3.1. Polycrystalline Rhodium Sample
As a sample, a 10 × 12 mm2 rectangular piece of polycrystalline rhodium film with a purity of 99.99% (MaTecK, Juelich, Germany) and with a thickness of 0.2 mm was used. The sample was mounted on a sample holder transferred to a ultra-high vacuum (UHV) station equipped with a spatially-resolved VG-Scienta, Uppsala, Sweden R3000 hemispherical analyzer. Cleaning was carried out by argon sputtering, annealing, and chemical treatment with oxygen. Ion cleaning was carried out at 300 K using argon ions (10 µA, 3 keV). The sample was then annealed for 30 min to 1123 K. To remove carbonaceous impurities, additional chemical treatment was carried out using oxygen at 773 K until the XPS spectra did not contain a carbon signal.
Polycrystalline rhodium foil, consisting of many crystallites of different sizes with different structures, is an example of a set of randomly distributed crystallites with different crystallographic orientations and provides promising insight into structure dependence and the heterogeneity of catalytic reactions.
Figure 5 shows images of the polycrystalline surface of the rhodium sample obtained from an optical confocal microscope. The clean surface presented in
Figure 5a had many scratches and imperfections, but clear grain boundaries could also be observed. The fact that they are indeed crystallites of different orientations becomes apparent after the high oxidation process. Such an attempt was made for the same sample in a separate experiment, using a flow-through high pressure reactor. The sample was exposed to 1.5 bar of oxygen at 873 K for 30 min. As a result, the crystallites were colored with at least six different shades, as shown in
Figure 5b. Most grains were rather small, but one could also find some crystallites of about 1 mm in size. In our study, four such larger grains were selected to minimize the risk of skipping (e.g., due to the slack on the sample manipulator mechanism). Four chosen areas are marked in
Figure 5 and differ in hue after high oxidation, which may indicate that they had different reactivity. To set the sample during XPS measurements, two tantalum and two silver markers were spot-welded in the corners of the sample. By scanning the sample along X and Y axes, the coordinates of the center point of each marker were found and were then used to scale the area analyzed by XPS.
3.2. Crystallographic Orientation of Rh Foil Domains
Identification of the rhodium foil crystallography was obtained by EBSD. It is a scanning electron microscope-based microstructural-crystallographic characterization technique commonly used in the study of crystalline or polycrystalline materials [
18]. The technique provides information about the structure, crystal orientation, and phase of the material. A flat surface was prepared by polishing and rinsing with acetone. EBSD measurements were carried out using the Field Electron and Ion Co. Hillsboro OR, USA (FEI) Quanta 3D Field Emission Gun (FEG) and scanning electron microscope (SEM), equipped with an EDAX Inc. Mahwah NJ, USA Genesis Orientation Imaging Microscopy (OIM) back-scattered electron diffraction recording and analysis kit.
An EBSD analysis was carried out from the supplied rhodium sample in the area of about 7 × 9 mm
2. The SEM (scanning electron microscopy) micrograph recorded using a back-scattered electron detector from these conditions is shown in
Figure 6. Crystal orientation maps are often displayed in the so-called inverse pole figure (IPF) coloring [
18].
Figure 6a shows the inverse pole figure map (IPF) of the rhodium sample.
Figure 6b shows all Euler RGB maps, where the face-centered cubic (fcc) matrix grains have been marked in red, while a continuous network of green dots decorating the grain boundaries has been indexed as a body-centered cubic (bcc) phase. The fcc and bcc phases exhibit an orientation relationship that can be established from the EBSD results. The pole figures corresponding to the phases are shown in
Figure 6c.
3.3. Spatially-Resolved X-ray Photoelectron Spectroscopy (XPS) Analysis
In order to perform the XPS experiment, the sample was introduced into the UHV system consisting of preparation and analysis chambers. The cleaning, annealing, and oxidation were done in a preparation chamber (base pressure ≤ 5 × 10−10 mbar). Photoemission studies were performed in an analysis chamber at the base pressure ≤ 2 × 10−10 mbar. Non-monochromatized Al Kα radiation (1486.6 eV) was used as an excitation source. The main axis of the source was oriented at 55° with respect to the norm of the sample surface. The photoemission spectra were obtained) at the normal angle with respect to the surface. In order to calculate the rate of oxidation, experimental data were fitted to asymmetric line-shapes by using CasaXPS® software (Casa Software Ltd., Teignmouth, UK) after modeling standard Shirley background.
In spatially-resolved XPS studies, we took advantage of the imaging properties of the energy analyzer [
16]. Photoelectrons emitted from the surface due to the X-ray irradiation are focused and the linear image is created in the non-dispersive plane of the analyzer, while at the same time the XPS spectrum corresponding to each point in the line is established in the dispersive plane. Data acquisition takes place by using a two-dimensional detector (micro-channel plate). XPS data were collected from multiple channels of the detector for the chosen regions of the Rh sample. The advantage of such a spatially separated method in oxidation studies is that all test areas are exposed to exactly the same oxygen pressure and temperature for the same time in one experiment.
Applying spatially resolved XPS to individual stepped Rh (hkl) domains of a polycrystalline Rh foil used as a set of randomly distributed crystallites with different crystallographic orientations, we demonstrated the dependence of the Rh oxidation rate on the surface structure. Measurements performed at an oxygen partial pressure in the 10−5 mbar range and at temperatures from 423 K to 668 K enable determining the activation of the initial oxidation of Rh.
The sample was oxidized in an O
2 atmosphere (1∙× 10
−5 mbar) for 7, 15, 30, 45, 60, 75, and 90 min. The temperatures of oxidation were 423, 493, 578, 623, and 668 K. After sample oxidation, we began analyzing the sample composition using spatially-resolved XPS. For the analysis, we selected four different high-index crystallites, whose positions on the sample are shown in
Figure 5 and
Figure 6. The crystallographic orientations of these four crystallites found in EBSD are as follows: (−17 −8 11), (−24 2 −29), (9 5 −6) and (−4 14 1). Orientations of particular crystallites with respect to unit cell as well as their structure models are shown in
Figure 7.
3.4. Statistical Multivariate Analysis to Examine XPS Data for Chosen Rh Crystallites
In the process of matching the curve to the registered spectrum, and thus obtaining information on the proportion of occurrence of each chemical state, there is always some uncertainty. The uncertainty of the XPS curve fitting results is due to the choice of background subtraction method, statistical fluctuations of the registered intensity, etc. Therefore, to make quantitative analysis more reliable, multivariate statistical insight can be used to examine XPS data. The main advantage of the multivariate statistical analysis (MVA) applications to spectral analyses such as XPS is that it includes a complete set of data at once, and no mathematical function describing the shape of the line is required. This approach allows, therefore, to create a global model, in particular, it is a very effective way to compare spectra obtained in different experimental conditions and at different spatial coordinates on the surface, which simplifies time-consuming analysis of large datasets.
Principal component analysis (PCA), one of the most popular method for multivariate analysis, is fundamentally different from standard curve-fitting methods. It gives the possibility of obtaining a quick differentiation of the share of chemical components in the set of obtained photoelectron spectra by applying the geometric interpretation of the spectrum as a vector in n-dimensional space. The number n denotes the number of measurement points (e.g., binding energy values) for which XPS spectra were recorded, usually a large number like 200 or more. PCA assumes that any set of spectra described by a linear combination of two or more pure chemical components can be decomposed into orthogonal abstract components: the principal components. The principal components create a new, low-dimensional space. A detailed description of PCA can be found elsewhere (see e.g., [
19,
20]). Typically, when PCA is applied to spectral data with different composition, or from different stages of the chemical/physical process being investigated, the goal is to reduce dimensions by using space with one or two principal components (PCs). Thus, the n-dimensional problem goes from n to the 1D or 2D-principal component model. The PCs model “well distributes” the most diverse spectra in the PC1–PC2 space, which can be displayed graphically.
Thanks to the use of PCA, the dataset is reduced to two cross-product matrices. These matrices, commonly called scores and loadings, can be used to visualize the quantitative relationship between spectra and the relationships between the original variables (characteristic binding energies of spectral peaks), respectively. More specifically, each spectrum has its own score value (contribution) in PC1, PC2, and so on. A small score value means that the spectrum is the most typical (like average spectrum), whereas the high score value means that it is extreme.
The most common is the graphical visualization of the score vector for PC1 versus the score vector for PC2. These are the two directions along which the data swarm exhibits the largest and the second largest “spread”. However, in some cases, the use of 1-dimensional score vector (i.e., plotting the score vector for PC1 versus the spectrum number) could give valuable information in a simple way. For spectroscopic data, one can see the quite high % variance captured by PC1 (i.e., 99.8%), which means that the set of spectra is composed of a very strong systematic component. If it is a case, the score value could have a semi-quantitative chemical interpretation (i.e., contains chemical composition information about the Rh3d doublet in this case).
In spectroscopic application, there are several options to normalize the data to compensate for individual spectral differences due to experimental errors. In this study, before using PCA, each raw spectrum was normalized to the sum of the intensities. Then, PCA was performed using PLS Toolbox v.8.8.1 (Eigenvector Research, Manson, WA, USA) for MATLAB (MathWorks, Inc., Natic, WA, USA).