Photoelectron spectroscopy of ceria: Reduction, quantification and the myth of the vacancy peak in XPS analysis

Primarily due to its inherent redox chemistry, ceria (CeO2) is of use in many diverse areas of research. However, there is a wealth of misinterpretation of the oxygen spectra when discussing the result of damage or reduction to the CeO2 lattice, especially with regard to a signal in this region attributed to oxygen vacancies. In this paper, it is shown that this peak cannot be due to vacancies and that a better understanding of the changes in stoichiometry of CeO2 is best viewed from that of the Ce(III) component when considered in tandem with the O 1s signal.

novice users, there can still be a barrier to question quantified data which can lead to the propagation of errors in the literature.An example of this is a fundamental misunderstanding in what XPS can tell us about the oxygen states within a metal oxide.
In the O 1s spectrum, there are usually at least two oxygen signals.The first and typically most intense is the signal for the lattice oxygen atoms, with a binding energy generally in the range of 529-530 eV; however, there are exceptions such as SiO 2 where the signal is approximately 533 eV.The second higher energy peak, in the range 531-532 eV, is assigned as hydroxide or in many studies as 'oxygen vacancies'.
The published literature is littered with examples of researchers discussing oxygen vacancies in various materials through the presence of a supposed defect peak in the O 1s core-level spectra, which can be traced back to a paper by Fan and Goodenough discussing In 2 O 3.   11   However, the entire basis of XPS as a technique is the measurement of the kinetic energy of a photoelectron ejected from an orbital, the proposition therefore that a specific peak is representative of a vacancy is preposterous, the very fact of there being a vacancy within the lattice means that there is simply no electron to be ejected and hence measured.
There have already been papers published addressing this erroneous assignment.For example, Idriss 12 has elegantly shown that through a review of the literature, the assignment of a peak as an oxygen vacancy is wrong, while recent computational studies by Frankcombe and Liu 13 again beautifully illustrate the fallacy of vacancy assignment in the case of ZnO.
5][16] However, it is because of this redox chemistry that the presence of Ce(III) states is thought of as oxygen vacancies.In this paper, it is shown experimentally how to address the role of Ce(III) reduced states in ceria and related materials and quantify the stoichiometry of the defective ceria.

| EXPERIMENTAL
All data were collected using a Kratos Axis Ultra DLD photoelectron spectrometer using Al kα radiation (photon energy 1486.6 eV) operating at 10 mA Â 15 kV (150 W).To maximise sensitivity, while lowering acquisition time per cycle, the system was operated in the hybrid mode with the slot aperture, giving a total analysis area of approximately 700 Â 300 μm.
The Ce 3d, O 1s and C 1s spectra were recorded at a pass energy of 40 eV and step size of 0.1 eV.The spectra were acquired sequentially over a total of 600 min, with each of the three regions taking approximately 2 min to acquire, giving approximately 6 min per iteration.
Charge compensation was performed using low-energy electrons flooding the ceria surface.The compensation system was operated with a filament current of 0.42 A, charge balance of 4.2 V and a filament bias of 1.4 V. Spectra are typically overcompensated by approximately 2-3 eV and hence corrected to the C 1s peak of adventitious carbon taken to be 284.8eV.However, over the course of the experiment, the C 1s peak was found to vary by approximately 0.4 eV becoming an unreliable calibration point; therefore, the O 1s peak of the lattice oxygen was used and taken to be 529.7 eV.The C 1s spectra are shown in the supporting information Figure S1.
Data were calibrated and analysed using CasaXPS v2.3.26 rev1.1 K. 17 Quantification was performed using the Kratos F1s RSF library as supplied within the software.Background removal was achieved using a Shirley background, and peak fitting where required was performed using a Voigt-like function, defined within the software by LA(α,β,m) where broadly the first two parameters control the asymmetry of the high and low binding energy sides of the peak, while the third parameter specified the width of the Gaussian used to modify the Lorentzian curve, omission of the β parameter means it is equal to the α parameter.Generally, Ce(III) and O 1s components were fitted using LA(1.53,243), while the more asymmetric like components of the Ce(IV) spectra were fitted using LA(0.9,2,50).Peak shapes for CeO 2 were derived using a stoichiometric single crystal in which peak areas were constrained using the expected peak ratios, although the full width at half maximum (FWHM) were allowed to vary.
The CeO 2 powder (Alfa Aesar, 99.9% purity) was used as received and mounted in standard laboratory conditions by compacting into a recess of an in-house modified standard Kratos Axis Ultra sample bar using a clean glass microscope slide.The CeO 2 single crystal was obtained from PI-KEM (Tamworth, UK) and cleaned within the spectrometer using argon sputtering using a Kratos Minibeam 1 operating at 4 keV over a 5 Â 5 mm raster area with an argon pressure of 10 À6 mbar.A stoichiometric crystal (as confirmed by XPS quantification) was regenerated through heating to approximately 750 Ci n 10 À5 mbar of oxygen.

| The photoelectron spectra of cerium oxides
Well-resolved cerium oxide spectra were first published by Burroughs et al., 18 who showed, as have many authors since, the origin of the complex satellite structure inherent within these f-electron systems derived from ligand-to-metal charge transfer. 18,19To discuss the reduction of ceria, we can familiarise ourselves with the spectra of CeO 2 and Ce 2 O 3 .The spectra shown in Figure 1 were acquired from a ceria single crystal and are labelled according to the nomenclature introduced in Burroughs et al. 18 Note that due to the absence of carbon on these surfaces, both spectra have been calibrated to the O 1s peak taken to be 529.7 eV.
For the CeO 2 spectrum, there is a total of six peaks, where the u iii and v iii peaks (917.0 and 898.3 eV, respectively) come from a Ce 3d 9 4f 0 O2 p 6 final state, while peaks u, v, u ii , and v ii (901.2, 882.5, 907.8 and 889.2 eV, respectively) are a consequence of Ce 3d 9 4f2 O 2p 4 and Ce 3d 9 4f 1 O2p 5 final states.
For the representative Ce 2 O 3 spectrum, the Ce(III) oxidation state exhibits four intense peaks; those labelled u i and v i have binding energies of 903.9 and 885.6 eV and attributable to a Ce 3d 9 4f 1 O2p 6 final state, whereas the peaks u 0 and v 0 at 899.2 and 880.7, respectively, are a consequence of a Ce 3d 9 4f 2 O2p 5 final state.
It is emphasised at this juncture that the spectra of that we are considering to be Ce 2 O 3 are not a bulk or stoichiometric material and FIGURE 1 The Ce 3d spectrum for CeO 2 and sputtered CeO 2 to mimic the Ce 2 O 3 spectra.The peaks are labelled according to the nomenclature used by Burroughs et al. 18 were created by sputtering the single crystal used for the CeO 2 spectrum.While this does yield Ce(III) spectra which are very similar to those previously attributed to Ce 2 O 3 based mixed oxides, 20 CePO 4 21 and ion irradiated CeO 2, 22 care must be taken with regard to the intensity ratios between the v i + v 0 peaks (and u i + u 0 peaks) as variations have been noted in chelated Ce(III) species. 18[23] 2.2 | The myth of the oxygen vacancy peak in the XPS of ceria and related materials 4][35][36] It is emphasised that these papers are not being singled out as the only ones that are erroneous in their assignment; indeed, as stated earlier, there are multiple examples in the literature.However, in these texts, little or no discussion is given to, for example, the C 1s spectrum which may indicate the presence of surface carbonate (although this may be somewhat complicated by the overlapping Ce 4s peak), or hydroxyl formation has not been considered.Certainly, metal hydroxides and carbonate typically have an O 1s signal in the binding energy range specified. 37,38Furthermore, there is no reason to why Ce 2 O 3 would have a higher O 1s binding energy than CeO 2 , and both experimental 39,40 and computational 41 measurements have indicated little or no shift for the O 1s lattice signal.

| Assignment of chemistry through the controlled reduction of CeO 2
The reduction of CeO 2 during XPS analysis is not unknown, 19,20,42 and the formation of reduced states can be attributed to a number of physical and experimental phenomena. 43However, the assignment of Ce(III) states is typically reported as defects or Ce 2 O 3 , without thought for actual assignment of the chemistry through a thorough analysis of the complimentary, but intrinsically linked Ce 3d and O 1s spectra.
Figure 2 shows Ce 3d and O 1s spectra taken at three different times from the prolonged exposure to X-ray analysis of a powdered CeO 2 sample.Without fitting, the increase in Ce(III) is clearly evident through the changes in the Ce 3d spectra, but note that there is little change in the O 1s spectrum.If, as claimed by some authors, the peak at 532 eV was due to Ce 2 O 3 , then the concentration of this would increase relative to the major O 1s lattice oxide component at 529.7 eV, so we can rule out this as an assignment.
6][47] This latter method provides an alternative to fitting spectra, by the inclusion of information from physical processes which affect the sample (such as X-ray degradation) without the need to identify synthetic line shapes.
In these cases, spectral forms are extracted through continued analysis which are a representative shape constraint for reproduction of the spectral data.It is this latter method, which is used here, and the fitted spectra are shown in Figure 3.It should be noted that use of synthetic line shapes derived from the spectra in Figure 1 yields consistent results and is shown in Figure S2 of the supporting information.With fitting, the increase in Ce(III) is even more evident, but the question remains, how we can assign the peak at 532 eV so it is representative of the chemistry of the evolving ceria system?Considering the increase in the concentration of Ce(III) and the stability of the aforementioned O 1s peaks, the Ce(III) content cannot be solely a consequence of Ce 2 O 3 ; instead, by using the complimentary data in the Ce 3d and O 1s spectra, it can be shown that the Ce(III) spectrum is a consequence of both Ce 2 O 3 and Ce(OH) 3 .
Table 1 shows the derived atomic concentrations for the cerium and oxygen chemical states, together with the calculated stoichiometry for the three times.By means of example, the derivation for T = 300 min is presented, while full derivations for the three times are given in the supporting information.Of course, the reasoning and analysis can be applied to the full reduction data set.
The peak above 533 eV is assigned to a small amount of surface contamination, which we note decreased between the first and remaining spectra which may suggest desorption of some surface species.Given its small concentration, and unlikely due to a Ce(III) or Ce(IV) species, this signal is ignored in the calculations discussed below.
Earlier, we discussed the binding energy of hydroxides and noted that they fall directly in the region 531-532 eV, so it is logical to first assume that this signal is due to surface hydroxide.Given the stoichiometry of the 1:3 ratio of Ce to OH and assuming that the signal at 532 eV is all hydroxide, then we can derive the Ce(III) contribution through the relationship: Ce III Therefore in our example, the concentration of Ce(III) due to the hydroxide is 4.02%.Our 'corrected' Ce(III) concentration is therefore  Repeating this calculation for the first and last sets of spectra, we find stoichiometries of CeO 1.9 and CeO 1.6 , respectively, which support the reduction of CeO 2 through the increase in Ce 2 O 3 , but as shown by the full derivations given in the supporting information, it shows the majority of the Ce(III) states in the as-received oxide that are due to hydroxide groups.
In this model, we have of course assumed that the Ce(III) signal is composed of both stoichiometric hydroxide and oxide.Nevertheless, using the informed sample approach, we capture the total Ce(III) signal without the need for fitting synthetic peaks, and hence, we do not have to worry about any subtle differences which may exist in the ratios of the v 0 (u 0 ) and v i (u i ) peaks which may be present in the Ce 3d spectrum of Ce 2 O 3 and Ce(OH) 3 .It is noted that the change in the stoichiometry of the CeO 2 is small; however, it does show that prolonged analysis can lead to changes in the spectra and expected ratios and that this method of analysis reveals an explainable trend in the increase of Ce(III) within the Ce(IV) oxide lattice with the surface terminated by Ce(III) hydroxide and hence a more consistent measure of chemistry at the surface and the degree of nonstoichiometry in the near surface region of CeO 2 .

| CONCLUSIONS
This work has sought to highlight the now increasingly documented erroneous assignment of peaks as oxygen vacancies.It has been shown that through careful consideration of the Ce 3d and O 1s levels together, coupled with an appreciation of the different oxygen states possible, the decrease in the O:Ce ratio in ceria can be followed and the Ce(III) chemistry can be assigned to both the oxide and hydroxide.
It is hoped that this work, together with the papers of Idriss 12 and Frankcombe, 13 will facilitate an improved interpretation of O 1s spectra for oxide materials.

FIGURE 2 (
FIGURE 2 (A) Ce 3d and (B) O 1s spectra taken from a series of spectral acquisitions totalling 600 min.

FIGURE 3 (
FIGURE 3 (A) Ce 3d spectra fitted using the shapes derived from the informed sample model approach, where blue = Ce(III) and yellow = Ce (IV), and (B) O 1s spectra fitted with lattice oxide (green), hydroxide (magenta) and organic (orange) components.

TABLE 1
The derived at% concentrations for Ce 3d and O 1s species at the three selected time frames.Note: Note that carbon has been omitted for clarity.(9.37-4.02)=5.35%.If we assume that this 5.35% is due to Ce 2 O 3 , then given the 2:3 stoichiometry, the percentage of oxygen in Ce 2 O 3 is 1.5 time this concentration, in this case 8.03%.Given the total lattice oxide concentration (51.84%), then the percentage due to Ce(IV) is (51.84-8.03)= 43.81%,and hence, the O/Ce (IV) ratio is 43.81/25.23= 1.7 consistent with defective CeO 2 , which could be described as CeO 1.7 .