Doped Ceria Nanoparticles with Reduced Solubility and Improved Peroxide Decomposition Activity for PEM Fuel Cells

Ceria nanoparticles (NPs) have unique catalytic properties which make them suited to scavenge degrading radical species and their precursor peroxides during PEM fuel cell operation. However, in the acidic environment of the fuel cell, ceria dissolves and the resulting cations migrate within the MEA, causing performance and durability losses. In this work, ex situ testing was used to evaluate the peroxide decomposition, selectivity towards radical generation, and solubility of Gd, Pr, and Zr-doped ceria NPs over a range of crystallite sizes and dopant levels. These doped materials exhibit better peroxide scavenging activity and dissolution resistance than undoped ceria. In these materials, activity is largely governed by increased surface area due to high internal porosity at smaller crystallite sizes compared to undoped ceria. Of the compounds tested, ceria NPs doped with 15 at% Zr (10 nm) and 5 at% Pr (17 nm) exhibited greater dissolution resistance than undoped ceria. Stabilization of the former doped NPs is attributed to crystallite agglomeration, while the increased stability of the latter is proposed to be due to its internally-porous, mesoscale structure suggested by its sorption isotherm. Both materials are more dissolution-resistant and active peroxide decomposers compared to undoped ceria but exhibit increased byproduct radical generation.

Polymer electrolyte membrane (PEM) fuel cells are electrochemical energy conversion devices that are more efficient than internal combustion engines and, when fueled with renewably generated hydrogen, can produce zero-emission electricity which can reduce greenhouse gas emissions and fossil fuel dependence. Presently, however, widespread adoption of PEM fuel cell systems is limited by material cost and durability issues. 1 Specifically, the membrane electrode assemblies (MEAs), which consist of the PEM and Pt -based catalyst layers (CLs) which are bonded to it, are heavily stressed during PEM fuel cell operation. 2 Incidentally, MEAs are costly components, as well, accounting for approximately 50% of the total stack cost. 3 Furthermore, enhancements to MEA durability that improve lifetime system efficiency directly reduce the cumulative amount of wasted hydrogen (i.e. that converted to heat instead of electricity). Therefore, improving MEA durability is a high-value target for reducing lifetime fuel cell costs.
Chemical degradation of the ion-conducting, or ionomer molecules (typically, a broad class of perfluorosulfonic acids [PFSAs] 4 ) which compose the PEM and are present in the CLs, is a multi-step process which involves (i) the formation of hydrogen peroxide (H 2 O 2 ), (ii) its decomposition by metal ion impurities present in the MEA, such as iron, copper, and titanium, 5 (iii) the generation of radical species, and (iv) their subsequent attack on vulnerable functional groups of the PFSA molecule. 6 Chemical degradation increases the PEM's susceptibility to premature failure by causing global ionomer thinning and diminishing its mechanical properties. 7,8 While chemical degradation is highly interdependent on MEA composition and operating conditions, it is generally understood that peroxide (H 2 O 2 ) and hydroperoxyl radicals (HOO•) are formed at the electrodes and in the membrane where the H 2 and O 2 diffusion fronts converge and subsequently diffuse throughout the MEA, where peroxide is present at 0.1 to 1.6 mM concentrations. [9][10][11] Hydrogen (H•) and hydroxyl radicals (HO•) are believed to be locally present at the electrodes and Pt band in the PEM at <0.1 μM concentrations. 11 The use of cerium (Ce)-based scavenging systems has received significant attention because of its ability to rapidly switch between oxidation states, which consumes both radicals and peroxides. 12 While involved in an array of reaction pathways, 6 in the most kinetically favorable reactions, Ce 3+ neutralizes hydroxyl and hydroperoxyl radicals according to Reactions  Ce ions may be directly exchanged with protons in the ionomer 14 or incorporated into the membrane and/or the catalyst layers (CLs) in its oxide form, CeO 2 (referred to as "ceria" throughout the text). 15,16 When ion-exchanged with the acidic PFSA, modeling studies suggest that >99.9% of Ce cations exist in the reduced Ce 3+ form. 13 X-ray photoelectron spectroscopy (XPS) studies indicate that 2 to 4 nm ceria nanoparticles (NPs) contain a ∼10/90% distribution of Ce 3+ /Ce 4+ ions on its surface. 17  Ionic Ce reduces MEA fluoride emissions rates (FERs), a measure of PFSA degradation, by three orders of magnitude, 14 whereas ceria NPs typically improve FERs by one to two orders of magnitude. 15 This disparity could be caused by the higher concentrations of radical-scavenging Ce 3+ in ionic form compared to the lower surface concentrations present on an oxide. Ceria NPs may be introduced into the MEA at higher concentrations than in ionic form because it does not initially block conduction pathways. 15 0.5 to 3 weight percent (wt%) ceria added to the membrane demonstrated negligible beginning of life performance losses compared to ceria-free MEAs. 15 However, when ceria NPs are added to the acidic PFSA membrane or catalyst layer ionomer, Ce dissolves and diffuses through the MEA after exposure to humidified reactant gases, 18,19 resulting in effects similar to direct Ce ion exchange. 20 Upon the introduction of water, concentration, and potential gradients present during cell operation, these Ce ions can mobilize, [21][22][23][24] which can generate losses by diminishing proton and O 2 transport in the cathode CL. [25][26][27][28] The mobilization of Ce 3+ may also cause local depletion of the radical scavenger, leaving ionomer regions vulnerable to radical attack. 22 If made with reduced solubility, however, ceria NPs have the potential to be localized as effective peroxide decomposers and radical scavengers at higher concentrations than ions, alone. Thus, the ideal additive is one that maximizes these catalytic properties while resisting dissolution over an MEA's lifetime.
Metal-doped ceria compounds have received significant attention in the literature for their improved O 2 conductivity and stability compared to undoped ceria. 29 These properties are attributed to the formation and mobility of additional O 2 vacancies, which are generated by doping ceria's fluorite lattice structure with aliovalent cations. 30 These vacancies are also related to the concentrations of Ce 3+ moieties on the surface of ceria NPs, which govern antioxidant activity. 17 Trivalent rare earth metals are promising candidate dopants because of their high solubility in ceria and promotion of O 2 vacancy formation. 29 Of these compounds, Ce 1−x Gd x O 2−δ (CGO) and Ce 1−x Pr x O 2−δ (CPO) are particularly interesting because they both display exemplary O 2 conductivity and favorable antioxidant behavior relative to other rare earth-doped cerias. 29,31 CGO has demonstrated improved peroxide decomposition activity through the formation of surface O 2 vacancies. 32 The mixed 3+/4+ valance state of Pr in CPO further suggests that it wouldn't preferentially exchange with or migrate to radical-scavenging surface Ce 3+ moieties. 31 In addition to rare earths, a range of other dopants have been investigated, however, many phase segregate instead of forming solid solutions. [33][34][35] Of these compounds, however, Ce 1−x Zr x O 2−δ (CZO) is attractive because of its reduced formation energy of O 2 vacancies and its greater resistance to thermal sintering and deactivation relative to undoped ceria. 36,37 While its degree of phase segregation has not been definitively agreed upon in literature, preliminary studies have shown enhanced antioxidant effects. 19 Because of these promising characteristics, CGO, CPO, and CZO NPs were assessed for their suitability in PEM fuel cell systems. The peroxide decomposition activity, selectivity towards radical generation, and solubilities were measured in these compounds across a range of crystallite sizes and dopant levels and compared to undoped ceria NPs. The solubilities of the most promising peroxide-scavenging compounds were investigated and correlated to variations in crystallite size, surface area, and pore structure. The results are applied to develop a framework for optimizing ceria-based scavenging systems to extend durability and limit performance losses, collectively reducing the lifetime costs of fuel cell stacks.

Experimental
Ceria and M-doped ceria nanoparticle synthesis.-Undoped and doped ceria NPs were synthesized using a wet chemical precipitation technique. In order to mitigate the deleterious effects of impurities, 33,38 all precursor salts were ⩾ 99.9% pure, which contain <150 ppm of rare earth metals and <25 ppm (typically, <5 ppm) of other elements, while all other reagents were ACS Reagent Grade. Ce acetate (Strem Chemicals) and Gd, Pr, or Zr acetates (Sigma Aldrich) were dissolved at the desired stoichiometric ratio in 18 MΩ deionized (DI) water to obtain a solution of 0.1 M total Ce + dopant ions. To the mixture, concentrated nitric acid (Fisher Scientific) was added until the constituents were dissolved (corresponding to a pH of ∼1). Once both metal acetates were fully dissolved, a 50 vol% ammonium hydroxide solution (Alfa-Aesar) was added to the solution while manually agitated until the solution turned white and opaque (pH ∼ 10). The solution was centrifuged, and then washed with DI water three times to remove acetate, ammonia, and nitric acid contaminants. Samples were then dried in an oven at 140°C for 1 h to remove moisture. The dried samples were subsequently transferred to a ceramic boat and heated under air between 200 and 800°C for 1 h to achieve a range of particle sizes from 5 to 40 nm. Throughout the text, doped ceria NPs are referred to as "CMO-x," where M is either G, P, or Z, corresponding to the dopant element Gd, Pr, or Zr, respectively at a concentration of x (either 5 or 15 at%). Commercial ceria nanopowder (Sigma Aldrich), with a specified crystallite size of <50 nm, was also analyzed in select experiments.
Nanoparticle characterization.-Phase purity, crystal structure, lattice parameters, and crystallite size of the synthesized NPs were determined by powder X-ray diffraction (XRD) with a Siemens D5000 diffractometer using a CuK α source operating at 40 kV and 35 A. Powder samples were dispersed across a zero-background quartz sample holder using acetone. 2 mm anti-scatter slits were used to define the illuminated sample area, with a 0.6 mm detector slit. Samples were measured from 2θ = 10 to 120°with a step size of 0.02°. The analysis was based on whole profile fitting refinement using Jade 9 software (Materials Data, Inc.). Error bars in the corresponding figures represent the typical fitting uncertainty of ±2%.
Nitrogen sorption using a QuantaChrome Autosorb IQ2 was used to determine the porosity and surface area of the synthesized pure and mixed cerias. Initially, 300 to 500 mg samples were placed in a 4 mm bore glass absorption tube. To prepare the samples for nitrogen sorption measurements, the samples were put under vacuum at 30°C overnight to remove absorbed water and contaminants. The full nitrogen isotherm was used, with atmospheric pressure being 580 torr. For surface area quantitation, a multipoint analysis was used based on 11 points in a relative pressure (p/p o ) range of 0.025 to 0.3535, according to the BET theory. Desorption points from a p/p o from 1.0 to 0.3764 were used to determine pore volume using the BJH method. Such N 2 sorption techniques are typically accurate within ±5%, with a reproducibility of ±1%, 39 which is reflected in the error bars of the corresponding figures.
Hydrogen peroxide scavenging and radical generation of nanoparticles.-Specific hydrogen peroxide decomposition rates (i.e., normalized to BET surface area) were measured by volumetric determination of catalytic oxygen production. Less than 50 mg of sample was added to 1 ml of deionized water and stirred with a magnetic stir bar for 1 min to disperse the NPs. The mass used was varied to have the same amount of accessible surface area across the experiments. An excess (∼15 ml) of 30 wt% hydrogen peroxide (Sigma Aldrich) was added to the nanoparticle-water mixture, and evolved oxygen was measured by water displacement from an inverted burette. The amount of oxygen evolved was measured every five minutes, and the resulting oxygen volume versus time was fit to a first order exponential decay to calculate the rate at time zero. The observed rate constant was normalized to the surface area of the catalyst used.
To determine the peroxide decomposition to free radical production selectivity, a process similar to the method used by Prabhakaran et al. was employed. 40 Summarized briefly, 23.5 mg of 6-carboxyfluoroscein (6CF, Sigma-Aldrich) was added to 250 ml of 1 wt% hydrogen peroxide. 30 ml of the 6CF-peroxide solution was introduced to an open beaker, to which less than 50 mg of sample was added, which was also scaled to maintain the same accessible surface area between samples. A multidiode UV-vis spectrometer (Hewlett Packard) was used to quantify the amount of 6CF remaining in solution by measuring the decay of the 492 nm peak. Measurements were taken every five minutes for an hour, and the loss of 6CF was fit to a first order exponential decay. This rate was normalized to the nitrogen BET accessible surface area of the catalyst used. The inverse selectivity, i.e., the normalized rate of 6CF decomposition divided by the normalized rate of oxygen evolution, is reported because the selectivity would go to infinity for catalysts which show zero radical production rates.
Dissolution of nanoparticles in acidic media.-For studies to simulate the solubility of NPs in acidic fuel cell environments, ∼40 mg samples were dispersed in 5 ml of 1 M sulfuric acid and allowed to settle for 12 h. UV-vis was performed using a Hitachi U-2900 spectrophotometer and Hellma 10 mm cuvettes at a wavelength range of 250 to 600 nm. Following initial characterization, the samples were heated at 50°C for 72 h.
As shown in the previous work of Banham et al., the resulting UV-vis spectra can be deconvoluted to yield a linear combination of Ce 3+ and Ce 4+ ion concentrations. 25 In accordance with that work, five solutions with unique, known Ce 3+ and Ce 4+ contents were measured (Fig. 1a), yielding the following calibration functions for the concentrations of Ce 3+ and Ce 4+ (denoted as C Ce 3+ and C , and I nm 320 are the peak intensities at the 253 and 320 nm wavelengths, corresponding to the concentrations of Ce 3+ and Ce 4+ ions, respectively (Fig. 1b). Using the known sample volume and molecular weight of Ce, the dissolved masses of the samples were calculated. All dissolution rates, measured in mg/hr, were based on the UV-vis profiles obtained at 72 h.
Because the calibration curves yielded R 2 values of ⩾0.998, we expect the UV-vis technique to have a negligible effect on measurement accuracy. Therefore, the uncertainty of all UV-vis experiments (i.e., peroxide reactivity, radical generation, and dissolution rate) is assumed to be based on variations between samples within a single batch. This uncertainty was determined by measuring the dissolution of two distinct commercial ceria samples from the same batch every 24 h for 72 h, which yielded an average deviation of ±10%.

Results and Discussion
Effects of Gd, Pr, and Zr dopants on peroxide decomposition and byproduct radical generation.- Figure 2 shows the effect of Gd, Pr, and Zr doping on the peroxide decomposition activity of ceria. Compared to pure ceria (Fig. 2a), these dopants, incorporated at 5 and 15 at%, all increased the catalytic activity across the entire range of crystallite sizes measured, with the exception of CPO-15 (Figs. 2b-2d). In these doped materials, the activity increase was greatest at crystallite sizes of <10 nm. For CGO and CPO, the optimal crystallite size for peroxide decomposition is ∼6 nm, with particles <6 nm demonstrating sharply decreased activity (Figs. 2b,  2c). In undoped ceria and CZO, however, there appears to be no lower bound for increasing activity across the range of crystallite sizes measured (Figs. 2a and 2d).
In 9M peroxide solutions, the reactivities of particles >6 nm decrease monotonically for all samples as a function of crystallite size. According to Reaction 3, only Ce 4+ neutralizes peroxides, 13,14 however, previous XPS studies demonstrated that the surface concentration of Ce 4+ slightly increases as a function of crystallite size: Stewart showed Ce 4+ surface concentrations increased from 74% to 77% between crystallites of 5.4 to 37.7 nm, 41 while using a different synthetic route, Trogadas et al. observed Ce 4+ concentrations of 94% at 5 nm crystallite sizes and >99% for crystallites >10 nm. 17 These results imply that Ce 4+ surface concentration is not a significant factor affecting peroxide reactivity under these experimental conditions. Instead, exposed Ce 4+ catalytic surfaces offered by smaller particles with greater surface areas appear to govern activity. Figure 3 shows the selectivity towards radical generation for the undoped and doped ceria NPs. Undoped ceria demonstrated near zero radical production at crystallite sizes of >7 nm (Fig. 3a). Byproduct radical production in 0.33 M peroxide solution decreases sharply with crystallite size for all particles <6 nm, which is consistent with a larger amount of Ce 4+ surfaces available to scavenge hydroperoxyl radicals. 42 While both 5 and 15 at% CGO and CPO had greater activity than undoped ceria across all crystallite sizes tested (Figs. 2b, 2c), radical generation increased significantly for crystallites >10 and >6 nm in the 5 and 15 at% samples, respectively (Figs. 3b, 3c). These narrow windows of optimal crystallite sizes could be detrimental because smaller particles with higher surface area could dissolve more easily during fuel cell operation, which would lead to greater radical generation than ceria alone. Thus, it is likely desirable to utilize the largest particle size possible. These results imply that both CGO-15 and CPO-15 are less suitable candidates to enhance fuel cell durability compared to the other materials. In contrast, both CGO-5 and CPO-5, had a wider range of crystallite sizes which exhibited both high activity and low radical selectivity. Because of its higher overall activity and lower radical generation at both smaller and larger crystallite sizes, the latter compound was selected for further study. Notably, both 5 and 15 at% CZO showed only a slight increase in radical generation at smaller crystallite sizes (Fig. 3d) while maintaining superior activity relative to the other samples (Fig. 2d). Because of the lower minimum selectivity enjoyed at larger crystallites, CZO-15 was selected for further study.
It should further be noted that in these experiments, peroxide concentrations were ∼2-4 orders of magnitude higher than typical operating MEAs 9-11 and do not contain controlled amounts of iron, which promotes hydroxyl radical formation. While useful for screening compounds, in future experiments, these parameters should be carefully examined and maintained at relevant concentrations (0.1-1.6 mM peroxide 9-11 and typically ∼1-10 ppm Fe 42 ) and coupled with a more detailed surface analyses to further explore the interdependence of parameters described above.
Characterization of selected nanoparticles.-To assess their suitability in acidic fuel cell systems, XRD, BET, and ex situ solubility experiments were performed to compare their crystallite sizes, porosities, and dissolution rates to baseline homemade and commercial ceria NPs. The results for the selected samples are summarized in Table I. As shown in Fig. 4, surface area decreases with increasing XRD crystallite size for all the nanoparticle samples. The dotted line in Fig. 4 is the calculated specific surface area (SA ideal ), in m 2 g −1 , for non-agglomerated, monodisperse particles with no internal porosity, and assuming spherical or cubic shape: where d is the nanoparticle diameter in nm, and ceria r is the density of ceria (7.2 g m −3 ). Compared to the undoped ceria, all doped samples, except for the 17 nm CPOx-5, show a rapid decrease in surface area as crystallite size increases. This suggests that the doped ceria compounds are more highly agglomerated than the undoped ceria, which dominates the contribution of their relatively small internal porosities, resulting in decreased surface areas. On the other  hand, the 17 nm CPOx-5 and undoped ceria samples had surface areas that were closer to the SA ideal curve. This implies additional internal porosity, that is consistent with their order-of-magnitude larger pore volumes measured from BET (Table I). Additional microscopic analysis, especially in conjunction with multiangle light scattering, would be useful to further quantify the degree of agglomeration and particle size. 43 Because of this, all analysis in this work is based on the nanoparticle crystallite size from XRD.
Solubility of selected M-doped ceria in acid.-To assess their stability in the acidic fuel cell system, ex situ dissolution tests were performed on the selected CPO-5 and CZO-15 compounds and compared to undoped homemade and commercial ceria NPs. Figure 5 shows a direct comparison between the dissolution amount that occurred as a function of crystallite size (Fig. 5a) and surface area (Fig. 5b). Generally, the dissolution rate is inversely proportional to crystallite size and directly proportional to surface area for each doped and undoped material. As shown in Fig. 5a, CZO-15 and the 17 nm CPO-5 are less soluble at smaller crystallite sizes compared to the undoped ceria samples. The increased dissolution rates of the smaller CPO-5 NPs may also explain the decrease in peroxide reactivity for particles <6 nm (Fig. 2c). The 50 mg samples used in these experiments likely completely dissolved in solution to Ce 3+ , which is not active towards peroxide decomposition according to Reactions 1 and 2. This reduction was not observed in CZO-15 (Fig. 2d), where the smallest NPs only lost, at most, 20% of their mass during this experiment.
Notably, for samples which contained relatively low amounts of internal porosity, as shown in Table I, dissolution rates are expected to increase as a function of surface area, as shown in Fig. 5b. While this appears to be true for the CZO-15 and undoped ceria samples, which increase exponentially faster with smaller crystallite size, the relation does not hold for the CPO-5 sample set. Specifically, despite   Table I. The dotted line is the ideal surface area (SA ideal ) calculated using Eq. 7. having additional internal porosity, the 17 nm CPO-5 shows far less dissolution that would be predicted based on surface area alone, which may imply a unique stabilization mechanism.
Mechanisms for solubility reduction.-The pore volumes reported in Table I suggest that the porous structure of the 17 nm CPO-5, commercial ceria, and the 6 nm ceria samples, are distinct from the other samples. It is clear that the pore volume/surface area ratios for these samples are higher than those observed for the other samples, as shown in Fig. 6. For most of the samples, the pore volume increases linearly with the BET surface area, as would be expected if the particles all had similar internal (primary) porosities but different particle sizes. However, the 17 nm CPO-5 and the commercial ceria are outliers from this trend, which implies that they have different porosities and pore structures compared to other samples. The 6 nm ceria sample slightly deviates from this trend, as well. Based on Table I, the additional porosity of the 17 nm CPO-5 and commercial ceria can be explained based on their internal mesopores which are absent for the other samples, in the order of 17 nm CPO-5 > commercial ceria > 6 nm ceria » other samples, which is consistent with the deviation from the trend in Fig. 6.
Comparing the gas sorption isotherms (Fig. 7), it is clear that the 17 nm CPO-5 (Fig. 7c) and commercial ceria (Fig. 7d) samples have a similar porous structure that is markedly different from the other samples, which may provide insight into their decreased solubility, beyond the size effect alone.
All sorption isotherms indicate mesoporous structures, which are formed from the agglomeration of crystallites into larger particles. The majority of the samples display H4-type hysteresis, which is exhibited by aggregated zeolite crystals, some mesoporous zeolites, and micro-mesoporous carbons. 44 This behavior is exhibited more strongly in samples predicted to be more agglomerated in Fig. 4 (i.e., greater deviation from the calculated SA ideal curve). These samples include the larger ceria (Fig. 7b) and CZO-15 (Fig. 7d). In contrast, the 17 nm CPO-5, the commercial ceria, and the 6 nm ceria show H3-type hysteresis, which is characteristic of aggregations, plate-like particles (e.g., some clays), as well as mesoporous structures with no pore condensate. 44 TEM results across a variety of synthetic routes and dopant compositions suggest that ceria is not expected to take 2-D forms 41 , implying the latter structure.
We assume that cations are primarily dissolved by and mobile through acidic liquid phases. Therefore, it is possible that the high internal porosity and absence of mesopore condensate in the H3-type samples leads to less connected liquid domains than their counterparts, which may be responsible for lower dissolution and mobility of the dissolution products, resulting in the higher observed stability. Further sorption analysis, especially after quantifying the agglomeration state, would be useful to validate this proposed stabilization mechanism and confirm the repeatability of samples exhibiting this behavior.
The 10 nm CZO-15 exhibited the highest stability but had low internal porosity (Table I) and an H4-type sorption isotherm (Fig. 7d), contrasting the other samples with relatively high stability. Therefore, its insolubility is attributed to the agglomeration of crystallites and reduced surface area, as indicated by the largest deviation from the SA ideal curve in Fig. 4. Dopant optimization.-Of primary interest for PEM fuel cells is to maximize nanoparticle radical and peroxide scavenging while limiting their solubility, which can lead to mobility of Ce and subsequent effects on performance and durability. The proposed "unfilled mesopore" mechanism appears beneficial to stabilize less agglomerated, yet internally-porous 17 nm CPO-5 and 19.7 nm ceria samples, resulting in dissolution rates of ∼0.25 and ∼0.4 mg hr −1 , respectively. In contrast, the high agglomeration of the CZO-15 sample leads to its order-of-magnitude larger stability of ∼0.01 mg hr −1 , even in the absence of internal porosity. The commercial ceria still has a relatively high absolute dissolution rate of 6.2 mg hr −1 , making it less practical for long term fuel cell operation, thus, eliminating it from further analysis.
The peroxide reactivity and byproduct hydroperoxyl radical generation of the most stable compounds tested are shown in Fig. 8. From this figure, the 10 nm CZO-15 and 17 nm CPO-5 emerge as the least soluble compounds, however their byproduct radical generation is higher than the 19.7 nm undoped ceria. The 10 nm CZO-15 exhibits ∼2x peroxide reactivity compared to the 17 nm CPO-5 and 19.7 nm ceria, which are similar in magnitude. We must also note that there is evidence for Pr and Zr surfacesegregation in NPs. 41 While their dissolution resistances are promising from an engineering perspective, further microscopic characterization is warranted to understand the effect of surface segregation on porosity, aggregation, solubility, and/or catalytic properties.
Implications for fuel cell durability.-Ceria is typically added to the CLs at area densities of ∼0.025 mg cm −2 , 25 while effective membrane concentrations can be as low as 0.5 wt% 15 (corresponding to an area density of ∼0.25 mg cm −2 for a 25 μm membrane with a density of 2 g cm −3 ). Thus, employing scavengers at the higher membrane area density value yields 75 mg for an automotive-scale MEA. Since the 10 nm CZO-15 and 17 nm CPO-5 samples dissolve at rates of ∼0.01 and 0.25 mg hr −1 in liquid, they would be completely ionized in 7,500 and 300 h, respectively, if these rates were identical in MEAs. It is possible, however that these rates could be significantly slower in MEAs, where the hydration is lower than in the ex situ experiments.
Owing to the low concentration of anionic counterions produced during ionomer degradation, it is not expected that Ce ions will significantly "wash out" of the MEA in the cell product water. 21 Therefore, it is possible that, when used in an MEA, an equilibrium will be built up between the ceria NPs and Ce-exchanged CL ionomer domains, slowing dissolution. Over the course of longer tests, however, the membrane ionomer could serve as an ion sink for complete dissolution of the oxides, resulting in a detrimental increase in Ce ion concentrations.
In order to protect the ionomer health during the anticipated 8,000 and 30,000 h lifetimes of light and heavy duty automotive FC systems, 1,45 respectively, the efficacy of doped ceria NPs should be validated using in situ MEA testing to quantify their dissolution and migration under relevant operating conditions. Such conditions may also be optimized to decrease scavenger dissolution. For example, by changing the anode RH from 100% to 50% RH, Banham et al. showed major mitigation of performance loss in MEAs with ceria in the CLs after accelerated stress test cycles. 25 This result underscores  Table I. The dashed line is linear fit to all samples aside from the 17 nm CPO-5 and the commercial ceria.
the critical role of water in the solubility and transport of Ce ions from the oxide NPs. If additional stabilization is needed, internally porous particles could be sintered into agglomerated, mesoporous forms, or the particles themselves could be encapsulated in a hydrophobic matrix, such as in the MPL, as believed to have been done in the MEAs of the Toyota Mirai fuel cell electric vehicle. 46 However, the latter approach appears most effective in maintaining Ce 3+ concentrations in MEA, as opposed to utilizing the catalytic abilities of the ceria NPs, themselves, to protect the ionomer.

Conclusions
In this work, Gd, Pr, and Zr-doped ceria NPs were synthesized in various compositions and sizes and evaluated against undoped ceria for their suitability as PEM fuel cell additives. Their peroxide decomposition activity and radical selectivity were measured in ex situ experiments. The most active and selective doping compositions, 5 at% Pr and 15 at% Zr were selected for measurement of morphological properties and solubility. All the samples showed consistent exponentially-increasing solubility with decreasing crystallite size. However, when assessed vs surface area, certain compositions were distinct, favorable outliers, implicating unique stabilization mechanisms. Pore volume per surface area is significantly enhanced in these samples, which corresponds to a proposed mesopore structure without pore condensation, according to BET. This structure could be more dissolution-resistant, while high catalytic activity is maintained through its high surface area due to internal porosity. Agglomeration of NPs is also proposed as a stabilization mechanism, however, it is not clear how break-up of agglomerates during catalyst layer ink preparation would impact their resulting scavenging abilities and solubilities.
Of the most stable compounds, the 10 nm 15 at% Zr-doped and 17 nm 5 at% Pr-doped ceria samples showed the most optimal performance in terms of stability and activity, while exhibiting slightly higher byproduct radical generation compared to undoped ceria. In the liquid test conditions employed here, however, typical concentrations in fuel cells would dissolve ∼1-2 orders of magnitude faster than fuel cell lifetime targets. This motivates the analysis of radical scavenging NPs in MEAs operated under realistic fuel cell conditions, to assess their durability enhancement effect and solubility.