Nanometric Fe-Substituted ZrO₂ on Carbon Black as PGM-Free ORR Catalyst for PEMFCs

All commercially available chemicals were used without further purification. Zirconium tetrachloride (≥99.5%), 2-methylnaphthalene (97%), 2-bromonaphthalene (97%), 1,8-diazabicyclo[5.4.0]-5-undec-7-ene (≥99.0%), and 2-ethoxyethanol (99%) were purchased from Sigma-Aldrich. 4-tert-butylphthalonitrile (≥98%) was obtained from TCI. Zirconium (IV) tetra-tert-butyl-dichlorophthalocyanine (ZrCl₂Pc(t-Bu)₄) and iron(II) tetra-tert-butyl-phthalocyanine (FePc(t-Bu)₄) were synthesized similarly to what is reported in previous publications,^29,30 except that 4-tert-butylphthalonitrile instead of 1,2-dicyanobenzene was used as a starting precursor in order to obtain soluble phthalocyanines. ZrCl₂Pc(t-Bu)₄ and FePc(t-Bu)₄ were characterized by Fourier transform infrared spectroscopy (to confirm metal insertion into the Pc)³¹ and by thermogravimetric analysis under air (to quantify the metal content).²¹ We would also like to mention that the obtained precursor contained a significant fraction of organic impurities, resulting in a Zr content ≈3 times smaller than expected for an ideal ZrCl₂Pc(t-Bu)₄. Unfortunately, we were unable to remove the impurities from the precursor even after rigorous purification and then we used in the synthesis an amount ≈3 times higher. The exact origin of the presence of organic impurities in the precursor is not yet known, but it could be due to the presence of water impurities in a reactant, leading to the catalysis of an organic side reaction.

For the present study, we synthesized four carbon-supported catalysts, viz., Fe_0.17Zr_0.83O_1.91 [denoted as Fe-ZrO₂ (1.0 wt%)], Fe_0.07Zr_0.93O_1.97 [denoted as Fe-ZrO₂ (0.36 wt%)], Fe, and ZrO₂. All samples were supported on graphitized Ketjenblack carbon [denoted as KB_graph] (EA-type from Tanaka Kikinzoku Kogyo). A graphitized carbon was chosen in order to minimize the presence and effect of surface oxygen groups, thus to have a better control of the partial oxidation during the heat-treatment step in the catalyst synthesis procedure. The ZrO₂ content was kept constant (theoretical 12 wt% ZrO₂/KB_graph) in all samples, except for the pure Fe catalyst [denoted as Fe], where no ZrO₂ is present. The synthesized catalysts Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 (Fe_xZr_1−xO_2−δ) are those reported in our past research, where the Fe/ZrO₂ atomic ratios are 7/93 (Fe-ZrO₂ (0.36 wt% of Fe)) and 17/83 (Fe-ZrO₂ (1.0 wt% of Fe)), whereby wt% of Fe is referred to the mass of metallic Fe in the final catalyst.²⁷ Catalysts were synthesized by first depositing the corresponding amounts of ZrCl₂Pc(t-Bu)₄ and/or FePc(t-Bu)₄ onto KB_graph using a similar impregnation process as described previously.³² After impregnation, the solvent (chloroform, ≥99.9%) was removed by rotovaporation (Heidolph, Hei-VAP Value). The mixture was further dried overnight at 70°C to remove all residual chloroform. The dried mixture was then transferred to a quartz tube furnace (HTM Reetz, LK 1300-150-600-3). Initially, the temperature was ramped up to 800°C in 5% H₂/Ar gas atmosphere. A heating rate of ≈10°C min⁻¹ was maintained until ≈700°C, which was then decreased to 2°C min⁻¹ until 800°C to avoid temperature overshoot. The temperature was then held for 2 h in the same gas-mixture, before switching to a mixture of 2.5% H₂ and 0.5% O₂ in Ar (partial oxidation, PO) (5.0 grade, all gases supplied by Westfalen AG) for another 1 h at the same temperature. Finally, the furnace was cooled to RT in 5% H₂/Ar. The as-prepared catalysts were not subjected to any acid washing step after their synthesis.

Catalysts were subjected to inductively coupled plasma atomic emission spectrometry (ICP-AES; iCap 6500, Thermo Fisher Scientific) in order to determine the exact amount of Fe and ZrO₂. The samples were first digested in hydrofluoric acid (40 wt%), and then in aqua regia by using a microwave digester. In addition to ICP-AES, the total metal oxide content (ZrO₂ + Fe₂O₃) was evaluated by thermogravimetric analysis (TGA, Mettler-Toledo, TGA/DSC 1) from the residual sample weight after complete combustion of the catalyst at 1000°C in a mixture of O₂/Ar (67/33%) atmosphere. The particle size distribution of ZrO₂ was evaluated by transmission electron microscopy (TEM) measurements with a JEOL JEM 1400-Plus transmission electron microscope, operated at an acceleration voltage of 120 kV. Holey carbon-coated TEM grids were used for sample mounting. A CCD camera was employed to collect several images at 200,000× magnification. The software ImageJ³³ was used to measure the diameter of at least 100 individual particles. From this data, the number averaged particle size (D_Average) and the standard deviation (SD) were computed. The Sauter diameter (D_Sauter) (surface−volume averaged diameter) was calculated as in Eq. 1, where l_i is the number of particles with diameter (d_i).

Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 were subjected to X-ray photoelectron spectroscopy (XPS) to investigate the Fe coordination and oxidation state. Analyses of the samples were performed on a Kratos Axis Supra spectrometer using monochromated Al K_α radiation at an energy of 1486.6 eV, operated at a total power of 525 W, 15 kV and 35 mA anode current. The samples were previously outgassed (overnight) in an ultrahigh vacuum chamber in order to remove atmospheric moisture and contaminants, so that the pressure in the chamber during the analysis was less than 1.0·10⁻⁸ Torr. All the binding energy values were calibrated using the carbon signal from the KB_graph carbon support (C 1s = 284.8 eV) as reference. The narrow Fe 2p spectra were collected between 702.5 to 745 eV binding energy (BE) using a step size of 0.1 eV and a pass energy of 40 eV. The XPS data analysis was performed by Casa XPS software. The represented data are an average of 10 spectra recorded for 10 min each.

Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 were also characterized by near-edge X-ray absorption fine structure (NEXAFS) data at the Fe L_2,3 edges. A pure ZrO₂ sample was also measured to demonstrate the absence of Fe. Spectra were collected at IFP's (Institut für Festkörperphysik) soft X-ray analytics facility WERA at the Karlsruhe synchrotron facility ANKA, Germany. Partial fluorescence-yield (FY) detection was used, both for its quasi-bulk probing depth of about 50 nm and for its independence from charging effects in non-conducting samples. For the Fe concentrations in this study, self-absorption and saturation effects were small and were not corrected. The photon-energy resolution was set to about 340 meV at the Fe L edges. The photon energy was calibrated to better than 30 meV by simultaneously measuring a NiO reference sample at the Ni L₃ edge (853.0 eV).³⁴ The collected data are the sum of two distinct scans (30 min each). To obtain the simulated data, the code developed by Thole and van der Laan, Butler, and Cowan,^35–37 and maintained and further developed by Stavitski and de Groot³⁸ was used to calculate spectra for different values of the crystal-field splitting (Delta_CF) and charge transfer energy (Delta_c). Hund's rule for exchange interaction was also taken into account. Charge-transfer effects were included for Fe²⁺ (Fe³⁺) by admixing transitions of the type 2p⁶ 3d⁷L → 2p⁵ 3d⁸L (2p⁶ 3d⁶L → 2p⁵ 3d⁷L), where L denotes a hole at the oxygen ligand. Multiplet parameters (in eV) are: i) Fe²⁺ HS: Delta_CF (10Dq) = 1.0, Delta_c = 4.0; ii) Fe²⁺ LS: Delta_CF (10Dq) = 2.5, Delta_c = 4.0; iii) Fe³⁺ HS: Delta_CF (10Dq) = 0.8, Delta_c = 4.5; and, iv) Fe³⁺ LS: Delta_CF (10Dq) = 3.0, Delta_c = 4.5. The correlation energy U_dd, the core-hole potential U_pd, as well as the hopping parameters T_sigma and T_pi were set for all spin and valence states to 5.0, 6.0, 2.0, and 1.0, respectively. The Slater integrals were renormalized to 80% of their Hartree-Fock values. According to the experimental resolution and the lifetime broadening, the simulations have been convoluted with a Gaussian with σ_G = 0.2 eV and with a Lorentzian with σ_(L_3) = 0.2 eV (σ_(L_2) = 0.3 eV) for the L_3 (L_2) edge. Finally, the position on the energy scale of the modelled spectra was adjusted by matching the white-line energy of experimental spectra with that of the Fe³⁺ high-spin modelled spectrum; the correctness of this adjustment was confirmed by comparison of the experimental data with the literature (see Results and Discussion).

In addition to XPS and NEXAFS, the Fe_0.17Zr_0.83O_1.91, Fe_0.07Zr_0.93O_1.97, and Fe catalysts were also subjected to ⁵⁷Fe Mössbauer analyses to ascertain the exact nature of Fe coordination in the catalysts. Mössbauer spectra were recorded on powder samples placed in poly(methyl methacrylate) based holders (absorber thickness: 80–90 mg/cm², sample holder diameter = 10 mm) at 4.2 K. For this, the absorber and the source of ⁵⁷Co in rhodium (ca. 1 Giga Becquerel) were cooled in a liquid helium bath cryostat. The spectra were fitted with a superposition of Lorentzian lines grouped into sextets using the MOS90 software (version 2.2). The fitted components often show broadened lines and have to be considered as representing distributions of magnetic hyperfine fields. Isomer shifts were measured with respect to the source having the same temperature as the absorber. Lastly, 0.245 mm/s was added to each isomer shift so that it can be referenced to α-Fe.

Rotating (ring) disk electrode measurements

Proton exchange membrane fuel cell measurements

In order to describe the adsorption of molecular oxygen on ZrO₂, we considered the most stable surface of tetragonal t-ZrO₂ i.e., the surface, which is equivalent to the (1 1 1) in the cubic system. As proven by Christensen and Carter,³⁹ this is the lowest energy surface of ZrO₂, thus we expect that it is the most exposed surface in our samples. Our structural model consists of a surface slab cleaved from t-ZrO₂ bulk at the theoretically determined lattice constants. We included in this slab three oxygen-terminated tri-layers, each with 16 ZrO₂ formula units, for a total system of 144 atoms (please see supporting information (SI), Fig. S1a). This model is sufficient to converge the O₂ adsorption energy with respect to slab thickness within 20 meV. Moreover, we used a 2 × 2 lateral supercell to ensure a realistic concentration of Fe surface species in the substituted zirconia and to avoid spurious image interactions upon formation of surface oxygen vacancies or adsorption of molecular oxygen. Each tri-layer presents oxygen atoms that points up (toward the electrolyte) and down (toward zirconia bulk), as shown in Fig. S1b and c; both these oxygen-atom types have been considered when modeling oxygen vacancies. All the coordinates of the bottom-most tri-layer (B in Fig. S1a) have been kept fixed at the bulk values during geometry optimizations, while the other two topmost tri-layers (S and SS in Fig. S1a) have been fully relaxed to their minimum-energy structure. Fe-substituted zirconia has been modeled by substitution of one Zr by one Fe atom on the surface (or on the sub-surface tri-layer), for a total Fe content of ≈2 at% with respect to all Zr atoms and of ≈6 at% considering only the surface atoms.

With these structural models, we carried out spin-polarized DFT calculations with the Vienna Ab-Initio Simulation Package (VASP).⁴⁰ We chose as level of theory the Perdew-Burke-Ernzerhof (PBE)⁴¹ exchange-correlation density functional and the on-site Hubbard U-J term for correcting SIE⁴² associated to Zr and Fe (partially) occupied d states. In particular, as in previous works, we have used the rotationally invariant DFT+U scheme as implemented in VASP with an effective U-J value of 4.0 eV for both Zr and Fe d electrons. This is an average U-J calculated from the values derived via ab initio calculations (Unrestricted-Hartree-Fock embedded cluster) for FeO (Fe(II) U-J = 3.7 eV) and Fe₂O₃, (Fe(III) U-J = 4.3 eV).^43,44 This value has been extensively validated in previous works on defective Fe-containing oxides.^45–50 We have chosen the same U-J value for Zr d electrons in order to avoid artificial electronic overlocalization on the transition-metal center with the highest U-J parameter. Moreover, this choice is also consistent with the DFT+U results from recent studies on ZrO₂: this U-J value delivers the correct electronic structure of bulk ZrO₂⁵¹ and, besides, it has been shown that adsorption energy of O₂ is qualitatively unchanged upon variation of the Zr d U-J value from 0 to 5 eV.⁵²

Nuclei and inner core electrons were replaced by projector augmented wave potentials (PAW) as obtained from the VASP repository. The valence/outer-core electrons that are included in the KS-DFT SCF cycles are listed in parentheses for each atom: Zr (4s², 4p⁶, 4d², and 5s²), O (2s² and 2p⁴ with the intermediate core radius PAW potential), C (2s² and 2p²), Fe (3d⁶ 4s²), and H (1s¹). These electronic variables are described via a plane-wave basis set, with a kinetic energy cutoff of 800 eV and a gamma-centered 3 × 3 × 1 k-point mesh. These numerical parameters are required for convergence of the total electronic energies within a threshold of 1 meV per formula unit. Bader's Atom-In-Molecule partial charges⁵³ have been computed within the super-cell-based approach. Oxygen vacancies, O₂ adsorption, and OOH species have been considered only on a side on the slab (S in Fig. S1), thus dipole corrections have been applied to avoid long-range polarization from the periodic images along the 'z' direction. Oxygen-vacancy formation energies have been computed in vacuum, while for O₂ adsorption and OOH formation energy calculations have been performed considering a PCM-like implicit solvent model for water as implemented in VASP_sol.⁵⁴

Quantitative analysis was performed to obtain the amount of Fe and Zr (as Fe₂O₃ and ZrO₂) in the carbon-supported catalysts. Table I summarizes the ICP and TGA results. In both Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 samples, atomic percent of Fe and Zr are calculated from the ICP quantitative analysis shown in Table I, while the oxygen content is a nominal content calculated based on the amount of Fe. For Fe and ZrO₂ samples, it is clearly seen that they do not contain any ZrO₂ and Fe, respectively, indicating that the catalyst synthesis did not result in any cross-contamination. It is also observed that in the pure Fe sample (first row) a higher Fe content is found analytically, compared to the nominally expected amount based on the synthesis. This is likely due to the loss of some carbon (likely from the support) during the partial oxidation (PO). This decrease in sample amount further translates into an effective increase in the Fe wt%. In addition, the agreement between the results for 1^st and 2^nd row of Table I from two different techniques i.e., ICP and TGA, further support their validity. On the other hand, the metal content in ZrO₂, Fe_0.17Zr_0.83O_1.91, and Fe_0.07Zr_0.93O_1.97 (2^nd, 3^rd, and 4^th row) is only approximately half of the nominally expected value. This discrepancy is likely due to the carbon produced by decomposition of the Zr precursor, which is significantly larger in comparison to that produced in the Fe sample, since the amount of FePc(t-Bu)₄ is ≈70 times lower than the ZrCl₂Pc(t-Bu)₄. Although the absolute amount of Fe and ZrO₂ in Fe_0.17Zr_0.83O_1.91, and Fe_0.07Zr_0.93O_1.97 is 50% of the expected, the atomic ratios between them are still the same as theoretical.

The last row in Table I shows the analytical results from a non-heat treated Fe_0.07Zr_0.93O_1.97 sample, which served as a reference to determine whether there is any loss of metal precursors before the heat-treatment step. The TGA analysis reveals that the theoretical and actual metal loadings are in very close agreement with each other, which clearly confirms that no precursors were lost until the heat-treatment. On the other hand, we could not completely exclude some loss of precursors during the heat-treatment due to sublimation.

XRD characterization of the above samples can be found in a previously published article by Madkikar et al.,²⁷ which clearly indicated the formation of nanometric ZrO₂ in all samples. From the broadening of reflexes, a crystallite size of <10 nm was clearly confirmed after comparing the XRD data with our past study.³² From the here conducted TEM image analysis (Fig. 1), it was found that the (number-based) average particle size of ZrO₂ in Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 is very similar (≈3 nm) (black and red bars in Figs. 1a, 1b). On the other hand, the ZrO₂ particle size in pure ZrO₂ sample was ≈6.5 nm (yellow bar in Fig. 1c), which is in accordance with our past study on nanometric carbon-supported ZrO₂ catalysts.³² The smaller ZrO₂ particle size in the samples where Fe is introduced could be a combined effect of increased nucleation rate and/or hindered growth due to crystal strain generated in Fe-substituted ZrO₂. These hypotheses are based on the past papers by Leite et al.⁵⁵ and Chen et al.,⁵⁶ which also observed that oxide nanoparticles with a dopant (Nb-doped SnO₂ and Al-doped ZnO) have smaller particle size than undoped oxide nanoparticles.

Zoom In Zoom Out Reset image size

XPS analysis

High-resolution XPS spectra of the Fe 2p region for the Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 samples are shown in Fig. 2. Since the signal-to-noise ratio is unfortunately rather low due to the low amount of iron in the samples, reliable deconvolution of the spectra is impossible. Instead, we compare the binding energy of the Fe 2p_3/2 signal to known binding energy values of Fe³⁺, Fe²⁺, and Fe⁰ species reported in the literature (cf. green, blue, and brown markers in Fig. 2, respectively).^57,58

Zoom In Zoom Out Reset image size

The observed binding energy range for the Fe 2p_3/2 signals reveals clearly that the dominant oxidation state of Fe in both samples is 3+ (711.8–710.8 eV, green region).⁵⁷ Owing to the rather low signal/noise ratio, however, the simultaneous presence of Fe²⁺ as minor phase cannot be completely ruled out (709.5–707.1 eV, blue region). Metallic Fe (Fe⁰, 707.2–706.7 eV, brown region) is clearly absent in both catalysts. Additionally, ZrO₂ formation was also confirmed (data not shown) by comparing the narrow Zr 3d spectra to a previous study from our group.³²

NEXAFS analysis

From Fig. 3 it is evident that the experimental spectra at the Fe L_2,3 edges of Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 (light and dark blue lines, bottom panel of Fig. 3) closely resemble in shape the simulated multiplet structure of high-spin Fe³⁺ (bold red line, middle panel). This is further corroborated by the close match of the position of the L₃ white-line energy of our experimental spectra with previously-reported Fe in hematite (Fe³⁺).⁵⁹ It is also observed that the shapes of the L₃ and L₂ edge are similar to that of goethite FeO(OH) and hematite α-Fe₂O₃, where Fe is present as high-spin Fe³⁺ at ambient temperatures and pressures.^60,61 Thus, it can be stated quite unambiguously that Fe exists for the most part as high-spin Fe³⁺ in both Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97. Lastly, the pure ZrO₂ sample (containing negligibly small amount of Fe, ICP (Table I)) serves as a baseline and confirms the absence of significant amounts of Fe (Fig. 3).

Zoom In Zoom Out Reset image size

Mössbauer analysis

Based on the Mössbauer spectra (Fig. 4), only one type of Fe is present in the Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 catalysts (top and middle panel). The isomer shift (δ) and quadrupole splitting (QS) of the primary doublet are 0.5 and 1.0 mm/s, respectively. These values indicate that Fe is present in high-spin isolated Fe³⁺ state.⁶² The δ and QS of the secondary doublet also fits to the previous assignment. In addition, some paramagnetic relaxation related to the same Fe³⁺ are also identified. On the contrary, Mössbauer spectra of the catalyst sample containing only iron (denoted as Fe) looks completely different (bottom panel in Fig. 4).

Zoom In Zoom Out Reset image size

In this case, three phases, viz., metallic Fe (bcc and fcc), along with γ-Fe₂O₃ and Fe₃C are identified, while isolated Fe³⁺ is not detected. This clearly indicates that the formation of metallic Fe, iron carbide, and iron oxide is hindered when ZrO₂ is present. The finding that iron is present for the most part in form of isolated Fe³⁺ in our Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 catalysts can be compared to reports in the literature, for which iron in the active sites of Fe-N-C catalysts is mainly in the Fe²⁺ oxidation state.^63–65 Thus, since we do not detect any Fe²⁺ in our NEXAFS and in our XPS analysis, we can exclude the presence of ORR active Fe-N-C moieties in our Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 catalysts. The high-spin state of Fe³⁺, evident in NEXAFS data, indicates that Fe is likely coordinated by oxygen, typically observed for this spin configuration.^60,61 On the other hand, from the Mössbauer analysis we observe only isolated Fe³⁺. Combining these two observations, we can conclude that Fe³⁺ substitutes Zr⁴⁺ in the ZrO₂ structure, forming the solid solution Fe_xZr_1−xO_2−δ.

Oxidation state of Fe from DFT

In the following, we compare the ORR activity of our catalysts by comparing the potential at a fixed current (E_iR-free @ −1 A/g_cat) at which the faradaic current is 3–5 times higher than the capacitive current but still <5% of the diffusion limited current. This implies capacitive correction errors will be small and oxygen transport resistances are negligible at our chosen mass-specific reference current of −1 A/g_cat.

Determination of ORR currents is always performed after subtraction of the capacitive current contribution (measured in Ar-saturated electrolyte) from the overall currents measured in oxygen-saturated solution (Fig. 5a). Clearly, Fe_0.17Zr_0.83O_1.91 and Fe_0.07Zr_0.93O_1.97 are much more active than the catalysts containing only Fe or ZrO₂ (Fig. 5a, inset). In addition, we would also like to mention that the difference in the activity between ZrO₂ and Fe_0.07Zr_0.93O_1.97 does not originate from the difference in particle sizes (Fig. S6). This undoubtedly indicates a strong synergism between Fe and ZrO₂. It should be also noted that, since Fe is confirmed to be in Fe³⁺ oxidation state (see above), it is assumed that Fe is present as Fe₂O₃ for the ease of quantification. In addition, from our previous work, we concluded that the optimum ORR activity lies between 0.3–0.75 wt% Fe₂O₃ (7–17 at% Fe).²⁷ However, based on durability considerations, we decided to use the catalyst with lowest amount of Fe (i.e., 0.3 wt% Fe₂O₃ (7 at% Fe) ≡ Fe_0.07Zr_0.93O_1.97) as a benchmark catalyst for further analyses and characterizations. This is because Fe could be leached from the catalyst in the strongly acidic environment of a PEMFC. These liberated iron cations cannot only replace protons in the electrode layer and in the membrane, but can also act as Fenton's reagent which leads to the decomposition of locally formed hydrogen peroxide to hydroxyl and hydroperoxyl radicals which lead to membrane degradation and are thus detrimental to PEMFC long-term performance.^67,68

Zoom In Zoom Out Reset image size

From our DFT-based results, the much higher mass activity of Fe-substituted ZrO₂ with respect to ZrO₂ can be ascribed to the different surface patterns and, hence, potential oxygen (intermediate) adsorption sites. Besides surface Zr and Zr/Fe atoms in pure and Fe-substituted ZrO₂, respectively, we must consider the presence of oxygen vacancies (V_O). In pure ZrO₂, the concentration of V_O is very small, since the oxygen vacancy formation energy (ΔE_form) both in the bulk⁶⁹ and in the surface^51,70 is quite high. It is well known that ΔE_form can be dramatically decreased, and hence, V_O can be increased by aliovalent substitution of Zr⁴⁺ cations with trivalent species like Y, La, or Al.^71,72 While Sangalli et al. have studied oxygen vacancies in bulk Fe-substituted ZrO₂,²⁶ we focus here on surface vacancies, due to their possible implication on the ORR catalysis. As detailed in Fig. S3 and Table S1 of the SI, the presence of Fe atoms dramatically decreases ΔE_form. For any possible surface vacancy position in ZrO₂, ΔE_form ranges between 5.6 and 5.9 eV, while in Fe-substituted ZrO₂ ΔE_form decreases to a value between 0.07 and 0.8 eV. It is noteworthy here that ΔE_form decreases not only for the generation of oxygen vacancies which are directly linked to Fe (along Fe-O-Zr bonds), but also for the other Zr-O-Zr bonds along the surface slab. This effect arises from the strong hybridization of Fe d-e_g states and O p states, which significantly lowers ΔE_form by allowing the convenient delocalization of the extra charge (2 e⁻ left by the leaving oxygen atom) onto the O sublattice (Table S1 in SI).⁷³ This is also a further indication that, different from pure ZrO₂ where vacancies induce the formation of Zr³⁺ centers, Fe exists mainly as Fe³⁺ without a formal reduction to Fe²⁺ in Fe-substituted ZrO₂.

Fig. 5b shows mass activities at 0.3, 0.4, 0.5, and 0.6 V_RHE as a function of temperature in an Arrhenius representation. Via a linear regression, the activation energy (E_act) was determined and is reported in the last part of this manuscript. A Tafel analysis (Fig. 5b, inset) of the Fe_0.07Zr_0.93O_1.97 catalyst in the range of ≈0.4–4 A/g_cat reveals a Tafel slope (TS) between 200–160 mV/decade (from 10–40°C). These TS values are slightly lower compared to pure ZrO₂ based catalysts (180–210 mV/decade), which were reported in an extensive study by Mittermeier et al.²¹

The influence of catalyst loading on H₂O₂ yield and the apparent mass activity of the Fe_0.07Zr_0.93O_1.97 catalyst was also investigated by RRDE. Four catalyst loadings were tested (30, 72, 288, and 576 μg/cm²) and the H₂O₂ yield at each loading was evaluated (Fig. 6). The loadings were selected such that the difference between consecutive loadings should be at least a factor of two, with the lowest and highest loading differing by a factor of ≈20. Peroxide yield is only shown within the potential window where a meaningful signal-to-noise ratio is obtained, i.e., where capacitive corrections to the disk currents were small (see above). It is readily observed that an increase in the catalyst loading is accompanied by a decrease of the H₂O₂ yield.

Zoom In Zoom Out Reset image size

At the lowest loading of 30 μg/cm², the fraction of oxygen being reduced to hydrogen peroxide amounts to ≈37%, while at the highest loading of 576 μg/cm² it decreases to <10%. Bonakdarpour et al.⁷⁴ and Muthukrishnan et al.⁷⁵ have reported a similar trend for Fe-N-C type catalysts. It is also worth mentioning that, for similar loadings, the H₂O₂ yield from Fe_0.07Zr_0.93O_1.97 is approximately half compared to the Fe-N-C catalysts in the study by Bonakdarpour et al.⁷⁴ When compared to pure ZrO₂ catalysts which produce mainly peroxide, it is hypothesized that Fe_0.07Zr_0.93O_1.97 has two sites for the ORR:²¹ ZrO₂ producing mainly H₂O₂, while Fe and/or oxygen vacancies convert the generated H₂O₂ to H₂O. This hypothesis is underpinned by the fact that the H₂O₂ yield decreases with increasing catalyst layer thickness, i.e., as the residence time of generated H₂O₂ within the catalyst layer increases, providing more time for H₂O₂ decomposition inside the layer.

Another interesting point to discuss here is that when the ORR mass activities of Fe_0.07Zr_0.93O_1.97 are compared for different loadings (30–576 μg/cm²), an increase in mass activity (≈2.3 times at 0.65 V_RHE) for thick films (576 μg/cm²) is observed (Fig. 7). When a similar analysis is being performed on the data for the Fe-N-C catalyst by Bonakdarpour et al., a mass activity increase of ≈2.4 times at 0.65 V_RHE can be noted when the loading is increased from 80 and 400 μg/cm².⁷⁴ It is worthwhile to mention that we performed the above analysis at a potential greater than half-wave potential (0.65 V_RHE) to avoid the use of high measured currents, which would introduce significant artefacts. The origin of these artefacts is the possible total consumption of oxygen in part of the catalyst layer at high currents, since, due to its tortuosity and porosity, the oxygen diffusion in the catalyst layer is much slower than in solution.

Zoom In Zoom Out Reset image size

This increase of the mass activity of PGM-free catalysts with film thickness is counter-intuitive if compared to Pt-based catalysts, where the mass activity would decrease with increasing catalyst layer thickness due to mass transport limitations.^76,77 Together with the lower hydrogen peroxide yield observed at higher electrode loadings (equivalent to a higher number of electrons utilized per O₂ molecule), this behavior may be characteristic for ORR catalysts which primarily reduce oxygen to hydrogen peroxide, but at the same time catalyze the chemical decomposition of hydrogen peroxide to O₂ and H₂O.^21,74

In order to elucidate the origin of the different products obtained with pure ZrO₂ and Fe-substituted ZrO₂ (defective and non-defective), we have considered adsorption of O₂ on all possible catalytic active sites for both materials. In non-defective ZrO₂, we consider only Zr sites due to the very low concentration of V_O expected from the high ΔE_form value (see Table S1 (SI)). In Fe-substituted ZrO₂, we have considered Fe and Zr and non-equivalent Fe-V_O-Zr and Zr-V_O-Zr, which present a very low ΔE_form and are likely to occur under operating conditions. As shown in Table S2 (SI), O₂ adsorbs weakly on top of Zr atoms both in non-defective pure and Fe-substituted ZrO₂ with a negligible associated surface-molecule charge transfer. Without oxygen vacancies, O₂ is not expected to be adsorbed on top of Fe atoms in non-defective Fe-substituted ZrO₂. Regarding defective Fe-ZrO₂, O₂ adsorption is favored on either Fe-V_O-Zr or Zr-V_O-Zr, due to a direct charge transfer from the reduced surface to O₂ antibonding orbitals, which results in stable bridge-like peroxide/superoxide species (Fig. S4b, c, and d in the SI). We must note that only for molecular oxygen adsorption on the Fe-V_O-Zr we obtain an overall favorable energy balance (i.e., negative enthalpy) for the two subsequent processes of oxygen vacancy formation and O₂ adsorption (data listed in Table S2 in the SI). Such low affinity of ZrO₂ for O₂ and, on the other hand, the convenient adsorption of O₂ on the easily-formed oxygen vacancies in Fe:ZrO₂ represent the first step and a necessary condition for a much higher ORR activity of Fe-doped zirconia.

On these stable species, we have considered the two proton-coupled electron transfer (PCET) steps of the ORR, leading to the formation of H₂O₂ or H₂O via a first ^*OOH intermediate. In pristine zirconia, where V_O are difficult to form, our calculations predict a stable Zr-bound ^*OOH species (Fig. 8a). On the other hand, by considering the active sites made by the formation of different oxygen vacancies created around a Fe cation, our results show the formation of both a stable ^*OOH species (Fig. 8b) and a dissociated intermediate made of ^*OH and ^*O species (Fig. 8c). This dissociation does not occur for Zr-V_O-Zr sites in Fe-substituted ZrO₂ (Fig. 8d).

Zoom In Zoom Out Reset image size

The species in Fig. 8c can conveniently form only water after the second PCET step, because the oxygen-oxygen bond has already been dissociated in the first PCET. When the ^*OOH species is stable, the second protonation step can lead to H₂O₂, if the second proton goes to the un-protonated oxygen atom, or to H₂O via the dissociation of the oxygen-oxygen bond and the formation of a surface bound ^*O species. We computed the total energy variations related to the reactions R1 and R2 for each catalytic site (CS) surveyed:

considering the equivalence, at zero applied bias, between the energy of the proton-electron couple and half the energy of molecular hydrogen (H⁺ + e⁻ ⇆ ½ H₂) as in the theoretical standard hydrogen approach.⁷⁸ Thus, in the calculation of energy variations for the R1 and R2 reactions the term E(H⁺ + e⁻) has been substituted by ½ E(H₂). The aim of these energy variations is to assess the intrinsic thermodynamic balance between R1 and R2 for the catalysts under study at the different CS conditions. The computed energy landscape is shown in Fig. 9.

Zoom In Zoom Out Reset image size

From these energies, we can discern whether the formation of H₂O₂ (R1, red lines in Fig. 9) or H₂O (R2, blue lines in Fig. 9) is preferred on each catalytic site. Negative/positive energy variations (below/above the reference gray line in Fig. 9) correspond to thermodynamically favorable/unfavorable processes. We must note here that these energy variations can only assess the selectivity toward the final products at a given CS, but they are not sufficient to quantitatively assess the overall different activities of the two catalysts.

According to these results, pure defective ZrO₂ leads mainly to H₂O₂ production, which is in agreement with our previous findings with Fe-free ZrO₂/C catalysts.²¹ On the other hand, in Fe-substituted ZrO₂ all oxygen vacancies around Fe will lead to a strong preference for the formation of H₂O, while at these active sites the formation of H₂O₂ is unfavorable. A lower amount of H₂O₂ production in Fe-substituted ZrO₂ is to be ascribed to oxygen vacancy active sites in the proximity of Zr atoms. This is in agreement with the experimental selectivity data from RRDE (see Fig. 6), where we observe higher production of H₂O even at the lowest loading in Fe-substituted ZrO₂ compared to pure ZrO₂, but still together with H₂O₂.

Fig. 10a depicts the measured geometric currents (I_geo) recorded in anodic scans (at 10 mV/s) in H₂/O₂ configuration in a 5 cm² active area single-cell PEMFC for the Fe_0.07Zr_0.93O_1.97 catalyst (solid lines) vs. HFR-corrected potential (E_iR-free). These data are not corrected for the capacitive contribution (measured in N₂), since it is very small over the relevant potential window between 0.1–0.7 V (see black solid line in Fig. 10a). Furthermore, a closer inspection of the H₂/N₂ data shows that Fe_0.07Zr_0.93O_1.97 does not seem to catalyze the hydrogen oxidation in this potential range, as is evident from the absence of oxidative currents for cross-over H₂ (i.e., H₂ permeating from the anode to the cathode compartment). The oxidative currents are <0.5 mA/cm², while typical Pt/C cathode catalysts would show an oxidative current in the range of a few mA/cm² depending on the pressure, temperature, and relative humidity.⁷⁹ After comparison of I_geo and I_m (capacitively corrected, dotted lines in Figs. 10a, 10b taken from Ref. 21), we can conclude that the Fe_0.07Zr_0.93O_1.97 is clearly more active than the Fe-free ZrO₂ catalyst from our past study,²¹ as one would have already expected from the published RDE data.²¹

Zoom In Zoom Out Reset image size

Fig. 10b inset depicts the Tafel analysis for mass-specific ORR currents from Fe_0.07Zr_0.93O_1.97 between 40–100°C determined between ca. 2–20 A/g_cat. The Tafel analysis yields TS values of 170–130 mV/decade, which are slightly lower than those determined by TF-RDE measurements. From this data, the mass activity at any given iR-free potential is extracted and compared in Fig. 10c for PEMFC and RDE measurements in an Arrhenius type representation. At first glance, it is easily observed that the mass activities determined from PEMFC measurements are higher compared to the ones obtained from RDE measurements. When compared at same temperature of 40°C, I_m from the PEMFC measurement is ≈3-fold higher compared to the RDE results at a potential of 0.6 V_iR-free (black symbols); this factor increases to ≈5-fold as the potential decreases to 0.2 V_iR-free (green symbols). From Fig. 7, an estimate of the difference in mass activity with respect to the catalyst loading, i.e., between 72 μg/cm² (in RDE) and 380 μg/cm² (in PEMFC) can be obtained and would be around 1.8 times at 0.65 V_iR-free. However, the RRDE analysis reported in Fig. 7 was performed at 20°C, while the above PEMFC and RDE data are compared at 40°C. Thus to confirm whether there is any additional increase in mass activity at higher temperature for a thicker film, a further RDE measurement was performed at 40°C with a higher catalyst layer loading of 576 μg/cm². Fig. 11 compares the mass activity at 40°C for the thin- and thick-film RDE with the results obtained from PEMFC measurements. It becomes evident that I_m from a thick-film RDE measurement closely matches the I_m from PEMFC measurement with a similar loading at 0.65 V_iR-free.

Zoom In Zoom Out Reset image size

This demonstrates that the increase in apparent mass activity of the Fe_0.07Zr_0.93O_1.97 catalyst correlates with the increase in catalyst loading. We cannot find any reason for a change in pure electrochemical ORR mechanism only due to a loading increase. But we know from RRDE that the Fe_0.07Zr_0.93O_1.97 catalyst generates a significant amount H₂O₂ (Fig. 6) and we have experimentally observed (Fig. S5) that Fe_0.07Zr_0.93O_1.97 is a good catalyst for H₂O₂ chemical decomposition. Combining these observations, it is likely that a change in the catalyst loading can affect the residence time of produced H₂O₂ (Fig. 6) within a catalyst film, which relates to the local H₂O₂ concentration. Thus, we can hypothesize that the increase of the mass activity with loading is due to the shift of to higher potentials because of the depletion of produced H₂O₂ via chemical decomposition (in thicker-films), leading to an increase in the rate of electroreduction of O₂ to H₂O₂. The origin of this phenomenon is the same as reported by Mittermeier et al.;²¹ however, in that case the activity of Fe-free ZrO₂ catalysts in PEMFC was lower than in thin-film RDE due to the opposite reason, namely higher local H₂O₂ concentration in PEMFC.

The activation energy (E_act) of the Fe_0.07Zr_0.93O_1.97 catalyst at different cathode potentials (E_iR-free), was determined using the Arrhenius representation from PEMFC and thin-film RDE measurements (Fig. 10c). It is observed that E_act of Fe_0.07Zr_0.93O_1.97 from RDE and PEMFC measurements lie in a similar range (12–20 kJ/mol), which are comparable to the E_act of Fe-free ZrO₂ from RDE (15–25 kJ/mol) and lower from PEMFC measurements (20–30 kJ/mol).²¹