In Situ Oxidation and Reduction of Cerium Dioxide Nanoparticles Studied by Scanning Transmission Electron Microscopy

Abstract

Cerium dioxide nanocubes and truncated octahedra were reduced and oxidized in the scanning transmission electron microscope. The reduction process was stimulated by the electron beam and oxidation was supported by background gases in the microscope environment. High-angle annular dark field imaging is sensitive to local lattice distortions that arise as oxygen vacancies are created and cerium cations reduce enabling high spatial resolution characterization of this process with temporal resolution on the order of seconds. Such measurements enable us to differentiate and infer that the observed behavior between the nanocubes and truncated octahedra may be due to the difference in crystallographic termination of surfaces. In situ measurements taken with different partial pressures of oxygen reveal the cerium oxidation state and the dose rate threshold for the onset of beam reduction are influenced by the environment. Increasing oxygen partial pressure reduces the Ce³⁺ content and decreases susceptibility to electron beam driven reduction.

1. Introduction

In situ scanning transmission electron microscopy (STEM) offers the opportunity to observe materials under dynamic conditions with high spatial resolution. Here we observe the oxidation and reduction of cerium dioxide (CeO₂, ceria) nanoparticles using STEM and electron energy loss spectroscopy (EELS). CeO₂ is reduced by highly-energetic electron beams.(Garvie and Buseck, 1999) This behavior has been demonstrated to occur when the beam current is increased above a critical dose-rate threshold (Johnston-Peck et al., 2016a). Upon decreasing the beam current, the reduced nanoparticle begins to oxidize through interactions with the background gases in the microscope environment. To study this transformation requires both adequate spatial and temporal resolution. STEM-EELS spectrum imaging has produced datasets quantifying the oxidation state of cerium cations in ceria with atomic spatial resolution (Turner et al., 2011). Unfortunately the dwell time in a spectrum image is typically on the order of milliseconds compared to microseconds for a bright field or annular dark field (ADF)-STEM image. This means the temporal resolution of spectrum imaging is worse for a given set of sampling conditions and would face limitations in tracking dynamic processes. One method to circumvent this limitation is to extract additional information related to the reduction-oxidation (redox) process directly from the ADF images to supplement the EELS data.

To this point, when the sample is oriented to a high-symmetry-crystallographic axis, electron channeling becomes pronounced and local changes in the sample can manifest as contrast within the resulting ADF images. In CeO₂ specifically, atomic static displacements due to oxygen vacancies and the different ionic radius of Ce³⁺ and Ce⁴⁺ cations will generate contrast (Johnston-Peck et al., 2016b). The relative simplicity of a single-crystal nanomaterial, without additional features (e.g., line or planar defects) that can also be sources of contrast, presents a situation where the image contrast can be interpreted beyond the mass-thickness relationship of high-angle annular dark field STEM (HAADF-STEM) images to being qualitatively sensitive to relative differences in the extent of reduction. Therefore, in addition to being able to identify different crystallographic phases, we can also discriminate between materials that are the same phase but are anion-deficient (i.e., fluorite CeO₂ and fluorite CeO_2-x). This enables tracking of the conversion from CeO₂ to Ce₂O₃ and then back to CeO₂ with a combined temporal and spatial resolution that a spectroscopic technique alone (e.g., STEM-EELS) is currently unable to provide.

With HAADF-STEM imaging we observe nucleation and growth of the cubic sesquioxide phase (C-Ce₂O₃) in nanocubes, as well as the spatial dependence of the oxidation process in nanocubes and truncated octahedra to CeO₂ from CeO_2-x. Analyzing this information along with EELS data we correlate differences in redox activity between particle geometries as a function of surface termination. To appreciate the influence of environment on CeO₂ and the role it plays on the measurement process we measure the critical-dose rate threshold necessary to reduce CeO₂ and quantify Ce³⁺ content as a function of gas pressure using environmental STEM (ESTEM). We observe increasing oxygen pressure affects the results by increasing the critical-dose rate threshold. Also, the Ce³⁺ content of ≈ 5 nm nanoparticles were found to be responsive to environment suggesting that the changes in oxygen chemical potential can influence the outcome of measurements made in the electron microscope.

2. Materials and Methods

Image and EELS data shown in Figures 1–6 and 8a was recorded on an FEI Titan 80–300 equipped with a spherical-aberration corrector for the probe-forming lens. The microscope was operated at an accelerating voltage of 300 kV. The probe convergence angle was ≈ 13.5 mrad; HAADF-STEM images were acquired with collection angles of ≈ (88 to 175) mrad or ≈ (112 to 168) mrad. CeO₂ nanocubes were synthesized using wet chemical methods, details can be found in previous reports (Johnston-Peck et al., 2016a; Johnston-Peck et al., 2016b). CeO₂ truncated octahedron and nanoparticles were acquired from Sigma-Aldrich and Strem Chemicals, respectively. Samples were deposited onto a carbon support film and plasma cleaned by a low energy RF plasma (Ar/O₂) to prevent the build-up of amorphous carbon during electron beam irradiation, as a carbon shell inhibits reduction of the CeO₂ particles (Johnston-Peck et al., 2016a).

Figure 1.

The ratio of the Ce M_4,5 peaks as a function of time as the dose rate was changed. Increases in the M₅/M₄ ratio reflect an increased Ce³⁺ concentration. When the dose rate is above the critical-dose rate threshold (red points) the particle reduces, conversely when the dose rate is below the critical-dose rate threshold (black points) the particle oxidizes. HAADF-STEM images of the particle before (b) and after (c) the exposure experiment in (a).

Figure 6.

Time series of a CeO₂ truncated octahedron oxidizing from CeO_2-x to CeO₂. Select frames are shown. The increase in intensity from (a) is shown (b-f) below the original image. Contrast changes appear first at tips. Dose rate: 4.54 *10⁵ e·nm⁻²·s⁻¹

Figure 8.

A representative HAADF-STEM image of the CeO₂ particles. Representative EEL spectra of the Ce M_4,5 edge for each gas condition. The fitted weight of Ce³⁺ and Ce⁴⁺ reference spectra are shown in red and blue underneath each spectrum.

For irradiation experiments, a 1024 × 1024 raster with a pixel dwell time of 0.8 μs was used. During the EEL spectral acquisition, the electron beam was scanned multiple times over the entire square raster pattern; accordingly, the measurement indicates the spatially-averaged oxidation state of the entire particle. Dose rates (e·nm⁻²·s⁻¹) were calculated by dividing the probe current by the area of the raster. This represents a time-averaged value because each point of the specimen is illuminated only once during each complete raster, not continuously. A value based on the size of the probe would be more representative, however an accurate measurement of the probe profile is challenging and during the experiments the probe profile changes as the current was adjusted. Cumulative doses (e·nm⁻²) were calculated by multiplying the dose rate by the exposure time. The electron probe currents were measured using the microscope’s fluorescent screen that had been previously calibrated using a pico-ammeter.

EELS data shown in Figures 7 and 8b was recorded on an FEI Titan ESTEM without aberration correction for the probe-forming optics. The microscope is equipped with a residual gas analyzer (RGA). All experiments were conducted using an accelerating voltage of 300 kV. The data shown in Figure 7 was collected with a convergence angle of ≈ 10 mrad. For irradiation experiments, a 1024 × 1024 raster with a per-pixel dwell time of 0.8 μs was used. During the EEL spectral acquisition, the electron beam scanned multiple times over the entire square raster pattern. The data shown in Figure 8b was collected with a convergence semi-angle of approximately ≈ 3 mrad and a EELS collection angle of ≈ 12.5 mrad and the dispersion was 0.05 eV/ch. A Gatan DualEELS spectrometer was used to acquire the data, to ensure a precise energy-loss scale the zero-loss region and the Ce core-loss region were both recorded. In post-processing, spectra were then aligned relative to the zero-loss peak. The electron probe currents were measured using a pico-ammeter attached to the drift-tube of the EEL spectrometer.

Figure 7.

White-line ratio of the Ce M_4,5 edge from a CeO₂ nanocube as a function of cumulative dose. Under vacuum the nanocube reduces. When oxygen is introduced raising the system pressure by over an order of magnitude, no detectable reduction occurs.

The white-line ratio of the Ce M₅ to the Ce M₄ peak was calculated by taking the second derivative of the spectra and integrating the positive portion of each edge. This was implemented in Digital Micrograph (Mitchell and Schaffer, 2005). The uncertainty of these values is defined as,

$$\delta W=W\sqrt{{\left(\frac{\sqrt{N_1}}{N_1}\right)}^2+{\left(\frac{\sqrt{N_2}}{N_2}\right)}^2}$$

where W is the white-line ratio and N₁ and N₂ are the integrated intensities of the Ce M_4,5 edges.

Quantification of the EELS data in Figure 8b, whose results are displayed Table 1 was performed by least squares fitting, as employed in EELSModel (Verbeeck and Van Aert, 2004). CeAlO₃:2% Ge and a large-grain CeO₂ powder were used as controls to provide the characteristic Ce³⁺ and Ce⁴⁺ spectra. To reduce differences due to sample thickness, the Fourier-Ratio deconvolution, as implemented in Digital Micrograph, was applied. Backgrounds were removed and deconvolution routines were applied to all spectra prior to fitting. The uncertainty of the values reported is the standard deviation of the measurements, where measurements were recorded from multiple regions of each sample.

Table 1.

Oxidation state of the ceria powder under different atmospheres

Gas	Pressure (Pa)	% Ce3+	σ
Oxygen	≈ 158	0.4	0.8
Oxygen	≈ 1	4.4	2.3
Vacuum	≈ 2*10−5	11.9	2.5

Simulated images were calculated by the multislice routine using QSTEM (Koch, 2002). Structural information including Debye-Waller factors were obtained from data published by Kümmerle and Heger (Kümmerle and Heger, 1999). Parameters of the simulation approximate the experimental conditions used here: 300 kV accelerating voltage, 0.05 mm Cs, 13.5 mrad convergence angle, 88 mrad inner collection angle. A total of 20 phonon configurations were calculated and a 0.1 nm Gaussian was applied to account for incoherent contributions.

3. Results and Discussion

3.1. Electron Beam-Sample Interactions

The electron beam can be used to reduce CeO₂ and measurements by EELS can observe the subsequent changes to the fine structure of the Ce M_4,5 edges (Garvie and Buseck, 1999). In Figure 1a, a particle was exposed to two different dose rates. When the dose rate is below a critical threshold, the particle will not reduce within the detection limits, or if it has already been reduced the particle will oxidize. When the particle is exposed to a dose rate above the critical-dose rate threshold the particle reduces as oxygen vacancies are created. This process is repeatable and the damage mechanisms active under these conditions do not significantly disrupt the Ce-sub lattice as the particle morphology does not significantly change after being exposed to the beam (Figure 1b and 1c). While no significant restructuring (e.g., formation of voids, protrusions, or gross changes in shape) of the particle morphology is perceptible, small changes in facet lengths were measured. Mobility of the surface atoms in CeO₂ nanomaterials under electron irradiation is consistent with previous reports (Bugnet et al., 2017; Möbus et al., 2011). Momentum transferred from the incident electrons to undercoordinated surface atoms likely imparts sufficient energy for surface atoms to diffuse (i.e., knock-on displacement) to more energetically favorable configurations thereby changing the facet sizes.

When the particle is oriented to a zone axis facilitating axial channeling of the electrons (orientations that support planar channeling may also enable this phenomenon but were not tested) additional information about the redox process is present in the ADF-STEM image contrast. Specifically, the image contrast becomes sensitive to regions with point defects (ordered oxygen vacancies would indicate a phase change and could identified using traditional techniques). This can be understood as atomic static displacements in proximity to oxygen vacancies affecting channeling of the incident electrons along the atomic columns while suppressing coherent scattering. The phenomenon of points defects influencing electron scattering has been covered in-depth elsewhere (Cowley, 1995; Grillo et al., 2008; Hall et al., 1966; Johnston-Peck et al., 2016b; Muller et al., 2004; Perovic et al., 1993; Rossouw et al., 1994). Practically this means under equivalent conditions a HAADF image of CeO_2-x (fluorite structure with oxygen vacancies or anion-deficient fluorite) will exhibit lower intensity (image grayscale values) than CeO₂ (fluorite structure without oxygen vacancies) thereby generating contrast. There is a detection limit dependent both on microscope settings and sample conditions and here a single point defect does not alter the electron scattering enough to generate perceptible contrast. Rather the presence of multiple defects per projected atomic column will be necessary to produce contrast.

This described contrast behavior is demonstrated in Figure 2 as the nanocube is reduced, changes in HAADF contrast with corresponding EELS measurements are shown. The nanocube is oriented to a <001> zone axis. From the image contrast, we observed the concentration of oxygen vacancies first increases adjacent to exposed surfaces and propagates to the particle core (Frame 251 to Frame 315). Note the nanocube face near the top of the image is in contact with another nanocube inhibiting oxygen desorption so the bottom half of the nanocube reduces more quickly than the top half. This leads to the asymmetric contrast behavior observed in the image series. The Ce white-line ratios increase, indicating reduction, coincident with the image contrast changes. As this process continues, the concentration of oxygen vacancies increases and anion vacancy ordering initiates as identified by the appearance of additional reflections in the fast Fourier transform (Frame 630). This results in the formation of the cubic sesquioxide phase (C-Ce₂O₃, bixbyite structure), first at regions adjacent to the surface and then in the core. The C-Ce₂O₃ assignment was consistent with the indexing results from two additional crystallographic orientations and EELS measurements indicating the oxidation state of the Ce ions were nominally Ce³⁺ (or slightly higher as C-Ce₂O₃ can form over a range of stoichiometries) (Adachi and Imanaka, 1998; Kümmerle and Heger, 1999). During this process the cation sub-lattice does not restructure and remains related (as what amounts to a distorted fcc lattice) to the original fluorite structure as illustrated in the unit cell models of Figure 3. From several particles analysis of FFTs comparing the lattice spacings of the {200}_CeO2 and {400}_Ce2O3 reflections measured an average expansion of ≈ 3.4%.

Figure 2.

A series of HAADF-STEM images with FFT of a CeO₂ nanocube being reduced to C-Ce₂O₃. As CeO_2-x forms contrasting regions can be observed in Frames 251, 275, 294, and 315 and are marked with dashed lines. In Frame 630 the onset of oxygen vacancy ordering is observed and C-Ce₂O₃ begins forming, arrows in the FFT indicate the presence of new reflections associated with the phase change to C-Ce₂O₃. Ce M_4,5 white-line ratio EELS measurements reported in the bottom left-hand corner, correlate the change in image contrast to a change in cerium oxidation state. Dose rates: Frame 26 – 3.24 * 10⁵ e·nm⁻²·s⁻¹; Frame 251, 275, 294, 315 – 5.56 * 10⁵ e·nm⁻²·s⁻¹; Frame 630 – 9.26 * 10⁵ e·nm⁻²·s⁻¹; Frame 881, 986 – 1.11 * 10⁶ e·nm⁻²·s^-1. The vertical streaking in the FFTs is due to scan noise.

Figure 3.

Unit cell models of C-Ce₂O₃ (a) and CeO₂ (b). Isolating a portion of the C-Ce₂O₃ unit cell highlights the similarities between the two phases (c). Simulated HAADF images of C-Ce₂O₃ (d) and CeO₂ (e) oriented to a <001> zone axis. The yellow and red markers indicate the position of Ce and O atomic columns, respectively.

After reducing the particle and as the electron dose rate decreases below a critical threshold the particle begins to oxidize, as the cubic sesquioxide (C-Ce₂O₃) phase is energetically unfavorable when compared to CeO₂ (Petit et al., 2005). As oxygen vacancies are annihilated the anion vacancy ordering in proximity to the nanocube surface is eliminated transitioning from C-Ce₂O₃ to CeO_2-x, as further time elapses the core converts to CeO_2-x. Next the image grayscale values of the surface regions increase contrasting with the core, indicating that the particle surface has reverted to approximately stoichiometric CeO₂ while the core remains CeO_2-x. The contrast between the core and surface eventually resides giving the nanocube a homogenous appearance indicating the entire nanocube has transformed back CeO₂. This process can be repeated many times with approximately the same behavior observed each time. The complete reduction and oxidation process of a CeO₂ nanocube can be viewed in movie “Cube_Redox” in the Supporting Information. The in-situ transformation to the C-Ce₂O₃ phase has been reported previously driven by electron beam reduction (Ding et al., 2016; Haigh et al., 2011; Johnston-Peck et al., 2016a; Sinclair et al., 2017) or a reducing gas environment (Crozier et al., 2008; Wang et al., 2009). We will focus on two parts of the oxidation and reduction process in greater detail, the formation of C-Ce₂O₃ and the oxidation process which occurs after reduction.

3.2. C-Ce₂O₃ Formation

Most recently the structure of C-Ce₂O₃ was refined using single crystal neutron scattering (Kümmerle and Heger, 1999). The unit cell is approximately twice the size of the fluorite structure and belongs to space group $Ia\overline{3}$. Oxygen vacancies align along <111> directions and result in displacement of adjacent Ce ions away from the vacancies and O ions toward the vacancies. Image simulation of the C-Ce₂O₃ and CeO₂ phases oriented to a <001> direction is shown in Figure 3. The striped appearance, due to the Ce ions at the 24d sites relaxing away from their original fcc sites, of the C-Ce₂O₃ phase in the simulated data is mimicked in experimental data shown in Figure 4c. As vacancies order during the transition between fluorite (CeO_2-x) and sesquioxide (C-Ce₂O₃) multiple domains were observed to nucleate. Relative differences in oxygen vacancy ordering with respect to the parent fluorite lattice create domains which can be related to one another through a translation or rotation, meaning that anti-phase boundaries and orientation variants are present within the particle. An example of a nanocube with different C-Ce₂O₃ domains is shown in Figure 4c. Repeated reduction-oxidation sequences of the same nanocube using the same dose rate did not produce identical C-Ce₂O₃ domain structures, rather the size and orientation of the C-Ce₂O₃ domains was different with each reduction (Supporting Information Figure S1). This suggests that under these conditions the nucleation and growth of C-Ce₂O₃ was not controlled by the applied dose rate, which influences rate of reduction, but other factors.

Figure 4.

HAADF-STEM image of a fluorite CeO₂ cube (a) taken at a dose rate of 2.74 *10⁵ e·nm⁻²·s⁻¹ (probe current of ≈ 30 pA). The FFT of the image is shown in (b). The same particle exposed to a higher dose rate of 1.82 *10⁶ e·nm⁻²·s⁻¹ (probe current of ≈ 200 pA) transforms into sesquioxide C-Ce₂O₃ (c). The FFT of the image is shown in (d) contains reflections not observed in (b). The C-Ce₂O₃ {200} and {020} reflections marked by red circles in (d) are masked and an inverse FFT is calculated (e) highlighting the C-Ce₂O₃ domain structure.

Oxygen vacancy ordering is a competition between the chemical potential of enthalpy and configurational entropy of oxygen vacancies. The abundance of free surfaces where oxygen vacancies are introduced and the stochastic nature of beam driven reduction creates a situation where fluctuations in vacancy concentrations develop. At critical oxygen vacancy concentrations C-Ce₂O₃ can form and grow. Because the spatial position of these critical concentrations is not controlled here and because C-Ce₂O₃ can only form and grow at critical concentrations, the domain structure does not appear identical with repeated reduction. In situations where the introduction and assembly of vacancies is more precise, the formation of C-Ce₂O₃ domains may show repeatable behavior. This may be achieved when oxygen vacancy injection is spatially controlled (e.g., by limiting the number of free surfaces) or creating a potential for vacancy diffusion through electrical bias. Strain in epitaxial films has been also suggested as a potential mechanism to dictate the domain structure (Sinclair et al., 2017).

3.3. Oxidation Behavior

Backgrounds gases in the microscope column oxidize the reduced ceria particles. The composition of background gases in our FEI Titan 80–300 is unknown because gas monitoring equipment (e.g., residual gas analyzer) is not present. In vacuum systems pumped by turbomolecular pumps it is reported that the concentration of molecular oxygen and carbon dioxide is low compared to that of water vapor (Postek, 1996). Data from a residual gas analyzer (RGA) attached to the ESTEM corroborates this result (Supporting Information Figure S2). Therefore, the oxidation process is assumed to be dominated by the absorption and dissociation of water vapor because water vapor will interact with nanoparticle surface at the highest frequency of all oxidizing species present, although molecular oxygen will also contribute.

As the nanocube oxidizes the phase transitions from C-Ce₂O₃ to CeO_2-x to CeO₂. The transitions initiate near surfaces and the core follows. In Figure 5 images from a nanocube oxidizing after being reduced by the electron beam are shown. At this point in the process CeO_2-x is present and the C-Ce₂O₃ phase is no longer detected. As time proceeds, the image intensity initially increases in the regions adjacent to the corners (Figure 5b and 5c) – which are terminated by {111} and {110} facets – reflecting further oxidation to CeO₂. This intensity increase is then observed in regions adjacent to faces (Figure 5d and 5e) – which are terminated by {100} facets – and then within the particle core (Figure 5f). A model and micrographs depicting the nanocube geometry and surface terminations are shown in Supporting Information Figure S3. To highlight these contrast changes, the difference in image intensity between the first image (Figure 5a) and each following image (Figure 5b–5f) is shown below each respective image. To do this the images were aligned using a rigid registration routine implemented in FIJI (Schindelin et al., 2012; Schneider et al., 2012; Thevenaz et al., 1998). To improve signal to noise and minimize the effect of non-linear distortions which are not corrected by rigid registration, the average intensity of three images (the image shown along with the preceding and following images of the time series) is calculated. This contrast behavior was similarly observed in subsequent redox cycles on the same nanocube (Supporting Information Figures S4 and S5) as well as in other nanocubes (Supporting Information Figures S6 and S7) suggesting this behavior is characteristic to nanocubes under these conditions.

Figure 5.

Time series of a CeO₂ nanocube oxidizing from CeO_2-x to CeO₂. Select frames are shown. The increase in intensity from (a) is shown (b-f) below the original image. Contrast changes appear first at corners. Dose rate: 3.70 *10⁵ e·nm⁻²·s⁻¹

This experiment was repeated on truncated octahedra and an example of reduction and oxidation can be viewed in movie “TruncatedOctahedron_Redox” in the Supporting Information. The truncated octahedra were oriented to <110> zone axes and further information on particle geometry and surface terminations are shown in Supporting Information Figure S8. In Figure 6 images from a truncated octahedron oxidizing from CeO_2-x to CeO₂ are shown. As time proceeds, the image intensity initially increases in the regions adjacent to the particle tips (Figure 6b and 6c) – which are terminated {100} facets – to regions adjacent to faces (Figure 6d and 6e) – which are terminated by {111} facets – and then the particle the core (Figure 6f). This contrast behavior was also reproducible (Supporting Information Figures S9 and S10).

In these image series because the image contrast reflects regions with different concentrations of oxygen vacancies, this means that regions adjacent to corners and tips are oxidizing more quickly than regions adjacent to faces. In nanocubes the corners are bound by {111} and {110} facets while in truncated octahedra the tips are bound by {100} facets. It can be hypothesized the cause of this observed behavior is influenced by geometric and facet effects. Near corners and tips of the nanocubes and truncated octahedra the portion of atoms at the surface increases per atomic column (Supporting Information Figure S11). As such, diffusion path lengths from the surface to oxygen vacancies in the bulk would be comparatively shorter in these regions and presumably would oxidize more quickly. Alternatively, differences between facets in the process of gas adsorption and subsequent absorption of oxygen into the lattice could also influence the oxidation process. On ceria water vapor adsorbs dissociatively as −OH and −H groups and these groups can either recombine and desorb as water or the hydroxyl group can cleave leading to the formation and desorption of molecular hydrogen.(Molinari et al., 2012) The latter will oxidize the reduced ceria while the former leaves the oxidation state unchanged. Differences in the thermodynamics or kinetics of this process as a function of surface termination would subsequently influence the rate of oxidation.

To deconvolute the geometric effect from the facet behavior, simulations and kinetic data comparing the behavior of cubes and truncated octahedra were used. In nanocubes the surface is predominately {100} terminated while in truncated octahedra the surfaces are predominately {111} terminated. {110} surfaces are present on the nanocubes but not the truncated octahedron so their contribution will be ignored from this point in the discussion. Therefore, if the relative activity of {111} and {100} surfaces differs, this will manifest as differences in the rate of oxidation displayed by nanocubes and truncated octahedra. Diffusion modeled by the finite element method (FEM) was compared to the spatially dependent oxidation behavior observed in HAADF-STEM (Figures S12 and S13). While kinetic data, comparing EELS oxidation state values as a function of time controlling for environment and electron beam exposure conditions, was analyzed to look for differences in redox rates as a function of particle geometry (Figures S14 and S15). Details on both these analyses are in the Supporting Information. The FEM diffusion models indicate that in nanocubes {100} surfaces alone do not account for the oxidation activity observed in HAADF images. While EELS data indicates that under for a given set of electron beam conditions nanocubes will reduce to a greater extent and at a faster rate than truncated octahedra. Also, truncated octahedra oxidize at a faster rate than nanocubes. This indicates that for these conditions {111} facets are more stable and have higher activity for the oxidation process.

These results of facet dependent behavior agree with prior DFT calculations. Molinari et al. indicated the heat of reduction for the {111} and {100} surfaces are 2.01 eV and 1.61 eV, respectively (Molinari et al., 2012). Surface studies comparing epitaxial thin films with {111} and {100} termination have reached mixed conclusions (Henderson et al., 2003; Mullins et al., 2012). Mullins et al. compared reduced {111} and {100} surfaces and demonstrated that reduced {111} surfaces tend to decompose adsorbed hydroxyls and form hydrogen thereby oxidizing while {100} surfaces promote water formation and no change of state.(Mullins et al., 2012) While Henderson et al. reports {111} surfaces are less favorable than non-{111} surfaces for oxidation by water (Henderson et al., 2003). Mullins et al. suggested differences originating in sample preparation and defect populations as a possible explanation for this discrepancy (Mullins et al., 2012). Using the methods presented here, STEM affords the possibility to directly relate specific structures with performance offering a potential route to elucidate the origin of such competing observations.

3.4. ESTEM Measurements

The oxidation of reduced ceria nanoparticles in the high vacuum environment of the microscope suggests that the particles have a low energy, equilibrium state that is kinetically accessible on the time scale of our experimental observations. Accordingly, to better understand how the environment of the analytical tool can influence the outcome of measurements we used ESTEM to understand how oxygen partial pressure influences the threshold for beam reduction and the oxidation state of the particle.

To investigate the role of atmosphere on the critical dose rate threshold a nanocube was exposed to different dose rates and pressures of oxygen. Under vacuum conditions the oxidation state of the cerium atoms was recorded as a function of cumulative dose for three different dose rates. Below the critical-dose rate threshold, no detectable change in oxidation state was observed. While above the critical dose rate, the particle was reduced (Supporting Information Figure S16) in accordance with prior results (Johnston-Peck et al., 2016a). The same experiment was then repeated with molecular oxygen introduced to the sample environment. Results in Figure 7 demonstrate the increase in oxygen pressure increased the critical dose rate threshold. A similar observation was recently made on CeO₂ films using TEM (Sinclair et al., 2017). This observation can be understood as the rate balance between the electron beam driven reduction process and the tendency of CeO_2-x to oxidize. As the pressure of molecular oxygen increases the frequency of collisions between the ceria particle and the oxidizing species increases. This increases the rate at which oxygen can adsorb and subsequently annihilate any oxygen vacancies that were created by electron beam-specimen interactions. Therefore, for a fixed dose rate, when the pressure of oxidizing species is low, vacancies are created at a rate faster than they are annihilated, the vacancies accumulate, and EELS can detect reduction of the cerium atoms. When the pressure increases, the vacancies are annihilated at the same approximate rate at which they are generated such that vacancies do not accumulate and EELS does not detect reduction of the cerium atoms.

In electron microscopy studies it is commonly reported that the surface of CeO₂ nanoparticles are reduced. This observation has been made with direct measurements using STEM-EELS and imaging techniques (Goris et al., 2014; Haigh et al., 2011; Johnston-Peck et al., 2016b; Lin et al., 2014; Spadaro et al., 2016; Turner et al., 2011; Wu et al., 2004), and can be inferred by observing the Ce³⁺ concentration in ceria particles is inversely related to diameter (Sims et al., 2018; Spadaro et al., 2016; Wu et al., 2004). To determine if the microscope environment influences the oxidation state of CeO₂ particles the oxygen chemical potential of the ESTEM was changed. Particles ≈ 5 nm in diameter, as shown in Figure 8a, were studied. The small size of the particles means the surface area to volume ratio is higher than the nanocubes and truncated octahedra and any surface effects will be greater. EEL spectra were recorded in vacuum and under two different pressures of oxygen. Representative spectra recorded from ensembles of particles under each condition are shown in Figure 8b and quantification of the Ce M_4,5 edge into Ce³⁺ and Ce⁴⁺ content is shown in Table 1. As the pressure of oxygen increases, the Ce³⁺ content decreases from 11.4% when measured in vacuum to 0.4% when measured with a pressure of ≈ 158 Pa oxygen.

To mitigate the reducing effect of the electron beam both dose rate and total dose were minimized by setting the beam current to ≈10 pA and recording both the individual core loss and low loss spectrum over a period of 1.1 s. Further, single particle studies were avoided and the electron beam was scanned over an area of 729 nm² during the spectral acquisition, so each spectrum reflects the oxidation state of many particles averaged. Control experiments in vacuum indicate that no additional reduction was observed using these conditions for exposures lasting longer than 1.1s. This demonstrates that the microscope environment influences the Ce³⁺ content, specifically the vacuum environment increases the Ce³⁺ content relevant to its state in environments with greater oxygen pressures. While the reduction of bulk CeO₂ is reported to occur at either lower pressures or higher temperatures than those used here (Bevan and Kordis, 1964; Ricken et al., 1984) this difference in behavior can be understood as the manifestations of the size of the particles and the gaseous environment. Both experimental studies and calculations indicate surface regions are more susceptible than bulk to reduction (Botu et al., 2014; Kim et al., 2004; Nolan et al., 2006). This is further exacerbated in redox (containing both reducing and oxidizing gases) environments (Botu et al., 2014), as is created by the background gases of the ESTEM (Supporting Information Figure S2). While these EELS measurements were not spatially resolved and therefore did not discriminate between the particle surface and core during acquisition presumably a majority of the Ce³⁺ atoms are located near the particle surface in accordance with the aforementioned findings (Goris et al., 2014; Haigh et al., 2011; Johnston-Peck et al., 2016b; Lin et al., 2014; Spadaro et al., 2016; Turner et al., 2011; Wu et al., 2004). Therefore, as oxygen is introduced, the oxygen chemical potential increases, stabilizing stoichiometric CeO₂ at the surface and reducing the Ce³⁺ content. Because the ≈ 5 nm particles studied have a large surface to volume ratio this behavior becomes pronounced even though the measurements were spatially averaged. Whereas spatially-averaged EELS measurements from large particles would be less sensitive to the surface region and this effect over the range of pressures considered here would presumably be negligible, as large particles have been shown to have only a few percent Ce³⁺ content as measured in STEM under vacuum conditions (Sims et al., 2018; Wu et al., 2004).

Because particle size and surface effects are often of interest, it is important to contextualize results to separate contributions due to the extrinsic, environmental factors from those that may be due to intrinsic effects. This is important when comparing and interpreting results that have been recorded from different instruments with different environments. As well as when addressing fundamental questions, such as, whether the lattice expansion inversely proportional to particle diameter observed in TEM measurements (Hailstone et al., 2009; Tsunekawa et al., 1999; Wu et al., 2004) is the result of the measurements being taken in a reducing environment (where lattice parameter dilates due phenomenon associated with reduction of Ce⁴⁺ to Ce³⁺) or others factors such as capillary pressure caused by surface stress (Diehm et al., 2012). Or how aliovalent dopants atoms spatially localize and influence the oxygen vacancy concentrations relative to bulk and surface (Collins et al., 2017). While the measurements here indicate that environment can influence the outcome of certain results, comprehensive measurements will be needed to address the aforementioned questions and quantitatively establish the spatial concentration of Ce³⁺ atoms in particles as a function of atmosphere, potentially extracting thermodynamic information.

Without the ability to accurately measure the makeup of the microscope environment certain data interpretation and comparison is hindered. As discussed by Miller and Crozier (Miller and Crozier, 2014), EELS and RGA have limitations when quantifying gas species. EELS lacks sensitivity at low pressures and limits what other measurements can be made simultaneously. RGAs may operate at different pressures than the sample environment and may not be attached in the direct vicinity of the sample. As a result, differences in gas diffusivities can skew quantification. Further, RGA measures the mass/charge ratio and that also poses challenges to accurate identification and quantification. Water vapor and oxygen sensors mounted in the immediate vicinity of the sample should be able to address some of these limitations and be of value for studies when knowing the environmental characteristics is critically important.

4. Conclusions

The methodology we have presented demonstrates the possibility of a single particle technique to provide insight on structural-property relationships. These techniques are useful because they can complement global measurements or first-principal calculations, which may not capture the complexity and variable nature of a dynamic system and its microstructure. Specifically, we have demonstrated that CeO₂ truncated octahedra terminated predominately with {111} surfaces compared to nanocubes terminated predominately with {100} surfaces are more resistant to reduction and once reduced will oxidize more quickly. These redox experiments were conducted in vacuum at room temperature using the electron beam to stimulate the reduction process. It is unknown if this reduction pathway is practically equivalent to that of a gaseous reducing environment and establishing equivalence between the two different reduction processes will be important if beam-mediated reduction is to be used as a stimulus for applied studies.

Conducting measurements like these under controlled temperature and gas environments may be able to provide additional insight into the catalytic behavior of CeO₂ as a function of structural parameters. Moreover, the HAADF-STEM image contrast mechanism responsible for the sensitivity to the anion-deficient fluorite phase is not necessarily limited to applications in the Ce-O system or even oxides. Point defects that locally distort the surrounding crystal lattice will alter the electron scattering behavior and potentially generate image contrast. This means that it may also be possible to use these techniques to monitor other reactions such as intercalcation. However, it is prudent to use independent verification, either through calculations or other measurement techniques, to verify that the physical change in the structure is associated with the observed contrast change.

ESTEM measurements link oxygen chemical potential of the microscope environment and the results of oxidation state measurements in ceria particles. Under certain conditions instrumental influence may become an important consideration for comparing results acquired from different platforms, as well as attempting to interpret functional properties and establish structure-property relationships. This highlights the importance of tools and techniques that accurately quantify the composition of the instrument environment.