on this matter. The deposition rates we employ when coupled with the size of the vanadium induced features observed in STM suggest that there are a few vanadium atoms in each of the circular features observed. The coverage calibration relies on calibration by AES using a characteristic mean free path of the Auger electrons and has an inherent uncertainty. At very low coverages of vanadium I do –nd it surprising that metal atoms can hop or migrate the 25 ” on the oxide surface required to ìnucleateœ.I think this issue requires further study. Prof. Freund said The study of diÜusion of metals on oxides is a hot topic. With STM the diÜusion of single atoms has not been studied in detail as yet. Field ion microscopy can be used for such a study and activation energies for diÜusion can be estimated (ref. 127 of the Introductory Lecture). More detailed studies are in progress. Dr Yubero communicated I would like to comment on some of your results related to the characterisation of vanadium oxide on alumina or titania. When you perform the electronic characterisation by XPS you observe (Fig. 3 of your paper) that there is an energy shift to lower binding energies on the V 2p3@2 peak as the amount of vanadium oxide deposited increases.I agree with you when you say that in spite of this energy shift the chemical nature of the vanadium atoms (i.e. V3` species) is always the same. You justify this energy shift by a Coulomb interaction between the positive charge left on the particle as the result of the photoemission process and the photoemitted electron. However you do not –nd the same behaviour when you deposit V2O3 on titania (Fig. 2 of your paper) which is also an insulator with more than 3 eV band gap. 2) I would like to point out that this kind of energy shift eÜect in the –rst stages of growth of oxides has been extensively studied by the group of Gonzaç lez-Elipe et al.1 In fact they have developed a theory based on bonding and polarisation eÜects to predict these energy shifts in the binding energy of the peaks of the deposit as well as in their Auger parameter.Note that variations on the Auger parameter of a given atom are directly proportional to the extraatomic relaxation energy and that the latter can be related to bonding and polarisation eÜects. Then according to their work,1 if the dielectric properties (i.e. refractive index) of an overlayer of a substrate are similar no energy shifts are expected. However when a high refractive index overlayer (e.g. TiO2) is deposited on a low refractive index substrate (e.g. MgO SiO the Auger Faraday Discuss. 1999 114 85»103 102 parameter of the overlayer atoms increases as the overlayer thickness increases. Besides in these particular cases (i.e.TiO on SiO and TiO on MgO) it is observed that the binding energy of 2 2 2 the Ti 2p peaks decreases with increasing coverage. On the other hand when TiO is deposited on 2 a metallic substrate as Ag (even higher refractive index) the observed behaviour for the Auger parameter and TiO binding energies is the opposite. Then the observed behaviour on the present on 2 systems i.e. V2O3 TiO2 and V2O3 on Al2O3 would be explained just by considering that the dielectric properties and in particular the refractive index of V2O3 is similar to that of TiO but 2 signi–catively higher than that of Al2O3 . Note that polarisation eÜects of diÜerent materials scale with 1/n2 with n the refractive index. 1 J. A. Mejïç as V. M. Jimeç nez G. Lassaletta A. Fernaç ndez J. P. Espinoç s and A. R. Gonzaç lez-Elipe J. Phys. Chem. 1996 100 16255. Prof. Madix communicated in response This is an interesting suggestion and it is certainly possible that these polarization eÜects contribute to the shift in binding energy observed. It is interesting to note that we see larger shifts of the V 2p peak when metallic vanadium is deposited but both vanadium and vanadia produce shifts toward lower binding energy as the surface coverage is increased. The direction of this shift is compatible with our interpretation that a metal-like oxide is formed though its refractive index is probably lower. I do not have the data required at this time to make a quantitative evaluation of the relative shifts to be expected however. Prof. M‘ller communicated I would like to remark that using diÜerent preparative conditions (successive cycles of submonolayer vanadium in UHV each followed by annealing in 10~6 mbar oxygen atmosphere at around 150 °C the exact procedure of Prof. Granozzi and co-workers being reported in ref. 1 and which was later applied in ref. 2) the authors have there been able to grow epitaxial layers of VO as demonstrated from XPD UPS ARPEFS and LEED data. 2 J‘rgensen and P. J. M‘ller Appl. Surf. Sci. 1 M. Sambi G. Sangiovanni G. Granozzi and F. Parmigiani Phys. Rev. B 1997 55 7850. 2 M. Sambi M. Della Negra G. Granozzi Z. S. Li J. HoÜmann 1999 142 146. 103 Faraday Discuss. 1999 114 85»103

ISSN:1359-6640    

DOI:10.1039/a908284k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

A study of the electronic magnetic structural and dynamic properties of low-dimensional NiO on MgO(100) surfaces William C. Mackrodt,*a Claudine Noguerab and Neil L. Allanc a School of Chemistry University of St. Andrews St. Andrews Fife UK KY 16 9ST b L aboratoire de Physique des Solides associeç au CNRS Universiteç Paris Sud 91405 Orsay France c School of Chemistry University of Bristol Cantockœs Close Bristol UK BS8 1T S 1. Introduction Over the past few years important advances have been made in the fabrication of ultra-thin crystalline oxide layers grown epitaxially on a variety of substrates.1,2 The methods commonly used are direct oxidation of a metallic surface or deposition of metal atoms followed by controlled oxidation. In this way –lms a few Angstroé ms thick may be grown the crystallinity and perfection of which depend strongly on the ratio of the lattice parameter of the oxide to that of the substrate and details of the preparative conditions such as deposition rate oxidising temperature etc.These ultra-thin –lms exhibit several unusual properties which are of interest both from a fundamental point of view3h5 and with regard to potential technological application. First as a result of their small thickness they do not charge when exposed to electron or electromagnetic radiation and can thus be submitted to detailed spectroscopic investigation. They may also exhibit surfaces that are not usually obtained by direct cleavage of the bulk which is the case for high energy surface orientations such as polar surfaces.The constraint imposed by the substrate may also lead these epitaxial oxide layers to adopt lattice symmetries which diÜer from those of the thermodynamically stable bulk and where structural phase transitions occur transition temperatures can depend strongly on the thickness of the epitaxial layer. The system NiO(100)»MgO(100) is prototypical of multilayered oxide –lms largely because the lattice mismatch between the two components is less than 1%. It has been shown that MgO grows on NiO(100) epitaxially and in a layer-by-layer mode,6h8 and that the same is true for NiO on MgO(100).5,6,9h13 Intermixing occurs only if –lms are prepared at high temperature. Several types of experiment indicate a strong dependence of the electronic and magnetic properties of these ultra-thin –lms on their thickness.EELS (electron energy loss spectroscopy) experiments6,10 have 105 Faraday Discuss. 1999 114 105»127 Received 25th May 1999 Recent developments in the growth of ultra-thin epitaxial layers of oxides and the fabrication of a diversity of nanostructures has led to current interest in and much speculation about the properties of low dimensional systems. In this paper we report recent calculations for low dimensional NiO on MgO(100) surfaces both from –rst principles electronic structure calculations and free energy calculations based on surface lattice dynamics. The results include surface structures and dynamics at a range of temperatures and electronic structures of ground excited ionised d]d and charge»transfer excitonic states in diÜerent spin alignments.This journal is( The Royal Society of Chemistry 2000 N revealed excitations at energies less than the bulk band gap which have been attributed to d]d transitions. For coverages of NiO on MgO(100) greater than 2.8 monolayers (ML) these excitations are identical to semi-in–nite NiO(100) whereas for NiO thicknesses less than 2.8 ML a new transition at 2.18 eV appears which has been assigned to surface or interface states and appears to show that the electronic structure of thin NiO –lms is modi–ed by the presence of the MgO substrate. Other bulk and surface transitions have been reported to shift in energy with coverage. Two explanations have been suggested to account for these changes. The –rst is the loss of octahedral symmetry at a surface Ni associated not only with the reduced nearest neighbour coordination but also with a lateral elongation of the NiO lattice and contraction of the interplanar separation.The second is a putative loss of octahedral symmetry at a Ni at the NiO/MgO interface resulting from the diÜerence in covalency between NiO and MgO. With regard to the existence or otherwise of NiO/MgO interface states it is of note that no such states have been detected by EELS8 for MgO layers grown on NiO. XPS experiments9,13 have shown that Ni and O core level spectra depend on –lm thickness which has been interpreted in terms of the number of Ni second neighbours. Such spectra are also consistent with a negative charge transfer from MgO to NiO at large thicknesses.A detailed Hartree»Fock interpretation of core level shifts in NiO/MgO layered systems in terms of surface lattice relaxation and other eÜects has been given previously.14 While magnetic data are generally more difficult to obtain Alders et al.5 have recently reported some very elegant linear polarised X-ray absorption experiments from which they obtain values of the Neç el temperature TN of NiO overlayers as a function of thickness and show that even for 20 ML –lms T has not recovered the bulk value. It is within this context that we have initiated a theoretical study of ultra-thin –lms and surface properties of NiO. NiO is a paradigm magnetic insulator whose bulk and surface properties have been the subject of extensive experimental and theoretical investigation.15 It has long been considered as highly ionic and early –rst principles calculations based on the local spin density approximation (LSD) described it as a Mott»Hubbard system in the AF spin arrangement with a 2 narrow gap spanned by Ni d-states.16 However seminal work by Sawatzky et al.17h19 showed that hole states in Li NiO were largely of O(p) character which suggested that the –rst ionised state of NiO is essentially d8L and the ground state of p]d charge»transfer type.Subsequent –rst principles calculations con–rmed the majority weight of the valence band edge to be O(p) including spin unrestricted periodic Hartree»Fock (UHF) calculations.20h22 The latter have shown that the insulating and (high spin) magnetic properties are the result of large on-site Coulomb and exchange interactions between essentially localised electrons with strong orbital polarisation resulting from the orbital dependence of the one-electron potential.This is determined principally by the non-local exchange interaction which is evaluated exactly within the Hartree»Fock approximation and implemented again exactly in the CRYSTAL code.23 UHF calculations have also provided direct evidence of O(p) holes in Li:NiO and NiO24,25 and Fe(d) holes in NiO,25 in agreement with experiment.26 Thus despite its approximate nature and inherent limitations the periodic UHF method would seem to be well suited to describing the electronic and magnetic properties of NiO in lower dimensions. Extending a recent UHF study27 of an unsupported NiO(100) monolayer which included both ground and d]d and charge-transfer excited states here we report similar calculations of several NiO/MgO slab con–gurations including NiO(100) monolayers on MgO(100) or sandwiched between two MgO(100) monolayers.While there appear to be no other calculations based on extended/periodic systems for comparison there have been high level cluster calculations of magnetic interactions28 and d]d 29h31 and charge»transfer excitations, 31 which provide a guide as to the in—uence of both electron correlation and non-local/ extensive eÜects. In view of the growing body of experimental data on the high-temperature properties of ultrathin –lms we also report quasi-harmonic lattice dynamics and free energies calculations of more extended NiO/MgO slab con–gurations at elevated temperatures using recently developed atomistic lattice methods based on pair potentials.32 Few calculations have included dynamic eÜects such as temperature largely because the full dynamical treatment of surfaces and extended defects even within the quasi-harmonic approximation presents severe computational demands if reasonably high precision is required.There are three main simulation techniques available for the calculation of surface lattice properties Monte Carlo simulation molecular dynamics and quasi-harmonic lattice dynamics. Of these only the last is capable of giving free energies (as well as Faraday Discuss. 1999 114 105»127 106 derived properties such as the entropy and the heat capacity) directly and to high precision.This method is consequently not only the most suitable for structure optimisation as a function of temperature but in many applications it has also been shown to be a valid approximation up to two-thirds of the bulk melting temperature.33,34 Accordingly the paper is organised as follows. In Section 2 we review brie—y the theoretical methods used and in Section 3 present the main body of our results which we discuss in Section 4. O 1s(8)2sp(4)3sp(1)4sp(1) 1 C O B A2 C O C A3 C O C 4 ì4 dxy ]dz2 respectively.27 Thus by ane 2. Theoretical methods 2a. Electronic structure calculations The all-electron ab initio LCAO Hartree»Fock method for periodic systems and its computational implementation in the CRYSTAL 95 computer code23 have been described in detail previously.35 The calculations reported here use extended Gaussian basis sets and are based on the spin unrestricted (UHF) procedure36 to describe open-shell electronic con–gurations.The numerical values of the tolerance parameters involved in the evaluation of the (in–nite) Coulomb and exchange series were identical to those used in recent studies :24,25,27 a detailed account of the eÜect of these tolerances is discussed elsewhere.37 The reciprocal space integration utilised the Monkhorst»Pack sampling,38 with shrinking factors that gave 15»36 k-points in the IBZ depending on the overall symmetry of the calculation and the SCF convergence criterion based on diÜerences in the total energy of the unit cell of less than 10~6 Ha.A posteriori corrections for electron correlation based on three generalised gradient approximations are included in the present study they are due to Perdew,39 Perdew and Wang40 and Perdew Burke and Ernzerhof 41 and are referred to P PW and PBE respectively. As in previous calculations,20h22,24,25,27 the localised crystal orbitals consisted of 25 atomic orbitals for Ni 15 for Mg and 14 for O of the type Ni 1s(8)2sp(6)3sp(4)4sp(1)5sp(1)3d(4)4d(1) Mg 1s(8)2sp(5)3sp(1)4sp(1) where the numbers in brackets are the numbers of Gaussian functions used to describe the corresponding shell e.g. 1s 2sp 3d etc. The exponents and contraction coefficients were identical to those used for the bulk.20h22 In this study we consider various slabs of NiO and MgO for which we use the notation (type)nlm N where N refers to the number of layers and nlm to the orientation.Thus NiO(100) monolayer is (NiO100 1 the bi-layer NiO/MgO (NiO/MgO)100 2 and the three tri-layers we consider here (NiO/MgO/MgO)100 3 (NiO/MgO/NiO)100 3 and (MgO/NiO/MgO)100 3 . Furthermore all our calculations are based on a constant cation A »anion distance of 2.1 é –xed by the MgO substrate while our axis system equates the p xy-plane with (100) so the and dz2 orbitals are perpendicular z to the slab planes. As in a previous study,27 to investigate the magnetic properties of these slabs we have considered four spin arrangements of a 2]2 surface unit cell shown below C C B C F C O C A C B B B where C and B represent high spin Ni2` ions and F A1 A2 and A the ferromagnetic anti- 3 ferromagnetic fully-frustrated and ferrimagnetic spin alignments respectively.27 In the case of (NiO/MgO/NiO)100 3 we consider slabs in which two ferromagnetic NiO layers are aligned both ferromagnetically and antiferromagnetically.In this way we have been able to distinguish superexchange coupling within and between NiO planes as a –rst step towards understanding the recent results reported by Alders et al.9 As described in detail below we have obtained converged UHF solutions for a number of excited states involving both single and quadruple i.e. complete local d]d excitations which we designate as e and e excited state œ we mean a state in which all four Ni ions of the unit cell have undergone a dxy ]dz2 1 excitation.107 Faraday Discuss. 1999 114 105»127 Our treatment of hole states follows that previously used for NiO bulk24 and (100) monolayer, 27 wherein a renormalisation of the (in–nite) inter-cell Coulombic interaction is eÜected by adding a uniform background charge of opposite sign and equal magnitude to the crystal potential in the plane of the monolayer. As our results indicate this has no eÜect on the densities of states other than a rigid shift in energy of the single particle spectrum. Furthermore we con–ne our attention to diÜerences in total energy only as between various electronic and spin states of the hole and small lattice distortions which again are invariant to the uniform background charge. 2b. Lattice simulations The methods we have used to carry out lattice simulations at –nite temperatures based on lattice statics and quasi-harmonic lattice dynamics have been described in full recently.31,42 For a slab of sufficient thickness to provide what in eÜect is a bulk-like region in the interior and two noninteracting free surfaces the Helmholtz free energy of the system F at a temperature T is minimised with respect to the collection of structural parameters MEAN that de–ne the slab.In the quasi-harmonic approximation it is assumed that F(MEAN T ) can be written as the sum of static and vibrational contributions, F(MEAN T )\Ustat(MEAN)]Fvib(MEAN T ) l Ustat(MEAN) is the potential energy of the static lattice in a given state of strain MEAN and is evaluis the sum of harmonic vibrational contributions j(q) of modes with wavevecis given by ated from interatomic pair potentials while Fvib from all the normal modes.For a periodic structure the frequencies tor q are obtained by diagonalisation of the dynamical matrix D(q). Fvib Fvib\ ; M12 hlj(q)]kB T ln[1[exp([hlj(q)/kB T )]N (hl S\ ; q j in which the –rst term is the zero-point energy and the associated vibrational entropy S exp(hlj(q)/kB T )[1 [kB ln[1[exp([hlj(q)/kB T )] j(q)/T ) \ ; 2 ] exp(hl AdF B G h A1 B H 2lj(q) j(q)/ 1 kB T )[1BAdlj2(q) dE vib dEA q j For a macroscopic crystal the sum over q becomes an integral over a cell in reciprocal space which can be evaluated by taking successively –ner uniform grids until convergence is achieved.Since the reciprocal space is two-dimensional the Brillouin zone summation requires a two dimensional mesh of wavevectors. The minimisation of F and subsequent thermodynamic manipulation can of course be carried out by brute force from numerical values of F. However for the slabs of any complexity such as those considered here there are large numbers of internal strains EA and it is much more efficient to use analytic expressions for the derivatives of F with respect to strain. The strain derivatives are given by E{ T A E{ where the subscript E@ denotes that all the E are kept constant except for the diÜerentiation variable. In a recently developed code40 the derivatives (dlj2(q)/dEA)E{ are obtained from analytic expressions for the derivatives (dD(q)/dEA)E{ by –rst-order perturbation theory.42,43 Full details of the particular derivatives of the Parry summation needed for the Coulombic interactions in ionic slabs are given by Taylor et al.;32 the derivatives required for pairwise short-range potentials are collected together in ref. 42. Since the perturbation is in–nitesimal the procedure is exact. In addition for thermodynamic properties no special consideration needs to be given to degeneracies in –rst-order perturbation theory for the trace of (dD(q)/dEA)E{ is invariant for any complete normal set of eigenvectors of D. To obtain the equilibrium structure a variable metric method is used for the minimisation of F with respect to the EA . In the initial con–guration the static energy Hessian (d2Ustat/dEA dEB) which is a good approximation to (d2F/dEA dEB) is calculated from its analytic expression and its inverse together with the (dF/dEA) are used to obtain an improved q j Faraday Discuss.1999 114 105»127 108 con–guration. In subsequent iterations the (dF/dEA) are calculated in the new con–guration and the inverse Hessian updated by the BFGS formula.44 An optimisation therefore requires one static Hessian calculation and a small number of dynamic gradient calculations. In practice it has been found that this is much more efficient than methods involving repeated evaluation of the Hessian or frequent line minimisations or methods in which the derivatives are determined numerically. [(dxz)2(dyz)2(dxy)2(dz2)1(dx2~y2)1] e(eg)a\e(t2g)a\e(t2g)b\e(eg)b (bulk) AF2[F[AF1 B all of which indicate that the highly ionic high spin character previously 100 1 27 is retained in multi-layered NiO/MgO systems.Furtherand F can be written to a –rst 1 A2 A3 3. Results 3a. Electronic ground states We begin with a brief resumeç of NiO bulk20 and (100) monolayer,27 which despite their diÜerent dimensionality and consequent nearest-neighbour coordination and diÜerent ligand-–eld splitting exhibit several common features. Both are predicted to be highly ionic high spin insulators with lattice constants of A B4.3 é and B4.0 Aé respectively. Moreover for both systems the d8 ground state con–guration is with for the bulk and e(dz2)a/e(dx2~y2)a\e(dxz)a/e(dyz)a e(dxy)a\e(dxz)b/e(dyz)b e(dxy)b for (NiO)100 1 where e are the single particle eigen values and the subscripts a and b refer to spin up (C) and spin down (B) electrons respectively.The local Ni spin magnetic moment is calculated to be B1.9 l for both systems and the stability of low index spin alignments B and d d) ( and indirect superexchange Ese) interactions UHF total energies lead to E and E of [1.6 meV and [8.2 meV respectively for NiO bulk at a lattice constant se and [1.0 meV and [6.9 meV for (NiO)100 1 at the same lattice constant. Another major A1[A3[F[A2 ((NiO)100 1 ) below the Neç el temperature. If to a –rst approximation it is assumed that the diÜerences in energy between the spin alignments for each system separately can be written simply in terms of direct spin»spin (E values for of 4.2 Aé diÜerence between the two systems is that from an analysis of the unoccupied O(p) density of states (DOS) of the self-trapped hole the band gap is estimated to be B4 eV for the bulk24,25 and B5 eV for (NiO)100 1 .27 Turning now to the various NiO/MgO(100) slabs we have obtained converged UHF solutions and A spin alignments.As before,20,27 Mulliken 3 of the ground electronic states of the F A1 A2 analyses45 yield eÜective atomic charges of B1.9 e 3d populations of B8.1 and local spin moments of B1.9 l found for the bulk20 and (NiO) more the d of 8 ground state con–guration [(dxz)2(dyz)2(dxy)2(dz2)1(dx2~y2)1] (NiO)100 1 is found to remain unchanged in all four spin alignments the stabilities of which are in the order A1[A3[F[A2 as shown in Table 1.As before,27 the relative energies of A approximation in terms of direct and superexchange interactions as E(A E(F)\B]2Ed E(A1)\B]Ed]2Ese 2)\B E(A3)\B]Ed]Ese Faraday Discuss. 1999 114 105»127 109 and A spin alignments (F and AF in the case of the bulk) Table 1 Comparison of the energies (meV/Ni) of the F A2 3 relative to A (AF for NiO bulk and various (100) NiO/MgO 1 1 slabs all at a 2) 0\4.2 Aé System 1) A3 F A(AF 2 19.7 12.8 13.6 13.5 13.6 15.4 » 6.9 7.4 7.4 7.4 8.5 26.2 14.8 16.0 15.9 16.1 18.6 Bulk (NiO)100 1 (NiO/MgO)100 2 (NiO/MgO/MgO)100 3 (NiO/MgO/NiO)100 3 (MgO/NiO/MgO)100 3 from which E and E can be obtained. These are given in Table 2 where E and E d se correspond- se d ing to direct and superexchange interactions within a single NiO(100) layer are seen to increase monotonically with Ni coordination to broadly that of the bulk in (MgO/NiO/MgO)100 3 .In addition our results for (NiO/MgO/NiO)100 3 in which the two ferromagnetic NiO layers are coupled antiferromagnetically suggest stronger superexchange between NiO(100) layers than within the layers at least for the simple tri-layer we have examined. The densities of occupied states of the A alignment of (NiO/MgO)100 2 over the –rst 12 eV 1 shown in Figs. 1a and 1c indicate further close similarities with 2D and 3D NiO. The upper valence band B5 eV wide is essentially O(p) with only a minor contribution from dxy states at lower energies and negligible Ni weight at the upper edge.The local O(p) DOS of the two planes shown in Fig. 1b are similar both in terms of band width and overall pro–le. The major diÜerence is the weight of states at the upper edge shown in Fig. 1b which derives predominantly from the NiO layer. This suggests though by no means guarantees that the low energy holes states will be associated with the NiO lattice. The Ni d states occur B1 eV below the O(p) band and are dispersed over B4.5 eV in three distinct sub-bands with e(dz2)a/e(dx2~y2)a\e(dxz)a/e(dyz)a e(dxy)a\e(dxz)b/e(dyz)b e(dxy)b as in the bulk. This is in marked contrast to the crystal-–eld splitting e8 (dxz)/e8 (dyz)\e8 (dz2)\e8 (dxy)\e8 (dx2~y2) which suggests that as in the bulk the single particle spectrum is determined to a large extent by the on-site Coulomb and exchange terms.3b. d«d excited states The lowest energy electronic excitations from the ground state in NiO correspond to orbitally forbidden (*l\0) local d]d transitions or Frenkel excitons which have been observed in both Table 2 Comparison of E and E (meV) for NiO bulk d and various (100) NiO/MgO slabs all at a0\4.2 Aé se System [Ese [Ed 1.6 1.0 1.2 1.2 » 1.3 1.6 8.2 6.9 7.4 7.4 7.4 10.5b 8.5 Bulk (NiO)100 1 (NiO/MgO)100 2 (NiO/MgO/MgO)100 3 (NiO/MgO/NiO)100 3 a (NiO/MgO/NiO)100 3 (MgO/NiO/MgO)100 3 a No superexchange coupling between NiO layers. b Superexchange interaction between two ferromagnetic NiO layers antiferromagnetically aligned. Faraday Discuss.1999 114 105»127 110 Fig. 1 a Valence band DOS of A (NiO/MgO)100 2 . b Comparison of O(p)MgO and O(p)NiO valence band DOS 1 100 F 2. A c Comparison of the valence band DOS of II bulk NiO A (NiO/MgO)100 2 and 1 of A (NiO/MgO) A (NiO) 1 100 1 . 1 and the optical46,47 and EEL29,48h51 spectra of the pure material and in EEL spectra of NiO/ MgO(100) layered systems.6,10 We have obtained variationally minimised solutions corresponding to the complete range of one-electron (e1) d excitations of the type xy ]dz2 dxy ]dx2~y2 etc. to x2~y2 ]dz2 dz2 ]dx2~y2 d the spin-forbidden d xy/dyz ]dz2/dx2~y2 states and the two electron 111 Faraday Discuss. 1999 114 105»127 100 2 to con–rm that they are essentially independent 100 3 to examine 100 1 are col- (NiO)100 1 (b) (NiO/MgO)100 2 (c) and excited state for F(NiO/MgO)100 2 .These states remain highly ionic high spin insulating with changes in the ionic charges 3d populations and local spin moments of O1%. The corresponding excitation energies derived from direct total energy diÜerences with respect to the ground state are listed in Table 3 where we include UHF values and those derived from post-Hartree»Fock corrected energies. We have also obtained converged e solutions for a number of these excited states 4 for the F alignment. As found previously for (NiO)100 1 e e 1 ,27 the diÜerences between and 4 energies are substantially less than 0.1 eV so that we have not attempted to obtain excitation energies extrapolated to zero concentration. We have calculated excitation energies of the spin-allowed transitions for the A alignment of (NiO/MgO) 1 of the magnetic state of the lattice and excitation energies for F(MgO/NiO/MgO) the eÜect of nearest neighbour coordination.These together with results for F(NiO) lected in Table 4 from which the excitation energies for (MgO/NiO/MgO)100 3 (e) are plotted as a function of Ni coordination in Fig. 2. 3c. First ionised state p The removal of an electron from fully symmetric NiO/MgO multilayers in whatever spin alignment leads to conducting states of essentially d8L character exactly as suggested by the ground state DOS where the unpaired electron/hole is delocalised over the O sites no matter what starting electronic con–guration is chosen. As reported previously for NiO bulk24,25 and (100) monolayer,27 the removal of this symmetry constraint allows the electronic con–guration to relax to non-degenerate insulating states of lower energy in which the unpaired electron/hole is localised in a p orbital at a single O site.We have obtained converged solutions for states in which the hole is localised (separately) in both layers of (NiO/MgO)100 2 and in diÜerent spin con–gurations. The relative energies of the states are collected in Table 5 in which the nomenclature XO S` and XO S`(x) correspond to localised holes in the XO layer of (NiO/MgO)100 2 where S is the spin con–guration of the NiO layer and x the alignment (f»ferromagnetic a»antiferromagnetic) of the unpaired p electron relative to S. As suggested by the valence band DOS of the ground state (Fig.p Table 3 Comparison of UHF PBE PW and P91 d]d excitation energies (eV) in F(NiO/MgO)100 2 P91 PW PBE UHF Excitation 1.95 0.83 0.83 1.17 2.88 4.08 1.95 0.82 0.83 1.17 2.93 4.13 1.94 0.82 0.83 1.17 2.94 4.14 1.98 0.83 0.83 1.16 2.97 4.24 xy]z2 xz]z2 xy]x2[y2 xz]x2[y2 x2[y2]z2 a z2]x2[y2 a 1.85 1.85 1.85 1.87 xz/yz]z2/x2[y2 a Spin forbidden. 4.0 Aé ) Table 4 Comparison of d]d excitation energies (eV) in (a) F(NiO)100 1 (a0\ (b) F(NiO)100 (c) 1 (a0\4.2 Aé ) F(NiO/MgO)100 2 A (d) 1(NiO/MgO)100 2 and (e) F(MgO/NiO/MgO)100 3 (e) (d) (c) (b) (a) Excitation 2.66 » 0.81 » 3.68 3.54 1.99 0.84 0.86 1.20 » » 1.98 0.83 0.83 1.16 2.97 4.24 1.27 0.31 0.85 » 2.24 5.00 1.18 » 1.11 » 1.86 » xy]z2 xz]z2 xy]x2[y2 xz]x2[y2 x2[y2]z2 a z2]x2[y2 a 2.14 1.90 1.87 1.57 1.81 xz/yz]z2/x2[y2 a Spin forbidden.Faraday Discuss. 1999 114 105»127 112 and Fig. 2 Comparison of calculated d]d excitation energies in F(NiO)100 1 (NiO/MgO)100 2 (MgO/NiO/MgO)100 3 as a function of Ni coordination. g corresponding to the various holes states given in Table 5 and 100 3 and (MgO/NiO/MgO)100 3 . The later are given in (NiO)100 1 from which the variation of Eg in NiO as a function of Ni 1b) the lower energy states correspond to the hole localised in the NiO layer. Mulliken population analyses indicate that in all these states B90% of the hole density is localised at a single O site with a moment of B0.9 l but with no signi–cant change in local moments at the cation sites.B Localisation of a hole at a single O site leads to the creation of a narrow band of unoccupied O(p) states at the top of the valence band with only minimum changes at the conduction band edge as shown in Fig. 3 for A1 `(NiO/MgO)100 2 and reported previously for NiO bulk24,25 and (100) monolayer.27 This is the exactly equivalent to the changes in the oxgygen k-edge spectra obtained originally by Kuiper et al.17 for LixNi1~xO where the energy between the extrinsically controlled unoccupied O(p) states and the conduction band edge approximates the band gap in NiO. We have estimated the band gap E for the NiO F`(f) state in (NiO/MgO/MgO) Table 6 together with that for coordination can be obtained.This is shown in Fig. 4 where the increase in E is seen to be close g to linear. 3d. Charge transfer states In a recent study of NiO(100) monolayer27 we have calculated the energy of the pz ]dz2 charge transfer excitonic states as an independent check on estimates of the band gap Eg from the unoccupied DOS of the –rst ionised states. Here we report similar calculations for (NiO/MgO)100 2 . As before,27 we have con–ned our attention to the computationally more convenient F alignment and obtained variationally minimised solutions for the e4 pz ]dz2 charge transfer state (Ni`O~) g Ni Table 5 Relative energies *E (eV) and associated gaps E (eV) of the –rst ionised states of (NiO/MgO)100 2 and V in F(NiO/MgO)100 2 State *E Eg 5.7 5.9 6.0 6.0 6.0 3.6 0.0 0.028 0.161 0.642 0.268 » NiO F`(f) NiO A1 ` NiO F`(a) MgO F`(f) MgO F`(a) NiO F`(f) (VNi) 113 Faraday Discuss.1999 114 105»127 Fig. 3 Empty gap states of A1 `(NiO/MgO)100 2 . in both triplet and singlet spin states that is to say con–gurations of the type and Ni(d9) O(p5) Ni(d9) O(p5)a a a (NiO/MgO) Eg (MgO/NiO/MgO) Fig. b These can be viewed as fully-condensed Mott»Wannier (M»W) exciton states so that in Ni(d9) O(p5) the excited electron is b-spin (which it must be) and the hole b-spin giving a triplet Table 6 Comparison of E (eV) associated with NiO F`(f) –rst ionised states g 100 2 and 5.1 5.7 5.7 6.6 100 3 as a function of Ni coordination.a in 100 1 (NiO) (NiO/MgO/MgO)100 3 (MgO/NiO/MgO)100 3 System (NiO)100 1 (NiO/MgO)100 2 (NiO/MgO/MgO)100 3 (MgO/NiO/MgO)100 3 100 1 (NiO/MgO)100 2 and a 4 E in (NiO) g Faraday Discuss. 1999 114 105»127 114 Table 7 Energies (eV/Ni) of the e4 charge»transfer states of F(NiO/MgO)100 2 Energy State 5.175 5.412 5.874 Triplet p p ]dz2 Singlet p p ]dz2 Triplet p p ]dx2~y2 exciton. The corresponding formation energies are obtained from direct energy diÜerences between the ground and charge transfer states and are lower bounds to the optical band gap thereby providing an alternative estimate of E to that from the single-particle eigen values.They are given g in Table 7 which shows that the triplet state as expected is the lower of the two by B0.24 eV which is similar to that found found previously.27 We have also calculated the –rst excited con- –guration of the e triplet excitonic state corresponding to pp ]dx2~y2 charge transfer which we 4 –nd to be B0.7 eV above the ground state. 3e. Mg substitutional states In view of the negative deviation from ideality of the enthalpy of mixing of (bulk) NiO MgO,52,53 low levels of anti-site defects would be expected to occur in ultra-thin –lms of NiO/MgO particularly at high temperature and this is indeed found to be the case. As a prelude to a more extensive investigation of this we have examined the electronic structure of the (NiO/MgO)100 2 bilayer containing 25% MgNi substitutional defects in the NiO layer.Mulliken analysis indicates that the substitution of an Mg2` ion for Ni2` in the NiO layer leads to only very minor changes in the electron distribution with deviations in population of \1% by comparison with the unsubstituted system. Further evidence of the minimal change in electronic structure is contained in the valence band DOS shown in Fig. 5 where it is compared with that of the non-defective (NiO/MgO)100 2 bilayer. From this it is evident that while there are small deviations in the DOS which might well be accounted for by the change in symmetry of the substituted system and consequent modi–cation of the k-space sampling it remains largely unchanged. This accords well with previous UHF calculations for substitution in the bulk.21 Fig.5 Comparison of the valence band DOS of F(NiO/MgO)100 2 and MgNi in F(NiO/MgO)100 2 . 115 Faraday Discuss. 1999 114 105»127 Table 8 Electron re-distributions (e/atom or e/MgO) associated with V (2~) Ni and VNi (0) and their diÜerences dq in (NiO/MgO)100 2 Ni MgO Mg Vacancy ONiO OMgO V (2~) Ni V(0) Ni dq [0.048 [0.911 ; [0.075 [0.863 ; [0.027 [0.019 ]0.031 ]0.050 ]0.067 ]0.002 [0.066 ]0.044 [0.018 [0.063 ]0.023 ]0.020 [0.003 3f. Cation vacancy states (NiO/MgO)100 2 bilayer. They are doubly charged VNi(2`) which (NiO/MgO)100 2 bilayer with a transfer of B0.07 e from NiO to MgO as shown in The principal oxidative disorder in (bulk) NiO consists of Ni vacancies VNi in diÜerent charge states and (free) holes,54 with the reasonable likelihood that similar disorder will prevail in NiO/MgO ultra-thin –lms principally at high temperature.We have considered two charged states of the Ni vacancy in a corresponds to the removal of an Ni2` ion and neutral V(0) which corresponds to the removal of Ni an Ni2` ion plus two electrons. As above our computational resources have limited us to a concentration of 25% vacancies in the NiO layer. Furthermore we have not considered any lattice relaxation associated with either defect state. Mulliken analysis of the VNi(2`) state which is insulating indicates that the electron distribution in both layers remains largely unchanged from a non-defective Table 8. The removal of two electrons from this state results in the formation of the neutral V(0) Ni state which is also insulating with the unpaired electrons/holes localised in p orbitals at two next-nearest-neighbour O sites adjacent to the vacancy.Once again Mulliken analyses collected in Table 8 indicate that there is very little re-distribution of electron density between the two layers and that B91% of the hole density is localised in the NiO layer with local moments close to 1 l at the two O sites. In both vacancy states there is complete retention of the local Ni B moments. As Figs. 6 and 7 show despite the minimal re-distribution of electron density within and between the two layers there are substantial changes in the valence band DOS of both vacancy states. In the doubly charged state the Ni sub-bands retain largely the same pro–le as in the non-defective state but are shifted to lower energy relative to the top of the valence band by B1 eV.There is a more drastic re-arrangement of the oxygen states with a substantial drift of states from the main body of the non-defective O(p) band down to B10 eV below the valence band p F(NiO/MgO)100 2 and VNi2~ in F(NiO/MgO)100 2 . Fig. 6 Comparison of the valence band DOS of Faraday Discuss. 1999 114 105»127 116 Fig. 7 Comparison of the valence band DOS of F(NiO/MgO)100 2 and VNi in F(NiO/MgO)100 2 . upper edge. In the neutral state the re-distribution of states is even more dramatic with the three Ni sub-bands now further split as a result of the reduced symmetry of the neutral vacancy and shifted to lower energy again by B1 eV.The pro–le of the O2~ p DOS is changed to a more even distribution of states across the main part of the band width which remains close to 6 eV in width with the p states associated with the two O~ oxygens indicated as ìholeœ in Fig. 7 shifted to lower energy. As in the case of the free hole the bound holes of the neutral vacancy give rise to a narrow band of unoccupied states below the conduction band edge. However as shown in Fig. 8 these states are shifted to higher energy by B2 eV compared with the free-hole state leading to a gap of B3.6 eV between the hole and conduction bands. 3g. {100} Surface free energies Fig. 9 shows the dynamically relaxed M100N surface free energy of MgO at 700 K as a function of slab thickness which indicates that approximately ten layers are required for convergence to 0.001 F`(f)(NiO/MgO)100 2 in F(NiO/MgO) Fig.8 Comparison of the empty gap states of 100 2 . and VNi 117 Faraday Discuss. 1999 114 105»127 Fig. 9 Variation of the MgO M100N surface free energy at 700 K with slab thickness. J m~2. This is more than twice the number of layers (4) required to converge the static energy. All the calculations reported here are for slabs comprising twelve layers and 1728 q-vectors used in the Brillouin zone summation.55 The M100N surface free energy decreases slightly with temperature by 0.05 J m~2 over the temperature range 0»2600 K. Above this temperature the quasiharmonic approximation breaks down with the appearance of imaginary frequencies for the bulk on the otherhand imaginary frequencies appear at B2900 K.Thus the quasiharmonic approximation fails at somewhat lower temperatures for the M100N surface than for the bulk due to the presence of modes with large vibrational amplitudes which raises the possibility that surface melting occurs at temperatures below that for the bulk. For NiO our results can be compared directly with those obtained by Mulheran56 using an Einstein approximation. Fig. 10 shows the calculated temperature dependence of the unrelaxed and fully dynamically relaxed M100N surface energies. Here it is interesting to note that our value of the surface energy in the static limit 1.18 J m~2 compares favourably with that of 1.23 J m~2 from ab initio Hartree»Fock calculations.57 As shown in Fig.10 both the relaxed and unrelaxed M100N surface energies decrease by B0.1 J m~2 over the temperature range 0»2000 K. The change in the dynamically relaxed surface energy with temperature over this range is only one-third of Fig. 10 Calculated temperature variation of the dynamically relaxed and unrelaxed free energy of the M100N surface of NiO. Faraday Discuss. 1999 114 105»127 118 that reported by Mulheran56 for the same surface which indicates of the limitations of the Einstein approximation. There are no experimental data for direct comparison although the temperature variation appears to be consistent with that noted by Benson and Yun58 for rocksalt M100N surfaces. The calculated relaxations are also in good agreement with ab initio Hartree»Fock calculations.57 These found a contraction of the –rst inter-layer spacing of 0.53% which provides additional support for our use of two-body potentials in free energy simulations.We note also that the reported theoretical relaxations are well within the upper bound of B2% suggested by LEED measurements.59 3h. Surface vibrational densities of states Bulk and surface vibrational densities of states (VDOS) at 300 K for the M100N surface of NiO are shown in Fig. 11 together with the excess VDOS (surface minus bulk) which is responsible for the dynamic contribution to the surface free energies. The form of the bulk VDOS is in good agreement with that reported by Coy et al.60 Once again there does not appear to be any experimental data for direct comparison.The form of the excess VDOS with a decrease in intensity at B6 THz diÜers signi–cantly from that presented previously for the M100N surface of MgO due essentially to the diÜerent atomic masses of Ni and Mg. 3i. Free energies of Ni2ë segregation We have calculated heats of segregation *h of Ni2` to the surface of an MgO slab assuming the formation of ordered fully relaxed structures. At a given temperature *h is obtained from the free energies of fully relaxed twelve-layer slabs with two or four cation surface sites which may be occupied by Mg or Ni ions and the free energy of a bulk supercell containing one Ni2` ion which is an excellent approximation to an isolated impurity ion in the bulk. The top and bottom halves of slabs are the same in any calculation.The heat of segregation is therefore not determined as a continuous function of coverage but at speci–c coverages here 25% 50% 75% and 100%. The calculated heats of segregation at a range of temperatures are shown in Fig. 12 where for convenience individual points are connected by straight lines. The negative values of the segregation energy denote the segregation of Ni2` ions from bulk to surface sites. The magnitudes of *h B0.15 eV re—ect the small diÜerence in size between Ni2` and Mg2` ions and at each temperature the variation of segregation energy with coverage B4% is considerably less than that noted previously for larger cations such as Ca2` at the M001N surface of MgO in the static limit.61 The variation of the segregation energy with temperature over the range studied is somewhat Fig.11 Calculated bulk surface and excess (surface minus bulk) vibrational densities of states at 300 K for the M100N surface of NiO. 119 Faraday Discuss. 1999 114 105»127 Fig. 12 Calculated heat of segregation *h of Ni2` at the M001N surface of MgO at a range of coverages and temperatures. greater than the variation with coverage decreasing as it does by B10% from 100 K to 1900 K. Segregation energies to layers other than the surface layer are found to be negligible. As in the case of the surface energy and lattice relaxation comparisons with Hartree»Fock calculations are instructive. DiÜerences in energy between the unrelaxed structures (NiO/MgO/MgO)100 3 and (MgO/NiO/MgO)100 3 range from [0.113 eV for the A alignment to [0.103 eV for A1 which 2 compare with a value of B[0.15 eV deduced for the segregation energy from atomistic simulations.3j. Vibrational densities of states of defective surfaces It is straightforward within a lattice dynamics approach to evaluate the vibrational densities of states of defective surfaces at –nite temperatures again based on fully relaxed structures at the Fig. 13 Vibrational densities of states at 700 K for the M100N surface of NiO with 50% Mg2` coverage. The densities of states of the undefective M100N surface and the excess DOS (defective»undefective) are also plotted. Faraday Discuss. 1999 114 105»127 120 Fig. 14 Vibrational densities of states at 700 K for the M100N surface of NiO with 50% Ni vacancies compensated by oxygen holes.The densities of states of the undefective M100N surface and the excess DOS (defective» undefective) are also plotted. temperature of interest. Fig. 13 presents two examples both at 700 K the –rst (Fig. 5) corresponds to a 50% surface coverage of NiO by MgO the second (Fig. 6) to a 50% surface coverage by neutral Ni vacancies VNi (0) i.e. half the Ni surface sites vacant with each vacancy charge compensated by two oxygen holes in the surface layer. The diÜerence between the densities of states of non-defective and defective slabs in Fig. 13 indicates the depletion of most low frequency modes (\10 THz) on the introduction of the lighter Mg ions with marked decreases in intensity at B9 THz and B11 THz. These are accompanied by a pronounced shift of density to higher frequencies.Fig. 14 which admittedly relates to a somewhat unrealistic surface vacancy concentration shows a pronounced decrease in intensity for modes around 12 THz with increases at \5 THz B10 THz and B17 THz. E for 4. Discussion 4a. Ground state properties As expected the ground states of the NiO/MgO multi-layers we have considered here are all ionic high spin and insulating with electronic con–gurations Mulliken populations and valence band DOS that are very close to those of the bulk20 and M100N monolayer.27 While there is some evidence from XPS5,13 of charge transfer from MgO to NiO for sufficiently thick NiO layers our results are inconclusive in this respect as shown in Table 9 due mainly to the limited sizes of slab we have examined.The order of the spin alignments is also identical to that reported for the monolayer. However there are diÜerences in the direct and superexchange coupling energies within an NiO layer both of which increase with Ni coordination as shown in Table 2 leading to values for (MgO/NiO/MgO) decrease of B13% in superexchange coupling from 100 3 that are similar to that for the bulk. We note in particular our (MgO/NiO/MgO)100 3 to (NiO/MgO)100 3 and (NiO/MgO/MgO)100 3 which compares with a decrease of B20% reported by de Graaf et al.28 based on cluster calculations. The diÜerence in superexchange coupling within ([8.5 meV) and between ([10.5 meV) NiO layers is particularly interesting in relation to the recent report by Alders et al.5 that somewhere in excess of 20 ML of NiO on M100NMgO are required before the bulk value of the Neç el temperature is recovered.As the number of layers increases there is a competition between the A spin alignment of the M100N layered structure and the AF alignment 1 II of the bulk which the present results suggest might be tilted in favour of the bulk by the interlayer superexchange since (MgO/NiO/MgO)100 3 is calculated to be B4% greater than that for se 121 Faraday Discuss. 1999 114 105»127 the bulk. An important point worth noting here is that our interest here is primarily in the changes in the direct and superexchange coupling energies as the local Ni coordination changes and not their absolute values the reliability of which is the subject of continuing investigation.62,63 4b.d«d excitations Turning now to d]d excitation energies Table 3 shows that corrections derived from correlation only functionals39h41 based on Hartree»Fock densities are negligible. We emphasise that this does not indicate that electron correlation is unimportant or that the use of more sophisticated treatments of electron correlation within the framework of periodic calculations will come to similar conclusions. A comparison of columns (c) and (d) of Table 4 indicates that the spin alignment also appears to have a negligible eÜect on d]d excitation energies which is reasonable since the diÜerences in energy between the various spin alignments are at least two orders of magnitude less than those between diÜerent electron con–gurations.What Table 4 and Fig. 2 do show quite clearly however is that the local Ni coordination can be an important factor depending on the initial and –nal states of the excitation. Thus the energy of the dxy ]dx2~y2 excitation is more or less independent of the local Ni coordination while that of the two-electron excitation dxz/dyz ]dz2/dx2~y2 increases by B36% from (NiO)100 1 to (MgO/NiO/MgO)100 3 . On the otherxy ]dz2 hand the energies of the d dx2~y2 ]dz2 excitations increase by B109% and spin-forbidden and B64% respectively while that of the spin-forbidden dz2 ]dx2~y2 decreases by B29%. This pattern is readly explained in terms of steric hindrance eÜects. Since a dxy ]dx2~y2 excitation involves a transition entirely within a M100N layer it might reasonably be expected that the presence of planes above and below would in—uence the energy only to a very minor extent and this is what UHF calculations predict.On the other hand the dxy ]dz2 and dx2~y2 ]dz2 excitations which increase by 1.39 eV and 1.44 eV respectively from 4 to 6 coordination involve ìout-ofplaneœ transitions which might reasonably be expected to be restricted by planes above and below leading to increases in energy and again this is what we –nd. While individual points of detail await further clari–cation there is general agreement29,30,51 as to the assignment of the reported optical absorption and EEL spectra below B3 eV to speci–c bulk and M100N surface d]d excitations. These can be compared with our values for (NiO/MgO)100 2 and (MgO/NiO/MgO)100 2 which we take to be representative of the 5-fold coordination of the M100N surface and 6-fold coordination of the bulk.With reference to Table 10 absorptions at B1.0 eV B1.9 eV and B2.8 eV have been assigned to spin-allowed 3A2g ]3T2g 4h [(t2g 6 eg2)](t2g 5 eg3)] 3A2g ]3T1g [(t2g 6 eg2)](t2g 4 eg4)] and 3A2g ]3T1g [(t2g 6 eg2)](t2g 5 eg3)] excitations D in bulk NiO. In our calculations the (MgO/NiO/MgO)100 2 removes part of the symmetry of 3T and degeneracy of the 3T states from which we obtain energies of 0.81 eV 2.14 eV and 2.66 2g 1g eV corresponding to the bulk excitations. We have not obtained a converged solution for the dxz ]dz2 excitation but with reference to Fig. 2 note that were this transition to follow the linear relationship between energy and coordination number the energy for the trilayer would be 1.30 eV.These compare with values of 0.79 eV 1.40 eV and 3.40 eV reported by de Graaf et al.30 and energies of (0.86»1.04) eV and (1.50»1.81) eV reported by Freitag et al.29 for the –rst two excitations based on cluster calculations which included electron correlation eÜects. As shown in Table 10 EEL excitations at B0.6 eV B1.0 eV B1.3 eV B1.6 eV have been assigned to the M100N surface of NiO,29,50,51 and that at B2.2 eV to a possible transition at a M100N NiO/MgO interface.6,10 These compare with our calculated values for (NiO/MgO)100 2 of Table 9 Comparison of interlayer electron transfer dq (e/NiO) in various NiO/MgO (100) multi-layers System dq (NiO/MgO)1002 ]0.006 (NiO/MgO/MgO)1003 [0.007 (NiO/MgO/NiO)1003 [0.005 (MgO/NiO/MgO)1003 ]0.032 Faraday Discuss.1999 114 105»127 122 Table 10 Comparison of EELS and theoretical d]d transition energies (eV) Experiment Bulk» Energy/eV Transition 3A2g ]3T2g 1.05,a 1.08,b,c 1.10,d (t2g 6 eg2)](t2g 5 eg3) 1.13,a,e 1.16,f 1.1g 1.79,a 1.86,b 1.87,d 3A2g ]3T1g (t2g 6 eg2)](t2g 4 eg4) 1.95,a,e 2.75,f 2.8,g B3a 3A2g ]3T1g (t2g 6 eg2)](t2g 5 eg3) M100N Surface» Energy/eV Transition 0.57,d 0.60g,h 1.0h 1.3h 1.6,g 1.62,d 1.63f 3B1 ]3E 3B1 ]3B2 3B1 ]3A2 2.18i 3B1 ]3E 3B1 ]3E a Ref. 48. b Ref. 47. c Optical absorption. d Ref. 29. e Ref. 46. f Ref. 49. g Ref. 50. h Ref. 51. i Refs. 6 and 10. 0.83 eV 0.83 eV 1.16 eV 1.87 eV and 1.98 eV with an average discrepancy of B0.2 eV.Furthermore as pointed out by Fromme et al.,51 the energy of the dxy ]dx2~y2 excitation would be expected to remain more or less unchanged from 6-fold to 5-fold coordination and that of the 100 2 is only an approximate representation of the non-defective dxz ]dz2 excitation to decrease which as Table 4 shows is exactly what we –nd. Thus bearing in mind that unrelaxed (NiO/MgO) semi-in–nite M100N surface of NiO the inclusion of which is largely to examine the eÜects of changing the local Ni coordination our results would seem to reproduce qualitatively the surface d]d excitation energies. Table 10 also shows that overall our results are at least comparable to those reported by Freitag et al.29 and Geleijns et al.,31 which suggests that for extended/periodic systems the neglect of correlation eÜects beyond Hartree»Fock would seem to be less important than it is for (NiO6)10~ and (NiO5)8~ clusters.4c. Hole states and band gaps In view of its p-type properties the nature of hole states in NiO has continued to attract considerable attention.15 Previous UHF studies of the bulk24,25 and M100N monolayer27 have provided direct evidence for localised holes with strong O(p) character in complete agreement with Sawatzky and coworkers.17h19 They have shown by analogy with the oxygen k-edge spectra of LixNi1~xO,17 that reasonable estimates of the band gap Eg in NiO can be obtained from the gap between the polaron band and conduction band edge as shown in Fig. 3. Here we –nd that the calculated band gap of the unsupported M100N monolayer 5.1 eV which is B25% greater than that of the bulk increases in (NiO/MgO)100 2 to 5.7»6.0 eV for diÜerent hole states.As suggested by the ground state valence band DOS holes within the NiO layer are more stable than those within MgO by 0.27»0.64 eV with further diÜerences in energy between diÜerent spin con–gurations of 0.03»0.16 eV for the NiO layer and 0.37 eV for the MgO layer. As in the case of other quantities of interest we have estimated the variation of E with Ni coordination. Fig. 4 shows a near linear 100 1 g E from 5.1 eV in (NiO) increase in (MgO/NiO/MgO)100 3 where in each case the to 6.6 eV in g hole/unpaired electron is in a p orbital in the NiO plane. This trend indicates that the hole is less (MgO/NiO/MgO)100 3 p than it is in (NiO)100 1 which re—ects the change in Madelung stable in potential.Pauli repulsion between the unpaired electron and the neighbouring Mg2` ions might also contribute to this increase. Faraday Discuss. 1999 114 105»127 Theory Present Ref. 30 Ref. 29 0.81 0.79 0.86»1.04 2.14 1.40 1.50»1.81 2.66 3.40 » Present Ref. 31 Ref. 29 0.83 0.83 1.16 1.87 1.98 0.46 0.83 1.04 1.17 2.55 0.54»0.65 0.86»1.00 1.11»1.30 1.22»1.44 » 123 As in a previous study of the M100N monolayer,27 we have obtained variationally-converged solutions for e4 pp ]dz2 charge»transfer states of F(NiO/MgO)100 2 from which the corresponding excitation energies are obtained from direct energy diÜerences.The particular signi–cance of the latter within the context of this study is that they are lower bounds to the band gap and hence provide support or otherwise for the values obtained from the single-particle eigenvalues. The formation energy/Ni of the fully condensed triplet excitonic state is B0.5 to B0.7 eV less than Eg compared with B1.7 eV found previously for the monolayer.27 While there are no experimental values of exciton formation for direct comparison with our values the well known tail in the optical absorption coefficient of NiO from the onset of absorption at 3.1 eV to the maximum increase in intensity at 4.0 eV64 might reasonably be explained in terms of bound excitonic states below the conduction band edge. This is supported further by the energy of the –rst excited excitonic state B5.9 eV which suggests that the –rst excited state lies very close to the conduction band edge as it does for the F-centre in MgO and CaO.65 Structural defects play an important role in controlling the properties of all materials and here we have considered two classes of defect that might be expected to occur in ultra-thin –lms of NiO on MgO.They arise from intermixing and non-stoichiometry both of which are known to be prevalent in the bulk. Despite the fact that intermixing clearly destroys the structural integrity and low-temperature magnetic order of overlayers it appears to lead to only very minor changes in the electron distribution and densities of states. This in turn suggests that there will be no gross changes in the valence band spectra of layered NiO/MgO systems resulting from interdiÜusion.While our results for Ni vacancy states correspond to levels of non-stoichiometry far in excess of that normally encountered and make no allowance for lattice relaxation there seem to be no obvious reasons why they are inapplicable to lower concentrations. The most signi–cant features of these results are –rst that even at a concentration of 25% both the doubly-charged (VNi2~) and neutral (VNi) vacancy states are insulating ; second that in common with the free hole state there is strong localisation of unoccupied p density at two non-adjacent O sites of VNi with the development of local moments close to 1 p lB ; and third that in both cases there is only a minor re-distribution of electrons or screening compared with the non-defective lattice with changes of less than ^0.05 e in the atomic charges.However the comparatively minor changes in electron distribution lead to more substantial changes in the valence band densities of states as shown in Figs. 6 and 7. For both vacancies there is a downward shift of the Ni(d) states relative to the top of the valence band by B1 eV and in the case of VNi a noticeable broadening and splitting of these states due to a reduction in symmetry. The densities of O(p) states are also broadened with a signi–cant weight in the region of the Ni(d) levels for VNi2~. This reduction in energy might in part be attributed to a small delocalisation of O(p electrons towards the vacant Ni site which would lower their kinetic energy but leave the Mulliken populations more or less unchanged.By comparison with VNi Ni 2~ V the reduced O(p) density of would be expected to lead to less delocalisation and hence less re-distribution of states to lower energy other than those speci–cally associated with the two hole sites which are stabilised by a reduction in the on-site Coulomb repulsion energy as noted previously for the free hole.25 In addition to changes in the valance band DOS of VNi there is also a signi–cant change to the gap between the polaron band and the conduction band edge which is reduced to B3.6 eV compared with B5.7 eV for the free carrier. Since these gaps equate to the E of the corresponding holeless states this indicates that the O(p) states of the g doubly charged vacancy state are raised relative to the conduction band edge exactly as expected from the eÜective 2- charge of a Ni2` vacancy.4d. Lattice simulations s) The lattice simulations reported here use a recent code43 based on lattice statics and quasiharmonic lattice dynamics and designed to minimise the free energy of crystalline slabs which are –nite in one direction and in–nite in the other two. For simulations involving large unit cells numerical diÜerentiation of the free energy with respect to all the internal coordinates is normally prohibitively expensive which has led to approaches such as the zero static internal stress approximation (ZSISA) in which the free energy is minimised with respect to the lattice vectors only.In this more recent development the full set of free energy –rst derivatives is calculated analytically leading to a complete minimisation of the quasiharmonic free energy with respect to all the cell Faraday Discuss. 1999 114 105»127 124 variables. Furthermore since the sizes of slab used are sufficiently thick to provide a bulk-like region in the interior of the slab and two essentially free surfaces the simulations reported here represent a radical departure from the two-region strategy used by Tasker66 and Gay and Rohl,67 in which the positions (and polarisations) of the ions in the vicinity of the surface only are relaxed explicitly by minimising the internal energy of the system while the remainder are constrained to their bulk lattice positions.Two important points emerge from our surface free energy calculations. The –rst is the requirement that many more layers are required for convergence than for static simulations. This is due to the contribution to the free energy of long wavelength phonons perpendicular to the slab particulary at low temperatures. Similar considerations apply to molecular dynamics simulations where the unit cell size prevents the inclusion of small q vibrations. The second is the limitation of the Einstein approximation which again severely restricts the range of phonons that contribute to the free energy. Fundamental to all simulations is the quality of the interatomic potentials particularly where a paucity of experimental data limits the extent to which the validity of the simulations can be examined.Here support for the potentials used is provided by the close agreement between the static surface energy and lattice relaxation predicted from UHF calculations and lattice simulations. In the absence of data for direct comparison it is also re-assuring that the temperature dependence of the surface free energy is consistent with that for other rocksalt systems. With regard to NiO/MgO multilayers perhaps the most signi–cant of the simulation results concern the free energies of Ni segregation to the M100N surface of MgO for they appear to con–rm the stability of NiO overlayers even at high temperatures. However this is clearly a complex issue which involves both thermodynamic and kinetic factors particulary in view of the reported negative deviation from ideality of the enthalpy of mixing of NiO and MgO.52 5.Conclusions The overall conclusion of this paper is that useful information concerning the electronic magnetic structural and dynamic properties of NiO monolayers on MgO M100N surfaces can be obtained from a combination –rst principles unrestricted periodic Hartree»Fock calculations and atomistic lattice simulations based on quasi-harmonic lattice dynamics. From an examination of several semi-in–nite slab con–gurations including an unsupported NiO monolayer an NiO monolayer deposited on MgO M100N layers and an NiO layer sandwiched between two MgO M100N layers we have been able to obtain the dependence of physical quantities such as the magnetic exchange coupling constants and the d]d and charge transfer excitation energies on the Ni coordination number Z which increases from 4 to 6 in this series.Furthermore they are all obtained from diÜerences in total Hartree»Fock energies and show a remarkably monotonic variation with Z. Further work currently in progress compares these results with those of pure bulk NiO and systems with reduced dimensionality.68 In addition information of practical importance has been obtained for Ni/Mg substitutional defects and cation vacancies in the NiO layer both of which might reasonably be expected to occur in ultra-thin –lms of NiO on MgO as a result of intermixing and non-stoichiometry. Finally free energy calculations based on surface lattice dynamics have been used to estimate the surface contribution to the vibrational density of states of NiO M100N and the heat of segregation of Ni2` to the MgO surface as a function of temperature which appears to con–rm the stability of NiO overlayers even at high temperature.These show that potential-based approaches are capable of providing detailed dynamic and temperature-dependent information for which at present there is little experimental data for comparison. The results reported here also show that the calculation and subsequent minimisation of the free energy via quasiharmonic lattice dynamics is sufficiently rapid that an extension to more complex surface structures with more extensive disorder is perfectly feasible. Acknowledgements W.C.M. and C.N. wish to thank the British Council and the French Ministere des AÜaires Etrangeres for the award of a grant within the Alliance scheme which has facilitated the work reported here.125 Faraday Discuss. 1999 114 105»127 References 1 H. Kuhlenbeck and H.-J. Freund in Growth and Properties of Ultrathin Epitaxial L avers ed. D. A. King and D. P. WoodruÜ Chem. Phys. Solid Surf. 1977. 2 D. W. Goodman J. V ac. Sci. T echnol. A 1996 14 1526. 3 T. Fujii D. Alders F. C. Voogt T. Hibma B. T. Thole and G. A. Sawatzky Surf. Sci. 1996 366 579. 4 R. Hesper L. H. Tjeng and G. A. Sawatzky Europhys. L ett. 1997 40 177. 5 D. Alders L. H. Tjeng T. Hibma B. T. Thole G. A. Sawatzky C. T. Chen J. Vogel M. Sacchi and S. Iacobucci Phys. Rev. B Condens. Matter 1998 57 11623. 6 C. Xu W. S. Oh Q. Guo and D.W. Goodman J. V ac. Sci. T echnol. A 1996 14 1395. 7 S. Imadduddin and R. J. Lad Surf. Sci. Spectra 1996 4 194. 8 M. L. Burke and D. W. Goodman Surf. Sci. 1994 311 17. 9 D. Alders F. C. Voogt T. Hibma and G. A. Sawatzky Phys. Rev. B Condens. Matter 1996 54 7716. 10 Q. Guo C. Xu and D. W. Goodman L angmuir 1998 14 1371. 11 D. M. Lind S. D. Berry G. Chern H. Mathias and L. R. Testardi Phys. Rev. B Condens. Matter. 1992 45 1838. 12 S. D. Peacor and T. Hibma Surf. Sci. 1994 301 11. 13 J. M. Sanz and G. T. Tyuliev Surf. Sci. 1996 367 196. 14 M. D. Towler N. M. Harrison and M. I. McCarthy Phys. Rev. B Condens. Matter 1995 52 5375. 15 S. Hué fner Adv. Phys. 1994 43 183. 16 K. Terakura A. R. Williams T. Oguchi and J. Kué bler Phys. Rev. B Condens. Matter 1984 30 4734.17 P. Kuiper G. Kruizinga J. Ghijsen G. A. Sawatzky and H. Verweij Phys. Rev. L ett. 1989 2 221. 18 J. van Elp B. G. Searle G. A. Sawatzky and M. Sacchi Solid State Commun. 1991 80 67. 19 J. van Elp H. Eskes P. Kuiper and G. A. Sawatzky Phys. Rev. B Condens. Matter 1992 45 1612. 20 M. D. Towler N. L. Allan N. M. Harrison V. R. Saunders W. C. Mackrodt and E. Apra` Phys. Rev. B 21 M. D. Towler N. L. Allan N. M. Harrison V. R. Saunders and W. C. Mackrodt J. Phys. Condens. 22 N. M. Harrison V. R. Saunders R. Dovesi and W. C. Mackrodt Philos. T rans. R. Soc. L ondon Ser. A 23 R. Dovesi V.R. Saunders C. Roetti M. Causa` N. M. Harrison R. Orlando and E. Apra` CRY ST AL 95 Condens. Matter 1994 50 5041. Matter 1995 7 6231. 1998 56 75. User Manual Universita` di Torino and CCLRC Daresbury Laboratory 1995.24 W. C. Mackrodt N. M. Harrison V. R. Saunders N. L. Allan and M. D. Towler Chem. Phys. L ett. 1996 250 66. 25 W. C. Mackrodt Ber. Bunsen-Ges. Phys. Chem. 1997 101 169. 26 C. Springhorn and H. Schmalzried Ber. Bunsen-Ges. Phys. Chem. 1994 98 746. 27 C. Noguera and W. C. Mackrodt submitted for publication. 28 C. de Graaf R. Broer and W. C. Nieuwpoort Chem. Phys. L ett. 1997 271 372. 29 A. Freitag V. Staemmler D. Cappus C. A. Ventrice K. Al Shamery H. Kuhlenbeck and H.-J. Freund Chem. Phys. L ett. 1993 210 10. 30 C. de Graaf R. Broer and W. C. Nieuwpoort Chem. Phys. 1996 208 35. 31 M. Geleijns C. de Graaf R. Broer and W. C. Nieuwpoort Surf. Sci. 1999 421 106. 32 M. B. Taylor C. E. Sims G.D. Barrera N. L. Allan and W. C. Mackrodt Phys. Rev. B Condens. Matter 1999 59 6742. 33 D. Fincham W. C. Mackrodt and P. J. Mitchell J. Phys. Condens. Matter 1994 6 393. 34 G. D. Barrera M. B. Taylor N. L. Allan T. H. K. Barron L. N. Kantorovich and W. C. Mackrodt J. Chem. Phys. 1997 107 4337. 35 C. Pisani R. Dovesi and C. Roetti Hartree»Fock Ab Initio T reatment of Crystalline Systems Springer- Verlag Berlin Heidelberg New York 1988. 36 J. A. Pople and R. K. Nesbet J. Chem. Phys. 1954 22 571. 37 R. Dovesi M. Causa` R. Orlando C. Roetti and V. R. Saunders J. Chem. Phys. 1990 92 7402. 38 H. J. Monkhorst and J. D. Pack Phys. Rev. B Condens. Matter 1976 13 5188. 39 J. P. Perdew Phys. Rev. B Condens. Matter 1986 33 8822; J. P. Perdew Phys. Rev. B Condens.Matter 1986 34 7406. 40 J. P. Perdew and Y. Wang Phys. Rev. B Condens. Matter 1986 33 8800; J. P. Perdew and Y. Wang Phys. Rev. B Condens. Matter 1986 40 3399; J. P. Perdew and Y. Wang Phys. Rev. B Condens. Matter 42 M. B. Taylor G. D. Barrera N. L. Allan and T. H. K. Barron Phys. Rev. B Condens. Matter 1997 56 1992 45 13244. 41 J. P. Perdew K. Burke and M. Ernzerhof Phys. Rev. L ett. 1996 77 3865; J. P. Perdew K. Burke and M. Ernzerhof Phys. Rev. L ett. 1997 78 1396; J. P. Perdew K. Burke and M. Ernzerhof Phys. Rev. L ett. 1998 80 891. 14380. 43 M. B. Taylor G. D. Barerra N. L. Allan T. H. K. Barron and W. C. Mackrodt Comput. Phys. Commun. 1998 109 135. Faraday Discuss. 1999 114 105»127 126 44 W. H. Press S. A. Teukolsky W. T. Vetterling and B.P. Flannery Numerical Recipes in Fortran Cambridge University Press Cambridge New York Victoria 2nd edition 1992 p. 420. 45 R. S. Mulliken J. Chem. Phys. 1955 23 1833; R. S. Mulliken J. Chem. Phys. 1955 23 1841. 46 R. Newman and R. M. Chrenko Phys. Rev. 1959 114 1507. 47 V. Propach D. Reinen H. Drenkhaln and H. Mué ller Buschbaum Z. Naturforsch. B Anorg. Chem. Org. Chem. 1978 33 619. 48 P. A. Cox and A. A. Williams Surf. Sci. 1985 152 791. 49 S. Hué fner P. Steiner F. Reinert H. Schmitt and P. Sandl Z. Phys. B Condens. Matter 1992 88 247. 50 B. Fromme M. Schmitt E. Kisker A. Gorschlué ter and H. Merz Phys. Rev. B Condens. Matter 1994 50 1874. 51 B. Fromme M. Moé ller Th. Anschué tz C. Bethke and E. Kisker Phys. Rev. L ett. 1996 77 1548. 52 P.K. Davies and A. Navrotsky J. Solid State Chem. 1981 38 264. 53 K. D. Heath W. C. Mackrodt V. R. Saunders and M. Causa` J. Mater. Chem. 1994 4 825. 54 See for example P. A. Cox T ransition Metal Oxides Clarendon Press Oxford 1992. 55 It is crucial to use enough q-vectors to achieve convergence. Too small a number of q-vectors leads in general to a much smaller decrease in the surface free energy with temperature than is shown by the converged values. A grid of sufficient size must be used in the reciprocal space summation to take adequate account of long-wavelength modes. 56 P. A. Mulheran Philos. Mag. A 1993 68 799. 57 J. V. Mitchell M. Chem. T hesis University of St. Andrews 1995. 58 G. C. Benson and K. S. Yun T he Solid-Gas Interface ed. E. A. Flood Arnold London 1967. 59 M. R. Welton-Cook and M. Prutton J. Phys. C Solid State Phys. 1980 13 3993. 60 R. A. Coy C. W. Tompson and E. Gué rmen Solid State Commun. 1976 18 845. 61 W. C. Mackrodt and P. W. Tasker J. Am. Ceram. Soc. 1989 72 1576. 62 I. D. Moreira and F. Illas Phys. Rev. B Condens. Matter. 1997 55 4129. 63 C. de Graaf F. Illas R. Broer and W. C. Niewpoort J. Chem. Phys. 1997 106 3287. 64 R. J. Powell and W. E. Spicer Phys. Rev. B Condens. Matter 1970 2 2182. 65 T. M. Wilson and R. F. Wood J. Phys. Solid State Phys. 1976 7 190. 66 P. W. Tasker in Computer Simulation of Solids ed. C. R. A. Catlow and W. C. Mackrodt Springer-Verlag Berlin Heidelberg New York 1982 p. 288. 67 D. H. Gay and A. L. Rohl J. Chem. Soc. Faraday T rans. 1995 91 925. 68 W. C. Mackrodt and C. Noguera unpublished results. Paper 9/04185K 127 Faraday Discuss. 1999 114 105»127

ISSN:1359-6640    

DOI:10.1039/a904185k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Scanning tunnelling microscopy on the growth and structure of NiO(100) and CoO(100) thin �lms Ina Sebastian,a Thomas Bertrams,a Klaus Meinelb and Henning Neddermeyer*a a Martin-L uther-Universitaé t Halle-W ittenberg Fachbereich Physik D-06099 Halle Germany b L ettiner Str. 9 D-06120 Halle Germany I. Introduction Ordered thin –lms of oxides are of general importance and not much is known of their formation or of their structural and morphological details. In the case of NiO and CoO ordered thin oxide –lms can be grown by oxidation of metal single crystal surfaces (see for example ref. 1 where the growth of NiO(100) layers on Ni(100) has been described). The low-energy electron diÜraction (LEED) pattern of oxidised Ni(100) surfaces however mostly indicates the presence of a rather imperfect NiO(100) –lm on top of Ni(100),1 which might result from the large lattice mismatch (of nearly 20%) between Ni and NiO.Although the use of highly oriented Ni(100) as substrate leads to considerable improvement of the LEED pattern from the oxidised surface2 the large lattice mismatch always remains a problem with the possible consequence of defect structures in the oxide –lm. A diÜerent approach to prepare a well ordered NiO(100) –lm was used by Marre and Neddermeyer3 who evaporated Ni onto an Ag(100) surface in an O atmosphere. LEED and UV 2 photoelectron spectroscopy have shown that a well ordered and smooth NiO(100) can indeed be obtained in this way. The reason for this more perfect growth of NiO(100) layers by using an 129 Faraday Discuss.1999 114 129»140 Received 28th April 1999 We have prepared ordered thin –lms of NiO and CoO in (100) orientation by evaporating Ni (Co) in an O atmosphere onto Ag(100). The –lms have been analysed by scanning 2 tunnelling microscopy and low-energy electron diÜraction. In the initial stage (coverage up to a few monolayers) growth and structure of the grown –lms drastically depend on the preparation conditions (in particular on the temperature of the substrate during deposition and post-annealing). In this case we also observe strong interactions with the substrate. Ag atoms are partially removed from the substrate terraces and form islands or migrate to step edges. No indications for incorporation in the oxide thin –lms are seen. The oxidic features grow on top of the substrate or in the vacancy islands within the –rst layer of the substrate left behind by the removed Ag atoms.At low substrate temperatures (near room temperature) an essential part of the oxidic features corresponds to a precursor state rather than to the fully developed (100) oxide –lm which only develops after post-annealing to higher temperatures (typically around 500 K). I/U characteristics and the sample bias dependency of the contrast of the islands grown have been utilised for identi–cation of whether an oxide reaction had taken place or not. The surfaces of the oxide precursor show a typical defect structure similar to those found on cleaved NiO(100) (M. R. Castell et al. Phys. Rev. B Condens. Matter 1997 55 7859).This feature shows ìrandom walkœ at room temperature. This journal is( The Royal Society of Chemistry 2000 Ag(100) substrate is the smaller lattice mismatch (only 2%) between NiO and Ag. These results were later con–rmed by scanning tunnelling microscopy.4 It has been demonstrated4 that deposition of Ni onto Ag(100) in an O atmosphere leads to much better NiO(100) –lms than oxidation 2 of Ni(100) at least for surface orientation which is accessible by standard Laue techniques. It is the purpose of the present work to provide some additional details on the growth of NiO(100) –lms on Ag(100) details that have not been described previously,4 and to compare the results from the NiO(100) –lms with those from CoO/Ag(100) prepared and measured in the same way. The lattice mismatch between CoO(100) and Ag(100) is slightly larger (4%) than of NiO(100)/ Ag(100) and it might be interesting to look for defect structures associated with the diÜerences of the lattice mismatch.It should be mentioned that Castell et al.5 have observed at elevated temperatures (nearly 500 K) atomic defect states on the (100) cleavage planes of bulk NiO. Since similar defect states had already been detected in our previous study of NiO(100) –lms but had not been described explicitly we have also concentrated on the presence of such defect states for the CoO(100) –lms. Surprisingly defect states on the precursor CoO(100) layers which look similar to those on NiO(100) exhibit a ìrandom walkœ over distances of 1»2 lattice parameters during the observation time (typically 1 min) at room temperature.This implies a particular small activation barrier for surface diÜusion of such defects at least for the oxide precursor. An additional issue which will be discussed is the possibility of Ag atom removal from the substrate during growth of the oxide layers. In the previous work4 the large rearrangement of the Ag(100) substrate during post-annealing of the deposited –lms has been mentioned but without describing it in more detail. For the CoO layers we have observed particularly strong eÜects for the Ag(100) substrate with direct consequences for the growth mode of the oxide –lms. Although less drastic similar eÜects are present in the NiO/Ag(100) system. The present work will demonstrate that the growth phenomena and the structural changes of the oxidic –lms when changing from a precursor structure to a well ordered (100) oxide layer are very complicated and are also dependent on the layer thickness.Only for thicker –lms was a rather uniform appearance of the surfaces detected. In the very thin –lm limit (1»2 monolayers) the surprising variety of structures and island shapes indicates an unexpectedly large interaction between the growing –lm and the substrate. We should mention that comparable work has been done for NiO6 and CoO7 layers on Au(111). II. Experimental The instrumentation used for the preparation and characterisation of the oxidic –lms has been described in some more details by Berghaus et al.8 An overview of the apparatus during the measurements of the NiO layers is given in ref.4. In the meantime we have improved the STM equipment to obtain greater stability and reproducibility for the coarse approach and better positioning of the sample. In addition a spot pro–le analysis LEED system (SPALEED) has been added to the system which allows the measurement of high-resolution LEED patterns of the sample. Such measurements turned out to be important for the analysis of moireç structures that can result from the lattice mismatch of the oxide and the metallic substrate. The Ag(100) crystals used in the present measurements were cleaned by cycles of ion bombardment (Ar`) and annealing until LEED showed a satisfactory pattern. In some cases the spots from the substrate were rather broad due to imperfect polishing of the surfaces and the small terrace size.In order to explore the growth mode of NiO and CoO more systematically and to control whether metallic Ni or Co were no longer present within the oxide –lms we also studied the condensation of clean Ni and Co layers. It is known from the literature that neither Ni3 nor Co9 grows in a layer-by-layer-like growth mode. This has been con–rmed in the present experiments which show the growth of three-dimensional clusters of the pure metals in the monolayer regime. The NiO and CoO layers have been grown by deposition of Ni (Co) onto Ag(100) in an O2 atmosphere of 1]10~6 mbar O with the formation of the oxides at a rate of approximately 2/3 2 (0.2) monolayers (ML) per minute. This means that an O excess is present during the oxide –lm 2 growth.The in—uence of sample heating both during and following the deposition was been studied in more detail and showed drastic eÜes. The thicker –lms in particular those obtained after moderate annealing or by deposition on a slightly heated substrate showed the pattern of the oxides with fourfold symmetry along the surface normal. Faraday Discuss. 1999 114 129»140 130 The STM data were acquired in the form of constant current topographies (CCTs). To study the electronic eÜects on oxide formation we systematically varied the sample-bias voltage and its polarity. The stability of the measurements during measuring of the oxidised surfaces was not very good in general. The state of the tip changed frequently probably due to transfer of species from the sample to the tip.In some cases the tip changes caused contrast variations of the measured surface structures. By comparison with the results obtained after varying of the sample-bias voltage it turned out that the tip changes could be used for identi–cation of the oxidic structures on the surface. Unfortunately such tip changes occurred randomly and could not be obtained at will. In order to study electronic eÜects upon oxide formation more quantitatively we also measured local I/V characteristics. These characteristics showed characteristic diÜerences on the oxidic precursor and the –nal oxide structures and will be described below for NiO/Ag(100). III. Results and discussion During the measurements three eÜects mainly determined the oxide –lm growth for both NiO and CoO.Firstly in the initial stage and for deposition at room (or slightly elevated) temperatures Ag atoms were partially removed from the substrate surface and formed islands with a typical squarelike or rectangular shape. The attachment to step edges has not directly been proved but has also to be considered since the surface area of vacancy islands on the substrate was larger than that of the visible Ag islands. Secondly for room (or slightly elevated) temperatures the condensed material grew in the form of —at islands with a height of one monolayer (ML); this can probably be explained by a kind of oxide precursor. Thirdly by annealing the oxide precursor a transformation of the islands took place into the (100) oriented oxide –lm with a minimum thickness of 2 ML.These observations were made for both materials. DiÜerences exist in the details of the growth process and in the observed microscopic and atomic structure and will subsequently be described for both NiO and CoO. III.A. NiO(100) Some results of our experiments on the NiO(100) –lm growth have already been described in ref. 4 and we will concentrate here on the growth mode of the oxidic precursor on the defects of the oxide –lms and on the description of I/U curves measured on the substrate the precursor and the oxide –lm. In Fig. 1 an overview image on the precursor structure is reproduced. As has already been shown and discussed in ref. 4 the CCTs exhibit —at and connected islands on top of the substrate surface with edges mainly along the [110]-like symmetry directions.These directions were determined from atomically resolved STM measurements of the clean Ag(100) surface (not shown here) and from LEED. The growth mode of the features has completely changed in comparison to that of the clean metal. While in the latter case for a coverage of 2/3 ML three-dimensional Ni clusters with a typical lateral extension of 2»3 nm and a height of 0.3»0.5 nm were found the oxidic precursor grew in the form of two-dimensional islands. These islands show rows of atomically resolved features at a distance of 0.3 nm which agrees with the Ag lattice parameter. The precursor cannot be explained by an O/Ni adsorption structure which is known to develop either a p(2]2) or a c(2]2) symmetry.10 Since neither the observed distances nor the symmetry agrees with either of these adsorption structures we argue that the islands correspond to an oxidic precursor that already contains O ions in sufficient number to form NiO(100) islands in the approximately correct stoichiometry after annealing without needing further O exposure.Note that a 2 twofold symmetry has developed on the surfaces of the precursor. In agreement with the fourfold symmetry of the substrate two orthogonal orientations of the row structures are recognised on the islands. A more detailed inspection of the CCT reveals that the typical row structure on the surfaces of the islands has also formed within parts of the outermost Ag substrate layer. Its average grey tone level essentially agrees with that of the uncovered part of the Ag substrate.This observation can only be explained if one assumes that the condensation process gives rise to the reactive removal of Ag atoms in the substrate surface layer and that a preferential formation of the precursor is found in the vacancy islands of the substrate. Note that for O/Ni/Ag(100) ììemptyœœ vacancy islands are not seen on the surface. 131 Faraday Discuss. 1999 114 129»140 Fig. 1 2/3 ML Ni deposited in an O atmosphere (corresponding to an exposure of 50 L) onto Ag(100) at 2 room temperature. The surface has been measured at U\[0.5 V at I\2 nA the size is 30]30 nm2. The smooth dark parts of the surface correspond to initial Ag(100) surface. The incorporation of the c(1]2) O/Ni precursor structure in the Ag surface is seen on several parts e.g.at location denoted by ìì3œœ . The protruding islands have on most parts developed the c(1]2) O/Ni precursor structure (e.g. at ìì2œœ). The smooth parts on the islands (e.g. at ìì1œœ correspond to Ag islands of monatomic height on top of the substrate. They contain the Ag atoms removed from substrate during the deposition. Steps are not available on this part of the surface for the condensation of the migrating Ag atoms. Annealing the surface produces the –nal oxide state in the deposit (Fig. 2). The NiO(100) islands correspond to the protruding or depressed structures with rectangular shape in Fig. 2(a) and (b) when measured with larger or smaller sample bias voltage as given in the caption. The surface area covered by these islands is smaller by a factor of two compared to that of the precursor.This means that the minimum thickness of the NiO(100) islands is 2 ML. Two further points have to be mentioned and discussed. Firstly the step and terrace structure of the substrate surface after condensation and annealing of NiO is completely diÜerent to the clean Ag(100) sample. These large rearrangement eÜects of the substrate surface have already been described in ref. 4 but can be understood much better now. If one considers the fact that vacancy islands are already formed on the substrate during the initial stage of deposition then these rearrangements are the consequence of NiO growth on top of the substrate and condensation of the remaining vacancy islands.Note that the NiO islands are always surrounded by Ag sometimes only in the form of small ridges. The overall morphology of the sample surface is then determined by the position of the NiO islands and by minimising the step length of Ag terraces. A second point to be mentioned is the observation of a large number of defects on top of both NiO and Ag. These defects have already been mentioned4 but are represented here again (with improved visibility) in order to compare them with our results on CoO following below. Since the I/U characteristics are essentially determined by the electronic structure of the sample surface we measured them on characteristic parts of the surfaces. These measurements frequently gave rise to instabilities in the tunnelling current and could not be used for evaluation of local dI/dU dependences due to the large noise.However the non-diÜerentiated I/U curves did already show characteristic behaviour on the speci–c parts of the surface that supported the above assignments of observed features to the metal substrate Ag islands oxide precursor and NiO structure. In Fig. 3(a) the local tunnelling current I is plotted against the sample voltage U while keeping the sample»tip distance constant for smooth parts of the two-dimensional islands (e.g. at location ìì1œœ) shown in Fig. 1 and for the c(1]2) structure (position marked with ìì2œœ). The two characteristics are very similar and clearly show the typical behaviour of a metallic state namely an ohmic dependency at the Fermi level (corresponding to a sample bias U\0 V).For positive values of U the conductivity on top of the smooth parts of the islands (corresponding to Ag) is larger than on the reconstructed parts of the surface. This can probably be explained by the diÜerences in the electronic states of Ag and Ni. The d states of metallic Ni directly cut the Fermi level in contrast Faraday Discuss. 1999 114 129»140 132 Fig. 2 2/3 ML Ni deposited in an O atmosphere onto Ag(100) at room temperature (the same experiment as 2 shown in Fig. 1) and subsequently annealed at 450 K. (a) U\5 V I\2 nA image size\50]100 nm2. (b) The same part of the surface when measured with small sample bias (U\1 V) . The NiO islands now appear as depressions. Note that the surface area covered by the NiO islands is about half of that of the precursor structure in Fig.1. Individual Ag islands within the NiO double layer structures are no longer detected. to Ag where only the free-electron sp states may contribute to the tunnelling current. In addition the presence of the O atoms in the O/Ni structure will also in—uence the valence and conduction band states. Anyway the metallic behaviour on the c(1]2) ordered parts of the surface indicates that NiO has not yet formed on the surface. After the O/Ni structure has been transferred into the oxide structure the I/U characteristics (Fig. 3(b)) show drastic diÜerences. For negative and positive sample bias voltages up to about U\3 V the tunnelling current on top of the oxide islands is negligible while that on the Ag substrate again shows ohmic behaviour.For large positive sample bias voltages (around U\4 V) the tunnelling current on the oxide islands is comparable to that on the substrate. At this voltage the electronic states of NiO are accessible and obviously do contribute to the tunnelling current. As a consequence the oxide islands are imaged in form of protruding islands. III.B. CoO(100) The general growth and formation phenomena observed for the CoO islands were found to be rather similar to that for the NiO –lms. A closer close inspection of the results revealed a number of diÜerences which made it worth considering the results in the present work. Some of the results might re—ect the diÜerences in the possible oxidation states of Ni and Co which can lead to the formation of Co and the spinel structure in the latter case.11 It should be mentioned in this 3O4 context however that for (111) oriented thin –lms the spinel structure may also develop for NiO.12 Additional diÜerences might depend on the kinetics of the growth process or on the reactivity during –lm deposition as will be demonstrated by our experimental results.As in the case of NiO we consider the results obtained directly after room temperature deposition and after further 133 Faraday Discuss. 1999 114 129»140 Fig. 3 Local tunnelling current I against the sample bias voltage U on various parts of the samples as reproduced in Figs. 1 and 2. (a) Comparison between Ag islands (i.e. smooth parts of the islands) and the c(1]2) O/Ni precursor. The distance between sample and tip has been set at U\[2 V.(b) Comparison between NiO double layer islands and the Ag substrate. The distance between sample and tip has been set at U\4 V. annealing. Moreover for the O/Co precursor structure we have observed random walk of the defect states at an atomic level already at room temperature which might provide evidence for the unstable nature of the CoO precursor. We note that the electronic eÜects seen for NiO are also present on O/Co/Ag(100). They are even more pronounced than O/Ni/Ag(100) since they are already found in the precursor state in contrast to the insensitivity of the c(1]2) O/Ni structure against changes of the sample bias. In addition contrast changes on the O/Co/Ag(100) surfaces due to changes of the tip are also more frequently observed than for O/Ni/Ag(100).As –rst results we describe surfaces where a small amount of Co (of the order of 1 ML) was deposited in an O atmosphere onto Ag(100). In Fig. 4(a) the results for deposition at room 2 temperature without further annealing of the sample are shown. Four grey tone levels can clearly be recognised on the individual larger terraces. Note that a step edge of the initial Ag substrate runs from left to right about 1/3 of the image height from the bottom. It might be difficult to recognise the step edge since it no longer follows a straight line but has developed a large number of etched parts and is very rough therefore. Above the step edge (where the structures appear in a somewhat brighter grey tone level) the four grey tone levels characteristic for one terrace are clearly seen.The darkest features with a typical lateral extension of 2 nm and mostly of rectangular shape mostly correspond to vacancy islands in the Ag substrate formed during the deposition process. The Ag atoms removed are mostly condensed in the form of islands with similar lateral extension. They are represented in the brightest grey tone level. The remaining intermediate grey tone levels correspond to the initial substrate level (the darker one) and the brighter ones (whose Faraday Discuss. 1999 114 129»140 134 Fig. 4 (a) Deposition of 1.2 ML Co in an O atmosphere onto Ag(100). Substrate at room temperature. 2 U\[2 V I\0.2 nA image size\100]100 nm2. (b) Deposition of 1.2 ML Co in an O atmosphere onto 2 Ag(100) at a substrate temperature of 390 K (diÜerent experiment).U\[3 V I\0.1 nA image size\100]100 nm2. The nominal thickness of the –lm should be sufficient in both cases to cover the substrate completely. This is not the case in (a) and (b). We ascribe the observed discrepancy to inaccuracy of the calibration. However the relative scale of the thickness calibration is the same for all Co deposition experiments. lateral extension is by a factor of 2 larger than that of the Ag features) are due to the O/Co precursor. Note that Ag is now found at three atomic levels the original one and one deeper and one higher than the initial one. The O/Co precursor which does show distinct electronic eÜects in its contrast (not reproduced here) essentially occupies one height level on top of the initial substrate.A certain small fraction of the Ag vacancy islands may also contain the O/Co precursor. If the condensation is performed at a slightly elevated temperature (390 K) the resulting overall surface morphology completely changes (Fig. 4(b)). Ag vacancy islands are no longer found neither are individual Ag islands seen on top of the substrate terrace due to condensation of the removed Ag islands. The O/Co precursor structure is now seen in two distinct positions (very 135 Faraday Discuss. 1999 114 129»140 Fig. 5 (a) CoO(100) and Ag islands after deposition of nominally 1.2 ML Co (in O2) onto Ag(100) at 390 K and subsequent annealing to 470 K. The sample has been measured at U\[2 V I\0.3 nA and a size of 100]100 nm2. The directions have been determined from atomically resolved measurements.Note that the CoO islands are characterised by a seam of protruding defects on the step edges. (b) Atomic model of CoO(100)/Ag(100) assuming 4-fold hollow sites of the substrate for Co. similar to O/Ni). One part is condensed in the form of an adlayer (denoted 2 and 3) on the Ag(100) substrate (denoted 1) and the other part is incorporated in the topmost Ag layer of the substrate. Surprisingly and in contrast to O/Ni the O/Co features develop two kinds of island shape. One is more rounded (e.g. at 2) and the other one is more rectangular (e.g. at 3). The latter islands whose edges are oriented along [110]-like directions show a row structure of brighter features at a typical distance of a few nm which means that they are not of atomic origin (it could be a moireç pattern).If we compare the surface area of the O/Co features present on both (a) and (b) we see that it is not too much diÜerent which means that the O/Co precursor state (very similar to O/Ni) is formed by a dense O/Co array with a height of one atomic layer. The diÜerences between 2 and 3 are not yet known precisely but we believe that the more rounded islands correspond to a state that is closer to that of CoO than that of the rectangular islands. For example the rounded islands sometimes run over the step edges of the substrate without any visible interruption. This behaviour is also found for reacted CoO islands (see below) and can be described by a carpet-like coverage of the surface. The fact that Ag vacancy islands are no longer seen on the surface (in contrast to the results shown in Fig.4(a)) is the consequence of the slightly increased substrate temperature during deposition. Obviously small Ag islands and vacancy structures do not correspond to a thermodynamic equilibrium and are only found for the room temperature experiments due to the kinetic limitations of Ag migration. At slightly elevated temperatures the kinetic limitations can easily be overcome. If the O/Co islands shown in Fig. 4 are annealed further (470 K was found to be sufficient) they are transferred into the CoO(100) state (Fig. 5(a)). The nominal coverage is the same as in Fig. 4 Faraday Discuss. 1999 114 129»140 136 Fig. 6 (a) Formation of CoO(100) by deposition of 2 ML Co in O atmosphere on Ag(100) at 390 K and 2 subsequent annealing at 470 K.U\[3.5 V I\0.1 nA the size is 100]100 nm2. The assignment of the surface either to Ag or CoO is made by the presence of typical defects which could be identi–ed on the two kind of surfaces (in the reproduced image CoO appears to be somewhat rougher than the Ag metal). (b) CoO(100) layers as prepared by deposition of 10 ML Co in O atmosphere onto Ag(100) at a substrate temperature of 470 K. 2 (actually the image was obtained after annealing the layers as reproduced in Fig. 4(a)). Obviously the surface is covered by two kind of islands. One kind of island appears with a smooth surface these islands correspond to Ag formed during the O/Co to CoO transformation process (one is denoted by Ag on the image).The other islands correspond to CoO(100) structures. They exhibit similar electronic eÜects as described for the NiO(100) islands (not reproduced here). Their shape is essentially quadratic. However the orientation of the step edges is along the [100]-like directions in contrast to that of the NiO(100) islands. In Fig. 5(b) an atomic model for such islands is shown under the assumption that the Co atoms occupy 4-fold hollow sites of the substrate. It is qualitatively clear that the orientation of the step edges of the observed CoO(100) islands minimises the polar eÜects of the ionic charges of CoO. It is actually more surprising that the NiO(100) 137 Faraday Discuss. 1999 114 129»140 Fig. 7 Random walk to next neighbour and second-next neighbour positions of defects.The images have been obtained on a surface where 2 ML Co has been deposited in O atmosphere onto Ag(100) at 400 K. 2 U\[2 V I\0.1 nA the image size is 10]10 nm2. The time diÜerence between measurements of (a) and (b) is approximately 1 min. A movie corresponding to this –gure is available as electronic supplementary information. See http ://www.rsc.org/suppdata/fd/1999/129 islands are con–ned by the [110]-like steps. The total surface area covered by the CoO(100) islands is much lower than for the initial –lm structure. Therefore the height of the CoO(100) features is at least one double layer. The carpet-like overgrowth of the substrate by the CoO(100) islands can be seen on some parts of the low-coverage surface displayed in Fig.5(a). In Fig. 6 the growth mode of CoO(100) can be followed up to Co deposition with a nominal thickness of 10 ML and annealing at 470 K. CoO(100) clearly shows a two-dimensional growth mode as can be seen in Fig. 6(a). It is evident from the measured image that the step edges always follow [100]-like directions. The observed rounding of the CoO(100) islands may be explained by ìì freezing-in œœ of the high-temperature island shape. As a –nal example for a grown CoO(100) –lm the layered structure of 10 ML Co deposited in an O atmosphere onto Ag(100) at 470 K (without further annealing) is reproduced in Fig. 6(b). At 2 this stage Ag islands are no longer recognised. The growth mode of CoO(100) is nearly of the layer-by-layer type. However the fact that around 10 uncompleted layers of CoO can be identi–ed within the grown structure strongly indicates a transition to a three-dimensional growth mode.In the –nal part of our work on O/Co/Ag(100) we return to the intermediate state of CoO formation as represented in Fig. 4(b). Following the same preparation conditions (deposition at Faraday Discuss. 1999 114 129»140 138 400 K no further annealing) we made attempts to improve the resolution in order to understand the diÜerences between the more rounded and the rectangular O/Co precursor structures. It turned out that the atomic defect structures of both kinds of islands were diÜerent. While the ordered features seen in Fig. 4(b) on the rectangular islands are extended over a lateral size of typically 5 lattice parameters the more rounded islands showed atomic-like defect structures (which are not discernible in the overview image).Measuring these structures with better resolution we found that both kind of defects showed random walk at room temperature. Although migration of defects is not an unexpected eÜect its observation at room temperature seems to be surprising and further supports our qualitative assignment of the two kind of islands to a precursor state of the oxide. Since the oxide –lms are readily formed at temperatures of 200 K above room temperature the mobility of defects and constituents has to increase at a lower temperature and this is apparently the case. In Fig. 7 two images are reproduced which have been acquired subsequently (on the time scale of 1 min) and where a large number of defects have changed their position.In Fig. 7 the atomically resolved lattice of the O/Co precursors can clearly be identi–ed. The centre part showing no atomic corrugation corresponds to a small Ag island. Attached to this Ag island we –nd O/Co with a number of local defects (visible as depressions) which correspond to the rounded O/Co islands. The rest of the surface (for example the parts extending from the left bottom (right top) corner towards the inner part of the measured surface) shows the detailed structure of a rectangular island. Instead of local defects this part shows a kind of modulation on the length scale of several lattice parameters. Both kinds of defect features change their position with time and this can be seen by comparing Fig.7(a) with Fig. 7(b). In Fig. 7(a) the direction of movement to the resulting positions (as observed in Fig. 7(b)) are indicated by the arrows. The circles in Fig. 7(b) correspond to the initial positions of the defects. A more detailed analysis shows that six defects move to next neighbour and two to second-next-neighbour sites (probably via two next neighbour steps). The importance of this observation lies in the fact that the activation barrier for next neighbour hopping obviously is rather small and that the reason why the transition from the precursor to oxide state takes place at moderate temperatures may be qualitatively understood. IV. Conclusion We have demonstrated that by reactive deposition of Ni or Co in an O atmosphere onto Ag(100) 2 either by subsequent annealing (450»500 K) or by condensation of the –lms on a hot (450»500 K) substrate NiO or CoO –lms in (100) orientation can be grown.The overall morphology and the nature of the deposited –lms and of the substrate drastically depend on the deposition and annealing temperature. For room temperature deposition an O/metal precursor state is identi–ed that appears to have a height of one atomic layer both on top of the substrate and in the vacancy islands of the substrate which are generated during the deposition process. The oxide –lm can only be obtained at elevated temperatures and grows with a minimum thickness of a double layer. In the initial stage a two-dimensional growth of the oxide is found which changes to a more three-dimensional growth behaviour for thicker –lms.Acknowledgements This work has been supported by the Deutsche Forschungsgemeinschaft. The initial part of the work (growth of NiO layers) has been –nanced through the Forschergruppe ììModellkatœœ at the Ruhr-Universitaé t Bochum. References 1 M. Baé umer D. Cappus H. Kuhlenbeck H.-J. Freund G. Wilhelmi A. Brodde and H. Neddermeyer Surf. Sci. 1991 253 116. 2 D. Cappus C. Xu D. Ehrlich B. Dillmann C. A. Ventrice Jr. K. Al Shamery H. Kuhlenbeck and H.-J. Freund Chem. Phys. 1993 177 533. 3 K. Marre and H. Neddermeyer Surf. Sci. 1993 287/288 995. 4 Th. Bertrams and H. Neddermeyer J. V ac. Sci. T echnol. B 1996 14 1141. 139 Faraday Discuss. 1999 114 129»140 5 M. R. Castell P. L. Wincott N. G. Condon C. Muggelberg G. Thornton S. L. Dudarev A. P. Sutton and G. A. D. Briggs Phys. Rev. B Condens. Matter 1997 55 7859. 6 C. A. Ventrice Jr. H. Hannemann Th. Bertrams and H. Neddermeyer Phys. Rev. B Condens. Matter 1994 49 5773. 7 I. Sebastian M. Heiler K. Meinel and H. Neddermeyer Appl. Phys. A 1998 66 S525. 8 Th. Berghaus A. Brodde H. Neddermeyer and St. Tosch Surf. Sci. 1987 184 273. 9 I. Sebastian M. Heiler and H. Neddermeyer unpublished results. 10 G. Wilhelmi A. Brodde D. Badt H. Wengelnik and H. Neddermeyer in T he Structure of Surfaces III ed. S. Y. Tong M. A. Van Hove X. Xide and K. Takayanagi Springer Berlin 1991 p. 448. 11 P. A. Cox T ransition Metal Oxides Clarendon Press Oxford 1995. 12 A. Barbier personal communication. Paper 9/03416A Faraday Discuss. 1999 114 129»140 140

ISSN:1359-6640    

DOI:10.1039/a903416a

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Structure determination of molecular adsorbates on oxide surfaces using scanned-energy mode photoelectron di� raction M. Polcik,a R. Lindsay,§a P. Baumgaé rtel,a R. Terborg,a O. SchaÜ,a S. Kulkarni,a A. M. Bradshaw,a R. L. Toomesb and D. P. WoodruÜ*b a Fritz-Haber-Institut der MPG Faradayweg 4-6 D 14195 Berlin Germany b Physics Department University of W arwick Coventry UK CV 4 7AL . E-mail D.P.W oodruÜ=warwick.ac.uk 1. Introduction Despite the considerable growth in the body of quantitative data concerning the structure of surfaces and adsorbates on them during the last 20 years or so the great majority of these data are concerned with metal or semiconductor surfaces. By contrast there are very few such data for oxide surfaces ; indeed the most recent (1996) edition of the NIST Surface Structure Database1 contains no entries for adsorbates on oxides (although there are some more recent examples of such results including some on molecular adsorbates2,3).One of the reasons for this of course is the perceived difficulty in preparing well-characterised oxide surfaces which has held back surface science studies of oxides in general. Another is the problem of charging for insulating oxides which makes the use of electron diÜraction difficult. It is also not clear to what extent adsorbates on oxides and especially molecular adsorbates on oxides commonly involve the long-range ordering required for conventional diÜraction studies. One technique which has proved rather eÜective in the quantitative determination of adsorbate structures including molecular adsorbates on metals and semiconductors is scanned-energy mode photoelectron diÜraction (PhD).4 In this technique a core level photoemission signal from the adsorbate is measured in a –xed direction as a function of photon (and thus photoelectron) energy.Components of the emitted photoelectron wave–eld are elastically scattered by surrounding atoms (notably including backscattering from substrate atoms) and interfere with the directly emitted component of the same wave–eld leading to modulations in the measured intensity as the photoelectron wavelength changes and scattering pathways fall in and out of phase with the § Present address Chemistry Department University of Manchester Manchester UK M13 9PL 141 Faraday Discuss.1999 114 141»155 Received 19th March 1999 Using N 1s scanned-energy mode photoelectron diÜraction (PhD) combined with full multiple scattering simulations the local adsorption site of NO on NiO(100) has been determined. The molecule bonds through the N atom atop a surface layer N atom while the N»O axis is tilted away from the surface normal by 59(]31/[17)°. The special problems presented by adsorbates on compound surfaces for the direct inversion of PhD data to provide a –rst-order site determination are discussed and some alternative schemes tested. This journal is( The Royal Society of Chemistry 2000 directly emitted wave. These modulations thus provide local structural information which does not rely on long-range order and is element speci–c.Moreover by exploiting the ìchemical shifts œ in the photoelectron binding energy associated with diÜerent local environments of atoms of the same element chemical-state speci–c structural information is also obtainable.5,6 This is potentially of especial interest in the case of oxide surfaces to allow full structural studies of oxygencontaining molecular adsorbates (one demonstration of this idea albeit based on a very small data set has recently been published3). Of course this method is also only applicable to conducting samples but even in the case of insulating oxides oxide surface studies are possible by using thin epitaxial oxide –lms on conducting metallic substrates as has proved useful in the application of other forms of electron spectroscopy. In this paper we present in detail our –rst results of the application of our PhD methodology to determine the structure of a molecular adsorbate on an oxide surface namely NO adsorbed on a thin NiO(100) –lm grown epitaxially on Ni(100).In addition to presenting the detailed analysis of this problem (a brief report of which will appear elsewhere7) we also highlight some of the special problems presented by oxide surfaces to this methodology and brie—y review the prospects for further applications in the light of these results. 2. The NiO(100)/NO system and experimental details The adsorption properties of NO on NiO(100) have been previously characterised rather extensively notably by Freund and coworkers.8,9 On the basis of vibrational (ìhigh resolutionœ) electron energy loss spectroscopy (HREELS) and temperature-programmed desorption (TPD) it appears that there is only a single species NO on the surface.X-Ray photoelectron spectroscopy on the other hand shows two distinct N 1s peaks separated by approximately 5 eV but it has been proposed that these correspond to excitation to two diÜerent –nal states (strongly screened and weakly screened at lower and higher photoelectron binding energies respectively) from the same initial state of a single local adsorption geometry. The polarisation-direction dependence of near-edge X-ray absorption –ne structure (NEXAFS) indicates that the N»O molecular axis is tilted relative to the surface normal by approximately 45°. Much of this work was conducted on thin –lms grown epitaxially on Ni(100) to avoid charging problems and while such –lms can suÜer from a high density of surface defects it was demonstrated that the NO adsorption behaviour is dominated by ideal (100) terrace sites.The adsorption system has also been studied using total energy cluster calculations ; two such independent studies10,11 both –nd an energy minimum for a structure involving NO bonding through the N atom atop a surface Ni atom with an N»O molecular axis tilt of approximately 45°. There is however no experimental determination of the adsorption site. Our experiments were performed using NiO(100) –lms grown in situ epitaxially on Ni(100) according to the prescription used in the work of Freund and coworkers.8,9 The Ni(100) substrate was –rst prepared by the usual combination of X-ray Laue alignment spark machining mechanical and electrochemical polishing and in situ cycles of argon ion bombardment and annealing.The cleanness and long-range order of the Ni(100) surface and of the epitaxial NiO –lm were established by soft X-ray photoelectron spectroscopy (including Ni 2p and O 1s spectra) using the incident synchrotron radiation from the HE-TGM (high energy toroidal grating monochromator)12 on the BESSY electron storage ring in Berlin and by LEED respectively. NO exposure was conducted with a sample temperature of approximately 135 K and led to no additional LEED beams consistent with earlier studies. N 1s photoemission spectra recorded using the incident synchrotron radiation in the energy range used for the PhD measurements showed that exposure to the photon beam led to a progressive reduction in the emission intensity attributed to photon-stimulated desorption (although whether this due to direct photoexcitation or involves photoemitted and secondary electrons is unknown); this observation is consistent with previous work using conventional XPS with an Al Ka source.13 Somewhat surprisingly however the N 1s spectra recorded as a function of incident X-ray exposure showed selective attenuation of the higher binding energy state as shown in Fig.1. At –rst sight these spectra suggest that the two well-resolved N 1s peaks are attributable to diÜerent species with only one of them being strongly sensitive to the X-radiation. As we shall show however our PhD data show clearly that both components of this initial doublet are associ- Faraday Discuss.1999 114 141»155 142 Fig. 1 N 1s photoelectron energy spectra recorded at a nominal phon energy of 500 eV after increasing exposure times ; in the upper panel the spectra were recorded under ultra-high vacuum (UHV) conditions in the lower panel a constant partial pressure of 7]10~9 mbar NO was maintained. The (nominal) photoelectron kinetic energy axis is uncalibrated. ated with the same surface species and that the –nal single peak is due to an entirely diÜerent species which has coincidentally a similar photoelectron binding energy. Moreover this timedependence of the N 1s spectra in the X-ray beam could be overcome by working in a partial pressure of NO of approximately 7]10~9 mbar which led to time-independent spectra (see Fig.1). Our interpretation of these results is that the incident X-rays stimulate two processes NO desorption and N»O bond scission the latter leading to oxygen desorption but leaving chemisorbed atomic N on the surface. The N 1s photoelectron binding energy of this atomic species is not distinguishable in our low-resolution experiments from that of the well-screened –nal state peak of NO. In the presence of a partial pressure of NO however any desorbed NO is replaced by new molecular arrival from the ambient gas but in addition chemisorbed atomic N is able to react with NO (either from the gas phase or from the coadsorbed species) presumably to form N2O which is desorbed from the surface.The results indicate that the NO partial pressure we have used is sufficient to observe a steady-state N 1s spectrum dominated by adsorbed molecular NO. Notice incidentally that this state is governed not only by the relative efficiencies of the NO/N reaction and NO photodissociation processes but also by the relative arrival rates of the incident NO molecules and photons. Indeed detection of the atomic N signal in photoemission requires that a second photon photoionises the adsorbed N atom resulting from photodissociation by a photon (or the associated secondary electrons) that arrived at this same site earlier before this fragment is removed by chemical reaction. In order to determine the kinetic energy dependence of the N 1s photoemission signal which forms the basis of the PhD modulation spectra short energy range photoelectron energy distribution curves (EDCs) around the N 1s peak were recorded at steps in the photon energy of 2 eV in an energy range corresponding to photoelectron energies of 80»400 eV.These spectra were recorded using a 150° concentric spherical sector analyser (VG Scienti–c) –tted with threechanneltron parallel detection and mounted at a –xed angle of 60° to the incident radiation in the 143 Faraday Discuss. 1999 114 141»155 horizontal plane which also is the plane of polarisation of the synchrotron radiation. The angular acceptance of this analyser was set to 5°. Sets of spectra of this kind were recorded at a number of diÜerent polar emission angles in the two principal azimuths [011] and [010] in order to provide a sufficiently large data set to maximise the probability of establishing a unique structural solution.In order to obtain the PhD modulation spectra from these raw data each individual EDC was –tted by a sum of two Gaussian peaks plus Gaussian-broadened steps and an experimental background template that accounts for inelastic and secondary electrons. The integrated intensities of these peaks as a function of photoelectron kinetic energy I(E) were then normalised to produce s(E)\((I(E)[I the PhD modulation function de–ned as 0(E))/I0(E) where I0(E) is the intensity in the absence of diÜraction which is assumed to be described by a smooth spline passing through the measured I(E) data. It is these modulation spectra which form the basis of the quantitative surface structure determination.3. NiO(100)/NO surface structure determination multiple scattering simulations Our standard methodology for surface structure determination through the technique of PhD consists of a two-stage approach.14 In the –rst stage an approximate adsorption site can often be obtained by the use of some form of direct inversion of the experimental data. This stage is intended to play a similar role to Fourier transform methods (such as the Patterson function) in conventional X-ray diÜraction although there are problems associated with this kind of treatment of low energy electron scattering data due to complex scattering factors and the potential importance of multiple scattering. The speci–c method we have used with considerable success in the past especially for high symmetry adsorption sites is the so-called projection method,15,16 but this approach is formally only applicable to elemental substrates.For this reason we have omitted this stage in the present analysis although in the following section we discuss this problem in more detail. The more quantitative stage in PhD analysis involves performing full multiple scattering simulations17h20 of the data for possible model structures and adjusting the model to optimise the –t. In the absence of a direct data inversion this trial-and-error approach included consideration of a more complete set of distinct structural models than would otherwise prove necessary. In all cases however assessment of the quality of the agreement between experiment and theoretical simulations is aided by the use of an objective reliability (R)-factor.An appropriate de–nition has been given previously,21 but we note here that the R-factor is normalised such that a value of zero corresponds to perfect agreement and a value of unity to experimental and theoretical curves that are wholly uncorrelated. Using this R-factor it is also possible to de–ne a formal test of the precision associated with the structural parameters values obtained in the best-–t structure by de–ning a variance in the minimum value of the R-factor Rmin .22 In view of the fact that a PhD modulation spectrum is characteristic of a local emitter geometry one important –rst step in the analysis of the NiO(100)/NO data was to demonstrate that the two distinct N 1s photoelectron peaks do indeed correspond to a single adsorption site and thus also to a single species.Fig. 2 provides this evidence by comparing the experimental s(E) curves obtained for each of the two spectral components at two diÜerent emission directions normal emission and a polar emission angle of 30° in the [011] azimuth. Notice that since the two modulation functions are plotted as a function of photoelectron kinetic energy E at any one such energy the two functions were recorded at diÜerent photon energies. The spectra for the two states are clearly closely similar. Indeed if the R-factor normally used to compare experiment and theory is used to compare the two experimental data sets recorded from the two spectral components the values obtained (0.09 and 0.12 for the 0° and 30° spectra respectively) are comparable to the very best theory»experiment –ts we see in PhD.We therefore conclude that the original assignment of these two components was correct and that the two spectral components do originate from the same local site. To optimise the signal-to-noise quality of the experimental data for the structural analysis the sum of the two intensities was used in this stage of the analysis. Note incidentally that a similar PhD modulation spectrum recorded in normal emission from the single N 1s peak seen when the NO-covered surface was exposed to the synchrotron radiation beam under UHV Faraday Discuss. 1999 114 141»155 144 Fig. 2 Comparison of the experimental PhD modulation spectra obtained from the two separate components of the N 1s photoemission spectrum in two diÜerent emission directions.conditions (Fig. 1) did not show these same strong modulations providing further proof that this peak does correspond to a diÜerent surface species. Although we have not made use of any direct method of data inversion to provide a –rst estimate of the adsorbate structure a visual inspection of the raw PhD spectra (Fig. 3) strongly favours an atop adsorption site for the N atom. In particular we –nd strong modulations (around ^40%) at normal emission but the modulation amplitudes fall oÜ rapidly as the emission angle is increased being less than ^10% at 30° emission in the [001] azimuth. In many previous studies we have shown that PhD typically shows the strongest modulations when a near-neighbour lies directly behind the emitter relative to the detector placing this neighbour in the favoured 180° scattering geometry (and indeed this forms the basis of the direct inversion methods15,16,23h25).For this reason atop emitter geometries are characterised by strong PhD modulations only close to normal emission.21,26,27 The strength of these modulations also suggests the near-neighbour backscatterer is Ni rather than O for which the backscattering cross-section is much smaller. Fig. 3 Comparison of the experimental N 1s PhD modulation spectra (thin lines) for several diÜerent emission directions with the results of the best-–t theoretical simulation (bold lines) based on the geometry shown in Fig.3. 145 Faraday Discuss. 1999 114 141»155 In our search for the optimum structure we have therefore concentrated on the following geometries all of which involve the N atom being directly above either a Ni or O backscatterer atom NO bonded through the N atom in sites atop a surface Ni or O atom and NO bonded through the O atom either atop a surface Ni or O atom or in bridge or hollow sites. In all cases in these initial searches the N»O axis was assumed to be perpendicular to the surface but the in—uence of diÜerent N- and O-substrate layer spacings in increments of 0.05 Aé was explored. Notice that while intramolecular scattering does have some in—uence on the –nal PhD spectra the technique operates under conditions which stress the in—uence of backscattering so for the adsorbate atom closest to the surface this is dominated by the substrate scattering and intramolecular scattering in—uences the results more weakly mainly through multiple scattering.Of course if the NO were to be bonded with the O atom down against the surface intramolecular scattering would include the potentially important N 1s photoelectron backscattering oÜ this O atom. The results of this preliminary search of possible structures strongly favoured the geometry in which the molecule bonds through the N atom which is atop a surface Ni atom; the R-factor for this case was 0.39 whereas the lowest value for any of the other structures was 0.74 vastly greater than this minimum value plus its variance which was found to be 0.07.Having identi–ed this basic adsorption geometry as favoured a more detailed optimisation of the associated structural parameters was undertaken with the aid of an adapted Newton»Gauss algorithm to perform automated searches of the multi-parameter space. The speci–c parameters optimised were the Ni»N and N»O nearest neighbour bond lengths dNiN and dNO the tilt angles of these two bonds relative to the surface normal hNiN and hNO the azimuthal polar angle of these tilts / relative to the [011] direction and the NiO outermost layer spacing z12 (see Fig. 4). Deeper NiO layer spacings were assumed to have the bulk value. In addition the mean square Fig. 4 Schematic diagram of the best-–t geometry for NO adsorbed on NiO(100) showing de–nitions of the main structural parameters investigated.Faraday Discuss. 1999 114 141»155 146 é Table 1 Best –t structural parameter values for the NiO(100)/NO adsorption geometry (Fig. 3) obtained in this analysis Valuea Parameter d h d h NiN NO NiN 12 NO A 2 T z S S u u 1.88^0.02 Aé 3]3/[8° 1.12]*/[0.15 Aé 59]31/[17° * 2.07^0.04 Aé (1.8]4.2/[1.8)]10~2 Aé 2 (3.8^1.9)]10~3 Aé 2 M 2 T a For a discussion of the estimated precision (especially in the cases marked *) see the text. vibrational amplitudes of the N atom parallel and perpendicular to the surface were optimised to give the lowest R-factor. The best-–t structure resulting from this optimisation is shown schematically in Fig. 4 while the actual comparison between experiment and the results of the theoretical simulations is included in Fig.3; the R-factor value of 0.09 is very low re—ecting the excellent –t seen visually. Table 1 lists the structural parameter values associated with this structure together with estimates of their precision. Notice in particular that this best-–t structure clearly does involve a signi–cant tilt of the N»O axis away from the surface normal consistent with earlier experimental and theoretical determinations of this parameter. As remarked earlier intramolecular scattering is commonly much less important than substrate scattering in the PhD technique and in the present case the role of the scattering from the O atom is clearly weak. This leads to the poor precision in the N»O tilt angle (with the near-normal orientation clearly excluded but a signi–cant possibility of very large tilt angles) and also in—uences the precision of the N»O distance.Indeed while the optimum value for this distance is 1.12 Aé A and values less than 0.97 give unacceptable –ts one can increase this bondlength without limit (ultimately therefore removing the O atom completely) Fig. 5 Contour map showing the variation of the R-factor as a function of the values of the two parameters dNO and hNO which de–ne the position of the O atom within the adsorbed NO molecule. The bold contour at a value of 0.11 (corresponding to the sum of the minimum value and its estimated variance) de–nes the limits of those values estimated to fall within one standard deviation of the best-–t structure.Faraday Discuss. 1999 114 141»155 (SuA 2 T S and uM 2 T) 147 without increasing the R-factor value above the sum (0.11) of the minimum value (0.09) and its variance (0.02). For this reason we have marked this positive error estimate in Table 1 with an asterisk. Of course this ambiguity in the N»O bondlength impinges on the error estimate for the N»O bond orientation ; if the O atom is removed this angle is meaningless! Fig. 5 shows a map of the variation of the R-factor as a function of these two parameters the N»O bondlength and the N»O bond angle which eÜectively de–ne the precision with which we can locate the O atom position ; the contour corresponding to an R-factor of 0.11 is shown in bold and as described above all geometries within this contour fall within the estimated limits of our precision.The error for the N»O bond angle quoted in Table 1 is actually a worst-case value for any N»O bondlength of less than 1.43 Aé (i.e. falling on the map of Fig. 5) ; the angular error around the optimum bondlength is signi–cantly smaller (]22/[8°). Notice incidentally that previous NEXAFS and HREELS studies8,9 show clearly that the surface species is molecular NO so although the PhD data are relatively insensitive to the presence of the O atom we do know that this atom is present. One parameter not given in Table 1 is the azimuthal angle / of the N»O tilt. In the PhD experiment one necessarily averages over all symmetrically equivalent tilt azimuths and this is accounted for by a general summing over symmetrically equivalent ìdomainsœ which must be included when any structural parameter lowers the local symmetry below that of the point group of the substrate.The combination of this domain averaging and the insensitivity of the present data to the intramolecular scattering meant that R-factor diÜerences between diÜerent azimuthal orientations of the N»O tilt were minimal and certainly well within the variance although the actual best-–t structure was one which included additional azimuthal averaging to simulate total disorder of this parameter. The statistics are inadequate to make any clear statement about this aspect of the structure although we believe that an azimuthally disordered geometry is most probable. 4. Adsorbate structure determination on oxides using PhD applicability of direct methods In the previous section we remarked that the present NiO(100)/NO structure determination was performed without using the –rst stage of our standard methodology the direct data inversion to provide a –rst-order structural estimate because this approach is formally not applicable to a compound substrate.We now consider this problem and possible ways of circumventing it in more detail. The basic objective of the direct data inversion stage is to render more efficient the essentially trial-and-error approach to solving structural problems using time-consuming multiple scattering simulations to search the whole of parameter space. Ideally one would wish to have a method that provides an accurate three-dimensional ìimageœ of the atoms surrounding the emitter.Because of the eÜects of the energy angle and mass-dependent scattering phase shifts involved in low energy electron scattering from atoms combined with the complication of multiple scattering simple Fourier transform methods of data inversion are certainly not capable of yielding quantitatively meaningful results for photoelectron diÜraction (or indeed for any other low energy electron scattering method such as LEED). However the attractiveness of developing a direct inversion method often referred to as holographic reconstruction has led to considerable discussion and quite a number of alternative schemes for tackling this problem [see for example refs. 28 29] other than our own.15,16,24,25 None of these methods claims to be exact and as such can only provide an approximate structural solution although regrettably there have been several examples of publications that omit the structural optimisation by quantitative modelling and implicitly claim to have solved structures using only the direct method.30h32 The ìprojection methodœ which we have found to be rather successful as a means of providing an approximate ìimageœ of the real-space structure from PhD data is based on a relatively minor modi–cation of the Fourier transform approach although it speci–cally accounts for the role of the scattering phase shifts (but takes no account of multiple scattering).The basic idea can be best appreciated by considering a simpli–ed formulation of PhD,4,33 which includes only single scattering and leads to an expression for the modulation function in terms of the electron wavevector Faraday Discuss.1999 114 141»155 148 k as s(k)\; (aj(Hj k)/r)cos(krj(1]cos Hj)]dj(Hj k)) j in which H is the scattering angle of the r jth scatterer at a distance from the emitter. aj(Hj k) is j j an eÜective scattering amplitude for the jth scatterer which includes the modulus of the scattering factor and also a polarisation angular term and damping due to inelastic scattering and thermal vibrations while dj(Hj k) is the phase of the scattering factor of the jth scatterer. Evidently in the absence of the extra phase term dj(Hj k) a Fourier transform u(r)\Ps(k)exp(ikr) dk sth(k r) for a scatterer atom at a speci–c location r and to then compute the would yield peaks at the correct scattering pathlength diÜerences krj(1]cos Hj) and could therefore be used to invert the data.The presence of the scatterer phase shifts however can displace the atomic images by several tenths of an Aé ngstrom. We note however that the essential function of a Fourier transform used in this way is to pick out the periodic components of an oscillatory function and in the present case we are mainly interested in the dominant period which corresponds to scattering from the nearest neighbour(s). One way of doing this in a way which compensates for the scattering phase shifts is to compute a simple form of the theoretical single scattering modulation function projection integral of this onto the experimental modulation function sex(k).15 When the test position corresponds to a real near-neighbour atom position this integral c(r)\Psex(k)sth(k r) dk should peak.Functions of this type can then be computed for each of a number of experimental modulation functions measured in n diÜerent directions and the results combined according to i r) n C(r)\ ; exp(Sc i/1 which produces a three-dimensional ìimageœ of the scatterer positions. The exponential loading is used to pick out only the very strongest features which correspond to the near-180° backscattering near neighbours r is included to prevent the (1/r) term in sth(k r) causing the function to diverge as r]0 while the arbitrary scaling factor S can be adjusted to vary the ìcontrastœ in this image.Evidently this procedure can only be applicable when all the substrate backscatterers are of the same element and have the same dj(Hj k). Compound surfaces therefore present a fundamental problem to the correct application of this method. Of course if the compound comprises elements of very similar atomic number (for example in GaAs 32) one may have some success by ignoring this diÜerence although the resulting image does not then distinguish the diÜerent types of scatterer. In the case of oxides one commonly has elements with very diÜerent scattering factors ; the scattering factors basically scale with the atomic number although the detailed behaviour is signi–cantly more complex but oxygen is a rather weak backscatterer while Ni for example is relatively strong.The importance of this distinction in the projection method will be discussed further below. One rather straightforward way in which one might try to gain preliminary information from direct inversion of the data is –rst to simply apply the standard projection method assuming –rst that all the substrate atoms are Ni (i.e. to describe their scattering through the use of Ni scattering phase shifts) and secondly that they are all O (using O scattering phase shifts). This procedure would give rise to two alternative images of the emitter surroundings; in one of these those features attributable to Ni atoms should be correctly positioned whereas any due to O atoms would be incorrectly represented while in the other image the roles would be inverted.Of course the projection method is designed to pick out selectively the strongest backscattering near neighbour so one possible outcome is that both images will be dominated by the Ni backscatterers although 149 Faraday Discuss. 1999 114 141»155 the true positions of these should be more precisely determined when Ni scatterers are used in the projection integral calculation. The two panels on the left of Fig. 6 show the results of a test of this idea. The projection method of course de–nes a three-dimensional ìimageœ in the amplitude of C(r) but it is simplest to show two-dimensional cuts through this function perpendicular or parallel to the surface. In the case of the results presented in Fig. 6 cuts are shown parallel to the surface at a distance below the emitter of 1.9 Aé based on inversion of the experimental data using either Ni or O scattering phase shifts and amplitudes; these cuts pass through the maximum intensity for both sets of calculations and no other features of signi–cant intensity are seen in cuts perpendicular to the surface.The results in both cases show an intense feature directly below the emitter (which is located at (0,0,0)) consistent with the N emitter lying atop a near-neighbour substrate scatterer at a distance of approximately 1.9 Aé . The slight splitting of the feature in the Ni scatterer image may indicate slight oÜ-atop positioning but the signi–cance of this kind of detail in projection maps is marginal. Notice that the use of the O scatterer properties has not signi–cantly displaced the feature so these maps do not formally distinguish between atop Ni or atop O.Of course the actual value of C(r) at the peak is much larger in the Ni inversion because the theoretical modulation amplitudes are larger for Ni scatterers and this leads to larger projection integrals when the correct modulation period and phase is achieved. Potentially therefore the projection integral images of Fig. 6 might have been useful. They strongly suggest that the N atom of the adsorbed Fig. 6 Standard projection method maps based on the experimental N 1s PhD data from the system NiO(100)/NO (left-hand panels) and on simulations for NO atop a surface O atom of NiO(100) (right-hand panels) of cuts parallel to the surface but passing through the strongest projection method feature (1.9 Aé below the emitter).The inversions have been performed assuming either that all scatterers are Ni atoms or that all scatterers are O atoms. Faraday Discuss. 1999 114 141»155 150 NO does lie atop a surface layer atom at a distance commensurate with N bonding to the surface and even if the chemical identity of this atom is unknown the number of possible structural models is greatly reduced. In reality of course we actually drew these conclusions from visual inspection of the raw data. To explore further the possible utility of this idea we now extend our consideration to include simulated data for the case of NO adsorbed atop the O atoms of the NiO(100) surface ; we might after all argue that this test on the experimental data only works well because the nearest neighbour backscatterer to the N emitter in the real surface is a strongly scattering Ni atom.The right-hand panels of Fig. 6 show similar projection method maps applied to these simulated data; in this case too the only strong features found in the projection maps were centred directly below the emitter and the cuts shown in Fig. 6 are again parallel to the surface and centred on these maxima in C(r). Somewhat surprisingly the dominant features of the two sets of projection maps are very similar ; even when the nearest neighbour substrate backscatterer is the more weakly scattering O atom it is this scatterer which is found in the projection method. Of course this success in locating (with modest precision) the nearest neighbour substrate atom even without establishing its chemical identity would greatly reduce the number of trial structures in a practical application of our structural optimisation.On the other hand the absence of ìimagesœ of nonnearest neighbour (but more strongly scattering) Ni atoms in the maps obtained from these simulated data is somewhat surprising. Fig. 7 compares the actual theoretical simulations for the two diÜerent adsorption geometries being considered with the NO molecule placed at the same height above the outermost NiO layer. For the case of NO atop the Ni atom the theoretical curves are the same as those matching the experimental data; at normal emission a strong short-period modulation is seen associated with the backscattering from the Ni nearest neighbour although some higher frequency components are also discernible probably from second layer scattering.By contrast the equivalent normal emission spectrum from NO atop a surface O atom shows some sign of a similar long period oscillation but actually appears to be dominated by a period of about half of this value. In NiO(100) a surface O atom lies directly above a Ni atom in the second layer 2.08 Aé below so scattering from this atom (and perhaps others in this layer) is the most probable origin of this contribution ; separate calculations performed using only single scattering actually indicate that the shorter period modulations are substantially enhanced by multiple scattering but a likely event of this kind is forward scattering by the nearest neighbour O atom onto this second layer Ni.Whatever the detailed interpretation of these features is the very marked diÜerence in these two sets of simulated spectra highlights the value of modelling in distinguishing between diÜerent types of structures for a compound surface of this type. Notice however that the simulated Fig. 7 Comparison of simulated N 1s PhD spectra for adsorption atop Ni (»») and atop O (… … …) on NiO(100). Apart from the lateral displacement of the emitter all other aspects of the two structure (including their vibrational amplitudes) are identical. 151 Faraday Discuss. 1999 114 141»155 spectra for the NO-atop-O geometry do not show strong modulations at 50° emission in the [001] direction which should place one of the four top layer Ni nearest neighbour atoms in the favoured 180° scattering geometry.This can largely be attributed to the fact that in conducting these simulations we have used the same large vibrational anisotropy for the N emitter atom as was found for the optimum –t to the experimental data in the atop-Ni site. It is the large vibrational amplitude parallel to the surface which suppresses the importance of scattering from these nearest top layer Ni atoms. If this anisotropy had not been included the projection method might be expected to distinguish the two atop geometries more clearly yet our experience is that atop sites are characterised by these large parallel vibrational amplitudes. This detail highlights the difficulty of drawing general conclusions from speci–c model calculations.These results do however highlight the problem of distinguishing the elemental character of a nearest neighbour backscatterer by the projection method if this scatterer dominates either due to scattering strength or relative vibrational (Debye»Waller) damping. Nevertheless it is interesting to consider ways in which one might hope to recover information on the location of both substrate scatterer species atoms and to distinguish between them. In this regard the core problem is clearly related to the fact that the stronger scatterer atoms are prone to dominate the actual PhD spectra. However in the projection method the problem is exacerbated by the fact that the projection integral contains no normalisation to the intensity of the theoretical modulations but simply picks out the common periods and phases of the theoretical and experimental modulations the intensity of a feature in the transform actually being a product of the modulation intensities of the experimental and theoretical functions at the particular period and phase.This means that if theoretical modulation spectra based on a weak scatterer are inserted into the projection integral and the experimental data comprise modulations from both weak and strong scatterers the resulting integral will still peak when the theoretical modulations most closely match the period and phase of the dominant components in the experimental data which probably arise from the strong scatterers. If it is possible to pick out the weak scatterer components therefore it is probably important to normalise the projection integral to peak when the amplitude as well as the period and phase of the relevant modulation component match.One way to achieve this is to renormalise the projection integral to Psex(k)sth(k r) dk c(r)\SPsex(k)sex(k) dkSPsth(k r)sth(k r) dk sth(k r) is too large in This normalisation attenuates c(r) when the test modulation spectrum amplitude relative to the component of the same periodicity in the experimental modulation function. Of course this reduces the peak value of the projection integral and potentially greatly lowers the peak intensities and thus the contrast of the –nal images based on C(r) although a modi–cation of the scaling factor S can partially rectify this problem.Nevertheless it is important to recognise that this change and indeed its primary objective shifts away from the original intent of the projection method approach to concentrate only on the strongest features in the PhD data and the location of the near-neighbour scatterers that give rise to them. We may therefore anticipate that some of the spurious artefacts which characterised the early attempts at holographic inversion of photoelectron diÜraction data [see for example ref. 28 ] may be encountered. These fears prove well-founded. We have run tests on this modi–ed projection method both on the experimental data and on simulated data based on both NO atop surface Ni atoms and atop surface O atoms. Some features in the resulting images designed to isolate the Ni and O contributions do correspond approximately to the true positions of these scatterers but considerably more spurious features are seen than in the conventional projection method.Fig. 8 provides one example of the results in the form of two-dimensional maps perpendicular to the surface in the [010] azimuth based on simulated data for the N emitter atop an O atom in the outermost NiO(100) surface layer. The map intended to identify the O scatterer is dominated by features centred directly below the emitter but these are split and very spread out. The map based on the Ni scattering similarly shows the dominant features oÜset as expected for the four-fold coordination relative to these atoms but here too the locations of these features are not at all sharp or Faraday Discuss.1999 114 141»155 152 Fig. 8 Modi–ed projection method maps of cuts perpendicular to the NiO(100) surface (with the emitter at (0 0,0)) based on theoretically simulated data for which the N emitter lies atop a top layer O atom at a layer spacing of 1.88 Aé . The upper and lower maps are based on the use of Ni and O scatterers respectively in the projection integral. The renormalised projection integral described in the text has been used for these calculations. well-positioned. These preliminary tests thus suggest that identi–cation of the weaker scatterer locations by this direct inversion approach is far less promising than the rather eÜective application of the standard projection method to adsorption on elemental surfaces. 5.General discussion and conclusions Using the scanned-energy mode photoelectron diÜraction technique we have determined the local geometry for NO adsorbed on NiO(100); in particular we –nd that the molecule bonds through the N atom atop a surface Ni atom with the N»O axis strongly tilted relative to the surface normal. These –ndings are consistent with the results of earlier theoretical modelling and with prior determination of the molecular orientation by NEXAFS although there was no prior experimental determination of the adsorption site. This is the –rst complete structure determination of a molecular adsorbate on an oxide surface ; an earlier study3 of the formate species on TiO (110) was based on a single low quality PhD spectrum and falls far short of a true experimen- 2 tal adsorbate site determination.Although this structure determination was conducted exclusively through the use of trial-anderror modelling using multiple scattering calculations we have also considered the problems and potential of the use of the projection method of direct data inversion as a method of obtaining a –rst estimate of the structure for subsequent optimisation by full modelling. This approach is formally only applicable to elemental substrates. Attempts to modify the method to locate both the strongly scattering Ni and the more weakly scattering O atoms do appear to achieve this aim but only at the expense of introducing other spurious features that would render their interpretation ambiguous; this re—ects an essential misuse of an approach designed speci–cally to locate the dominant scatterers.On the other hand application of the standard projection method to adsorption on NiO does appear to give the approximate location of the nearest neighbour substrate backscatterer atom (but not its elemental identity) and so may be helpful in identifying reasonable 153 Faraday Discuss. 1999 114 141»155 starting structures for full multiple scattering analysis. By contrast at least in the case of NiO in which the two constituent atoms have very diÜerent scattering properties quite simple simulations distinguish the identity of near neighbour scatterers rather readily. Further studies of other oxide surfaces will be required to establish the generality of the utility of the direct imaging component of the methodology but it seems clear that the PhD method has considerable potential for the determination of adsorption geometries on these surfaces.Acknowledgements This work has been supported by the German Federal Ministry of Education Science Research and Technology (contract no. 05 SF8EBA 4) by the Engineering and Physical Science Research Council (UK) and by the European Commission through the Large Scale Facilities component of the HCM programme. The authors thank Volker Fritzsche for providing the multiple scattering computer codes used in this work. References 1 P. R. Watson M. A. Van Hove and K. Hermann NIST Surface Structure Database version 2 NIST Standard Reference Database 42 NIST Gaithersburg MD 1996. 2 D. Ferry P. N. M.Hoang J. Suzanne J.-P. Biberian and M. A. Van Hove Phys. Rev. L ett. 1997 78 4237. 3 S. A. Chambers S. Thevuthusian Y. J. Kim G. S. Herman Z. Wang E. Tober R. Ynzunza J. Morais C. H. F. Peden K. Ferris and C. S. Fadley Chem. Phys. L ett. 1997 267 51; S. Thevuthusian G. S. Herman Y. J. Kim S. A. Chambers C. H. F. Peden Z. Wang R. X. Ynzunza E. D. Tober J. Morais and C. S. Fadley Surf. Sci. 1988 401 261. 4 D. P. WoodruÜ and A. M. Bradshaw Rep. Prog. Phys. 1994 57 1029. 5 K.-U. Weiss R. Dippel K.-M. Schindler P. Gardner V. Fritzsche A. M. Bradshaw A. L. D. Kilcoyne and D. P. WoodruÜ Phys. Rev. L ett. 1992 69 3196. 6 K.-U. Weiss R. Dippel K.-M. Schindler P. Gardner V. Fritzsche A. M. Bradshaw D. P. WoodruÜ M. C. Asensio and A. R. Gonzaç lez-Elipe Phys. Rev.L ett. 1993 71 581. 7 R. Lindsay P. Baumgaé rtel R. Terborg O. SchaÜ A. M. Bradshaw and D. P. WoodruÜ Surf. Sci. L ett. 1999 425 L401. 8 H. Kuhlenbeck G. Odoé rfer R. Jaeger G. Illing M. Menges Th. Mull H.-J. Freund M. Poé hlchen V. Staemmler S. Witzel C. Scharfschwerdt K. Wennemann T. Liedtke and M. Neumann Phys. Rev. 9 M. Baué mer D. Cappus G. Illing H. Kuhlenbeck and H.-J. Freund J. V ac. Sci. T echnol. A 1992 10 B Condens. Matter 1991 43 1969. 2407. 10 V. Staemmler in Adsorption in Ordered Surfaces of Ionic Solids and T hin Films ed. H.-J. Freund and E. Umbach Springer-Verlag Berlin 1993 p. 169. 11 L. G. M. Petersson T heor. Chim. Acta 1994 87 293. 12 E. Dietz W. Braun A. M. Bradshaw and R. Johnson Nucl. Instrum. Methods Phys. Res. Sect. A 1985 239 359.13 Th. Mull H. Kuhlenbeck G. Odoé rfer R. Jaeger C. Xu B. Baumeister M. Menges G. Illing H.-J. Freund D. Weide and P. Anderson in Desorption Induced by Electronic T ransitions DIET IV ed. G. Betz and P. Varga Springer-Verlag Berlin 1990 p. 169. 14 D. P. WoodruÜ R. Davis N. A. Booth A. M. Bradshaw C. J. Hirschmugl K.-M. Schindler O. SchaÜ V. Fernandez A. Theobald Ph. Hofmann and V. Fritzsche Surf. Sci. 1996 357/358 19. 15 Ph. Hofmann and K.-M. Schindler Phys. Rev. B Condens. Matter 1993 47 13941. 16 Ph. Hofmann K.-M. Schindler S. Bao A. M. Bradshaw and D. P. WoodruÜ Nature (L ondon) 1994 368 131. 17 V. Fritzsche Surf. Sci. 1989 213 648. 18 V. Fritzsche J. Phys. Condens. Matter 1990 2 1413. 19 V. Fritzsche Surf. Sci. 1992 265 187. 20 V.Fritzsche and J. B. Pendry Phys. Rev. B Condens. Matter 1993 48 9054. 21 K.-M. Schindler V. Fritzache M. C. Asensio P. Gardner D. E. Ricken A. Robinson A. M. Bradshaw D. P. WoodruÜ J. C. Conesa and A. R. Gonzaç lez-Elipe Phys. Rev. B Condens. Matter 1992 46 4836. 22 N. A. Booth R. Davis R. Toomes D. P. WoodruÜ C. Hirschmugl K.-M. Schindler O. SchaÜ V. Fern- Gieêel P. Baumgaé rtel and A. M. Bradshaw Surf. Sci. andez A. Theobald Ph. Hofmann R. Lindsay T. 1997 387 152. 23 R. Dippel D. P. WoodruÜ X.-M. Hu M. C. Asensio A. W. Robinson K.-M. Schindler K.-U. Weiss P. Gardner and A. M. Bradshaw Phys. Rev. L ett. 1992 68 1543. 24 V. Fritzsche and D. P. WoodruÜ Phys. Rev. B Condens. Matter 1992 46 16128. Faraday Discuss. 1999 114 141»155 154 25 K.-M. Schindler Ph. Hofmann V. Fritzsche S. Bao S. Kulkarni A. M. Bradshaw and D. P. WoodruÜ Phys. Rev. L ett. 1993 71 2054. 26 R. Dippel K.-U. Weiss K.-M. Schindler P. Gardner V. Fritzsche A. M. Bradshaw M. C. Asensio X.-M. Hu D. P. WoodruÜ and A. R. Gonzaç lez-Elipe Chem. Phys. L ett. 1992 199 625. 27 R. Davis X.-M. Hu D. P. WoodruÜ K.-U. Weiss R. Dippel K.-M. Schindler Ph. Hofmann V. Fritzsche and A. M. Bradshaw Surf. Sci. 1994 307»309 632. 28 J. J. Barton Phys. Rev. L ett. 1988 61 1356. 29 S. Y. Tong H. Huang and C. M. Wei Phys. Rev. B Condens. Matter 1992 46 2452. 30 J. G. Tobin G. D. Waddill H. Li and S. Y. Tong Phys. Rev. L ett. 1993 70 4150. 31 H. Wu G. J. Lapeyre H. Huang and S. Y. Tong Phys. Rev. L ett. 1993 71 251. 32 H. Ascolani J. Avila N. Franco and M. C. Asensio Phys. Rev. L ett. 1997 78 2604. 33 J. J. Barton S. W. Robey and D. A. Shirley Phys. Rev. B Condens. Matter 1986 34 778. Paper 9/02212K 155 Faraday Discuss. 1999 114 141»155

ISSN:1359-6640    

DOI:10.1039/a902212k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

What can we learn on the structure and morphology of metal oxide/metal interfaces by measurement of X-ray crystal 157 truncation rods in situ during growth G. Renaud O. Robach and A. Barbier CEA-Grenoble Deç partement de Recherche Fondamentale sur la Matie` re Condenseç e / SP2M / IRS 17 rue des Martyrs 38054 Grenoble cedex 9 France I. Introduction Oxide surfaces1 and metal/oxide2,3 interfaces are present in numerous technological areas such as thin –lms composite materials microelectronics catalysis and protection against corrosion or industrial glasses. The thermal mechanical chemical magnetic or electrical properties of these materials often depend on the atomic structure of the internal interface they contain. From a theoretical point of view the properties of metal/oxide interfaces are difficult to predict because the interaction is very complex at the atomic scale.4h6 The interfacial energy contains several terms7 that are of the same order of magnitude.Their relative weights are difficult to estimate because of the lack of experimental data. The Ag/MgO(001) and Pd/MgO(001) interfaces have been chosen by numerous theoreticians as prototypical metal/oxide systems8 because they are relatively simple. They have four-fold symmetry the epitaxy is cube-on-cube,9h11 and the contribution of epitaxial strains to the interfacial energy is often neglected. For Ag this approximation is justi–ed because of the moderate lattice parameter mismatch [2.98% between fcc Ag and Faraday Discuss. 1999 114 157»172 Received 6th April 1999 The crystal truncation rods (CTRs) of the substrateœs surface were measured during the very –rst stages of in situ deposition of three fcc metals Ag Pd and Ni on the MgO(001) surface.These interfaces are known to form via nucleation growth and coalescence of islands. We show that quantitative analysis of the interferences between the waves scattered by the substrate and the wave scattered by a fraction of the metal –lm that is long-range correlated via the substrate allows the determination of the adsorption site the interfacial distance parameters that are important for theoretical calculation. Some other parameters of the metal/oxide interface are also deduced in particular information concerning the morphology. We show that in the cases of Pd and Ni the analysis is rather straightforward because most of the signal arises from a few atomic planes that are lattice-strained by the substrate parallel to the interface.Much more complicated is the case of Ag which is never fully strained by the substrate whatever the amount deposited i.e. the islandœs size. In the three cases the epitaxial site is shown to be unique above the oxygen ions of the MgO(001) surface. The evolutions of the interfacial distance during growth are compared. The results are discussed in view of the similarities and diÜerences between the three systems especially in view of the strongly diÜering lattice parameter mismatches and the strength of the metal oxide bound at the interface. General trends on the interfacial structure and morphology are deduced.This journal is( The Royal Society of Chemistry 2000 rocksalt MgO and for Pd despite the larger mis–t [7.64% because numerous experimental studies show that the –rst Pd monolayer at the interface is lattice-matched with the substrate. In addition Ag is a noble metal and hence no chemical reaction takes place at the interface. The Pd/MgO(001) interface is also considered as a model system to study the elementary processes in heterogeneous catalysis,12 which depend on the exact shape and distribution of the metal clusters that are spread over a ceramic substrate. The Ni/MgO(001) interface has the same symmetry but with a much larger mis–t of 16.4%. It has also been the subject of recent theoretical investigations that show that Ni is expected to strongly interact with MgO(001).13,14 In addition the Ni/ MgO(001) system has interesting magnetic properties since Ni is a ferromagnetic element.DiÜerent studies15h21 showed that substrate defects and –lm strain have large eÜects on the magnetic domain structure. The MgO(001) surface has been chosen because the relaxation is very small and thus the surface can be considered as a simple truncation of the bulk. Two important questions for theoreticians are the determination of the adsorption site among the three possible ones above O ions of the substrate above Mg ions or in between above the ììoctahedral site œœ and the determination of the interfacial distance between the last MgO(001) plane and the –rst plane of the metal. However experimental data are scarce because the insulating character of the substrate hampers most usual experimental surface science methods.To our knowledge although most calculations estimate the interfacial distance there are only very few experimental determinations of this parameter. For Ag a high resolution transmission electron microscopy (HRTEM) study concluded to the co-existence of two alternating epitaxial sites above Mg and O ions. However many theoretical calculations favour the O site. For Pd some recent theoretical results concluded that Pd adsorbs on top of the O at the surface,13,14,23,24 while others25,26 concluded to adsorption above Mg ions. A SEELFS (surface electron energy-loss –ne-structure spectroscopy) investigation27 concluded to adsorption above the Mg site.For Ni the O site is predicted. In the present study we show that these structural parameters can be determined by quantitative measurements and analysis of the MgO(001) crystal truncation rods (CTRs). In simple cases such as for Ni and Pd where part of the metal is lattice-strained by the substrate the analysis can be straightforward and a good accuracy obtained. In more complicated cases such as for Ag for which the metal is not lattice-strained but at least partially relaxed we show that in general these parameters can also be deduced but with a lesser accuracy. Note that despite the fact that in most theoretical investigations the contribution of epitaxial strains to the interfacial energy is neglected the relaxation of the lattice parameter mismatch is believed to yield an important contribution to the interfacial energy.28 The state of strain in the metal deposit critically depends on the magnitude of the mis–t as well as on the strength of the metal»oxide bonding at the interface.It is thus interesting to compare similar systems such as Ag/MgO Pd/MgO and Ni/MgO with very diÜerent mis–ts and strengths of the interfacial bond. Grazing incidence X-ray scattering (GIXS)29 is well suited for characterising the structure and morphology of metal/oxide interfaces during their growth because it is insensitive to the insulating character of the substrate and can be used in situ in ultra-high vacuum (UHV) without perturbing the deposit. This paper presents GIXS results obtained in situ during the room temperature growth of Ag Pd and Ni on MgO(001) from the very early stages and up to fairly large thickness.The results concerning the morphology of the deposit and the accommodation of the lattice parameter mismatch at the metal/MgO interface have been described elsewhere for the Ag/ MgO(001),30,31 Pd/MgO(001)32,33 and Ni/MgO(001)34 interfaces. We just recall here that in all three cases the growth is three-dimensional of Volmer»Weber type with nucleation growth and coalescence of islands followed by the formation of a continuous –lm. For Pd most of the deposit is partially relaxed excepted for the –rst monolayer which is partially lattice-matched parallel to the substrate. For Ag all the metal is partially relaxed and the interfacial plane is not latticematched to the substrate.For both Ag and Pd mis–t dislocations enter at the edges of the islands around 5»6 equivalent monolayers of metal deposited and they next reorder into a square network when the –lm becomes continuous. For Ni most of the deposit consists of relaxed Ni with cube on cube epitaxy but some other orientations with Ni(110)//MgO(001) are also present in much smaller amounts. Faraday Discuss. 1999 114 157»172 158 The experimental conditions are –rst described. Then the results of the measurements along CTRs of the MgO(001) surface during growth are analysed in order to deduce the epitaxial and interfacial sites together with the evolution of several other parameters with the deposited amount. These results are discussed in a last part and a detailed comparison is made between the three interfaces.The diÜerences between these three systems are discussed according to the magnitude of the lattice parameter mis–t and the strength of the interfacial bonding. a\0.12°\aMgO the critical angle c a\2/3aMgO and aB2 . Working at a\2/3aMgO was manda- c II. Experiments For Ag and Ni the GIXS experiments were performed using the SUV (surfaces in ultra-high vacuum) surface diÜraction set-up of the BM32 beamline at ESRF (European Synchrotron Radiation Facility Grenoble France).35 For Pd the w21v surface diÜraction set-up36 of the ID32 undulator beamline was used. The UHV chambers (base pressure 2]10~11 mbar for Ag and Ni; 10~10 mbar for Pd) equipped with two Be windows are mounted on diÜractometers that allow simultaneous deposit and the diÜraction measurements.The SUV chamber is also equipped with an electronic bombardment furnace an ion gun re—ection high energy electron diÜraction (RHEED) and auger electron spectroscopy (AES) systems. In all cases the substrate was kept at room temperature and the measurements were performed on cumulative deposits the growth being interrupted during X-ray scans. Ag was deposited by means of a Knudsen cell with a deposition rate of 0.73 Aé min~1. Pd and Ni were evaporated from electron-beam-heated rods of 99.99% purity. The Pd and Ni deposition rates were 1 Aé min~1. In each case the calibration was performed with a quartz microbalance prior and after the X-ray measurements and was checked by in situ X-ray re—ectivity measurements on the last deposit.The 15]15]0.5 mm3 MgO(001) single crystals were supplied oriented ^0.1° and both sides polished by Earth Chemical (Japan). The preparation of an MgO(001) surface with a quality adequate for GIXS is difficult and has been described in detail elsewhere.22 It leads to MgO(001) surfaces that are very —at and of high crystalline quality free from any impurity with in-plane domain size larger than 1 A lm an average terrace size of 6000 é A and an RMS roughness of 2.4 é . The X-ray beam energy was set at 18 keV which allowed measurements of CTRs37,38 over an extended range of perpendicular momentum transfer thus yielding a high accuracy on out-ofplane parameters. All measurements were systematically performed with three –xed values of the incident angle a of the X-ray beam with respect to the surface for total external re—ection of MgO c tory for small metal thickness in order to optimise the surface signal over the noise ratio.The sample surface was vertical. The incident X-ray beam size was 0.05 mm (H)]0.5 mm (V) for Pd and 0.42 mm (H)]0.39 mm (V) for Ag and Ni. The opening of the two pairs of detection slits was –xed at 1 mm (H)]1 mm (V) for the measurements of the MgO Crystal CTRs (corresponding to an angular acceptance of 0.1°) in all three cases. The Miller indexes (h k l) are expressed in reciprocal lattice units (r.l.u.) of MgO using the bulk (a fcc unit cell MgO\4.2117 Aé ). The l index is the component of the momentum transfer perpendicular to the surface. III.Results and analysis III.A. Quantitative measurements of the CTRs The sharp truncation of a surface is known to produce CTRs37,38 extending perpendicular to the surface and connecting bulk Bragg peaks. Between Bragg peaks the intensity variation is very sensitive to the structure of the interface. On the MgO(001) surface two kinds of non-equivalent CTRs are present ììstrongœœ ones (whose intensity is proportional to the square of the sum of the atomic form factors of O and Mg) with h and k even and ììweakœœ ones (whose intensity is proportional to the square of the diÜerence of the form factors) with h and k odd. They yield complementary information on the adsorption site.39 For the three metals we have quantitatively measured the (20l) and (11l) MgO CTRs on the bare substrate and during the –rst stages of metal deposition between hB0 and B12.5 ML where h is the amount of metal deposited in equivalent monolayers (ML).For Pd an additional 159 Faraday Discuss. 1999 114 157»172 Fig. 1 Modulus of the structure factor of the (20l) and (31l) CTRs for the bare substrate and for 1 ML of Pd deposited. (a) (20l) CTR obtained by integration and correction of rocking scans measurements for the bare MgO(001) substrate (open squares with the line showing the best –t) ; and for h\1 ML of Pd deposited (open circles with error bars ; the continuous line is the best –t). The best –ts yielded the following parameter values PdhMgO\2.22^0.02 Aé oonvsite\0.5^0.1 Pd\2.7^0.3 Aé d PdhPd\1.86^0.03 Aé ; (b) l-scan mea- h ML d and surements of the (31l) CTR for the bare substrate (open squares) and after deposition of h\1 ML (open circles) ; (c) calculated (31l) CTR for the bare substrate (open squares) and for 1 ML of Pd deposited with Pd above either Mg ions (dashed line) or O ions (thick continuous line).CTR the (31l) was also measured. For the bare substrate and for h\1 and 2 ML (and also 10 ML for Ag) the CTRs were measured by rocking the sample around its surface normal for each l value. These rocking scans were background subtracted integrated normalised to constant monitor counts and corrected for polarisation and Lorentz factors.40,41 Most other measurements were performed directly in ìì l-scansœœ i.e. along the CTRs and corrected with a diÜerent Lorentz factor,41 which is possible because the intrinsic width of the MgO CTRs is always (except close to l\0) much smaller than the width induced by the experimental resolution and because this width is not modi–ed by the deposition of the metal deposition at least for h\10 ML.It is important to note that this last observation implies that the possible modi–cations of the CTRs that may be induced upon metal deposition arise from metal atoms that are correlated via the substrate over very long lateral distances and have the substrateœs correlation length (resolution limited). This corresponds to pairs of atoms whose in-plane separation is equal to exact multiples of the MgO lattice vectors i.e. that have the same internal co-ordinate in the MgO unit cell (in prolonging of the MgO lattice inside the metal).We call this fraction of the deposit ììsubstrate-correlated metal fraction œœ (SCF or SC metal) or for simplicity ìì on-site œœ metal. Note that the atoms of the SCF are not necessarily located above the MgO sites and do not necessarily form a continuous crystal. III.A.1. Pd/MgO(001) interface. Fig. 1 shows the (20l) and (31l) CTRs for the bare substrate and for h\1 ML of Pd deposited. Pd deposition drastically modi–es the shape of the CTRs. On the (20l) CTR it induces a pronounced decrease in the intensity between l\0 and l\2 together with a sharp minimum around l\2.35. Along the (31l) CTR a destructive interference is present both on the low-l and high-l sides of the (311) Bragg peak and a positive interference on the right of the (333) peak.These modi–cations arise from the interference between the waves scattered by the substrate and those scattered along the CTR by that fraction of the Pd –lm made of Pd atoms correlated via the substrate or ìì on-site œœ Pd. By –tting the experimental MgO CTRs with an Faraday Discuss. 1999 114 157»172 160 oonvsite(z) of the ìì on-site œœ Pd plane located at a distance z above the last MgO plane appropriate model it is possible to determine the position of the adsorption sites of ìì on-site œœ Pd atoms and several structural parameters of the ìì on-site œœ fraction. The (20l) and (31l) CTRs were simultaneously –tted using a least-squares –tting procedure. The occupancy was described by a complementary error function (Fig. 2) oonvsite(z)\oonvsite 2N 1 erfcA z B J2 hPd where 1 2 erfcA zi .B N\ ; Metalplane J2 hPd nbi The –tting parameters are the total amount of ìì on-site œœ Pd (oonvsite in ML) the average height hPd of the ìì on-site œœ Pd with respect to the MgO substrate (which includes the root mean squared roughness) the interfacial distance dPdhMgO and the mean interplane (002) distance in the Pd –lm dPdhPd which was supposed to be independent of the height with respect to the interface. The scale and the substrate roughness (2.4 Aé RMS) were –rst determined by a –t of the clean substrate CTRs (Fig. 1) the substrate roughness being modelled by a Gaussian distribution of terrace heights.42 The MgO(001) substrate was assumed to be unaÜected by the Pd deposit. The best –ts of the (20l) CTR for h\0 and 1 ML are reported in Fig.1. Qualitatively the sign of the interference along this CTR (with h and k even) allows discrimination between epitaxy on top of either oxygen or magnesium on the one hand or above the octahedral site on the other hand. The strong destructive interference on the right-hand side of the (222) Bragg peaks allows the exclusion of the possibility of the octahedral site. The parameters of the quantitative –t of the (20l) CTR for h\1 ML were dPdhMgO\2.22^0.02 Aé oonvsite\0.5^0.1 ML hPd\2.7^0.3 Aé d and PdhPd\1.86^0.03 Aé . The measurement of a CTR with h and k odd is necessary to distinguish between the two remaining possible epitaxial sites oxygen or magnesium. Fig. 1(c) shows the simulated (31l) CTR using these parameters for a Mg site and for an O site.Clearly the epitaxial site is above the oxygen atoms of the substrate and not Mg ones. Fig. 2 Schematic drawing of the model used for –tting the CTRs. Right atomic positions. The metal (\Pd Ni or Ag) atoms are represented by grey circles Mg ions by black disks and oxygen ions by open circles. Left shape of the pro–le describing the occupancy of the metal planes as a function of the co-ordinate z perpendicular to the surface. In this –gure the substrate was supposed to be perfectly —at. 161 Faraday Discuss. 1999 114 157»172 Then for all deposits between h\0 and 12.5 ML the (20l) and (31l) CTRs measured in l-scans were simultaneously –tted over the ranges l\1»3.7 and 0.5»3.7 respectively assuming the oxygen site.The best –ts of the experimental data are reported in Fig. 3 and the corresponding parameters in Fig. 4. For all deposits the agreement is good which shows that the chosen model is adequate. oonvsite [Fig. 4(a)] and hPd [Fig. 4(b)] are found to –rst increase quickly with h and then slowly reach asymptotes around A B1.3 ML and 6 é respectively. dPdhPd [Fig. 4(d)] decreases from 1.895 Aé for A h\0.2 ML to 1.79 é for h\4 ML and then stays nearly constant with only a very slight decrease to 1.785 Aé for h\12.5 ML. Finally dPdhO [Fig. 4(c)] shows a peculiar behaviour it –rst decreases from A A B2.23^0.03 é at h\0.5 ML to 2.15^0.03 é for h\4 ML and then increases to reach a steady state value of 2.22 A ^0.03 é above 10 ML. However all these dPdhO values are very close to each other.The evolution with h of the structural parameters describing the ìì on-site œœ fraction contains a lot of information in particular on the nature of this ìì on-site œœ fraction. Let us –rst look at the average height (or RMS roughness). As shown in Fig. 4(b) below h\2 ML the average height of the ìì on-site œœ fraction is equal to the equivalent height of Pd deposited. It then remains much smaller than this equivalent height reaching only the height of 3 atomic planes for h\12.5 ML. This shows that the ìì on-site œœ fraction is con–ned near the interface between Pd and MgO(001). In addition at the very beginning (h\0.5 ML) the ìì on-site œœ occupancy [Fig. 4(a)] is nearly equal to the amount deposited the fraction of Pd that is already relaxed is much smaller than the pseudomorphic fraction.For h\1 ML oonvsite B0.5 ML half of the amount deposited is ìì on-site œœ and located at the interface. The ìì on-site œœ occupancy next Fig. 3 Comparison between the measured (rough line) and calculated (smooth line) (20l) and (31l) CTRs during the room temperature growth of Pd on MgO(001). The modulus of the structure factor is reported as a function of the out-of-plane momentum transfer. Both CTRs have been simultaneously –tted over a large range of out-of-plane momentum transfer. The amount h (in ML) of Pd deposited is indicated in the –gure. The curves were vertically shifted for clarity. Faraday Discuss. 1999 114 157»172 162 Fig. 4 Evolution with the amount h (in equivalent ML) of metal deposited of (a) the total amount of ììonsite œœ metal expressed in number of ML for Ag Ni and Pd compared with the total amount deposited (dashed line).(b) The ìì on-site œœ metal thickness for Ag Ni and Pd. The total equivalent thickness of metal deposited is also represented (solid line). (c) The interplane distance in the metal perpendicular to the surface for Ag (–lled squares) Ni (–lled circles) and Pd (triangles) compared with the distances expected for the bulk materials and to the distance calculated according to isotropic elasticity for metal strained in-plane to the MgO lattice squares) Ni (–lled circles) and Pd (triangles) and average interfacial distance (dashed line). parameter (horizontal lines). (d) Interfacial distance dMhMgO deduced from the –ts of the CTRs for Ag (–lled increases only much more slowly than h saturating at around oonvsite\1.2 ML for large deposited amounts.From these observations we conclude that for all deposited amounts most of the ìì on-site œœ Pd is composed of a pseudomorphic Pd layer i.e. a Pd layer that is lattice-matched with MgO parallel to the surface and is located either at the interface or in the next two Pd layers. This is con–rmed by the fact that the same scale factor is obtained when –tting CTRs of diÜerent orders which would not be the case if the interferences along the CTRs were arising from relaxed Pd as will be shown below for the case of the Ag/MgO(001) interface. Our conclusions are in agreement with previous HRTEM,43,44 as well as SEELFS45,46 measurements which show that the –rst Pd layer at the interface is laterally expanded to adopt the lattice parameter of MgO.M The out-of plane Pd-Pd distance [Fig. 4(d)] may be compared to the (002) bulk Pd interplanar dM distance Pd,B\1.945 Aé and to the value of dPd,S\1.71 Aé calculated in the framework of linear elasticity for Pd strained in plane to the MgO(001) substrate. This latter is deduced from the lattice parameters according to d dMPd,S[dPd,B \[2] C12] aAPd,S[aAPd,B M Pd,B M a C11 Pd,S A where C11 and C12 are the elastic constants of Pd (2.271]1011 Pa and 1.761]1011 Pa respectively) Pd,S is the in-plane lattice parameter of Pd strained by the MgO substrate (i.e. the aA 163 Faraday Discuss. 1999 114 157»172 Fig. 5 (11l) CTR vs. Ni amount deposited.From bottom to top 0 0.2 0.4 0.6 0.8 1.0 1.5 2.0 3.0 4.0 5.0 6.0 7.0 8.0 10.0 and 12.5. Comparison for each thickness between experimental data (Ö) and the best –t (»»») obtained with an oxygen epitaxial site and the 4 parameters reported in Fig. 4 for the Ni fraction which sits ìì on-site œœ. M MgO lattice parameter) and aPd,B is the bulk Pd lattice parameter. For all deposits the interplane A distance is intermediate between these two values. At the beginning of deposition it is closer to the bulk value and decreases down to a nearly constant value closer to that of elastically strained Pd for h[4 ML. At the beginning the ìì on-site œœ Pd has a thickness limited to one or two atomic layers so that the elasticity theory cannot be applied. Since it is covered and surrounded by relaxed Pd32,33 a value close to the bulk one is not surprising.As the thickness increases below h\4-5 ML the elasticity theory becomes a better approximation. However the islands are now composed mainly of relaxed Pd33 and a small amount of pseudomorphic Pd. This explains the decrease of the interplane distance although it remains larger than dPd,S. The fact that both the interfacial distance and the Pd interplane distance show a discontinuity around h\4-5 ML can be correlated to the onset of plastic relaxation33 for this amount of Pd deposited. Above 4-5 ML dislocations are introduced at the island edges relaxing the mis–t and the ìì on-site œœ fraction becomes con–ned in the regions of ììgood matchœœ between the Pd and MgO lattices i.e.laterally located between two mis–t dislocations lines. At these locations the ìì on-site œœ Pd is close to truly pseudomorphic which explains the value of the Pd interplane distance. However the out-of-plane interplanar distance should no longer be compared to elastic calculation of strained Pd but rather to a full calculation of the atomic positions between two mis–t dislocations which is beyond the scope of the present study. Above h\4»5 ML the increase of the interfacial distance would correspond to a weakening of the metal»oxide bond as the interfacial atoms become less isolated i.e. as they are more and more involved in the metallic bounding of the Pd –lm. An increase of the interfacial distance with h is actually what is predicted by ab initio calculations47,48 and is the tendency that is intuitively expected.Why the interfacial distance decreases at the very beginning of deposition is less clear. A tentative explanation would be correlated with a decreasing aspect ratio (height over lateral extension) of the islands with increasing h. As this aspect ratio would decrease the in—uence of Faraday Discuss. 1999 114 157»172 164 Fig. 6 Logarithm of the measured intensity along the (11l) (a) and (20l) (b) MgO(001) CTRs as a function of the out-of-plane co-ordinate l for diÜerent amounts of deposited Ag. Incident angle 0.08° from 0 to 8 ML 0.12° at 11 ML 0.15° at 17 ML and 0.22° at 19 ML. (a) (11L) CTR deposits 0 ML (open circles) 0.2 0.3 0.4 ML (open squares) 0.5 0.6 0.7 0.8 0.9 1 1.5 2 4 6 and 19 ML.(b) (20l) CTR deposits 0 ML (open circles) 0.2 0.5 1 2 3 ML (open squares) 4 6 8 11 and 17 ML. ììmetallicœœ bonding vs. interfacial bonding would also decrease resulting in a shorter interfacial distance. III.A.2. Ni/MgO(001) interface. The (20l) and (11l) (Fig. 5) CTRs were also measured during the room temperature deposition of Ni on MgO(001).34 The (20l) CTR shows a strong intensity decrease of half an order of magnitude between 0 and 1 ML. A similar intensity decrease is observed between 1 and 125 ML. Such a strong signal decrease is not consistent with the X-ray absorption from one Ni monolayer. It thus arises again from the existence in the submonolayer regime of a large part of the Ni atoms sitting on top of the substrate sites. Fig. 5 shows that as in the Pd case that the (11l) CTR is well reproduced with the above model of ìì on-site œœ Ni.Again in agreement with the theoretical calculations13,14,23,24 the evidence of interference along the (11l) CTR unambiguously shows that the Ni atoms sit above the oxygen atoms of the last MgO plane. Fig. 5 shows that within this model the agreement between the experimental and calculated structure factors along the (11l) CTR is good up to 4 ML and fair up to 12 ML. The values of the four previously described parameters for all thicknesses are reported in Fig. 4. In the very early stages of growth the distance between two successive Ni layers is close to that of bulk Ni (most likely isolated atoms or small clusters) [Fig. 4(d)] ; it decreases rapidly with increasing amount of Ni deposited down to a value close to that expected from the elastic theory.The fraction of the Ni layer that sits ìì on-site œœ increases –rst slowly and then more –rmly with the amount of Ni deposited and reaches about 1 ML for a total amount of Ni deposited of 10 ML afterwards it saturates [Fig. 4(a)]. The amount of ìì on-site œœ Ni remains always well below the amount of Ni deposited. The average height of the Ni layer passes through a minimum at 0.6 ML and then essentially increases with respect to the amount of Ni deposited [Fig. 4(b)]. 165 Faraday Discuss. 1999 114 157»172 On the overall as concerns the ìì on-site œœ fraction the case of the Ni/MgO(001) interface is very similar to that of the Pd/MgO(001) one except for the values of the distances while the structures of the remaining of the metal deposit are markedly diÜerent.III.A.3. Ag/MgO interface. III.A.3.a. Simpli–ed model. For Ag the evolution with h (Fig. 6) of the intensities along the (20l) and (11l) MgO CTRs looks at –rst glance very similar to that in the Pd case. However in-plane grazing incidence small and wide angle scattering32,33 showed that from the very beginning of deposition the Ag grows in the form of islands with an in-plane lattice parameter close to that of bulk Ag. In other words there is no pseudomorphic Ag. However there are residual strains in the Ag islands which implies that all interfacial Ag atoms are closer to a preferential substrate site which we therefore call the ììadsorption site œœ. When atoms are displaced too much with respect to this site interfacial dislocations are introduced yielding ììgoodmatchœœ regions in which the adsorption site can still be de–ned and ììbad-matchœœ regions in which the Ag atoms do not sit on top of any particular site.In a –rst crude approximation we could use the same model as for the Pd and Ni cases i.e. to consider that only the Ag atoms that are perfectly ìì on-site œœ contribute. The interferences would thus be due either to fully lattice-matched Ag or to separated ìì on-site œœ columns that are located at the centre of the islands during the –rst three stages of the growth or exactly halfway between two dislocation lines when the –lm is continuous. In this last case the ìì on-site œœ fraction does not form a continuous crystal in the directions parallel to the surface even on the scale of the interatomic spacing but the MgO CTRs can be modelled as before.Using this model a qualitative analysis of the sign of the interference observed along the MgO CTRs again shows that the Ag atoms of the –rst plane sit atop of the oxygen atoms of the last MgO plane. The parameters of the model (oTOTAL hM dAghMgO and dAghAg) determined by a simultaneous least-squares –t of the (11l) and (20l) CTRs are represented in Fig. 4 as a function of h. Fig. 7 shows the comparison between the experimental CTRs and the best –ts for selected deposited amounts h. The –rst striking feature is the very small amount of ìì on-site œœ Ag [Fig. 4(a)]. It may be surprising that large eÜects are observed on the MgO(001) CTRs with such a small ìì on-site œœ Ag amount.However the MgO CTRs are extremely sensitive to the presence of Ag because its scattering power is 24 times that of MgO on the ììintense CTRsœœ and 150 times that of MgO on the ììweak onesœœ. The second important result is the large values of the height h of the ìì on-site œœ fraction. The very small amount of ìì on-site œœ Ag (0.02 ML at h\0.2 ML) and also the large values of h (about 3 planes at h\0.2 ML) both con–rm that the ìì on-site œœ Ag does not consist of latticematched Ag. If some lattice-matched Ag was present like in the Pd and Ni cases it should be con–ned near the interface so its thickness should not exceed one or two planes and the major part of the –rst Ag plane would be lattice-matched which would yield like in the Ni and Pd cases an amount of ìì on-site œœ Ag close to the equivalent amount deposited h at least for small h.The values found for h indicate that Ag is already in the form of islands with a height of several planes at 0.2 ML there is no stage of two-dimensional growth. However although the ìì on-site œœ Ag is not lattice-matched and extends over a signi–cant height it is not only composed of the central ìì on-site œœ columns either. Indeed the amount of ìì on-site œœ Ag obtained with this simple model indicates that whatever the island size there is more than one ìì on-site œœ Ag column per island. III.A.3.b. General model origin of the interference. This observation led us to a more general model in which at least for h\4 ML we suppose that all the islands of Ag are approximately identical and are ììpinnedœœ in the same way on the substrate so that the whole Ag deposit contributes to the modi–cations of the MgO CTRs.In other words we propose a model of partially relaxed islands correlated via the substrate. We also come back to the original more exact notation of ììsubstrate-correlated Agœœ (SC Ag) instead of ìì on-site œœ Ag which is misleading. In such a model where the SC Ag fraction is the whole deposit all the Ag atoms contribute to the MgO CTRs but with a weight that decreases with increasing lateral separation from the central ììonsite œœ column. In a small island of relaxed Ag the central atomic column which is perfectly ììonsite œœ fully contributes because all the ìì on-site œœ columns are not only above MgO ones but are in addition fully correlated via the substrate.As interfacial Ag atoms lie farther away from this Faraday Discuss. 1999 114 157»172 166 Fig. 7 Comparison between the measured (crosses) and calculated (solid lines) MgO(001) CTRs during the room temperature growth of Ag on MgO(001). The logarithm of the modulus of the structure factor is plotted vs. l. The dashed lines correspond to the clean MgO(001) substrate. The (11l) and (20l) CTRs have been represented on the same l-scale although they are at diÜerent h k values. The curves corresponding to the diÜerent amounts of deposited Ag are shifted vertically for clarity. central column their contribution to the MgO CTRs decreases for two reasons.The –rst is that they are more and more displaced from the ìì on-site œœ position. The second is that pairs of atoms located in diÜerent islands far from the ìì on-site œœ column are less likely to be correlated by the substrate because the exact atomic distribution presumably diÜers between diÜerent islands especially if they are of diÜerent sizes. Let us –rst illustrate this model by supposing that all the islands are exactly identical and are ììpinnedœœ in the same way on the substrate. In this case the MgO CTRs can be modelled by calculating the intensity scattered by a supercell that comprises a semi-in–nite MgO column with a square basis whose lateral size is equal to the inter-island distance and a Ag island that is either fully relaxed or slightly strained containing at its centre a perfectly ìì on-site œœ column.A very simple estimation of the contribution of Ag to the MgO truncation rods allows the prediction of the diÜerences that will appear in the diÜracted intensity between this general model and the above simpli–ed one. Let us assume for simplicity that all the Ag has its bulk lattice parameter. The supercell comprises a semi-in–nite column of MgO yielding a CTR centred on the parallel momentum transfer value of MgO and a Ag island yielding a rod centred on the parallel momentum transfer value of Ag. The Ag rod of full width at half-maximum width 2p/d (where d is the in-plane island size) yields a contribution below the MgO CTR that decreases as the order of the CTR increases since the spacing between the MgO and Ag rods increases with h and k while the width of the Ag peak remains constant.We experimentally observed this decrease of the Ag 167 Faraday Discuss. 1999 114 157»172 é contribution along the CTR as a function of the rod order which con–rms the validity of this last model. By contrast for lattice-matched or ìì on-site œœ Ag the respective weight of the Ag and MgO contributions to the MgO CTR would not vary with the order of the rod. This last case corresponds to the experimental observation for Ni and Pd deposition which validates the use of the simpli–ed model in these two cases. III.A.3.c. V alidity of the interfacial parameters. One drawback of this general model is that it contains too many parameters for a quantitative analysis.Even if we were able to calculate the atomic positions in a Ag island given its size and shape we would have to introduce as free parameters a mean island size a mean shape a mean inter-island distance and the dispersion over all these parameters in order to estimate the Ag-Ag correlation function and derive the intensity. For the quantitative analysis of the MgO CTRs the simpler model of ìì on-site œœ Ag was therefore used. Of course this inadequately models the lateral position of the Ag atoms but it allows deriving the interesting parameters the height of the SCF the adsorption site the inter-plane distance in Ag and the interfacial distance. Let us –rst show that the adsorption site found with this simpli–ed analysis corresponds to the real adsorption site.For this we have calculated the intensity scattered by a supercell for the extreme case where Ag has its bulk lattice parameter. The lateral size of the island was taken to be equal to 20 Aé A the inter-islands distance equal to 66 and the interfacial distance was –xed at the experimental steady state value 2.52 Aé [Fig. 4(c)]. Fig. 8 shows the (11l) and (20l) CTRs calculated with the central atomic column of the symmetric island set either on top of O or Mg or the octahedral site. The (11l) CTR allows us to distinguish between either the O site on the one hand or the Mg or octahedral sites on the other hand while the (20l) CTR allows us to distinguish between either the octahedral site on the one hand or the O or Mg sites on the other hand. The clear destructive interference experimentally observed (Fig.6) on both sides of the Bragg peaks along both CTRs is consistent only with the O site. The simulation with the O site (Fig. 8) Fig. 8 Calculation of the intensity scattered along the (11l) (a) and (20l) (b) MgO CTRs by a supercell composed of a semi-in–nite MgO(001) substrate and a small hemispherical island of Ag at its bulk lattice parameter with a 20 Aé diameter. The logarithm of the intensity is plotted vs. l. The inter-islands distance (i.e. the lateral size of the supercell) was –xed at 66 Aé and the interfacial distance was –xed at the experimental steady-state value 2.52 Aé (Fig. 4). The MgO substrate contribution is shown as black squares and the Ag scattering as open triangles. For the (11l) CTR (a) open circles show the CTR intensity for either a Mg or octahedral epitaxial site and the thick line shows the intensity for an oxygen adsorption site.For the (22l) CTR (b) the open circles correspond to the octahedral site while the thick line shows the intensity for either an O or Mg site. Faraday Discuss. 1999 114 157»172 168 é qualitatively reproduces most of the observed interference at 0.5 ML. For the other sites the simulated CTRs strongly diÜer from the experimental ones. Hence this very simple simulation shows that in this system a qualitative inspection of the sign of the interference along the CTRs using the ìì on-site œœ model allows us to determine correctly the adsorption site. Given this veri–cation we have to consider that the values deduced for dAghAg [Fig.4(d)] and dAghO [Fig. 4(c)] are average values over the SCF because these distances are probably nonuniform within a given island and may vary slightly between islands of diÜerent sizes and diÜerent strain states. The value of the average height [Fig. 4(b)] is probably representative of the average height of the islands since it is the central portion of the islands that contribute the most. The average interplane distance dAghAg\2.00^0.02 Aé is intermediate between the value for bulk Ag (dAgAg B \2.043 Aé ) and the value of dAgAg S \1.950 Aé calculated from the linear elasticity theory for Ag strained in-plane to the MgO lattice parameter. Finally we do not attach a particular meaning to the total amount of ìì on-site œœ Ag [Fig.4(a)] because this parameter serves to ììhideœœ everything that is not modelled properly like the real lateral position of the atoms in the islands and the dispersion on this parameter from one island to the other. IV. Comparison between the A Pd and Ni/MgO(001) interfaces In order to better understand the basic physical phenomena governing the interfacial structure i.e. the parameters of the interface and the residual deformations within the metal islands it is interesting to compare similar systems with systematic variations of some physically relevant parameters among which the lattice parameter mis–t and the strength of the interfacial bonding. Indeed the –nal state of the islands results from a competition between diÜerent mechanisms. The larger the lattice parameter mis–t the sooner the introduction of mis–t dislocation or other mis–t releasing defects is expected.Also intuitively the larger the mis–t and the larger the metal stiÜness the less likely the metal is to be strained by the substrate. At the opposite the stronger the interfacial bonding the stronger the tendency of the metal to be strained by the substrate and possibly to form a fully lattice-matched (pseudomorphic) interfacial layer. According to many recent theoretical calculations the strength of the metal»oxide bond varies signi–cantly between the three interfaces. For Ag the binding is weak of physisorption type between 0.1 and 0.3 eV atom~1 and mostly of electrostatic origin while for Pd and Ni the bond is mostly polar covalent and ranges from 0.65 to 0.81 eV atom~1 for Pd and from 0.88 to 1.24 eV atom~1 for Ni.4,8,13,14,23,24,28,49h56 Comparison of the elastic constants of the three metals also shows that Ag is much softer than Pd and Ni.On the other hand diÜerences between the surface stress of the three metals can be neglected in a –rst approximation.57 Apart from the common three-dimensional growth mode which is expected by simple thermodynamic arguments these three interfaces indeed exhibit fairly diÜerent characteristics. For Ag and Pd thick –lms the lattice parameter mis–t is simply relaxed by a network of interfacial mis–t dislocations while for Ni the relaxation is performed by the growth in addition to cube-on-cube Ni of Ni clusters in Ni(110)// MgO(001) epitaxy with four diÜerent in-plane orientational relationships.34 The absence of ordered mis–t dislocations at the Ni/MgO(001) interface is not surprising because the large value of the mis–t implies alternative relaxation processes.The above analysis unambiguously shows that for Ag Pd and Ni the atoms of the –rst metal monolayer sit on top of oxygen ions of the substrate. The present results are in agreement with all recent theoretical predictions.4,8,13,14,23,24,28,48h55 and of a recent surface X-ray absorption spectroscopy study in the case of Ag.58 We thus believe that the controversy concerning possible epitaxial sites is close to being over in all cases the epitaxial site is unique atop the oxygen ions. For Pd whatever the deposit the interfacial distance lies between 2.15 Aé and 2.23 Aé with a steady-state value of 2.22^0.03 Aé .A This is very close to the values of 2.18 for h\1 ML and 2.225 A Aé for h\2 ML calculated recently,47 as well as to the theoretical value of 2.15 é obtained for a single Pd atom adsorbed on top of the oxygen ions.13,24 For Ag the average value of the interfacial distance is dAghMgO\2.52^0.1 Aé which is very close to the most recent theoretical Faraday Discuss. 1999 114 157»172 169 é Aé ,8,23 Aé ,8,28 Aé ,48 Aé ,49,55 Aé ,56 values of 2.34 Aé ,51 2.38 2.47 2.49 2.50 2.45 to 2.64 depending on the Ag amount deposited 2.64 A A é ,52 and 2.6953 as well as to the experimental value of 2.53 é found both by HRTEM59 and SEXAFS.58 For Ni the interfacial distance remains almost insensitive to the amount of Ni deposited in the validity domain of the analysis.The average distance on top of O sites is found to be 1.82^0.05 Aé A which is close to the value of 1.87 deduced from recent ab initio calculations.14,23,24 These results on the epitaxial site and interfacial distance show that the most recent ab initio calculations predict these parameters correctly. One important remark is that most published theoretical calculations neglect the lattice parameter mismatch between the metal and the substrate and assume perfectly ìì on-site œœ metal atoms. By contrast in the real situation the metal is not completely strained to the MgO in-plane lattice parameter. For Ag in particular even if most of the interfacial atoms are close to a particular adsorption site there are only few that are perfectly ìì on-site œœ.This means that to be accurate theoretical calculations of the interfacial parameters should take into account the lattice parameter mismatch and allow slightly ììoÜ-siteœœ Ag atoms. V. Discussion and conclusions The diÜerences between the three metals are clear when comparing the parameters of the ììonsite œœ fraction. The signi–cant decrease of the interfacial distance from 2.52 A Aé for Ag to 2.22 é for Pd and 1.88 Aé for Ni directly re—ects the increasing strength of the interfacial bond. The most striking diÜerences however lie in the amount and height of the ìì on-site œœ fraction at the beginning of the growth between h\0 and 10 ML. For Pd and Ni the amount of ìì on-site œœ metal saturates around 1.2 ML for large h and its average height saturates around 2»3 atomic planes.This shows that in both cases a signi–cant fraction of the interfacial atoms is fully lattice-matched laterally with the MgO substrate. The larger mis–t for Ni is compensated by its stronger bonding with MgO yielding a very similar pseudomorphic fraction in both cases. Very diÜerent is the case of Ag with an extremely small amount of ìì on-site œœ Ag (less than 0.1 ML for h\0.5 ML) extending over a very large height reaching 15 atomic planes for h\10 ML. This together with the other scattering measurements could only be interpreted by the absence of lattice-matched Ag at the interface.30,31 Despite the small lattice parameter mis–t and the softness of Ag which implies a lesser cost of residual deformations as compared to Pd and Ni all the Ag within the islands has a lattice parameter close to that of bulk Ag even at the interface.This eÜect can only be explained by the very weak binding at the Ag/MgO interface. From all the above comparison we may conclude that the structure and morphology at these metal/MgO interfaces are mostly in—uenced by the strength of the bonding at the interface rather than by the lattice parameter mis–t. In summary for the growth of Ag Ni and Pd on a clean and —at MgO(001) surface at room temperature a new quantitative analysis of the MgO CTRs allowed the determination of the adsorption site and interfacial distance as well as a description of the morphology of the deposit for very small deposited amounts.In the three cases the epitaxial site is shown to be above the O ions of the last MgO(001) plane and the interfacial distance and its evolution during deposition have been determined. The average values are dAghMgO\2.52^0.1 Aé dPdhMgO\2.22^0.03 Aé and dNihMgO\1.82^0.05 Aé . These results are consistent with the most recent theoretical calculations. For Pd and Ni most of the monolayer closest to the interface is pseudomorphic while most of the Ag is already (partially) relaxed at all stages of the deposition. Comparison between the three interfaces shows that the main diÜerences between the interfacial characteristics in the three systems mostly arise from the largely diÜering strength of the metal» oxide bond rather than from the lattice parameter mis–t.Concerning the technique of surface X-ray diÜraction itself this study shows that measurements of the substrate crystal truncation rods can be used to characterise the deposit even when it does not contain a fully strained part. This arises because the scattering by the adsorbate always contains a large component with the spatial frequency of the substrate because the adsorbate is deposited on the substrate. Indeed if we consider for instance a growth mode starting with very small islands made of one or two Faraday Discuss. 1999 114 157»172 170 atoms all these islands are pinned on a particular site of the substrate with at least the central atom being ìì on-site œœ. When more atoms of the adsorbate are deposited they stick close to these ìì on-site œœ atoms starting to form islands which may contains diÜerent strains.The important point is that the atomic neighbourhood of the adsorbate atoms around the central ìì on-site œœ one is approximately the same in all islands which create a repeating unit in the deposit the diÜerent units being separated by an integer number of substrate lattice parameter parallel to the plane. All these islands thus yield a signi–cant contribution along the CTRs even if the metal has a lattice parameter that is very diÜerent from that of the substrate. Acknowledgements We would like to thank A. Bourret I. K. Robinson J. Villain C. Priester C. Noguera F. Lanc” on and T. Deutsch for valuable help or discussion and J. Jupille for his participation in one of the experiments. We would like to thank the help of A.Stierle and of the staÜ of the ID32 and BM32 beamlines during the measurements. References 1 V. E. Henrich Rep. Prog. Phys. 1985 48 1481. 2 H. Bialas and K. Heneka V acuum 1994 45 79. 3 F. Reniers M. P. Delplancke A. Asskali V. Rooryck and O. Van Sinay Appl. Surf. Sci. 1996 92 35. 4 M. W. Finnis A. M. Stoneham and P. W. Tasker Metal»Ceramic Interfaces ed. M. Rué hle A. G. Evans M. F. Ashby and J. P. Hirth Pergamon Oxford 1990 p. 35. 5 V. E. Henrich and P. A. Cox T he surface science of metal oxides Cambridge University Press Cambridge 1994. 6 Acta Metall. Mater. 1992 S40. 7 C. Noguera Physics and Chemistry at Oxide Surfaces Cambridge University Press Cambridge 1996. 8 J. R. Smith T. Hong and D. J. Slorovitz Phys. Rev.L ett. 1994 72 4021. 9 P. Palmberg T. Rhodin and C. Todd Appl. Phys. L ett. 1967 11 33. 10 A. K. Green J. Dancy and E. Bauer J. V ac. Sci. T echnol. 1979 7 159. 11 D. G. Lord and M. Prutton T hin Solid Films 1974 21 341. 12 D. W. Goodman Surf. Rev. L ett. 1995 2 9. 13 G. Pacchioni and N. Roé sch J. Chem. Phys. 1996 104 7329. 14 N. Roé sch and G. Pacchioni in Chemisorption and Reactivity on Supported Clusters and T hin Films ed. R. M. Lambert and G. Pacchioni Kluwer Academic Publishers Dordrecht 1997 p. 353. 15 H. Sato R. S. Toth and R. W. Astrue J. Appl. Phys. 1962 33 1113. 16 A. A. Hussain J. Phys. Condens. Mater. 1989 1 9833. 17 N. I. Kiselev Yu. l. Manœkov and V. G. Pynœko Sov. Phys. Solid State (Engl. T ransl.) 1989 31 685. 18 H. Maruyama H.Qiu H. Nakai and M. Hashimoto J. V ac. Sci. T echnol. A 1995 13 2157. 19 H. Qiu A. Kosuge H. Maruyama M. Adamik G. Safran P. B. Barna and M. Hashimoto T hin Solid Films 1994 241 9. 20 H. Bialas and L.-S. Li Phys. Status Solidi 1977 42 125. 21 M. R. Fitzsimmons G. S. Smith R. Pynn M. A. Nastasi and E. Burkel Physica B (Amsterdam) 1994 198 169. 22 O. Robach G. Renaud and A. Barbier Surf. Sci. 1998 401 227. 23 A. M. Ferrari and G. Pacchioni J. Phys. Chem. 1996 100 9032. 24 I. Yudanov G. Pacchioni K. Neyman and N. Roé sch J. Phys. Chem. B 1997 101 2786. 25 K. Yamamoto Y. Kasukabe T. Takeishi and T. Osaka J. V ac. Sci. T echnol. A 1996 14 327. 26 A. Stirling I. Gunji A. Endou Y. Oumi M. Kubo and A. Miyatomo J. Chem. Soc. Faraday T rans. 1997 93 1175. 27 C.Goyhenex and C. R. Henry J. Electron Spectrosc. Relat. Phenom. 1992 61 65. 28 T. Hong J. R. Smith and D. J. Srolovitz Acta Metall. Mater. 1995 43 2721. 29 L. K. Robinson and D. J. Tweet Rep. Prog. Phys. 1992 55 599. 30 O. Robach G. Renaud A. Barbier and P. Gueç nard Surf. Rev. L ett. 1997 5 359. 31 O. Robach G. Renaud and A. Barbier Phys. Rev. B 1999 60 5858. 32 G. Renaud and A. Barbier Appl. Surf. Sci. 1999 142 14. 33 G. Renaud and A. Barbier Surf. Sci. 1999 433. 34 A. Barbier G. Renaud and O. Robach J. Appl. Phys. 1998 84 4259. 35 http ://www.esrf.fr. 36 G. Renaud B. Villette and P. Gueç nard Nucl. Instrum. Methods Phys. Res. Sect. B 1995 95 422. 37 I. K. Robinson Phys. Rev. B Condens. Matter 1986 33 3830. 38 S. R. Andrews R. A. Cowley J. Phys.C Solid State Phys. 1985 18 6247. 39 G. Renaud Surf. Sci. Rep. 1998 32 1. 171 Faraday Discuss. 1999 114 157»172 40 E. Vlieg J. Appl. Crystallogr. 1997 30 532. 41 O. Robach PhD-thesis University of Grenoble 1997. 42 J. Harada Acta Crystallog. Sect. A 1992 48 764. 43 K. Heinemann T. Osaka H. Poppa and M. Avalos-Borja J. Catal. 1983 83 61. 44 S. Giorgo C. Chapon C. R. Henry and G. Nihoul Philos. Mag. B 1993 67 773. 45 C. Goyenex C. R. Henry and J. Urban Philos. Mag. A 1994 69 1073. 46 S. Bartuschat and J. Urban Philos. Mag. A 1997 76 783. 47 J. Goniakowski Phys. Rev. B Condens. Matter 1998 58 1189. 48 L. Spiess Surf. Rev. L ett. 1996 3 1365. 49 J. Purton S. C. Parker and D. W. Bullet J. Phys. Condens. Matter 1997 9 5709. 50 P. Bloé chl G. P. Das H. F. Fischmeister and U. Schoé nberger in Metal-Ceramic Interfaces ed. M. Rué hle A. G. Evans M. F. Ashby and J. P. Hirth Pergamon Press Oxford 1990 p. 9. 51 E. Heifets F. Y. Zhukovskii E. A. Kotomin and M. Causa Chem. Phys. L ett. 1998 283 395. 52 E. Heifets E. A. Kotomin and R. Orlando J. Phys. Condens. Matter 1996 8 6577. 53 C. Li R. Wu A. J. Freeman and C. L. Fu Phys. Rev. B Condens. Matter 1993 48 8317. 54 H. X. Liu H. L. Zhang H. L. Ren S. X. Ouyang and R. Z. Yuan Ceram. Int. 1996 22 79. 55 U. Schoé nberger O. K. Andersen and M. Methfessel Acta Metall. Mater. 1992 40 S1. 56 L. Spiess Surf. Rev. L ett. 1996 3 1365. 57 J. P. Hirth and J. Lothe T heory of Dislocations Krieger Publishing Co. Malabar FL Wiley & Sons 1992 Appendix 2 p. 838. 58 A. M. Flanck R. Delaunay P. Lagarde M. Pompa and J. Jupille Phys. Rev. B Condens. Matter 1996 53 1737R. 59 A. Trampert E. Ernst C. P. Flynn H. E. Fischmeister and M. Rué hle Acta Metall. Mater. 1992 40 S227. Paper 9/02735A Faraday Discuss. 1999 114 157»172 172

ISSN:1359-6640    

DOI:10.1039/a902735a

出版商:RSC    

年代:1999

数据来源: RSC