Mg clusters on MgO surfaces: study of the nucleation mechanism with MIES and ab initio calculations

Transcription

1 Mg clusters on MgO surfaces: study of the nucleation mechanism with MIES and ab initio calculations L. N. Kantorovich,a A. L. Shluger,a P. V. Sushko,ab J. Gu nster,c P. Stracke,d D. W. Goodmanc and V. Kempterd a Department of Physics and Astronomy, University College L ondon, Gower Street, L ondon, UK W C1E 6BT b T he Royal Institution of Great Britain, 21 Albemarle Street, L ondon, UK W 1X 4BS c Department of Chemistry, T exas A&M University, College Station, T X , USA d Physikalisches Institut der T U Clausthal, L eibnizstra e 4, D Clausthal-Zellerfeld, Germany Received 23rd April 1999 We combined experimental studies using ultraviolet photoelectron spectroscopy (UPS), metastable impact electron spectroscopy (MIES) and temperature programmed desorption (TPD) with ab initio calculations of metal adsorption on the perfect MgO surface and at defect sites in order to elucidate the role of surface defects in the initial stages of nucleation and growth of metal clusters at oxide surfaces. MgO Ðlms (2 nm thick) grown on Mo and W substrates were used as a prototype system. The MIES and UPS (HeI) spectra were collected in situ, and the growth of Mg clusters was observed by monitoring the dynamics of additional MIES peaks during Mg deposition. TPD experiments were made in order to monitor the surface coverage by Mg clusters and to determine the Mg desorption energies. Interpretation of the results was made on the basis of theoretical modelling using density functional theory (DFT) calculations in both periodic and embedded cluster models. The geometric and electronic structures of the surface terrace, F-centre, positively charged anion vacancy, and step edge at the MgO(001) surface were calculated, and their role in adsorption and clustering of Mg atoms on this surface was studied. The absolute position of the top of the surface valence band of MgO with respect to the vacuum was calculated and compared with the MIES results. The MIES spectra were modelled on the basis of surface density of states (SDOS). The calculated SDOS predicted the location of additional peaks in the band gap and their shift as a function of Mg concentration on the surface in agreement with the MIES data. The desorption energies of Mg atoms from small Mg clusters formed at step edges are found to be about 1.3 ev atom~1. Comparison between the theoretical results and the experimental data suggests preferential initial adsorption of Mg atoms at steps and kinks, rather than at charged and neutral vacancies. At larger exposures these Mg atoms serve as the nucleation sites. This journal is ( The Royal Society of Chemistry 2000 Faraday Discuss., 1999, 114, 173È

2 1 Introduction Understanding of the mechanisms of growth and parameters of the geometric and electronic structures of metal clusters and layers on metal oxide surfaces is important for a number of technological applications. In particular, metal addition to oxides leads to an enhanced reactivity via electron transfer to a variety of adsorbed molecules leading to the formation of radical anion species.1,2 The interaction between metal clusters and metal oxide supports plays a key role in catalysis3,4 and microelectronics.5 The interaction between metal atoms and oxide surfaces is important for understanding the mechanisms of their segregation,6,7 di usion of metal atoms on insulators,8 formation of metal-induced point defects on oxides,2,9 and the formation of nanoparticles inside semiconductors and insulators.10 Surprisingly, little is known about the structure of metal/oxide interfaces, in particular, the initial stages of the metal adsorption, types of the metal adsorption sites, the nature of bonding to oxides and between the adsorbed metal atoms.5,11 Although it is clear that the defect sites, such as surface vacancies, step edges and kinks play a signiðcant role, at least at the early stages of metal growth on oxides, the number, distribution and structure of these defects is very difficult to control experimentally. On the other hand, most of the existing theoretical calculations are concerned with metal adsorption on ideal oxide surfaces (see, for example, refs. 12 and 13), and only very few treat metal adsorption on defective oxide surfaces (for a review see ref. 14). In particular, Ferrari and Pacchioni15 performed cluster HartreeÈFock calculations of MgO surfaces with point defects (neutral and charged anion and cation vacancies), and studied the charge transfer between the Rb atoms and these defective surfaces. While the neutral point defects are not very reactive, the charged anion vacancies can ionize metal atoms provided the electron affinity of the defect is larger than the ionisation potential of the metal atom at the surface. Thus, the interaction of metal atoms with surface point defects can greatly alter their surface di usion behaviour and, consequently, can be responsible for cluster nucleation in the neighbourhood of defect sites. In this study we combined several experimental techniques with theoretical modelling in order to elucidate the early stages of Mg cluster formation on the (001) surface of MgO thin Ðlms. Metastable impact electron spectroscopy (MIES) using He* (1s2s) projectiles is particularly useful for these purposes because it probes the very top surface layer and is sensitive to quite small concentrations of adsorbed species. It is naturally combined with ultraviolet photoelectron spectroscopy (UPS) using HeI as the light source. Unlike UPS (HeI), MIES is extremely sensitive to features resulting from the charge density of s-electrons speciðc for adsorption of alkali and alkaline earth atoms. Correlation of the spectroscopic data with the results of the temperature programmed desorption (TPD) experiments is illuminating for the interpretation of the spectral features and understanding of the initial stages of the metal cluster growth. To understand better the experimental data and to construct a model of metal adsorption, we performed ab initio electronic structure calculations using the density functional theory (DFT), and the method of pseudopotentials. The embedded cluster model16,17 was employed for DFT calculations of the surface ionisation energies, which are compared with the position of the top of the valence band with respect to the vacuum level of the MgO Ðlm, determined from the MIES data. Using the periodic DFT calculations we treated the adsorption of up to Ðve Mg atoms on the perfect (001) surface, near an anion vacancy, and at a neutral surface F-center. In order to elucidate the role of extended surface defects in the Mg cluster growth we modelled the adsorption of up to four Mg atoms at a step edge. To facilitate comparison with the experiment, in all these cases we analysed the energetics and the geometric and electronic structures of the metal clusters adsorbed on the surface and near defects, and calculated the surface density of states (SDOS). What have we learned regarding the adsorption of metal on MgO thin Ðlms from this complex study? Both the experimental MIES spectra and the results of our calculations give similar energies for the position of the top of the valence band (about 6.5 ev) of the MgO Ðlm with respect to the vacuum level. This is an important parameter for adsorption, photochemistry and interface studies on MgO Ðlms. In the context of the present paper, one can compare this value with the 3s electron ionisation energy of the Mg atom (7.6 ev), and use a simple argument in order to predict the nature of the chemical bonding of Mg atoms on the MgO surface. This qualitative prediction 174 Faraday Discuss., 1999, 114, 173È194

3 was then conðrmed using the results of our DFT calculations, which demonstrate the formation of bonding and anti-bonding states due to the interaction of adsorbed Mg atoms with the surface oxygen ions. The electronic states due to Mg adsorption manifest themselves in the MIES spectra, which can be approximately interpreted using the calculated SDOS. The attachment energies of Mg atoms to Mg clusters formed at di erent surface sites are compared with the TPD data. The best agreement is achieved with the energies at the step edge. These are the largest adsorption energies we found. Combined with the MIES data, this suggests that the initial nucleation of Mg clusters happens at step edges. The paper is organised as follows. In Section 2 we give a brief account of the experimental techniques used in this study and present the experimental results. The theoretical methods and the results of calculations are described in Section 3. The discussion of the experimental and theoretical results and conclusions are presented in Section 4. A preliminary account of some of our results is presented in ref Experimental results 2.1 Experimental techniques The apparatus used in these studies has been described previously.19,20 BrieÑy, it is equipped with a cold-cathode gas discharge source, which also serves: (i) for the production of metastable He (3S/1S) (E* \ 19.8/20.6 ev excitation energy) with thermal kinetic energy required for MIES, and (ii) as a source for ultraviolet photoelectron spectroscopy UPS(HeI), E* \ 21.2 ev. The intensity ratio 3S/1S is found to be 7 : 1. Additionally, the apparatus is equipped with X-ray photoelectron spectroscopy (XPS), Auger electron spectroscopy (AES) and low energy electron di raction (LEED). Metastable and photon contributions within the beam were separated by means of a time-of-ñight method using a mechanical chopper. The MIES and UPS spectra were acquired with incident photon/metastable beams 45 with respect to the surface normal. The kinetic energy of the electrons emitted in the direction normal to the surface is measured by employing a hemispherical analyzer (Leybold EA10/100) with an energy resolution of 250 mev for MIES/UPS. Collection of each MIES/UPS spectrum requires approximately 140 s. A second apparatus, described elsewhere,21 is equipped with a MIES/UPS source of the same type as described above, and a setup for TPD was used to calibrate the Mg coverages. The TPD spectra were collected with a di erentially pumped quadrupole mass Ðlter in line-of-sight to the sample while ramping the sample temperature linearly by 3Ks~1. In addition, this second apparatus is equipped with XPS, AES and LEED. The qualitative interpretation of the results of the MIES experiments is based essentially on a model of refs. 22 and 23, which is shown schematically in Fig. 1. Excited He* atoms approach the surface with thermal velocities. At the distances of about 2.5È4.0 Ó, when there is a considerable overlap between the surface, t (n is a band, k the wavevector), and the He t wavefunctions, nk 1s surface electrons tunnel into the 1s He hole states. This may happen from all surface energy levels that are higher than the He 1s level. In the case of MgO, all electronic states of the O 2p VB participate (see Fig. 1). The energy gained in this so-called Auger de-excitation (AD) process is transferred to the electron occupying the He 2s level which is emitted in the same process. The kinetic energies, E, of the emitted electrons are measured in MIES experiments and their dis- kin tributions constitute MIES spectra. Conventionally, though, the electron spectra are presented vs. the binding energy scale which refers to the Fermi energy, E, of metallic substrate, as shown in F Fig. 1. Experimentally, the Fermi energy, E, is a Ðxed point on the energy scale, and corresponds F to the maximum kinetic energy at which electrons can be measured with MIES and UPS from a metallic substrate. Since substrate and analyser are in electrical contact, E appears at the same F kinetic energy, irrespective from the substrate work function, i.e., for all Mg exposures. Thus, presenting the spectra with a binding energy scale, with E as origin, allows the change of the F work function (due to, for example, adsorption or charging) to be determined from the shift of the high-energy cuto of the spectra. In addition, the absolute value of the work function can be determined from the energetic distance between this cuto and the point on the energy scale that equals the excitation energy (19.82 ev) of the probe atom. The maximum binding energy, with respect to the vacuum level, E, probed by He* equals its excitation energy minus the binding vac energy of the He* 2s electron. Faraday Discuss., 1999, 114, 173È

4 Fig. 1 Energy diagram for a He* probe atom in front of a surface of insulator. Left side: energy levels of the isolated He and Mg atoms and surface density of states in the valence band (VB). Also shown is the position of the Fermi level, E, in the insulator band gap; U is the work function of the surface. Middle: binding energies, F E \ E [ E, of electrons involved in the Auger de-excitation process are usually presented with respect to bin F this axis, which has its origin at E. Right side: schematic of the experimental spectrum of kinetic energies of F the electrons emitted in the AD process (E ). Zero kinetic energy corresponds to a binding energy of ev kin with respect to the vacuum level (or (19.82 [ U) ev with respect to E ). F 2.2 Electron spectroscopy MgO layers with an approximate thickness of 2 nm were prepared by evaporation of Mg on Mo(100) and W(110) substrates at room temperature, followed by a subsequent annealing at 800 K in oxygen ambient. The MIES and UPS spectra measured on as-prepared MgO Ðlms are very similar to those obtained on MgO single crystals.24 Fig. 2 shows the MIES and UPS spectra collected during the exposure of the MgO Ðlm grown on the Mo(001) surface to Mg atoms at 100 K; no signiðcant changes are found at 300 K. As discussed previously,24 the MIES spectrum acquired from the MgO Ðlm, prior to the Mg exposure (bottom spectrum), reñects the MgO SDOS as seen via an AD process. Thus, the spectral feature with binding energies between 4 and 10 ev with respect to E is due to the ionisation of the MgO valence band states with O 2p F character. The large peak in the MIES spectra located between 10 and 17 ev is, to some extent, a ected by secondary and scattered electrons and will not be considered in the following discussion. One important characteristic of the electronic structure of our MgO Ðlm is the position of the top of the valence band with respect to the vacuum level. It depends on the surface preparation and is difficult to determine using conventional methods (see, for example, the discussion in ref. 25). It can be determined using the MIES spectra and the following considerations. The distance (as seen in Fig. 2) between E and the top of the valence band is about 4 ev. The distance between F E and the vacuum level, i.e., the work function U, is determined from the high energy cuto of the F spectra. The work function measured for the MgO Ðlms grown on the Mo(100) and W(110) substrates is equal to 2.7 ev. Therefore, for the MgO Ðlms used in our study we Ðnd the top of the valence band at 6.7 ^ 0.4 ev with respect to the vacuum level. As a consequence of the Mg dosing, an additional peak, located at ^2 ev above the top of the valence band (2 ev binding energy), develops within the band gap. Since the work function of the MgO Ðlm (2.7 ev) is considerably smaller than the He 2s binding energy, no unoccupied states are available at the surface into which resonant transfer of the 2s electron can take place, even when taking into account an eventual shift of the 2s level during the He* interaction with the surface. Consequently, the band gap feature arises from the AD process, which involves electrons from the 176 Faraday Discuss., 1999, 114, 173È194

5 Fig. 2 MIES and UPS spectra acquired from the MgO surface as a function of the Mg exposure. In MIES, the bottom spectrum shows the clean MgO surface; the topmost, the fully covered one; UPS vice versa. Inset (top panel): work function change vs. exposure time. The dashed spectra were acquired near the work function minimum. occupied states below the Fermi level due to adsorbed Mg atoms. At larger Mg exposures the MgO valence band emission between 4 and 10 ev weakens considerably. The disappearance of the O 2p structure in the topmost MIES spectrum in Fig. 2 indicates that the entire surface is covered by Mg; the shape of this spectrum is very similar to that for Mg Ðlms. Also shown in Fig. 2 (see inset) is the work function change during the Mg exposure. When Mg atoms are dosed to the oxide, the work function decreases by ^0.5 ev while the valence band intensity decreases by ^10% (dotted MIES spectra). Simultaneously, the top of the valence band shifts towards larger binding energies by approximately the same amount. This coincidental shift of the valence band structures and the high binding energy cuto indicates a band bending e ect rather than a real change of the work function. Such a band bending can be attributed to the creation of additional states on the surface.5 At larger exposures the work function plateaus at a Faraday Discuss., 1999, 114, 173È

6 level typical for metallic Mg Ðlms (3.6 ev), which is consistent with a model in which Mg islands grow with lateral bonding similar to bulk Mg. In order to gain more detailed information about the changes in the low binding energy region during Mg dosage to the MgO surface, MIES spectra were also acquired using a lower Mg evaporation and a higher energy resolution. Fig. 3 summarizes the data obtained at the low evaporation rate. As shown in Fig. 3, both the energy position and the peak width depend (weakly) on the exposure time. An observed feature has a Gaussian shape with 1.8 ev FWHM over the entire studied exposure range. Since this feature appears well separated both from the valence band maximum and E, the species responsible for this structure exhibits nonmetallic F behavior. The band gap feature can be detected until its intensity falls below a level of 10~3 of that of the valence band O 2p emission of the clean MgO surface. Additional measurements show that the band gap feature is stable up to 500 K and has virtually disappeared upon heating to 600 K. In the valence band region, the UPS measurements (see Fig. 2) provide similar information to MIES, i.e., an attenuation of the MgO substrate intensity at increasing Mg coverages, and a shift of the O 2p structure, which follows, essentially, the work function change of the substrate. On the other hand, due to the fact that UPS probes the average character of several top layers, a signiðcant contribution from the MgO substrate is still noticeable at maximum Mg coverages. In addition, the valence band of the clean MgO surface reveals a two peak structure between 4 and 10 ev, which is discussed in detail ref. 24. However, the most obvious di erence is the absence of the Mg-induced band gap feature in UPS; only a small intensity increase in the binding energy range up to 4 ev is observed at high Mg coverages. This can be attributed to the fact that, unlike MIES, UPS probes not just the surface layer but the rather deeper layers, and in addition is very insensitive to metallic s states.26 A similar band gap feature was observed in the MIES spectra after dosing with Na atoms; however, it is less stable thermally and disappears between 350 and 400 K upon annealing. Li and Cs additives on the MgO surface also demonstrate similar band gap features Temperature programmed desorption TPD is an excellent technique for determining surface coverage and studying the interaction between adsorbates themselves and with the surface. Fig. 4 displays TPD and MIES spectra acquired from the Mg-covered MgO surface at the same Mg exposure. Starting at the uppermost MIES spectrum acquired for the clean MgO surface, the Mg exposure increases monotonically towards the bottom spectrum. In the MIES spectrum for the highest Mg coverage (bottom spectrum), the O 2p band has essentially disappeared, indicating complete coverage of the MgO surface by Mg. The relative peak area of the corresponding TPD Mg feature (m/z \ 24) is 264 times larger than the Mg feature in the uppermost TPD spectrum, which corresponds to the Fig. 3 Intensity, energetic position width (FWHM) of the band gap feature vs. time. 178 Faraday Discuss., 1999, 114, 173È194

7 Fig. 4 Comparison between TPD and MIES data. The TPD spectra were taken from the surface characterized with MIES. The topmost MIES spectrum shows the clean MgO surface; the Mg coverage increases monotonically towards the bottom spectrum. detection threshold of our mass Ðlter. However, even at this very low coverage, the Mg-induced band gap feature appears fully developed in MIES. Analysis of the Mg TPD peaks reveals an exponential increase in intensity towards higher temperature with increasing coverage (leading edge behaviour), indicating that the desorption follows zeroth-order kinetics. Since this behaviour is typically observed for desorption with scission of adsorbateèadsorbate bonds, this result suggests that even at the lowest Mg concentrations accessible to TPD, the formation of 3D islands occurs. Layer-by-layer growth would lead to a Fig. 5 Arrhenius plot of the desorption rates obtained in the complete analysis28 of the TPD data in Fig. 4. Inset: calculated desorption activation energy vs. relative Mg coverage. Faraday Discuss., 1999, 114, 173È

8 two-peak structure in TPD (as is actually observed for Na/MgO at 100 to 150 K27). The formation of 3D islands is also supported by a non-linear Mg uptake vs. exposure time. The TPD data have been analyzed using the so called complete analysis28 in which an Arrhenius plot of the natural logarithm of the desorption rates yields a straight line for a particular Mg coverage (see Fig. 5). The slope of each line corresponds to the activation desorption energy, E des, for that particular coverage. The inset in Fig. 5 shows that the desorption energy increases initially from about 1.0 ev at very low coverages up to 1.4 ev. This latter value is close to 1.45 ev, the heat of sublimation for bulk Mg. Even at the lowest exposure accessible for TPD, the band gap feature in MIES is fully developed, and the work function has traversed through its minimum. The band gap feature, as a function of exposure, smoothly transforms into the spectrum characteristics for the full metallic coverage. Our TPD results suggest that this feature is due to Mg clusters which, for sufficiently large exposures, acquire metallic properties. Because the same kind of feature is already present in MIES spectra at much lower exposures, it is reasonable to assume that at these low exposures MIES also detects the presence of clusters at the surface. Thus comparison of Figs. 4 and 5 also supports the formation of Mg clusters with MgÈMg bond strength similar to that in the Mg bulk metal. 3 Theoretical results 3.1 Theoretical models Before we go into a detailed description of the theoretical methods and the results of calculations, let us brieñy summarise the experimental data. Essentially we are presented with two types of data. The TPD results tell us that the adsorbed Mg clusters have quite large desorption energies of individual Mg atoms, which increase with the coverage. The MIES spectra demonstrate a feature that evolves with the Mg concentration, transforming at large Mg exposures into the spectrum characteristic of full metallic coverage. To construct an atomistic model of adsorption of Mg atoms and further growth of metallic clusters on the MgO Ðlm surface, we can also use the STM images29 of MgO Ðlms grown on Mo(001). They demonstrate that these surfaces are very rough and contain 3D MgO islands and a lot of steps within the islands. Such surfaces would normally also have a number of anion and cation vacancies. The anion vacancies can be Ðlled by electrons from the metal substrate forming charged or neutral F-centres. The latter can be ionised directly during the AD process or by trapping electron holes; in addition, holes can localise near cation vacancies (forming the well known stable V-centres). In this study, we assumed that the Mg atom adsorption takes place on terraces, near charged anion vacancies, neutral F-centres, and at step edges. Comparing the calculated Mg atom adsorption energies with the TPD data we can approximately deduce which sites are most favoured. The spectroscopic MIES and UPS data, though, require a much more complex analysis. The MIES spectra of the MgO Ðlms before Mg dosing contain information about their electronic structure. In particular, the largest kinetic energy of emitted electrons corresponds to the AD process, which involves electrons from the highest occupied states localised at the top surface layer, which we call for simplicity the top of the valence bandïï. The position of the top of the valence band with respect to the vacuum determined from the MIES data is 6.7 ^ 0.4 ev. We can calculate the lowest ionisation energy of the surface terrace and compare it with these data. Another prediction that one can deduce from the experimental results is that the top of the valence band and the 3s states of Mg atom have close energies. This suggests formation of partially covalent bonding of Mg atoms to the surface, which can also be checked theoretically. More detailed analysis of MIES spectra is less straightforward.30 In the AD process, which is the only one that we took into account in the present paper, a surface electron is transferred into the 1s hole state of the He* 1s2s atom and the excited He 2s electron is ejected. Since only one surface electron is involved in the process, it is often assumed that the AD MIES spectra to a good extent reñect the SDOS.30,31 More accurate static30,32 and dynamic33h35 theoretical models suggest that, like in the Terso and Hamann model of STM,36 the density of states projected on the 1s function of the He* atom and integrated over the incoming trajectory of that atom would better represent the probability of the electron tunneling from the surface to the He*. Our 180 Faraday Discuss., 1999, 114, 173È194

9 calculations31,37 have demonstrated that the shape of the experimental MIES spectrum of MgO is di erent from the bulk and the surface DOS, and is indeed similar to the SDOS projected on the 1s He* function. However, the relative energies of di erent features in the three DOS and in the experimental spectra remain very similar. Therefore, we believe that the MIES features due to localised band gap states of adsorbed atoms and their relative position with respect to the top of the valence band are likely to be well reproduced by SDOS. 3.2 Theoretical methods Density functional theory38 is widely used in surface studies (see, for example, ref. 39). Two models within the DFT method have been used in our work. Due to the importance of the electron correlation for the hole states, we employed the DFT to calculate the surface ionisation energies within an embedded cluster model.16,17 To model the Mg adsorption and MIES spectra, a periodic model and the realisation of the DFT in the VASP code40h43 are more appropriate. In the embedded cluster calculations, a cluster of up to 50 atoms (quantum cluster) was treated quantum mechanically using DFT. It was embedded into a Ðnite cluster (region I) of 12 ] 12 ] 6 ions treated in a polarisable ion model. Pair potentials44 were used to calculate the interactions between these ions, and the shell model45 to treat the polarisable oxygen ions. The quantum cluster and region I were embedded into an outer region of frozen ions, which makes the total number of ions in the system 20 ] 20 ] 8. The e ective charges on all classical ions were ^2e (e is the electron charge). This setup provides the correct values for the Madelung potential and its gradients on the ions in region I and in the quantum cluster. As an example, in Fig. 6 we present the largest quantum cluster Mg O The matrix elements of the electrostatic potential of the rest of the system, including the dipole contributions from the polarized oxygen ions in region I, are included in the KohnÈSham equations implemented in the modiðed Gaussian94 code.46 The B3LYP functional47 was employed to calculate the electronic structure of quantum clusters. All electrons of oxygen ions and those of the magnesium ions (shown as black in Fig. 6) were described using the 6-31G standard Gaussian basis set.48 To facilitate calculations of large quantum clusters, the magnesium ions (shown as gray in Fig. 6) were treated using the pseudopotentials of Wadt and Hay49 and the 1s function described by two contracted Gaussians. In smaller clusters, Mg atoms that had less than two nearest quantum oxygens were treated in the similar way. Periodic DFT calculations were performed in the slab geometry where an inðnite stack of slabs separated by vacuum gap (to suppress the mutual interaction between parallel slabs) is considered in the z-direction. Within each slab, the unit cell is periodically repeated in two dimensions. The Fig. 6 The top view on the Mg O quantum cluster used in the embedded cluster calculations of surface ionisation energy. Open circles represent oxygen ions, black and gray circles are magnesium ions with di erent basis sets (see text). Faraday Discuss., 1999, 114, 173È

10 KohnÈSham electronic orbitals t are expanded into plane waves and the plane wave coefficients nk are varied to obtain the minimum of the energy for the given ionic geometry. A number of unoccupied states are also included in the variational procedure as it speeds up the calculations and at the same time allows us to consider possible metallisation in the system. The whole system is brought to a mechanical equilibrium by minimising the forces acting on every atom in the unit cell until they become less than 0.1 ev Ó~1. Only valence electrons are treated explicitly, which is achieved by using non-local softïï Vanderbilt pseudopotentials.50,51 The advantage of using these pseudopotentials instead of the norm-conserving ones38 is that it is possible to have a relatively small cuto, E \ 400 ev, (especially for oxygen), which speeds up the calculations by at least a cut factor of four. The generalised gradient approximation (GGA)52 was used in all calculations, which is especially important for surfaces.39 The cell sizes, vacuum widths and system geometries for all systems studied in this paper are discussed in detail below. In all calculations the interionic distance of d \ Ó was used to 0 specify the surface unit cells. This was found in ref. 53 to be the equilibrium distance for the bulk MgO using the same GGA functional as in the present study. The calculations were performed in the following way. First, the singlet ground state of the reference system (perfect surface, step, etc.) without adsorbed metal atoms was considered. After, it was relaxed to mechanical equilibrium and Mg atoms were introduced into the calculation. The whole system was relaxed again, except for the atoms in the bottom layer of the slab, which were Ðxed in the positions corresponding to the perfect system to simulate the crystal bulk. From two to four k-points in the plane (2D) Brillouin zone have been normally used in all such ground-state calculations. This has been shown54,55 to be sufficient for the cell sizes considered here. The ground state calculations give adsorption energies, relaxed geometries and the electronic densities. The latter were analysed using the general visualisation tool LEV00,56 which facilitates construction and analysis of arbitrary 3D objects speciðed on a grid, such as electronic densities and wavefunctions. In addition to the usual contour maps of, for example, electron density, we have also found it very informative to integrate the charge density into spheres of di erent radii around various positions within the simulating cell, and to compare these results with those obtained for other positions of the same or other similar system (cf. ref. 54). The densities of states were calculated using a method of tetrahedra outlined in ref. 54. BrieÑy, using the point-group symmetry of the cell, a necessary mesh of k-points was generated for the plane Brillouin zone and the wavefunctions and energies of the surface electrons were recalculated again for all non-equivalent k-points using the VASP code. In most cases, our systems have no symmetry at all and a mesh of 13 k-points corresponding to 750 tetrahedra in the plane Brillouin zone was used. Then, the SDOS was calculated using LEV00. It is smeared by a Gaussian to simulate the e ect of phonon broadening57 at room temperature using a smearing parameter equal to 0.3 ev. All periodic calculations were performed on the T3E parallel supercomputers in the Edinburgh Parallel Computer Center and at the University of Manchester under the Computer Services for Academic Research (CSAR) initiative. 3.3 Calculation of the surface ionisation energy To calculate the surface ionisation energy, we used several quantum clusters of increasing size, as listed in Table 1. Four of them had quantum oxygens only in the surface plane, and in the case of Mg O four additional oxygen ions were added to the Mg O cluster in the second plane to check whether this will have any signiðcant e ect. After calculation of the perfect lattice, each cluster was ionised and the di erence in the total energy with the ionized state was calculated in two approximations: with, IP(I), and without, IP(0), a self-consistent account of the electronic part of the polarisation in region I, which corresponds to the verticalïï ionisation potential. Only the oxygen polarisation was included and treated classically in the shell model. As the hole delocalisation increases, one would expect the polarisation energy, *IP, to decrease. Except for the smallest cluster, which contains only one oxygen ion, in all the ionised clusters the hole was delocalised by all quantum oxygen ions. Increasing the cluster size we should approach the limit of a completely delocalised band hole state. For the smaller clusters, the hole was distributed almost evenly by all the oxygen ions. However, in the case of the Mg O this distribution was more complex, which reñects the fact that as the hole state becomes more delo- 182 Faraday Discuss., 1999, 114, 173È194

11 Table 1 The ionisation energies calculated using di erent quantum clusters, in ev Cluster IP(0)a IP(I)b *IPc W d Mg O Mg O Mg O Mg O Mg O a In the calculation of IP(0) the lattice polarisation outside the quantum cluster was not included. b IP(I) are calculated taking into account the classical lattice polarisation. c *IP is the di erence between the two which reñects the hole localisation and, consequently, the lattice polarisation energy. d W is the valence band width. calised its eigenvalue approaches the top of the valence band. As a result, several quasi-degenerate hole distributions become possible, which also hampers the convergency of calculations. As one can see in Table 1, this, however, does not signiðcantly a ect the calculated energies. The ionisation energies in both approximations Ðrst decrease sharply as the cluster size increases and then change slowly. The di erence in ionisation energies between the completely localised hole state in the smallest quantum cluster and the most delocalised state in the largest cluster should approximately correspond to half of the valence band width, W, which is also presented in Table 1. This approximately holds in our calculations. 3.4 Adsorption of Mg atoms on the perfect MgO (001) surface and near a surface F-centre A supercell consisting of three layers of oxygen and Mg atoms (eight surface unit cells in every layer) with the vacuum width between slabs equivalent to three additional layers was used to simulate the perfect MgO(100) surface. Up to four Mg atoms were added to this supercell to model the Mg adsorption. A similar setup was used to model the surface F-centre. The latter was created by removal of one oxygen atom from the topmost surface layer in the slab. The whole system was relaxed and its energy was used as the reference energy in the calculations of the corresponding adsorption energies. A detailed account of the surface F-centre calculations with the same method is given in ref. 54. In this paper we focus on the results related to Mg adsorption. The adsorption energies of one Mg atom on the perfect surface and near the surface F-centre are shown in Table 2. On the regular surface, the most stable position of the Mg atom is above the surface oxygen. The on top F-centre position is much less stable and, although there is a shallow energy minimum corresponding to the adsorption energy shown in Table 2, most of the Mg atoms would probably prefer the nearest oxygen sites to F centres. On top of the F-centre the Mg atom is pulling up part of the electron density from the vacancy. Above the oxygen site the adsorbed Mg atom is about 2.3 Ó above the surface plane, with the oxygen ion displaced by about 0.25 Ó towards it. No signiðcant relaxation of other surface ions was found. The calculated barrier for Table 2 Total adsorption energies for all systems studied, in ev Surface 1 Mg 2 Mg 3 Mg 4 Mg 5 Mg Perfect surface With F-center With step With anion vacancy Faraday Discuss., 1999, 114, 173È

12 di usion of an Mg atom along the perfect surface is only 0.26 ev, with the barrier point at about 3.2 Ó above the centre of the surface unit cell. The equilibrium distance between Mg atoms in a free Mg molecule, obtained in our calcu- 2 lations, is 3.75 Ó, which is much larger than the distance between two nearest oxygen ions and smaller than that between next-nearest ions. Therefore, when more than one Mg atom is added to the system, their lateral interaction does not allow them to occupy the most energetically favourable positions above the oxygen ions. As a result, we Ðnd that the potential energy surface in the lateral direction above the surface is very Ñat with many local minima. Typical geometries for adsorption of four Mg atoms on the perfect surface and near the F-centre are shown in Fig. 7(a), (b). As one can see, the geometries obtained are similar in both cases. Every Mg atom occupies a surface area containing one surface oxygen. It is positioned above the surface plane in the range 2.2È3.2 Ó, depending on the particular arrangement of the Mg atoms and on whether the nearest surface oxygens are displaced signiðcantly towards them from the surface plane. The adsorption energies found for the two systems are summarized in Table 2. When two Mg atoms are added to the terrace, the adsorption energy per adsorbed Mg atom does not increase more than twice, but is actually a little smaller than the double adsorption energy for one Mg due to the repulsion of Mg atoms. Then it grows slowly reaching 0.7 ev Fig. 7 Typical geometries for the adsorption of four Mg atoms on the terrace (a) and near the F-centre on the Ñat surface (b); for one (c) and two (d) Mg atoms at the anion vacancy. Oxygens are shown as open balls, surface Mg atoms as shaded smaller balls and adsorbed Mg atoms are shown as black circles. 184 Faraday Discuss., 1999, 114, 173È194

13 atom~1 when four atoms are added (in this comparison we are using an averaged parameter that corresponds to the dissociation of adsorbed cluster into free atoms). A similar tendency is also observed for the surface with the F-centre. This is due to the creation of mutual bonding between the adsorbed Mg atoms and between them and the surface oxygens. In order to demonstrate the character of this bonding, we show in Fig. 8 the contour plot of the valence electronic density of the system with one Mg atom adsorbed on the terrace above a surface oxygen. The strong contribution of the electron density from the surrounding oxygen ions into that of the adsorbed Mg atom is clearly visible. To analyse the bonding further, one can integrate the charge density around the adatoms. For one adsorbed Mg atom on the perfect surface, this does not indicate any signiðcant charge transfer to the surface. However, a detailed analysis of the occupied orbitals in this system reveals that the adatom participates in the states at the top of the O 2p valence band (VB) whereas there is a signiðcant contribution of the surface oxygens nearest to the adatom in the charge density localised on the Mg atom. This picture can be further clariðed by examining the calculated DOS for this system shown in Fig. 9 (the lowest curve). The last occupied state, n, manifests itself in the DOS as a feature about 1.2 ev from the valence band maximum. The corresponding partial density, o (r) \ & o w o2, contains contribu- n k nk tions from both the adsorbed Mg atom and the surface oxygens underneath. While integrating o (r) in spheres of di erent radii, we Ðnd that it does not account for the total charge around the n adatom, so that part of the density comes from the valence band states. On another hand, there is a considerable localisation of this density on the nearest surface O atoms. When more than one Mg atom is adsorbed, the electron density is easily shared between them. In all cases we found a rather di use density around the adatoms with strong highly localised peaks on the nearest surface oxygens. In the DOS these mixed states manifest themselves as a set of peaks in the gap, which (after smearing) show up as a broad peak around 2 ev above the VB maximum. One can also notice a considerable distortion of the O 2p valence band due to the Mg adsorption, this is seen as a bump at about [1.5 ev. With three Mg atoms per simulation cell, the partial densities, o (r), associated with the features n in the gap are still well localised. Adsorption of four Mg atoms in our setup corresponds to half of the monolayer. In this case the last occupied state in the system is very di use and is spread over most of the simulation cell. This state is mainly due to all four adsorbed Mg atoms and the surface oxygens nearest to them. In the DOS we Ðnd that this state has a considerable width (of over 2 ev) and overlaps both with other adatom-related local states at lower energies and with the Fig. 8 The contour plot of the valence electronic density of a Mg atom adsorbed above a surface oxygen atom on the terrace (in units of 10~2 electron Ó~3). To guide the eye, the surface atoms are connected by a dashed line. The cut has been made along [010] axes perpendicular to the surface plane. To avoid high peaks on oxygens, the density has been chopped at 0.2 electron Ó~3. Distances are in Ó. Faraday Discuss., 1999, 114, 173È

14 Fig. 9 DOS (arbitrary units) for the perfect surface with up to four adsorbed Mg atoms. The DOS is aligned so that zero energy corresponds to the unsmeared top of the VB. unoccupied states. Although, strictly speaking, the unoccupied states in the DFT do not have clear physical meaning, we believe, that this behaviour is an indication of the beginning of the system metallisation. This is because with four atoms it is already possible to construct a conðguration of adsorbed metal atoms in which the distances between the neighboring Mg atoms in the central and adjacent cells are of the same order of magnitude. Then some of the occupied states become very di use in the direction of the short distance between the metal atoms, and the system as a whole becomes conductive. This e ect is similar to the one in the percolation theory of conductance in disordered systems. Although some features of the DOS for the system with the F-centre are di erent, it nevertheless retains the same character. In particular, in the DOS for one adsorbed Mg atom (see Fig. 10, the lowest curve) there are two peaks in the gap: at 0.9 and 2.6 ev above the VB maximum. They correspond to the bonding and antibonding states between the Mg atom and the F-center electrons. These states also contain a signiðcant portion of the density localised on the nearest surface oxygens. Every new atom added to the system generates an additional peak in the gap of the DOS. After smearing (see Fig. 10), all these peaks form a broad feature in the gap approximately 2 ev above the VB maximum. The states that make up the defect band are quite di use. They spread over all Mg atoms (some states have bonding, some an antibonding character with respect to the adatoms) and have signiðcant localisation in the anion vacancy and especially on the nearest surface oxygens. While integrating the partial density associated with the states that form the broad feature in the gap, we have not been able to account for all the density, which means that the electrons from the VB themselves have signiðcant localisation on the adsorbed atoms. The VB states therefore are strongly perturbed. This manifests itself in the distortion of the O 2p band, seen in Figs. 9 and 10. Similar to the perfect surface, we found that metallisation starts to form when four or Ðve Mg atoms are added to the supercell with the surface F-centre: the defect states in the gap interact more strongly and are spread over larger energy intervals so that they overlap with unoccupied states. Thus, we conclude that upon adsorption of Mg atoms, chemical bonding is formed between the Mg atoms, and that at a coverage of roughly half a monolayer the adsorbed layer may become conductive. 186 Faraday Discuss., 1999, 114, 173È194

15 Fig. 10 DOS (arbitrary units) for the perfect surface containing one neutral F-center per simulation cell with up to Ðve adsorbed Mg atoms. The DOS is aligned so that zero energy corresponds to the unsmeared top of the VB. 3.5 Adsorption of Mg atoms near the anion vacancy Adsorption of Mg atoms near the F-center is similar to that on the perfect surface partly because the F-center bears roughly the same charge (two electrons) as the lattice oxygen O2~ ion. A doubly positively charged anion vacancy, V, may interact di erently with adsorbed metal atoms. a However, modelling of charged systems in the periodic model is less straightforward. Formal procedures (see, for example, refs. 58 and 59), which one can use to study charged systems in periodic boundary conditions, are not strictly applicable to surfaces and to slab geometries. In the bulk one can remove the Coulomb interaction energy between the extra charge across the simulation cells by dividing it by the relative permittivity, e. It is not that clear, however, how to model 0 the electronic polarisation in the surface case. Therefore, in this paper we adopted a di erent procedure. It is well known that in real systems charged defects tend to be compensated by other defects having the opposite charge. To compensate the anion vacancy at the slab surface we formed a cation vacancy, V, at the slab centre. This c makes the whole system neutral at the expense of introducing a dipole moment in the cell. The dipole moment is going to be large both along the surface and normal to the surface as we want to separate the two vacancies from each other as much as possible. To check the e ect of the dipole moment on the energetics, geometry and the DOS, we have run extensive tests. They have demonstrated that the e ect on the energetics and geometries of adsorption is insigniðcant, and leads only to a shift as a whole of the calculated DOS. Therefore, we used this setup and the 80-atom supercell modelling a Ðve layer slab in further calculations of the Mg adsorption. The cation vacancy was created in the middle layer of the slab, and the oxygen ion was removed in the top layer from the lattice site, which is most separated from the cation vacancy two layers underneath. The distance between the two vacancies is 3d \ 6.4 Ó in all calculations. After the geometry 0 relaxation, this system was treated as the reference for further calculations of the Mg atomïs adsorption. No signiðcant electron density is localised in the anion vacancy. One Mg atom is adsorbed between the vacancy and the nearest surface oxygen as shown in Fig. 7(c). The adatom is located 2.4 Ó above the surface plane and there is a considerable upward displacement of the nearest oxygen atom. The analysis of the electronic density revealed that two Faraday Discuss., 1999, 114, 173È

16 electrons of the Mg atom are strongly pulled towards the vacancy and form a di use electronic cloud localised in the area containing both the Mg atom and the anion vacancy. Because of this direct charge transfer from the adatom to the surface, the adsorption energy is more than 0.5 ev (see Table 2), which is much greater than in the case of adsorption on the F-centre. In the DOS for this system, shown in Fig. 11 (the lowest curve), one can notice a feature just above the VB top, which is due to this single di use state. Because the wavefunction associated with this state penetrates more into the surface than in the other two cases studied above, the extra peak in the gap was found only 0.35 ev above the VB maximum and after smearing appears as a shoulder in Fig. 11. When one more Mg atom is added to the system, at least two conðgurations are possible. If the Mg atom Ðnds a surface oxygen within the proximity of the anion vacancy, it shares its electrons with the anion vacancy and the Mg atom already adsorbed nearby. The total adsorption energy (Table 2) is more than doubled. This conðguration is depicted in Fig. 7(d). Another possibility is that the second Mg atom adsorbs further away from the vacancy. In this case the adsorption will happen on a terrace above one or two surface oxygens and the adsorption energy increases only up to 0.81 ev, which is less than one would expect from the results for the perfect surface. The corresponding DOS for these two conðgurations shown in Fig. 11 look very similar after smearing, and one can notice the development of a feature about 1 ev above the VB maximum. Addition of more Mg atoms leads to the formation of mutual electronic states between them and with the surface, and to a substantial gain in adsorption energy, as seen in Table 2. In the DOS shown in Fig. 11 (two upper curves) one can clearly see the development of a defect band around 1 ev above the VB maximum. It is closer to the VB edge because the defect is positively charged. 3.6 Adsorption of Mg atoms near a monolayer step Finally, let us turn to the Mg adsorption at a monolayer step. The simulation cell used in these calculations contained 44 lattice sites (for 22 oxygens and 22 Mg atoms) arranged in three layers, as in ref. 54. Each layer, as can be seen in Fig. 12(a), goes like a d -high staircase containing a 0 3d -long and inðnitely wide terraces. One Mg atom is adsorbed just in front of the step facing two 0 Fig. 11 DOS (arbitrary units) for the surface containing one anion vacancy (compensated by a cation vacancy in the middle of the slab) per simulation cell with up to four adsorbed Mg atoms. The DOS is aligned so that zero energy corresponds to the unsmeared top of the VB. 188 Faraday Discuss., 1999, 114, 173È194

17 Fig. 12 Three-layer slab system used in the study of the Mg adsorption at the monolayer step (a), the geometry for one (b) and four (c) Mg atoms adsorbed at the step. Notations as in Fig. 7. surface oxygens, as shown in Fig. 12(b), with the substantial energy gain of 1.26 ev. This is the biggest adsorption energy we obtained for a single Mg atom on the MgO surface. The two oxygens nearest to the adatom slightly displace towards it (by about 0.03 d ). However, the dis- 0 placements of the nearest surface Mg ions (up to 0.05 d ), shown by arrows in Fig. 12(b) are more 0 Fig. 13 DOS (arbitrary units) for the monolayer step system with up to four adsorbed Mg atoms. The DOS is aligned so that zero energy corresponds to the unsmeared top of the VB. Faraday Discuss., 1999, 114, 173È

18 Fig. 14 The contour plot of the partial charge density, o (r), associated with the defect state in the gap for a n single Mg atom adsorbed at the step. The cut has been made perpendicular to the direction of the step through the adsorbed Mg atom and the two nearest surface oxygens. To guide the eye broken lines indicate the surface structure. Other notations are as in Fig. 8. substantial. This conðguration is very stable: the system total energy is, by about 1 ev, higher if the Mg atom is adsorbed on the terrace or just above the center of the unit cell on the step edge. The DOS shown in Fig. 13 (the lowest curve) demonstrates a single peak at about 2 ev above the top of the VB. It is made of the orbitals of the adatom and the two surface oxygens nearest to it. The partial density, o (r), associated with this state is rather di use on the adatom, but forms n sharp peaks on the surface oxygens as shown in Fig. 14. The step simulation cell contains two equivalent positions at the step edge (see Fig. 12(b)). It is therefore not surprising that the second Mg atom prefers to stick at this position as well. The adsorption energy presented in Table 2 increases over 1.5 ev atom~1. The band at 3 ev above the VB in the DOS (see the second curve from the bottom in Fig. 13) becomes broader. However, as the third and the fourth Mg atoms are adsorbed, no such positions are available in the simulation cell, and the additional Mg atoms are forced to occupy the terrace sites. The typical geometry for four adsorbed Mg atoms is shown in Fig. 12(c). Nevertheless, the adsorption energy increases substantially (see Table 2). One can also notice that a second defect band in the DOS around 2 ev above the VB maximum is developed due to Mg adsorption on the terrace sites. This band consists of several states equal to the number of adsorbed Mg atoms. Note that the states responsible for the defect features in the gap of the DOS are localised in the direction perpendicular to the step. However, due to a relatively small size of simulation cell along the direction of the step, the states in question are delocalised in this direction. Nevertheless, we have not found any signs of the metallisation at this coverage. 4 Discussion Let us start with a brief discussion of theoretical results. First, we note a qualitative agreement between our results for the Mg adsorption on the MgO surface terrace and those obtained by Musolino et al.12 for adsorption of Cu (n \ 1,..., 4) on MgO using a DFT based method. n Interestingly, some of the calculated geometries and adsorption energies for clusters, and also the di usion parameters for one Cu atom on the surface are close, even quantitatively, to our results for the Mg adsorption despite the di erence in the Cu and Mg electronic structures. The results,12 though, demonstrate a richer variety of adsorption conðgurations, which we did not fully explore in this study. A qualitative agreement also exists with the embedded cluster DFT calculations of M clusters (M \ Cu, Ag, Ni, Pd) on MgO by Matveev et al.13 Both the results of Matveev et al Faraday Discuss., 1999, 114, 173È194

19 and our calculations predict polarisation of adsorbed metal atoms (see Figs. 8 and 14). However, our analysis suggests a more covalent character for chemical bonding of Mg clusters with the 4 MgO surface than suggested for other metals in ref. 13. We start our comparison with the experimental data from the perfect surface. Although the results of calculations for the surface ionisation energies (see Table 1) are in good agreement with the position of the top of the valence band with respect to the vacuum experimentally determined from the MIES data (6.7 ^ 0.4 ev), this agreement is not conclusive. Our results do not show a fast convergency with the quantum cluster size. Although its further increase is not feasible, we believe that the values given in Table 1 are already representative. Thermal Ñuctuations and the surface roughness broaden signiðcantly the band edge, which is reñected in the UPS spectra (see Fig. 2) and in the experimental error in the determination of the position of the top of the valence band from the MIES spectra. This broadening masks the electronicïï band edge, which can be obtained as a limit in our calculations (see, for example, ref. 60). Similar to refs. 18, 30, 31, we made the assumption that the MIES spectra due to the AD process reñect to a good approximation the SDOS. This, in fact, does not hold for the states in the lower part of the valence band and those below the valence band. As one can see in Figs. 9È11, the calculated surface DOS has a pronounced maximum at these energies, which is almost completely absent in Fig. 2(a). This is because the wavefunctions of lower energy states decay faster into the vacuum than those of the higher energy states, as discussed in detail elsewhere.37 Nevertheless we believe that the position of the defect states in the band gap with respect to the top of the valence band can be reproduced more reliably. The comparison of the numerical results and the experimental MIES spectra thus suggests that the band gap feature is due to adsorbed Mg atoms and small Mg clusters. The SDOS calculated for the Mg adsorption on the terrace, near the F-center and at the charged anion vacancy predict a shoulder at about [1.5 ev (see Figs. 9È11). However, despite the fact that the UPS spectrum for the perfect surface is well reproduced by the SDOS, such a shoulder is not seen in the UPS spectra after the Mg exposure, shown in Fig. 2(b). They also do not demonstrate any visible band gap features. We attribute this to the fact that UPS probes mostly deeper surface layers, which are not a ected by the Mg adsorption. Another reason is that UPS is very insensitive to metallic s states.26 Our theoretical results suggest that individual Mg atoms adsorbed on terraces are fairly mobile at room temperature (the calculated adiabatic barrier for di usion is only 0.26 ev). The largest adsorption energies were obtained for individual Mg atom adsorption at the step edge. These results indicate that at very low coverages one can expect more Mg atoms to be adsorbed at step edges than at terraces. However, as the Mg exposure increases, the energy gained due to attachment of each atom to existing clusters on the terrace also increases (see Table 2). So, for instance, to desorb one atom from the four atom cluster on the terrace requires 1.15 ev and from the terrace close to the step edge 1.36 ev (see Fig. 12(c)). As one can see in Fig. 5, the desorption activation energies, as derived from the TPD measurements, increase with the Mg exposure from about 1.0 ev at relatively low coverages, to 1.4 ev for almost metallised surface. As discussed in Section 2.3, the TPD data suggest that these energies correspond to desorption of individual Mg atoms from Mg clusters and metallic layer. Although it is tempting to directly compare the theoretical results with the TPD data, this is impossible due to the large variety of both adsorption sites and cluster geometries. Since TPD measures the smallest desorption energies, comparison with these data without proper simulation of desorption kinetics can be only qualitative. It suggests that the calculated desorption energies of individual Mg atoms are certainly in the range of desorption energies determined from TPD. Based on these results, it seems plausible to assume that Ðrst the metal atoms occupy all available sites immediately in front of the step edges (decoration of MgO islands or clusters). They form chemical bonds with the nearest surface oxygens accompanied by the considerable lateral interaction between the adsorbed Mg atoms. This assumption is supported by comparison of the dependence of the calculated SDOS on the character of Mg adsorption with the MIES spectra. According to the calculations, when Mg atoms are adsorbed at step edges, this should result in the development of the defect band in the band gap, about 2.5È3 ev above the VB maximum (see Fig. 13). As all such sites are occupied by the adsorbed atoms, the nearby terrace sites also become gradually Ðlled. A band about 2 ev above the top of the VB should develop due to newly adsorbed Mg atoms (see Fig. 9). Since the number of terrace sites is much bigger than that of the Faraday Discuss., 1999, 114, 173È

20 edge sites, the peak at 3 ev above the VB should soon become less visible. This implies that the Mg related peak in the MIES spectra should shift to lower energies as the Mg concentration increases. Careful analysis of the experimental spectra demonstrates that this indeed is the case. It is interesting to note a new structure formed by Mg atoms adsorbed at the step edge (see Fig. 12(c)): the oxygen vacancy created by the three adsorbed atoms AÈCÈB and the step Mg ion. The electron density plot shown in Fig. 15 clearly shows a considerable localisation of the electron density in the pocketsï created by the Mg atoms and in the vacancyï. The band gap feature corresponding to these states is seen in Fig. 13. In principle, a band gap feature of similar shape and energetic position could also result from Auger de-excitation of point defects, F-centers in particular. However, no indication of this feature is seen at the clean surface (which certainly is not free of point defects). We also note that, because of the proximity of the metal substrate and also because the position of its Fermi level is several ev above the top of the MgO VB, one can expect that in our experiments all electronic traps, such as anion vacancies and hole centres, will be quickly Ðlled by electrons, tunnelling from the metal. The speed of this should depend, though, on the Ðlm thickness. This will not be the case for other types of experiments, e.g., on single crystals. Our results for Mg/MgO allow for the following, more general qualitative considerations. One of the key parameters responsible for the type and strength of bonding between metal species and oxide surface is the ionisation energy of a metal atom with respect to the valence and conduction bands of the oxide surface. Provided the valence level of the metal atom is in resonance with the occupied valence band states, the bonding will feature covalent and polarisation contributions. This holds in the present case of Mg/MgO and for other metals on MgO with an ionization potential of about 7 ev and larger. On the other hand, if the metal level is in resonance with unoccupied states of the conduction band, as is the case for alkali atoms adsorbed on TiO, a charge transfer from the alkali metal 2 atoms to the oxide can be expected. For the alkali/tio case this leads to the reduction of the Ti 2 cation (and the appearance of band gap states due to Ti3` 3d formation). Thus, there is ionic chemisorption in this and similar cases. MIES studies on such systems are currently in progress at TU Clausthal. In cases like Mg/TiO, both charge transfer from the metal atom to the cation and hybrid- 2 isation of the Mg 3s and O 2p states might occur simultaneously, resulting in mixed covalent and ionic bonding. At present we are attempting to verify this prediction in a joint MIES and UPS study on Mg/TiO. 2 Fig. 15 The contour plot of the valence charge density for the step system with four adsorbed Mg atoms (labelled) as shown in Fig. 12. The cut is made parallel to the terrace in such a way that all four adsorbed atoms are crossed (note that they are not exactly at the same height). Broken lines indicate positions of the atoms at the upper terrace. The adsorbed atoms are in front of this terrace. Other notations as in Fig Faraday Discuss., 1999, 114, 173È194