Structure and phase stability of nanocrystalline Ce_1−x Ln _x O_2−x/2−δ (Ln = Yb, Lu) in oxidizing and reducing atmosphere

XRD patterns of the samples heated at 800 °C and 950 °C in O₂ or H₂ for 3 h were similar and revealed a continuous transformation of the structure from fluorite (Fmm) (F-type) to bixbyite (Ia-3) (C-type). Figure 1A shows as an example XRD patterns of Ce_{1− x} Lu _x O_{2− x/2−δ} heated in H₂ at 950 °C. F–C phase change, monitored as appearance of a weak (112) peak characteristic for C-type structure at 2Θ ~ 20⁰, occurred in this samples at x = ~0.5, i.e., at lower Lu content than in oxygen (x = ~0.6) (Małecka et al. 2008a). This shift is due to lowered stability of highly doped fluorite lattice under reducing conditions (see below). For Ce_{1− x} Yb _x O_{2− x/2−δ}, F–C transformation occurred at x = ~0.5 in both atmospheres. The XRD patterns of pure CeO₂, Lu₂O₃, and Yb₂O₃ fitted well the reference data: ICDD PDF Nr. 00-034-0394, 00-012-0728 and 04-001-2438, respectively. Addition of Lu and Yb decreased the cell parameter of parent cubic CeO₂ over the whole x range (for bixbyite type structure half of the cell parameter was taken), as expected from the effective ionic radii (Ce⁴⁺ = 0.0870 nm, Lu³⁺ = 0.0861 nm, Yb³⁺ = 0.0868 nm,—for coordination VI (Shannon 1976)) (Fig. 1B, C) and the fact that oxygen vacancies were formed in the ceria lattice upon Ce⁴⁺−Ln⁺³ substitution (one vacancy for two ions substituted). The formation of oxygen vacancies is necessary to ensure the electrical neutrality of the oxide. Hong and Virkar (1995), introduced a concept of an oxide vacancy radius, and the estimated oxygen vacancy radius was 0.1164 nm for trivalent, rare-earth oxide doped ceria, significantly less than the radius of O²⁻ ion (0.1400 nm Shannon 1976). For both dopants, deviations from Vegard’s rule were observed. In Ce_{1− x} Lu _x O_{2− x/2−δ}, the change of cell parameter with Lu content is slow and linear for x < ~0.35, but then the slope becomes steeper. This observation agrees well with dependence of the cell parameter on composition reported by Kimmel et al. 2006 at similar temperature and time of heating (900 °C/3 h) for Y-doped ceria. The authors assigned the deviations from linearity to coalescence of oxygen vacancies, which grows up with increasing dopant content (Bae et al. 2004). For x > ~0.5, where C-type structure occurs, the slope is higher again because bixbyite structure is less prone to doping with aliovalent ions. Lu³⁺ → Ce⁴⁺ substitution requires introduction of large O²⁻ ion into interstitial position, causing the lattice expansion. For Ce_{1− x} Yb _x O_{2− x/2−δ}, only one inflection point at x ~ 0.5 was observed in both atmospheres (Fig. 1C). The change in the slope of the plots is correlated with the F → C transformation of the structure.

Mean crystallite size and average maximum strain for the mixed oxides were calculated from XRD patterns with the profile fitting procedure (Roisnel and Rodriguez-Carvajal 2000). Results obtained for Ce_{1− x} Lu _x O_{2− x/2−δ} and Ce_{1− x} Yb _x O_{2− x/2−δ} heated at 950 °C in O₂ and H₂ are presented in Fig. 2A and B, respectively. Figure 2 also contains volume weighted mean crystallite sizes (Σn _i d _i ⁴/Σn _i d _i ³) calculated for the same samples from TEM images. The volume weighted mean was used instead of simple arithmetic mean to make it comparable with the mean crystallite sizes calculated from XRD data (Matyi et al. 1987). It appears that there is qualitative agreement between both sets of mean crystallite size data, indicating that there is no additional broadening XRD reflections due to contingent phase separation. For both mixed oxides strong dependence of crystallite size on Ln content (x) was found, which to some extent was affected by the heating atmosphere. In O₂ even small Ln addition (x ≤ 0.1) caused very strong inhibiting effect on crystallite growth, which saturated for x > 0.3. In samples heated in H₂, the mean crystallite sizes were bigger for small dopings but then became smaller for x > 0.3. The behavior observed in Fig. 2 agrees well with literature data for similar CeO₂–Ln₂O₃ systems heated in oxidizing atmosphere (Luo et al. 2006; Małecka et al. 2007). The possible explanation of the strong inhibiting effect of Ln addition on the crystallite growth could be segregation of the dopant ions at the surface of ceria crystallites reported for La (Belliere et al. 2006) and Pr (Luo et al. 2006) doped ceria. Data on crystallite growth in ceria-based nanocrystalline oxides heated in a reducing atmosphere are scarce. For Ce_{1− x} Zr _x O_2−δ strong inhibiting effect of Zr on ceria sintering, similar to that in oxygen was reported (Zhang et al. 2006).

Observed relation between the mean crystallite size and Lncontent correlated well with the data describing the average maximum strain (Fig. 2). For Ln content x < 0.3, the strain was bigger in the samples heated in O₂ than in H₂ but for higher x the opposite was true. Higher strain in low doped samples heated in O₂ may be related to the presence of thicker segregated layer enriched in Ln³⁺ ions. Further steep increase of strain with Ln content in the samples heated in H₂ is caused by creation of additional oxygen vacancies due to Ce⁴⁺ – Ce³⁺ reduction. At x ≅ 0.5 the transformation to C-type structure stops this process. It is seen, however, that for very high Ln content (Ln₂O₃ doped with Ce) the strain increases very steeply with doping. It indicates that C-type bixbyite lattice is much less prone to doping with aliovalent ions than fluorite lattice.

Extensive crystal growth was observed for all samples heated at 1100 °C in reducing and oxidizing atmosphere, irrespective of additive content. Due to limitations of experimental methods applied, we could not precisely quantify differences in crystallite size. Estimations by XRD performed for selected samples (where no phase separation occurred) confirmed the tendency seen in Fig. 2 that Lu or Yb addition decreases the mean crystallite size. Moreover, for high Ln dopant content mean crystallite size after treatment in H₂ was smaller than after treatment in air. This observation has been confirmed by EBSD (see below).

Phase separation was observed in Ce_{1− x} Lu _x O_{2− x/2−δ} heated at 1100 °C for 3 h in H₂ (Fig. 3A). As in oxygen (Małecka et al. 2008a), two main phases, F-type and C-type, and additionally a third C^#-type phase with bixbyite structure were observed (Fig. 3B). Reducing atmosphere promotes however the phase separation. It is seen clearly in the inset to Fig. 3A, which compares the XRD diagrams of the Ce_0.6Lu_0.4O_1.8−δ sample heated in H₂ and in O₂. The multi-phase region observed in a range 0.17 < x < 0.7 in the samples treated in H₂ atmosphere is broader than that in oxidizing one (0.35 < x < 0.7). The reason could be reduction of Ce⁴⁺ to Ce³⁺ in the mixed oxide, which generates additional oxygen vacancies and destabilizes fluorite structure. Contrary to what occurred in oxygen, the intensity of the XRD reflections of the intermediate C^# phase increased with increasing heat treatment time at 1100 °C. Besides above-mentioned phases, perovskite-type CeLuO₃ (Ito et al. 2001) was also found in hydrogen for narrow composition range 0.4 < x < 0.5. Analysis of the variation of the lattice parameter with Lu content indicates that at 1100 °C up to 17 mol% Lu dissolves in CeO₂ and the solubility of Ce in LuO_1.5 is 20 mol%. The solubility limit of Lu in CeO ₂ is narrower than 35 mol% observed in air at the same temperature (Małecka et al. 2008a). To our knowledge, there is no information in literature about structure stability of Lu-doped ceria in hydrogen.

Weak ~0.76 nm fringes were observed in HRTEM images (or more clearly in corresponding FFT patterns) and SAED patterns of Ce_0.5Lu_0.5O_1.75−δ heated at 1100 °C for 3 h in hydrogen (Fig. 4). Similar effect was found for samples heated in the air (Małecka et al. 2008a). The fringes could be assigned to superstructure ¼{220} reflections occurring due to doubling of the fluorite cell as a result of oxygen vacancy ordering (Ou et al. 2006), or due to modulation (commensurate) of F-type structure with a periodicity of 2a_F reported for CeO₂–Y₂O₃ system (Wallenberg et al. 1989). We think that the second explanation is more plausible because of the character of SAED patterns, where sharp, strong Bragg reflections due to F-type structure are observed together with sharp, low intensity satellite reflections. Particles exhibiting superstructure reflections belong to the C^# phase identified in XRD patterns (Fig. 4). We think that true structure model of the C^# phase may involve complex oxygen vacancy ordering, similar to that proposed by Ou et al. (2008).

Recently, it has been shown that exposure of CeO ₂ to intense electron beam in TEM causes partial reduction of Ce⁴⁺ to Ce³⁺ (Garvie and Buseck 1999) with formation and ordering of oxygen vacancies which results in superstructure reflections in SAED pattern (Akita et al. 2006). To exclude the possible effect of electron beam on observation of superstructure reflections in our study of Ce_{1− x} Ln _x O_{2− x/2−δ} samples, a blank experiment was performed. Figure 5 shows SAED patterns of CeO₂ crystallite taken after various exposure times to electron beam under normal working conditions. It appears that very weak additional satellite reflections became visible only after 30 min exposure while 60 min exposure time was necessary to produce well-developed superstructure reflections. Since the exposure time necessary for focussing and taking image is several minutes, we may exclude the possibility of decisive role of electron beam and/or vacuum conditions on superstructure formation in our samples. The much weaker beam effects in our case as compared with Akita et al. (2006) may be due to lower acceleration voltage used (200 kV instead of 300 kV) and smaller beam intensity (LaB₆ gun instead of FEG).

No F–C phase separation was observed for Ce_{1− x} Yb _x O_{2− x/2−δ} heated at 1100 °C in air and in hydrogen (Fig. 3C). This is an important difference relative to what occurred for Lu-doped ceria, where extensive phase separation occurred in both atmospheres at 1100 °C. However, as in case of Lu-doped ceria, the F-type structure modulation with a periodicity of 2a_F was observed in SAED and HRTEM images for samples heated at 1100 °C in both atmospheres. Presented results for Ce_{1− x} Yb _x O_{2− x/2−δ} heated in oxidizing atmosphere contradict those presented in Chavan and Tyagi (2005), Mandal et al. (2007), and Bevan and Summerville (1979) where phase separation into coexisting F-type and C-type phases over wide composition range 0.5 < x < 0.9 was reported. We think that the apparent difference is due to method of the sample preparation and temperatures (>1400 °C) used in Mandal et al. (2007), and Bevan and Summerville (1979). No data are available in literature on high temperature phase stability of Ce_{1− x} Yb _x O_{2− x/2−δ} in reducing atmosphere, but formation of the perovskite-type CeYbO₃ was reported (Ito et al. 2001). The conditions necessary for CeYbO₃ growth were much more severe than for CeLuO₃ (Ito et al. 2001), which explains why the former compound was not detected in our samples.

Differences in structure stability of Ce_{1− x} Yb _x O_{2− x/2−δ} and Ce_{1− x} Lu _x O_{2− x/2−δ} at high temperatures are clearly shown by the EBSD results (see maps in Figs. 6 and 7). Figure 6 shows an EBSD orientation map obtained for Ce_0.5Yb_0.5O_1.75−δ heated at 1100 °C for 45 h in air. The sample contains crystallites exhibiting fluorite type structure with mean size of 100 nm with a random distribution of orientations. Because the spatial resolution was ~25 nm all the grains from the map shown in Fig. 6 with an equivalent diameter smaller than 75 nm were not taken into account when the average grain size was calculated. This explains why the average grain size obtained from the EBSD data is significantly bigger than that estimated from XRD (32 nm).

EBSD analysis done on the Ce_0.5Yb_0.5O_1.75−δ sample heated at 1100 °C for 3 h in H₂ showed a similar microstructure but with finer crystallites compared with the sample treated in air for 45 h. Because the crystallite mean size was only slightly higher than the interaction volume (i.e., spatial resolution of EBSD) no reliable data could be acquired for calculating the mean grain size. XRD estimated this value as 13 nm.

Totally different picture emerges for the Ce_0.5Lu_0.5O_1.75−δ sample heated at 1100 °C for 3 h in H₂ (Fig. 7). In addition to relatively small crystallites with fluorite structure and average size of 470 nm, there are very large, micrometer size crystallites of second phase identified as orthorhombic, perovskite type CeLuO₃. This corresponds well with XRD results presented above, but raises a question on the possible reason of such drastic differences between Lu, and Yb-doped samples.

In most Ce_{1− x} Ln _x O_{2− x/2−δ} systems (Ln = Nd (Nitani et al. 2004; Chavan et al. 2005), Sm (Nitani et al. 2004), Yb (Mandal et al. 2007), Tm (Mandal et al. 2007)) reported in literature high temperature (>1200 °C) annealing in oxidizing atmosphere caused phase separation for x values close to 0.5 (Longo and Podda 1981; Nitani et al. 2004), though no phase separation was observed for Eu- (Mandal et al. 2006) or Gd-doped ceria (Grover and Tyagi 2004) even at very high temperatures (1400 °C). For CeO₂–Ln₂O₃ systems, where Ln is a heavy lanthanide (Dy, Tm, Yb) the tendency to phase separation in oxidizing atmosphere increases with decreasing radii of Ln³⁺ ion (Mandal et al. 2007; Bevan and Summerville 1979). The reason is decreasing lattice parameter of C-type Ln₂O₃ oxide leaving less free space for additional, interstitial O²⁻ ions that must be incorporated due to Ln³⁺ → Ce⁴⁺ substitution. This also explains why we observed phase separation in Ce_{1− x} Lu _x O_{2− x/2−δ} already at 1100 °C.

Much less is known on the phase stability of lanthanide-doped ceria in reducing atmosphere. Wang et al. (2004) studied interaction of nanocrystalline, fluorite type Ce_{1− x} Tb _x O_y (x ≤ 0.5) with hydrogen during heating up to 900 °C using in situ XRD. For x = 0.33 and 0.5, the authors observed phase separation with formation of bixbyite type Tb₂O₃. The process has been accounted for with the change of oxidation state of lanthanides ions from Ce⁴⁺/Tb⁴⁺ to Ce³⁺/Tb³⁺.

In Ce_{1− x} Ln _x O_{2− x/2−δ} samples heated at high temperatures, especially in hydrogen there is uncertainty concerning the valency of minority (dopant) lanthanide ions. Lu is not expected to appear in the valency other than 3+, but for Ce and Yb reduction from Ce⁴⁺ to Ce³⁺ and from Yb³⁺ to Yb²⁺ may occur. To get some insight into this, we performed XAS (X-ray absorption) and magnetic measurements for selected samples: Ce_0.1Lu_0.9O_1.55−δ and Ce_0.95Yb_0.05O_1.975−δ heated at 1100 °C in air or in hydrogen.

In Ce_0.1Lu_0.9O_1.55−δ heated in air XANES spectra collected at various temperatures from 5 to 283 K were practically identical with that of CeO₂ standard (Fig. 8) indicating the absence of any significant amount of Ce³⁺ ions. As displayed in Fig. 9, the sample is diamagnetic at room and elevated temperatures with the molar magnetic susceptibility of about −3.8 × 10^–4 emu/mol per Ce atom. A small upturn of the susceptibility observed at low temperatures arises likely due to the presence of very small amount of paramagnetic Ce³⁺ ions at the surface of nanocrystalline particles. This presumption is supported by the shape of the field variation of the magnetization taken at T = 1.71 K (see the inset to Fig. 9) being characteristic of weak diamagnets contaminated by minute amounts of paramagnetic impurities. Heating of Ce_0.1Lu_0.9O_1.55−δ in hydrogen at 1100 °C caused partial reduction of Ce⁴⁺ into Ce³⁺. XANES spectra collected at various temperatures (Fig. 10) clearly show a shift of the absorption edge toward lower energies indicating the presence of Ce³⁺. Figure 11 presents results of magnetic measurements for this sample, revealing a distinctly different behavior from that of the specimen treated in air. In contrast to the latter, the sample treated in hydrogen is paramagnetic at room temperature and shows a strong temperature dependence of the magnetic susceptibility. In the entire temperature range studied, the χ(T) curve can be well approximated by the formula\( \chi (T) = \chi_{\text{TIP}} + \frac{C}{T} \), where the first term represents temperature independent paramagnetism due to Ce⁴⁺ ions and the second one accounts for the presence of Ce³⁺ ions. Least squares fit of this expression to the experimental data yielded χ_TIP = 1.78 × 10⁻⁴ emu/mol_Ce and the parameter C = 0.0943 (K emu)/mol_Ce. Assuming that the Curie term results only from stable trivalent Ce atoms, one may estimate the content of such atoms to be about 12% of all the Ce atoms present in the sample studied. In line with this finding, the Curie paramagnetism of Ce³⁺ ions dominates the low-temperature behavior of the magnetization of the Ce_0.1Lu_0.9O_1.55−δ sample heated in hydrogen, as manifested by curved character of the field variation σ(B) (cf. the inset to Fig. 11).

These results answer the interesting question about the valency of Ce as dopant into bixbyite structure of Lu₂O₃. There are two possible scenarios. In the first, Ce as 3+ ion just replaces Lu³⁺ in the lattice, whereas in the second Ce⁴⁺ substitutes for Lu³⁺ and additionally one O²⁻ ion per two such events enters the lattice. In the former case, the strong lattice distortion is inevitable due to large differences in effective ion radii (Ce³⁺: 0.101 nm and Lu³⁺: 0.0861 nm). In the latter case, the difference in cation radii is smaller (Ce⁴⁺ = 0.0870 nm and Lu³⁺: 0.0861 nm), but large oxygen anion must occupy interstitial site. The most probable one is 16c site in Ia-3 (206) space group of Lu₂O₃ (Ou et al. 2008). It appears that the latter possibility is realized.

For the Ce_0.95Yb_0.05O_1.975−δ sample, the valency of Yb was studied by measurements of Yb L3 absorption edge. Figure 12 shows XANES spectra of the sample heated in air and in hydrogen. Comparison with the spectrum of Yb₂O₃ standard suggests that ytterbium exists as Yb³⁺ with no evidence of Yb²⁺. This result agrees with observed decrease of the lattice parameter of CeO₂ with Yb doping (Fig. 1C and 3C) since Ce⁴⁺ ions (0.0870 nm) are replaced with smaller Yb³⁺ (0.0868 nm); ionic radius of Yb²⁺: 0.102 nm is bigger than that of Ce⁴⁺ (Shannon 1976).