The reactive element effect of ceria particle dispersion on alumina growth: A model based on microstructural observations

The oxidation kinetics of alumina-forming metals can be affected by adding a small amount of a reactive (normally rare earth) element oxide (RExOy) and the segregation of the reactive element (RE) ions to the growing alumina grain boundaries (GBs) has been considered as a responsible reason. However, this interpretation remains a controversial issue as to how RE ions are produced by RExOy which is thermodynamically and chemically stable in metals. The question is answered by a model that is based on transmission electron microscopy (TEM) investigation of a CeO2-dispersed nickel aluminide oxidized in air at 1100 °C. The CeO2 dispersion is incorporated into the alumina scale by the inward growth of inner α-Al2O3, where it partially dissolves producing tetravalent Ce cations which then transform to trivalent cations by trapping electrons. The trivalent cations segregate to the α-Al2O3 GBs and diffuse outward along first the GBs and later the twin boundaries (TBs) in the outer γ-Al2O3 layer, being precipitated as Ce2O3 particles near surface.

Scientific RepoRts | 6:29593 | DOI: 10.1038/srep29593 GB diffusivity of aluminum cations (D GB Al ) and oxygen anions (D GB O ) by increasing the number of anion and cation vacancies. More recently, some investigators attributed the REE on the alumina growth to a modification of the electronic structure of alumina with GB donor and acceptor states to the extent that Al ionization at the interface is decreased 30,36 .
The RE segregation model has been supported by many experimental observations of RE segregation at the alumina GBs by means of TEM in a combination of X-ray energy-dispersve spectroscopy (EDS) 3,4,6,7,10,12,15,17,32 . In addition to the GB segregation, RE has been found to occur as RE x O y particles on the alumina scale surface 15,17 . On its basis, a dynamic segregation theory (DST) has been proposed, in which the RE segregants at the GBs are not static; they can transport outward along the GBs driven by the oxygen potential (i.e., oxygen chemical activity) gradient across a growing alumina scale and their high affinity for oxygen 15 .
Many literatures 15,17,20,24,25,28 also reported that addition of the RE x O y dispersions in an alloy plays a similar role as RE in decreasing the oxidation rate of alumina-forming metals. This effectiveness REE of the RE x O y on the alumina TGO growth is firstly attributed to the dissolution of RE x O y which is proposed to occur under the oxygen pressure gradient across the metal-oxide-gas system 15,24 , producing RE atoms to segregate to the oxide/ metal interface. However, there has been no experimental evidence for such dissolution (or dissociation) and the latter also appears to have no thermodynamic justification, because RE x O y (more stable than Al 2 O 3 ) have a very high thermal stability. The oxygen pressure in the metals, which decreases from the dissociation pressure of Al 2 O 3 at the interface to some low values (depending on the oxygen solubility and diffusivity) at some distance from the  interface, is normally not low enough to drive the RE x O y dispersions to dissociate and release RE atoms which can segregate to the interface. In view of this, the concerned REE exerted by RE x O y appears not be explained fully by DST, although it cannot be explained appropriately by PIM. A much more likelihood that the RE x O y dispersions  exert the REE is associated with the dissolution of RE x O y dispersions upon incorporation into the alumina scales, as described simply in 3,6 . This raises a question on how the oxide dispersions enter the alumina TGO.
Recently, an interpretation on the effect of the ceria particle dispersion on the growth process of alumina scale on an alumina-forming aluminide was proposed 32 . It highlights two points. First, the incorporation of the ceria particles into the alumina TGO results from inward growth of the inner part of the alumina in α phase. Second, the ceria particles do not exhibit the REE until they have been incorporated into the alumina scale, where they may dissolve to some extent to produce the cerium ions that can segregate to the alumina GBs and thus suppress the outward diffusion of Al cations along the short-circuit paths for the TGO growth. However, this interpretation is still lack of sufficient evidences. We further characterized the microstructure of the alumina TGO on the ceria-dispersed aluminide and traced cerium either in its elementary or oxide form from the metal to the TGO. There are new observations: (i) no dissolution of original CeO 2 particles in the metal during oxidation, (ii) the identification that the TGO scale is composed of the outward growing γ -Al 2 O 3 and inward growing α -Al 2 O 3 and the CeO 2 particles in the metal can be swept over by inward growing α -Al 2 O 3 , (iii) the detection of cerium ions segregated to the GBs of the inner α -Al 2 O 3 , and (iv) precipitation of novel Ce 2 O 3 particles along the twin boundaries in the outer γ -Al 2 O 3 . On these bases, we propose a model in the present work, which shows a scenario of a dynamic evolution of the ceria particles in the metal during oxidation. The model is helpful for better understanding of not only the REE on the alumina growth on the ceria-dispersed aluminide but also the concerned REE of the RE x O y dispersions in other alumina formers. In addition, it is useful for getting insight into the alumina growth on the metals alloyed with REs, which can be preferentially oxidized into RE x O y particles in the metals because the alloyed amounts of REs normally exceed their low solubility limits [1][2][3]7,23,25,26,51 .

Results
Ceria dispersion in aluminide before oxidation. The CeO 2 particles used, which have a typical CaF 2type crystal structure (space group:Fm m 3 ) with the lattice spacing of d (200) = 2.7 Å and d (111) = 3.1 Å on a basis of HRTEM investigation and FFT diffraction (see supplementary material 1), are in a size range of 15-30 nm. The particles were co-deposited with Ni, forming a ~35 μ m-thick Ni-based composite film, in which the CeO 2 particles with the content of 3.5 wt.% are in general uniformly distributed, as viewed previously by using SEM 37 . After aluminizing, the Ni-CeO 2 composite film was converted into a ~43 μ m-thick alumina-forming δ -Ni 2 Al 3 28 . The CeO 2 particles are uniformly dispersed throughout the thickness of the aluminide on a basis of the electron probe microanalysis (EPMA) 37 . Ceria dispersion in aluminide and its evolution in alumina scale after oxidation. Figure 1 shows the cross-sectioned aluminide for 30 min oxidation at 1100 °C. The aluminide forms an alumina scale. The inward growth of the alumina, as suggested by the non-planar interface, leads the metal to be either partially (as indicated by 1) or fully (as indicated by 2) enclosed by the oxide. The alumina scale viewed under TEM as seen in Fig. 2(a) displays a double-layered structure. The outer needle-like platelets, which exhibit a high density of lamellar nanotwins with coherent boundaries when tilted to the [110] zone axis, are γ -Al 2 O 3 as identified by HRTEM image and the corresponding SAED pattern in Fig. 2(b). Similar lamellar-twined structure has been observed in deformed fcc γ -grains of a single-phased austenitic steels 52 and in Au nanocrystal-seeded Si and Ge nanowires 53 . This suggests that the growth of γ -Al 2 O 3 platelets is controlled by outward diffusion of aluminum cations along the twin boundaries (TBs) in the [112] orientation. The oxide of the inner layer is α -Al 2 O 3 as unveiled in Fig. 2(c). Between the γ -Al 2 O 3 layer and the α -Al 2 O 3 layer appears a γ -and α -mixed area as seen in Fig. 3. The γ -and α -Al 2 O 3 grains are the smallest in the alumina scale and Ce-rich oxide nanoparticles (see the Ce X-ray mapping) can be sometimes observed. Similar Ce-rich oxide particles occur in the inner α -Al 2 O 3 layer. They are CeO 2 as identified in Fig. 4(a), displaying the shape and CaF 2 -type crystal structure similar to the original CeO 2 . The outer highly-twined γ -Al 2 O 3 layer is also doped cerium-rich oxide particles, which as circled in the TEM BF image in Fig. 4(b) are seemingly elongated along the twinning orientation, with respect to the particle shape of the original CeO 2 . The 3.3 Å lattice spacing of both (222) and (222) planes and (Mn 0.5 Fe 0.5 ) 2 O 3 (space group:Ia3)-similar structure (see HRTEM image and FFT diffraction in Fig. 4(b)) ascertain the particles as new Ce 2 O 3 rather than the original CeO 2 .
The aluminide has been degraded from δ -Ni 2 Al 3 into β -NiAl due to the aluminum consumption by oxidation, as shown in the inserted SAED pattern in Fig. 5(a). The aluminide contains the nano-dispersions, which have been characterized to be original CeO 2 . No Ce was acquired around the CeO 2 particles by the EDS detector with an incident beam spot size of 1.5 nm. Figure 5(b) shows an EDS result of a specific spot between two close CeO 2 particles at the GBs, showing no acquisition of Ce atoms there.
The particles of the CeO 2 , as inert oxide in the metal, actually act as the immobile markers for the direction of the alumina growth. They occur in the fine-grained γ -and α -mixed area (Fig. 3), suggesting that the area corresponds to the surface zone of the original aluminide. The CeO 2 particles in the α -Al 2 O 3 layer arises from the inward growth of the oxide. To further clarify this, the alumina scale formed only for 5 min has been observed. A CeO 2 particle which has been swept over by inward growing alumina is clearly seen in Fig. 6. In contrast, the Ce 2 O 3 particles in the outer γ -Al 2 O 3 layer should be newly precipitated. They can form, suggesting that there exist sufficient Ce cations which can be migrated from the inner α -Al 2 O 3 layer. The larger-sized ions as the Ce cations here doped in the alumina TGOs are easily segregated to and then migrate outward along the GBs 15,17 . As shown in Fig. 7, the Ce segregation at the α -Al 2 O 3 GBs can be clearly seen by using HADDF-STEM. The HADDF image presents the Ce segregated GBs presents as the lines with a light contrast similar to that of the CeO 2 particles (as arrowed in the BF image), because Ce has a higher atomic number than Al. The EDS analysis indicates the GBs containing a mean content of ~0.4 at.% Ce. The Ce at the oxide GBs unlikely originates from its atoms in the aluminide, because the latter have not been acquired in the metal (Fig. 5). It convincingly arises from the segregation of cerium cations, produced by partial dissolution of the CeO 2 particles incorporated in the α -Al 2 O 3 layer. The dissolved Ce cations also experience the charge transformation from tetravalent to trivalent in the alumina scale, on a basis of the precipitation of Ce 2 O 3 rather than original CeO 2 .
In sum, the TEM work presents several observations: (i) the aluminide during oxidation forms an alumina scale being composed of an inner α -Al 2 O 3 layer and an outer γ -Al 2 O 3 layer; (ii) the CeO 2 dispersions are incorporated into the α -Al 2 O 3 layer as the result of its inward growth; (iii) Ce ions segregates to the alumina GBs, and (iv) novel Ce 2 O 3 particles are precipitated in the near surface of the γ -Al 2 O 3 platelets.

Discussion
The precipitation of new Ce 2 O 3 in the outer γ -Al 2 O 3 platelets demonstrates a series of evolution of the original CeO 2 particles after they have been incorporated into the growing alumina, including their partial dissolution (since no evidence for such dissolution could be acquired in the aluminide (Fig. 5)), tetravalent-to-trivalent charge transformation and outward migration of the dissolved Ce cations. To unveil the dynamic evolution of the CeO 2 dispersion in the aluminide, a model is schematically illustrated in Fig. 8 based on the TEM observations and interpreted below. The highly-twined γ -Al 2 O 3 grains grow outward quickly on the aluminide at the onset of oxidation, and  the aluminide (Step II). α -Al 2 O 3 exhibits an n-type behavior with the principal defect of oxygen vacancy •• V O or free electron e′ (Vink-Kröger's notation). The CeO 2 partially dissolves into the α -Al 2 O 3 lattice through the reaction In combination of Eqs (1) and (2), the CeO 2 particle dissolution in the alumina lattice can be expressed by The reduction of CeO 2 to Ce 2 O 3 has been reported in the high temperature sintering of fine CeO 2 particles 54,55 . The Ce Al X (1.02 Å with the coordination number of 6 56 ) is larger than • Ce Al (0.87 Å with the same coordination number) in the ion size. Larger Ce Al X in the α -Al 2 O 3 grains yields higher lattice misfit microstrain, which drives the trivalent cations to segregate to the α -Al 2 O 3 GBs (Step III). Then, the segregated cations migrate from the α -Al 2 O 3 layer to the γ -Al 2 O 3 layer along the GBs in the α -Al 2 O 3 and the lamellar TBs in the γ -Al 2 O 3 platelets, under the driving force of the oxygen potential gradient across the oxide 15,17 . γ -Al 2 O 3 is a p-type oxide with the principal defect of Al vacancy ‴ V Al and electron hole • h . The TBs, although they are coherent, contain steps and kinks which can serve as the sinks for vacancy (as having been reported in 57,58 ) like ‴ V Al here. Steps and kinks in the lamellar TBs in the γ -Al 2 O 3 can trap ‴ V Al . Once Ce Al X and ‴ V Al are both oversaturated there, Ce 2 O 3 is precipitated (Step IV) through the reaction below, The Ce Al X cations prefer to diffuse outward along the TBs, causing the Ce 2 O 3 to be precipitated and elongated in the growth direction of the γ -Al 2 O 3 platelets (Fig. 4(b)). The precipitates are easily observed in the γ -Al 2 O 3 layer near the surface, because of higher •• V O there which promotes the precipitation reaction. The oxidation kinetics of alumina formers during 750-1200 °C is highly correlated with the diffusion of Al cations and O anions along the alumina GBs 30

. A decrease of D GB
Al by the RE segregations to alumina GBs has been proposed to be the reason why the RE-and RE x O y − doped alumina formers have a lower oxidation rate 3,4,6,7,10,12,15,17,32 . As illustrated in Fig. 8, D GB Al in the alumina layer here should be decreased when the Ce segregants outward migrate along the GBs in α -Al 2 O 3 layer and the TBs in the γ -Al 2 O 3 layer.
The model in Fig. 8 strongly suggests that the Ce segregants occur only when the CeO 2 dispersoids in the aluminide have been swept over by the inward moving alumina/metals interface. In other word, the REE on the alumina growth for the CeO 2 dispersion in the aluminide is intrinsically pertinent to the incorporation of the oxide particles into the alumina scale by its inward growth. This may be generalized to the REE of the other RE x O y dispersions on alumina TGO growth. In addition, many alumina formers are alloyed with a RE instead of its oxide. Theoretically, it is possible that the RE in an alumina-forming metal at and below the equilibrium partial oxygen pressure of Al 2 O 3 /metal at the interface can be internally oxidized to form RE x O y . For example, Hf in an alumina-forming CoCrAlHf was internally oxidized into fine spherical HfO 2 particles in an Al 2 O 3 /CoAl pack at 1200 °C 2 . The particle sizes of the formed RE x O y highly depend on the synergistic effect of several factors, e.g., RE amounts and solubilities in metals, metal compositions and microstructures, alloying and oxidation temperatures 25,38,51 . The particles have not been highlighted previously, plausibly because they are sometimes as small as nano-sized particles. Because of a very low solubility limit in metals (e.g., only 0.01 wt % for Y in the FeCrAl alloys 51 ), a RE overalloying is hard to avoid. Thus, RE-rich precipitates occur in metals [1][2][3]7,25,26,51 . They can be in-situ internally oxidized to form RE x O y particles at the front of oxidation. The RE x O y particles in the RE-alloyed metals, no matter whether they are formed by diffusional and non-diffusional oxidation of RE solute atoms and RE-rich precipitates, respectively, would not have the relative REE until they have been swept by the inward growing alumina.
In summary, the REE for the CeO 2 particle dispersion in the nickel aluminide is firstly correlated with the inward growth of the inner α -Al 2 O 3 in the TGO scale. The CeO 2 dispersion, after being swept by the inward-growing α -Al 2 O 3 , partially dissolves producing tetravalent • Ce Al . They then transfer to trivalent Ce Al X by capturing electrons and segregate to the GBs in the n-type α -Al 2 O 3 . The Ce Al X cations migrate outward along the GBs there and the TBs in the outer γ -Al 2 O 3, finally precipitating Ce 2 O 3 near the surface. The segregation and migration of the Ce cations along the planar defects would obstruct the diffusion of the Al cations for the growth of the alumina TGO. The explanation may be generalized to the related REE of other RE x O y dispersions in the alumina-forming metals. For the alumina formers alloyed with a RE, the RE, in the form of either solute atoms or RE-rich precipitates, is plausibly internally oxidized into RE x O y particles below the TGO/metal interface. They may then affect the alumina growth in a manner to the CeO 2 dispersion in the aluminide.

Method
The CeO 2 particles with a purity of 99.5%, a commercial product by Alfa Aesar company, were introduced to a nickel aluminide by using a two-step method 28,37 . First, the pure Ni samples with dimensions of 15× 10× 2 mm, after being abraded to a final 800 grit SiC paper, were electrodeposited with a Ni-CeO 2 composite film from the CeO 2 -loaded nickel sulfate bath (150 g/l NiSO 4 ·6H 2 O, 120 g/l C 6 H 5 Na 3 O 7 ·2H 2 O, 12 g/l NaCl, 35 g/l H 3 BO 3 ).
Scientific RepoRts | 6:29593 | DOI: 10.1038/srep29593 A mechanical agitation was maintained to mitigate the particle agglomeration and sedimentation during electrodeposition, as illustrated by the setup 59 . Second, the samples were aluminized at 620 °C for 5 h using a halide-activated pack-cementation in a powder mixture of Al (particles size: ~75 μ m) + 55 wt.% Al 2 O 3 (~75 μ m) + 5 wt.% NH 4 Cl in an Ar (purity: 99.99%) atmosphere. The characteristics of inward growth of the aluminide at the cementation temperature caused the CeO 2 in the electrodeposited film to be trapped, forming a ceria-dispersed nickel aluminide coating on the sample surface. After being ultrasonically cleaned in acetone, the aluminized samples were ready for oxidation.
The samples were not placed into a muffle furnace for oxidation until it was heated up to 1100 °C. The ceriadispersed nickel aluminide after oxidation were cross-sectioned for the scanning electron microscopy (SEM) investigations, and then ion sliced into thin foils by using techniques detailed elsewhere 32 , with the transparent areas desired for the TEM investigations under a JEOL 2100F TEM at 200 kV accelerating voltage, by using the techniques of bright field (BF) imaging and high resolution TEM (HRTEM) imaging and fast Fourier transformation (FFT) diffraction, scanning TEM (STEM) imaging as well as high-angle annular detector dark-field STEM (HADDF-STEM) imaging.