Surface core level shifts of clean and oxygen covered Ir(111)

We present the results of high resolution core level photoelectron spectroscopy employed to investigate the electronic structure of clean and oxygen covered Ir(111) surface. Ir 4f7/2 core level spectra are shown to be very sensitive to the local atomic environment. For the clean surface we detected two distinct components shifted by 550 meV, originated by surface and bulk atoms. The larger Gaussian width of the bulk component is explained as due to experimentally unresolved subsurface components. In order to determine the relevance of the phonon contribution we examined the thermal behaviour of the core level lineshape using the Hedin–Rosengren theory. From the phonon-induced spectral broadening we found the Debye temperature of bulk and surface atoms to be 298 and 181 K, respectively, which confirms the softening of the vibrational modes at the surface. Oxygen adsorption leads to the appearance of new surface core level components at −200 meV and +230 meV, which are interpreted as due to first-layer Ir atoms differently coordinated with oxygen. The coverage dependence of these components demonstrates that the oxygen saturation corresponds to 0.38 ML, in good agreement with recent density functional theory calculations.


3
with Ar + sputtering (T = 300 K, E = 3 keV) and annealing cycles (T = 1320 K) followed by oxygen treatments at 900-1200 K ( p = 1 × 10 −7 mbar). At the end of these cycles, hydrogen reduction was used to remove the residual oxygen. The whole procedure was repeated until the C1s and O1s regions of the core-level spectra did not show any trace of contaminants and a sharp (1 × 1) LEED pattern was observed. During the measurements the base pressure in the UHV experimental chamber was kept at 1 × 10 −10 mbar (residual gases consisting mainly of hydrogen and carbon monoxide). Nevertheless, in order to keep the surface contamination at a very low level (below 0.5% ML), the sample was annealed to 670 K at intervals of about 10 min. This results in desorption of the main contaminants, such as hydrogen, carbon monoxide and water. In fact, it is well known that even very small amounts of adsorbed impurities can modify the shape of the core-level spectra, as they generate new surface core level-shifted components at different BEs with respect to the clean surface.
The sample was cooled by liquid nitrogen and heated by electron bombardment from hot tungsten filaments mounted behind it. High energy resolution Ir 4f 7/2 core-level spectra were collected using photon energies from 140 to 200 eV with an overall energy resolution of about 40 meV, as determined by measuring the width of the Fermi level of a Ag polycrystal. In order to minimize the effect of phonon broadening, the sample was mounted on a liquid helium cryostat, which permits cooling down to 80 K when using liquid nitrogen. Temperature dependent core level spectra were measured only during cooling cycles in order to avoid intensity and lineshape modifications due to the heating current. The BE scale has been calibrated with respect to the Fermi level E F measured under the same experimental conditions. Different emission angles β were also used ranging from normal (β = 0 • , photon beam incidence of 70 • with respect to surface plane) to grazing emission (β = 80 • ).
We employed Fast-XPS [24], with energy resolution of 80 meV to monitor the Ir 4f 7/2 SCL shifted components in real-time during oxygen adsorption at 300 K. The oxygen doses, expressed in Langmuir (1L = 10 −6 Torr s), are corrected for the ion gauge sensitivity factor. During the uptake, molecular oxygen was dosed by background exposure at different pressures (from 1 × 10 −9 to 1 × 10 −7 mbar). The acquisition time of each Ir 4f 7/2 spectrum was of the order of 6 s.
All the photoemission spectra presented in this paper have been fitted using Doniach-Sunjić (DS) [25] functions, characterized by the Lorentzian width and the Anderson singularity asymmetry parameter α that takes into account the electron-hole pair excitations and/or shake-up states, convoluted with a Gaussian (full-width at half-maximum G) which accounts for the instrumental resolution and phonon broadening/vibrational fine structure. In the fittings, performed using procedures written within the IGOR software, the background was assumed to be linear.
The Ir 4f 7/2 core level spectrum corresponding to the clean surface, measured at hν = 160 eV and in normal emission conditions, is shown in figure 1(a). Two distinct components can be observed at 60.87 and 60.32 eV, which are assigned to photoemission from bulk and surface atoms, respectively.
Our analysis reveals that the best fit, obtained using two DS functions with different ( bulk = 200 meV and surface = 320 meV) and α (α bulk = 0.15 and α surface = 0.22), convoluted with Gaussians of different width (G bulk = 170 meV and G surface = 90 meV), is satisfactory, as indicated by the small modulation present in the residual. The deconvolution with just two components, shifted by 550 meV, is therefore sufficient to describe the lineshape of the Ir(111) 4f 7/2 core-level spectrum at this photon energy. However, these lineshape parameters are slightly High-energy resolution Ir 4f 7/2 core level spectra corresponding to the clean (1 × 1) Ir(111) surface measured at (a) hν = 160 K and β = 0 • (normal emission) and (b) hν = 140 K and β = 40 • . Both spectra, measured at 80 K, can be fitted with two components corresponding to bulk (black) and first-layer atoms (blue) with a SCLS of −550 meV. different from those found on the stepped Ir(210) [16], which were assumed to be the same for the three surface and bulk components ( B = 300 meV, α B = 0.12 and G B = 150 meV). What is more surprising in our results is that the Gaussian width of the bulk peak is larger than that of the surface, whereas the opposite behaviour is expected because of the higher vibrations at the surface giving additional phonon broadening.
In order to make an accurate determination of the SCLSs and of the lineshape parameters we adopted a procedure already used in other investigations [26]- [29]: we acquired spectra at different photon energies (hν = 140, 160, 180 and 200 eV) and emission angles (β = 0 • , 20 • , 40 • and 60 • ) in order to change the surface sensitivity and the photoelectron diffraction conditions to modify the surface to bulk intensity ratio R.
From a simple inspection of the data it is possible to observe that, while for higher photon energies the low BE component is suppressed, for the spectrum at β = 40 • and hν = 140 eV ( figure 1(b)), R is maximum. This effect cannot be explained only with inelastic mean free path arguments because the strong modulation of the lower BE peak is strongly affected by photoelectron diffraction effects.
The first analysis of the overall series of spectra was performed using two different DS functions, without constraints. Good convergence was achieved for the fits of the complete series, as confirmed by the structureless residual. We found a SCLS of −548 ± 15 meV, in very good agreement with recent theoretical calculations based on density functional theory, which give a value of −555 meV [30].
Although this result is quite similar to the previous findings (−0.50 ± 0.02 eV), it is not within the error bar. A possible reason for the discrepancy can be ascribed to the presence of contaminants in the previous experiments (hydrogen, water, CO), which are known to shift the surface component of a metal with a d-band more than half-filled towards higher BE. More recently, SCLS measurements of the 4f 7/2 core level of clean Ir(210) have evidenced that all surface features are shifted towards lower BE. In particular the component originated by secondlayer atoms, with a nominal coordination of 9, as for the Ir(111) surface atoms, presents a SCLS of −529 meV, which is closer to our findings. Nevertheless we have to keep in mind that slight modifications of the inter-atomic distances, due to first-to-second layer relaxation of the stepped surface, are known to induce modifications in the core level shift.
The behaviour of the lineshape parameters deserves further considerations. A summary of the results is reported in figure 2. A correlation between Lorentzian and Gaussian widths of the bulk and surface components is evident: larger values of correspond to minima of G B .
In order to rationalize the data analysis procedure, we mapped the evolution of the chi-square χ 2 as a function of each parameter, following the procedure proposed in [8,29]. The two-dimensional χ 2 contour plot ( bulk versus G bulk ), reported in figure 3(a), shows a pronounced and localized minimum, contrary to what observed in figure 3(b) for the surface component. Moreover, the fitting results revealed an additional feature, which is present in the whole data analysis: a high cross-correlation is present between B and G B , B and I B , G B and I B , while this does occur to a lesser extent for the surface peak. We interpret this effect as due to the higher sensitivity of the bulk peak to small variations of the DS parameters.
Possible reasons accounting for the larger Gaussian width of the bulk, compared with that of the surface, are given in the following paragraph.
The first relates to the possible limit of applicability of the DS function for fitting the Ir 4f core level. Contrary to Pt, where the Fermi level lies in a strongly varying density of states, the Fermi level of Ir lies in a region where the density of states is quite flat, which is the assumption implicit in the DS equation. Therefore the larger Gaussian width of the bulk component cannot be attributed to a wrong representation of the Ir 4f with the DS lineshape. The most plausible reason for this unexpected finding is the presence of subsurface layer components in the close vicinity of the bulk peak, which are not included in our fitting analysis: the slightly different BE of the atoms in the layers below the topmost surface results in inhomogeneous broadening of the bulk peak. The same behaviour was indeed found also for the Ta(100) surface [31]. Despite the close-packed geometry of fcc(111) surfaces, where the second layer atomic coordination is the same as that of the bulk, experimental and theoretical studies have reported SCLSs also for the second-layer atoms. Other systems where the secondlayer SCLS has been revealed are Ta(110) [8], W(110) [11], W(111) [12], Ru(1010) [27], Ru(0001) [28], Rh(111) [29], Rh(110) and Rh(100) [32].
Another interesting finding of our SCLS analysis is the different Anderson singularity index α of bulk and surface components, reported in figure 3, which can be described in terms of the orbital character of the conduction band charge. Theoretical studies have shown that final state effects at clean transition metal surfaces are dominated by intra-atomic d-electron screening [33] and are manifested in the spectra as a high-BE power-law tail described by the singularity index α. The α parameter found in the data analysis indicates that the conductionelectron screening is larger for the first layer (α = 0.225) than for the bulk atoms (α = 0.135). This behaviour has already been experimentally observed for other 5d transition metal surfaces. For example a singularity index α of 0.10 and 0.205 [31] was found for the bulk and surface components of Ta(100), respectively.
In figure 3 we report also the Lorentzian width . The larger of the surface component ( S = 310 meV) compared with the bulk ( B = 200 meV) is an indication that the increased localization of the more atomic-like d-electron charge density at the surface enhances the intra-atomic Auger core-valence probability. Similar trends were found also for Ta(100)  [32].
In order to univocally determine the relevance of the phonon contribution to the Gaussian width of bulk and surface components the temperature dependence of the Ir 4f 7/2 lineshape was measured in the temperature range between 90 and 890 K. The spectra, collected at hν = 180 eV and normal emission, are shown in figure 4. From the inspection of the raw data, one clearly observes that the two peaks broaden as the temperature increases. Within the temperature range investigated, the main contribution to the linewidth comes from the phonon distribution. A quantitative assessment of the temperature dependence of the lineshape was carried out with a least-squares analysis. Each spectrum was analysed by letting the BE of bulk and surface components and the Gaussian width be free parameters, the latter accounting for the phonon contribution and the temperature-independent instrumental resolution, while and α were fixed to their low temperature values. This is an adequate choice, because in a first approximation the core-hole lifetime and the electron-hole pair excitation probability do not depend on the temperature. Figure 5(a) shows the square of the Gaussian width as a function of temperature of the bulk and surface peaks: both curves show almost linear behaviour. These results confirm the larger Gaussian contribution of the bulk component at low temperature, while most notable is the difference of the slope of the surface and bulk curves. In parallel with broadening, both components shift towards lower BE with increasing temperature, as shown in figure 5(b), the bulk-peak by 31 meV, the surface-peak by 17 meV, between 80 and 780 K. Therefore the SCLS decreases by 14 meV.
We analysed the behaviour of the Gaussian width using the Hedin and Rosengren model [33] for the phonon broadening, based on the Debye model, where the squared G 2 is given by, Here, G res and G inh are the widths of the instrumental resolution and inhomogeneous broadening, respectively, G p (0) is the phonon broadening at T = 0 K and D is the Debye temperature. This equation can be applied for T / D > 0.8. Using a least-squares analysis we obtained D (bulk) = 298 ± 8 K and D (surf) = 181 ± 3 K, which are in very good agreement with previously determined values obtained using LEED, i.e. 315 ± 22 and 170 ± 12 K for bulk and surface Debye temperature, respectively [18].
The BE changes of bulk and surface components could be related to a change in the valence-band density of states as the lattice expands, even though final-state effects (changes in relaxation energy as the screening length changes) are expected to contribute.
The different behaviour of the surface and bulk thermal shifts is in strict contrast to the results obtained on alkali metals [34] where, despite the large thermal shifts, no difference was observed between the thermal shift of the bulk and surface core level components.
Using the Friedel model combined with a linear combination of atomic orbitals, Riffe et al [9] calculated a shift of −0.058 meV K −1 for the core level of bulk Ir. Our results seem to validate this simple model. First of all the experimental shifts are negative, as expected when the d-band is more than half-filled, as for Ir. Moreover, we find a thermal shift of −0.045 and −0.024 meV K −1 , for the bulk and surface components, respectively. While the thermal shift of the Ir 4f 7/2 of the bulk is in good agreement with the model, the difference found for the temperature behaviour of the surface peak can be ascribed to the different bulk and first-tosecond layer expansion. In the case of Be(0001) for example, where the first three topmost layers can be distinguished in the Be 1s core level spectra, the SCLS of the first-and secondlayer atoms decreases, while that of the third layer increases at higher temperature [35]: the comparison of the experimentally determined SCLS with density functional theory calculations was used to find the temperature-dependent interlayer distances. Another example is the Ta(100)  surface, where the thermal shift of the bulk Ta 4f component is ∼ 2.4 times larger than that of the surface peak [9].
The final step of our investigation was the study of the changes induced on the Ir 4f 7/2 SCLS by oxygen adsorption.
Previous LEED studies of O 2 dissociative adsorption on Ir(111) have evidenced the presence of a pattern with (2 × 2) periodicity [18]. XPS and HREELS (high resolution electron energy loss spectroscopy) data [21,22] suggested that the saturation coverage is 0.5 ML. This corresponds to a p(2 × 1) structure with three domains rotated by 120 • from each other, giving the same LEED pattern as a single p(2 × 2) domain, and with a single oxygen adsorption site, as suggested by the observation at the saturation coverage of one loss peak in HREELS at 550 cm −1 [19]. A LEED I-V investigation [18], although not decisive for the determination of the saturation coverage, concluded that the preferred adsorption site is threefold. Theoretical calculations [36,37] reported that the dissociation is nearly spontaneous, with a very small activation energy of 0.06 eV molecule −1 . Similarly to oxygen adsorption on other transition metal surfaces, as the coverage increases from 0.25 to 1.0 ML, the BE decreases substantially, because of the repulsive interaction between the adsorbed atoms [38]. Figure 6(a) shows a selection of the O 1s spectra collected during oxygen uptake at 80 K. The sequence of spectra, measured at hν = 650 eV, can be fitted using a single peak up to the saturation exposure of 1.5L. As shown in figure 6(b), the intensity of the O 1s peak increases with increasing oxygen coverage. The shift of the O 1s towards lower BE is rather small, less than 70 meV, and cannot be assigned to a change of the oxygen adsorption site. In an initialstate picture the O 1s BE variation can be understood by taking into consideration that, as the oxygen coverage increases, the distance between negatively charged O atoms decreases and a repulsive interaction builds up. This produces changes in O-Ir bond length which results in small variations of the BE. Therefore our results suggest that oxygen preserves the same adsorption site in agreement with previous experimental [19] and density functional theory [23,36] investigations.
In order to monitor the rearrangements in the adsorbed phase at increasing coverage, we have measured the Ir 4f 7/2 SCLSs induced by the oxygen atoms. These measurements offer the unique possibility of identifying the changes in the local environment of the substrate atoms, like the number of O atoms bound to each Ir surface atom.
The real-time SCLS method [23] has been applied to probe the evolution of the Ir 4f 7/2 core level during oxygen adsorption. Figure 7 shows the series of Ir 4f 7/2 spectra measured in the same experimental conditions used for the O 1s experiments. The spectrum of the clean surface consists of two well resolved Ir bulk and Ir 0 peaks representing the emission from the bulk and the surface atoms, respectively, as explained above. Oxygen adsorption at low coverage leads to the appearance of a new component Ir 1 which grows between the surface and the bulk peaks, while the intensity of the original Ir 0 surface peak decreases. At higher coverage, Ir 0 further decreases and a third surface component Ir 2 grows at BE larger than the bulk peak.
The deconvolution of the Ir 4f 7/2 spectra has been performed using for the extra components Ir 1 and Ir 2 , the same lineshape parameters of the clean surface component Ir 0 . Moreover, in the fitting procedure the BE positions of Ir 1 and Ir 2 were allowed to vary in a limited energy range, as explained below. A first attempt was performed leaving the BEs of all the peaks free, in order to detect acceptable ranges of variation. This fitting procedure achieved convergence only for some spectra; however, it allowed to determine the proper BE intervals for bulk, Ir 0 , Ir 1 and Ir 2 . The results of the fits of the whole series of spectra measured at different oxygen coverage, reported in figure 8, have been obtained by setting stronger constraints on the BE intervals as a function of the relative intensity of the peaks, the BE being more precisely identified when the intensity of a component is high. While the BE position of the bulk peak remains constant, as expected, a small shift of about 50 meV towards higher BE is experienced Figure 7. Evolution of the Ir 4f 7/2 core-level spectra during oxygen adsorption on Ir(111) at T = 80 K. Oxygen pressure ranges from 1 × 10 −9 to 1 × 10 −7 mbar, hν = 180 eV.
by Ir 0 at increasing oxygen coverage. The Ir 1 component was found to be at −200 ± 20 meV with respect to the bulk peak, while the SCLS of Ir 2 is +230 ± 30 meV. The larger error bar of the SCLS of Ir 2 is due to the low intensity of this component and the close vicinity to the bulk peak that makes the deconvolution more difficult. From a close comparison with the behaviour of the SCLSs induced by dissociative oxygen adsorption on other substrates [7,28,39,40] we assign the Ir 1 and Ir 2 components to first-layer Ir atoms single and double-bonded with oxygen atoms.
An important piece of information provided by these experiments concerns the oxygen induced changes in the local electronic structure of the surface atoms, obtained by the measurement of the BE difference E of Ir 1 and Ir 2 with respect to the clean component Ir 0 . Indeed, since E 1 = +350 meV and E 2 = +780 meV, it turns out that E 1 ≈ 2 × E 2 . This demonstrates the validity of the additivity rule of the oxygen-induced energy shift on the Ir surface, a property that was already found on other adsorbate [38]- [41] and coadsorbate systems [42]. The almost linear dependence of E on the number of O atoms bound to each Ir atom, reveals that the type of bonding of O remains the same in all the overlayers studied.
The oxygen-induced Ir 4f shift towards higher BE with increasing oxygen coverage can be qualitatively understood as due to charge transfer from the surface to the adsorbate. A complete understanding of this shift cannot be achieved only on these grounds because effects other than simple charge transfer play a crucial role in determining the SCLS. A strong correlation between E and the calculated valence d-band shift projected onto the surface atoms was found for many oxygen-transition metal systems. The oxygen adsorption typically induces a broadening of the d-band, along with a lowering of the density of states in the Fermi level region. This results in a d-band centre shift towards higher BEs with increasing oxygen coverage, which is reflected in the measured adsorbate-induced SCLSs. This effect applies also to the case of Ir. Figure 9 displays the behaviour of the integrated intensities of Ir 0 , Ir 1 and Ir 2 peaks as a function of the oxygen coverage. We show in the following paragraph that these functions contain detailed information on the kinetics of site occupation during chemisorption.
The Ir 0 surface component intensity is proportional to the number of first-layer clean Ir atoms, i.e. atoms that are not chemically bound to oxygen atoms [23]. At the beginning of the uptake, when the sample is clean, all Ir surface atoms are zero-coordinated to oxygen atoms. We normalize all intensities to this value.
In the early stage of oxygen adsorption the number of Ir 1 atoms linearly increases with oxygen uptake at the expense of the Ir 0 atoms. The Ir 2 component appears and grows later, becoming appreciable only when the oxygen coverage Q is about 20% of the saturation coverage Q max and reaches the maximum value just at saturation. At this coverage, the intensity of Ir 0 component remains significant, while the Ir 1 component has an intensity which is about 25% larger than the Ir 2 component.
The quantitative description of the adsorption phenomenon can be achieved through the observation of the clean surface signal, the only reliable indicator, since Ir 1 and Ir 2 intensities may well be affected by photoelectron diffraction effects. Therefore our analysis focuses on the Ir 0 curve, in particular on the slope of the initial linear behaviour. The decreasing intensity is a clear indication of adsorption: for each new oxygen atom adsorbed in threefold configuration, the intensity contribution of three surface Ir atoms should move from Ir 0 to Ir 1 .
From this observation we calibrated the surface coverage in a straightforward way by simply assuming that the intercept of the linear fit of the Ir 0 curve is 1/3, because of the already known adsorption in threefold configuration. This procedure was already applied on other adsorbate systems [23]. We found an oxygen saturation coverage of 0.38 ± 0.04 ML, a result that reconciles previous experimental and theoretical observations.
Indeed recently Stampfl et al [23] reported that the (2 × 2)-O structure with 0.25 ML coverage is the only chemisorbed phase that is predicted to be stable on the surface, and only for a very narrow range of the oxygen chemical potential, which can be achieved in UHV conditions. These results suggest that the saturation coverage of 0.5 ML, reported by the experiments, may only be metastable with respect to bulk oxide formation. Our results indicate that the p(2 × 1) structure cannot be experimentally obtained without a certain degree of surface disorder, like oxygen vacancies in the lattice, as reported in the structural model reported in figure 9. These findings underline the importance of considering the problem of pressure gap encountered in usual surface science studies, and its consequences on surface morphology. On Ir(111), when using oxygen exposures in the 10 −7 mbar range even for prolonged time, the coverage at saturation remains always below 0.5 ML: the increase of O 2 pressure and substrate temperature are therefore the key ingredients to produce a stable high coverage oxygen structure, which represents an intermediate step towards the subsurface penetration and the formation of Ir surface oxide, as predicted by theory [23]. Our results therefore confirm that the room temperature instability of a long-range oxygen ordered structure at 0.5ML is not due to reducing agents (CO, H 2 , etc) as supposed, but originates from the intrinsic properties of the O-Ir interface. The recent development of high pressure XPS technique with synchrotron radiation can be the base for testing the results with recent ab initio calculations about the stability of oxide species formed on Ir at higher temperature and pressure [23].
In summary, we have shown that the extreme sensitivity of core-level spectroscopy to the local electronic environment of surface atoms can be exploited to monitor the evolution of a wide range of phenomena taking place on solid surfaces. In particular the direct in situ and realtime measurement of core-level changes have proven decisive for an in-depth understanding of different properties of clean and oxygen covered Ir(111). The accurate determination of Ir 4f 7/2 core level BE and lineshape parameters allowed us to probe the electronic structure of first-layer Ir atoms. The larger Gaussian width of the bulk component compared with the surface, is explained as due to the presence of unresolved subsurface components (second and deeper layers). The temperature evolution of the Gaussian widths was used to determine surface and bulk Debye temperature, which are in good agreement with previous electron diffraction measurements. Finally the oxygen adsorption process was investigated by monitoring in situ the changes in the Ir 4f 7/2 core level spectra. The unambiguous identification of Ir surface atoms differently coordinated with oxygen and the behaviour of the corresponding intensities indicate that the oxygen coverage achieved in UHV conditions is below 0.5 ML, as predicted by theoretical calculations.