Highly active postspinel-structured catalysts for oxygen evolution reaction

The rational design principle of highly active catalysts for the oxygen evolution reaction (OER) is desired because of its versatility for energy-conversion applications. Postspinel-structured oxides, CaB2O4 (B = Cr3+, Mn3+, and Fe3+), have exhibited higher OER activities than nominally isoelectronic conventional counterparts of perovskite oxides LaBO3 and spinel oxides ZnB2O4. Electrochemical impedance spectroscopy reveals that the higher OER activities for CaB2O4 series are attributed to the lower charge-transfer resistances. A density-functional-theory calculation proposes a novel mechanism associated with lattice oxygen pairing with adsorbed oxygen, demonstrating the lowest theoretical OER overpotential than other mechanisms examined in this study. This finding proposes a structure-driven design of electrocatalysts associated with a novel OER mechanism.


Introduction
The oxygen evolution reaction (OER: 4OH À / O 2 + 2H 2 O + 4e À in alkaline conditions) plays an essential role in energyconversion applications such as water electrolysis and rechargeable metal-air batteries. [1][2][3] Since this reaction intrinsically involves large overpotentials causing colossal energy loss, precious-metal oxides (e.g., RuO 2 and IrO 2 ) are presently utilized as typical OER catalysts. [4][5][6] Despite their high performance, large-scale applications are restricted because of their scarcity and high cost. Accordingly, much effort has been directed toward the development of highly active transition metal oxide catalysts consisting of earth-abundant and low-cost elements. 7,8 Most of the transition metal oxide catalysts, such as spinel and perovskite, comprise tetrahedral and octahedral metal-oxygen units. 1,[9][10][11] The perovskite-structured oxides, one of the most well-studied experimentally and theoretically catalyst systems, 1,12 consist of vertex-sharing octahedra, in which a single-site adsorption/reaction mechanism is widely accepted as adsorbates evolution reaction (AEM). 12,13 The reactants are adsorbed on coordinatively unsaturated sites (CUS) formed by the extraction of oxygen at an octahedral vertex. Since the neighboring transition metal sites in vertex-sharing octahedra are far from each other, the bridging adsorption of adsorbates on two active sites is disturbed.
Several structures in transition metal oxides possess particular geometric conditions such as smaller transition-metal interatomic distances than that of vertex-sharing octahedra in perovskite, inducing interactions between adsorbates and multiple sites on the catalyst surface. Accordingly, dual-site adsorption/reaction mechanisms that the reactant bound to the CUS is also connected to another atom in the surrounding polyhedra are manifested by experiments and theoretical calculations. 5,[14][15][16][17] For example, the dual-site reaction mechanism bridging B-site octahedral CUS metal and A 0 -site pseudosquare coordinated transition metal has been suggested in the A-site-ordered quadruple perovskite (AA 0 3 B 4 O 12 ). 15,16 The dual-site reaction mechanism has been experimentally and theoretically examined in the rutile-structured RuO 2 . 5,18 RuO 2 is composed of one-dimensional edge-shared RuO 6 octahedral chains gathered by sharing vertices. The reaction mechanism reported in the RuO 2 (110) surface involves a reaction step where an oxygen atom adsorbed on the Ru CUS combines with the oxygen atom (O BRI ) bridging two Ru atoms in the octahedral chain neighboring to CUS. 5 Recently, Sugawara et al. reported that CaFe 2 O 4 exhibits higher OER activity than other Fe oxides, 17 suggesting a novel reaction mechanism in which three Fe atoms participate in direct O-O bond formation based on densityfunctional-theory (DFT) calculations. Although the geometric feature associated with multi-site adsorption and reaction is reasonably described, the unusual 3-step reaction via simultaneous adsorption of two OH À species on Fe CUSs (Fe CUS ) and two electrons transfer is assumed in this mechanism, in contrast to the ordinary 4-step reaction in which one OH À and one electron are sequentially involved at each step.
The crystal structure of CaFe 2 O 4 consists of edge-sharing FeO 6 octahedra chains like rutile. It can be classied as the postspinel structure, a high-pressure polymorph of spinel. Fig. 1 shows the crystal structures of perovskites, spinels, and postspinels. CaFe 2 O 4 -type postspinel structure has the onedimensional framework of octahedra with shared edges, including Ca ions in the voids, distinct from spinel and perovskite with the three-dimensional octahedral framework with shared edges and corners, respectively. Considering that the OER activity of CaFe 2 O 4 may be derived from the structural feature, the postspinel-related series of CaB 2 O 4 (B ¼ Cr and Mn) must exhibit higher OER catalytic activity than the spinel or perovskite oxides.
In this paper, we investigated the OER catalytic activities of postspinel-structured CaB 2 O 4 (B ¼ Cr, Mn, and Fe) and systematically compared activities with perovskite LaBO 3 and spinel ZnB 2 O 4 . Regardless of B-site transition metals, the OER activities in CaB 2 O 4 oxides are monotonically superior to those of ZnB 2 O 4 and LaBO 3 counterparts, which is supported by lower charge-transfer resistance in CaB 2 O 4 . We performed DFT calculations to reveal the origin of OER activity in CaFe 2 O 4 by remodeling the regular 4-step reaction mechanism from the previously reported 3-step mechanism 17 and compared with the comparison with several possible mechanisms. We eventually found a novel 4-step reaction mechanism with lower theoretical overpotential, where the adsorbed oxygen on the Fe CUS and the adjacent O BRI were desorbed to generate oxygen. This nding suggests a new design principle for improving catalytic activity in multiple crystal structures of transition metal oxides.  19 A mixture of ZnO (99.9%) and Fe(NO 3 ) 3 $9H 2 O (99.9%) at a molar ratio of 1 : 2 was dissolved in nitric acid solution ($5 M), to which a ve-fold excess of citric acid and one-fold excess of 1,2-ethanediol were added to the solution with stirring. The resulting solution was heated for 573 K and maintained at this temperature for 1 h to dry. Subsequently, the dried powder was red using a furnace at 673 K for 1 h and then 1273 K for 10 h in air with occasional grindings. LaCrO 3 and LaMnO 3 were also obtained using the polymerized complex method from mixtures of La(NO 3 ) 3

Basic characterization
The as-synthesized samples were identied by X-ray powder diffraction (XRD) with Cu-Ka radiation (Ultima IV, Rigaku, Japan). The synchrotron XRD (SXRD) patterns were collected using a Debye-Scherrer camera installed at the BL02B2 beamline in SPring-8, Japan. The wavelength was determined as 0.49968 A using CeO 2 as a reference. The SXRD data were analyzed using the Rietveld renement program RIETAN-FP. 20 Specic surface areas were determined by Brunauer-Emmett-Teller (BET) analysis of Kr gas adsorption data (BELSORP-max, MicrotracBEL, Japan). The morphologies of all the catalysts were conrmed by scanning electron microscopy (SEM) images (TM3030, Hitachi High-Tech, Japan). X-ray absorption near edge structure (XANES) spectra of Cr, Mn, and Fe K-edges were collected in the transmission mode at the BL14B2 beamline in SPring-8. The X-ray absorption spectra were normalized by spline functions between pre-edge and post-edge regions using Athena program of the IFEFFIT package. 21

Electrochemical characterization
Working electrodes were prepared using the drop-casting method of inks containing catalysts on glassy carbon electrode, referred to previous papers. 10 A 5 wt% proton-type Naon suspension (Sigma-Aldrich), 0.1 M KOH aqueous solution (Nacalai Tesque, Inc., Japan), and tetrahydrofuran (THF, Sigma-Aldrich) were mixed at a ratio of 2 : 1 : 97 in volume. The catalyst ink was prepared by mixing 5 mg of catalyst, 1 mg of acetylene black (Denka Co., Ltd, Japan), and 1 mg of the THF solution. A 6.4 mL of catalyst ink was taken with stirring and drop cast onto the glassy-carbon disk electrode with 4 mm diameter.
Electrochemical measurements were conducted using a rotating-disk electrode rotator (RRDE-3 A, BAS Inc., Japan) and a bipotentiostat (model-2325, BAS Inc., Japan). We used a Pt wire electrode and a Hg/HgO electrode (International Chemistry Co., Ltd, Japan) lled with a 0.1 M KOH aqueous solution (Nacalai Tesque, Inc., Japan) as the counter and reference electrodes, respectively. All electrochemical measurements were conducted under O 2 saturation at room temperature. This xed the equilibrium potential of the O 2 /H 2 O redox couple to 0.304 V versus (vs.) Hg/HgO. The disk potential was controlled between 0.3 and 0.9 V vs. Hg/HgO at a scan rate of 10 mV s À1 . The disk potential was represented in those vs. reversible hydrogen electrode (RHE), with IR-compensation (R ¼ 43 U). The capacitive effect was compensated by averaging the cathodic and anodic scans.
Chronoamperometry (CA) was conducted at 1.6 V vs. RHE, where IR-compensation was not made. The electrochemical surface area (ECSA) was determined by scanning non-faradaic region between 0.0 and 0.1 V vs. Hg/HgO, according to the literature. 22 Electrochemical impedance spectroscopy (EIS) measurement was conducted using an electrochemical analyzer (760E, BAS Inc., Japan) at 1.7 V vs. RHE at frequencies ranging from 0.1 Hz to 1 MHz.  [23][24][25] The generalized gradient approximation (GGA) parametrized by Perdew, Burke, and Ernzerhof (PBE) 26 were adopted to express exchange-correlation interactions. The strong on-site coulombic interactions on the localized 3d electrons were treated with the GGA + U approach. 27 The U eff ¼ 3.5, 4.0, and 3.9 eV were adopted for Cr, Mn, and Fe 3d orbitals, which were selected to reproduce the experimental oxidation enthalpy, as reported previously. 28,29 The PAW potential data-set with radial cutoffs of 2.3 A for Ca, Cr, Mn, Fe, Zn, and 1.52 A for O were employed, where Ca-3s, 3p, 4s, Cr-3p, 3d, 4s, Mn-3p, 3d, 4s, Fe-3d, 4s, Zn-4s, 4p, 3d, O-2s, 2p were described as valence electrons. Table S11 † summarizes the magnetic structures and nominal electron congurations considered in this work. The plane-wave cutoff energy was set to 500 eV for all calculations. The Brillouin zone was sampled using k 1 Â k 2 Â k 3 mesh points according to the Monkhorst-Pack scheme. 30 The mesh count for each direction was selected as the near natural number of 35 per lattice parameter (1 A À1 ). The lattice constants and internal coordinates were optimized until the total energy difference and residual forces converged to less than 10 À5 eV and 10 À2 eV A À1 , respectively. According to literature, 31-33 oxygen 2p band centers and unoccupied 3d band centers of transition metal atoms were computed from the projected density of states (DOS) as follows:

Density-functional-theory calculation
(1) and respectively. Here, f 2p (E) and f 3d (E) are DOS projected on O-2p and transition metal 3d orbitals, respectively; E F is the Fermi energy; and E max is the upper bound of unoccupied 3d bands. The E max value was set as 10 eV higher than that of E F . The number of conduction bands was increased until the shapes of projected DOS were converged.

Slab model surface energy and theoretical overpotential
The electronic structures of OER intermediates on the (001) surface of CaFe 2 O 4 terminated by exposed FeO 5 pyramids were investigated using DFT calculations. The slab models in Fig Fig. S6b †). The atomic positions were optimized except for atomic layers in the bottom of slab models (magenta areas in Fig. S6b †) to calculate surface energies. The other computational conditions, including the PAW data-set, U eff values, plane-wave cutoff energies, total energy differences, and residual forces, were identical with bulk calculations.
The surface energies of CaFe 2 O 4 under the equilibrium conditions in OER were calculated according to the procedure proposed in the literature. 15,18 The surface Gibbs free energy can be described for CaFe 2 O 4 as follows: where E DFT (slab) is the total energy of the slab model using DFT calculations; A is the surface area of the slab model. N Z and m Z (Z ¼ Ca, Fe, and O) are dened as the number of the atoms in the slab model and chemical potentials, respectively. The chemical potentials are determined under the equilibrium condition of water splitting. In agreement with the computational hydrogen electrode model described in the literature, 34,35 the chemical potential of oxygen can be expressed as a function of pH and f, the potential difference between the working electrode and the reference electrode, as follows: where E DFT (CaFe 2 O 4 ) is the total energy of bulk CaFe 2 O 4 . By solving eqn (3) for m Fe and substituting it with eqn (4) into (3), the surface energy G is obtained as a linear function dependent on m Ca . In this work, we constructed reaction mechanisms from reported AEM 12,13 and lattice-oxygen-mediated mechanism (LOM) 13 and conducted surface calculations for the mechanisms listed in Tables S12 and S13: † AEM-O BRI , the LOM-O BRI , AEM model, and dual-site AEM models referred by Sugawara et al. 17 In these reaction steps, the *X/*Y surface state of postspinel-structured CaFe 2 O 4 is determined as using the binding state *X for Fe CUS and the binding state *Y for adjacent Fe CUS with O BRI . Thebondings of *O-O BRI and *OOH-O BRI surfaces exhibit interactions between adsorbed oxygen and O BRI . For each of the individual surfaces, the free energy change DG *X/*Y (*X/*Y: adsorbed surfaces) was calculated using equations in Table S14. † For each of the six reaction mechanisms, the free energy change DG n (n: reaction steps) in the individual reaction was dened as each formula in Tables S15 and S16. † Using the largest DG n (n: reaction step), the value of theoretical overpotential (h th ) was calculated using the following equation: Results and discussion were assigned to orthorhombic phases, as reported previously. [36][37][38] Rietveld renement results obtained by using the SXRD data conrmed that the rened lattice parameters were similar to those previously reported in all samples ( Fig. S1 and Tables S1-S9 †  Fig. 3 shows the X-ray absorption spectra at Kedges of Cr, Mn, and Fe. The K-edge absorption positions of transition metals for perovskites, spinels, and postspinels are close to those of pure trivalent metal oxide references (B 2 3+ O 3 ) rather than aliovalent references (B 2+ O and B 4+ O 2 ), although the differences in local structures around the B sites appeared in shapes in higher energy ranges than absorption energies. The structural and spectroscopic analyses exclude the possible effects of valence on OER activities, 33 thus the effects of crystal structure on activity can be investigated in this study.
Specic surface areas determined by the BET analysis of Krgas adsorption data ranged in typical values between $1.1 and $2.5 m 2 g À1 , as listed in Table 1. The BET specic surface areas were adopted to normalize the current densities per surface area of catalysts in the electrochemical analysis. The values of ECSA in Table 1 were observed in the region of 10-27 m 2 g À1 , displaying the same proportion of the maximum to the minimum, compared to that of BET surface areas. The SEM images in Fig. S2 † indicate that the particle sizes roughly ranged between 0.1 and 10 mm for all. The crucial differences in grain size were not observed between CaB 2 O 4 and references containing the same B ion. The SEM observations roughly conrmed similar morphologies among all samples, compatible with BET and ECSA analyses. Fig. 4  Signicant differences between postspinels and other structures were observed in Tafel plots and EIS analyses. Fig. 6 shows the Tafel plots for LaBO 3 , ZnB 2 O 4 , and CaB 2 O 4 (B ¼ Cr, Mn, and Fe). The Tafel slope of CaFe 2 O 4 (53 mV dec À1 ) was much smaller than that of ZnFe 2 O 4 (102 mV dec À1 ) and LaFeO 3 (78 mV dec À1 ). Clear differences in Tafel slopes between postspinel-structured oxides and counterparts were also observed in Cr and Mn oxides. Since the Tafel slope varies in dependent on the rate-determining step (RDS), 39 the observed differences in Tafel slope indicate that the RDS is altered by crystal structures. Nyquist plots are displayed in Fig. 7  We investigated the long-term stability and surface crystalline states of CaB 2 O 4 . Fig. 8 shows the CA currents normalized by initial currents in CaCr 2 O 4 , CaMn 2 O 4 , and CaFe 2 O 4 . CaFe 2 O 4 exhibited no substantial degradation in OER activity. This    even in the as-cast sample and further evolution of the amorphous layer aer CA (Fig. 8). CaMn 2 O 4 also possessed the amorphous surface in the as-cast sample, which is probably the cause of the initial degradation in CA. Gradual increases in current density were observed for several oxides (CaMn 2 O 4 and CaFe 2 O 4 ), but the origin was unclear at the present stage. Since CaFe 2 O 4 did not exhibit severe amorphization, the intrinsic feature of the crystalline surface is predominantly reected in the electrochemical analyses.
Our electrochemical experiments elucidated that the OER catalytic activities in postspinels CaB 2 O 4 (B ¼ Cr, Mn, and Fe) are superior to those of perovskites LaBO 3 and spinels ZnB 2 O 4 , irrespective of B metal ions. The commonly observed properties in CaB 2 O 4 suggest that the edge-sharing one-dimensional octahedra in postspinel structures predominate the reaction mechanism. We conducted DFT calculations to discuss the reaction mechanism on the surface of CaB 2 O 4 associated with the geometric feature of the coordination polyhedra, in addition to the bulk electronic factors possibly affecting the OER catalysis. Fig. S4 † shows the DOS generated from bulk-model DFT   (Table S10 †). Unlike the linear tendency in the h 0.05 vs. D plots for the perovskite oxides, 32 there was no clear trend (Fig. S5 †) in the spinels and postspinels in this study. Hence, almost isoelectronic states in the bulk form cannot explain the origin of OER activity in the postspinel-structured oxides.
To evaluate credible information on the OER on postspinel, we calculated the theoretical overpotentials from the surface free-energies. In the surface-state calculations, CaFe 2 O 4 was selected because of its robust crystalline surface in OER conditions, as demonstrated in CA and HRTEM studies. The calculations of surface states were conducted on the (001) plane with O BRI ions bridged by two Fe ions, referring to the reaction mechanism in RuO 2 with one-dimensionally aligned octahedra. 4,5 We originally investigated three different reaction mechanisms AEM-O BRI , LOM-O BRI , and AEM. Fig. 9 shows the surface geometries aer structural relaxations. First, in the AEM-O BRI (Fig. 9a) Fig. 9b) in which OH À is adsorbed to Fe CUS (step 1: */-+ OH À / *OH/-+ e À ), and desorbed with O BRI to evolve O 2 (step 2: *OH/-+ OH À / */* + O 2 + H 2 O + e À ). Third, we considered the AEM mechanism, where the adsorbates are solely bound to Fe CUS because of the absence of the O BRI atom (Fig. 9c). This mechanism is similar to the conventional OER mechanism in single-site adsorption.
The free energies DG *X/*Y at each reaction step of the aboveexamined models are listed in Table S17 † dened with formulae in Table S14. † The value of DG *X/*Y tended to increase in correspondence with the number of adsorbed atoms in the *X/ *Y surface state. We calculated the values of energy change DG n of each reaction steps and h th for the six reaction mechanisms (Table S18 †) as following calculations in formulae in Tables S15 and S16. † Fig. 10 shows the energy diagrams for each reaction mechanism. In Fig. 10, the thick lines represent the potential determining steps (PDSs) with the largest DG n in each mechanism, accompanied by h th values calculated from DG n at PDSs. The PDSs for AEM-O BRI (h th ¼ 1.33 V) and LOM-O BRI (h th ¼ 0.85 V) were assigned to step 3 (*O-O BRI + OH À / *OOH-O BRI + e À ) and step 4 (*/*OH + OH À / */-+ H 2 O + e À ), respectively. In contrast, step 2 with a signicant large h th (2.04 V) was the PDS in the AEM (*OH/* + OH À / *O/* + H 2 O + e À ). LOM-O BRI demonstrated the lowest h th among the 4-step mechanisms. Due to the high variation in surface structures and types of adsorbed species in PDSs, we could not identify any consistency in adsorption states of PDSs among the three reaction mechanisms. The PDSs of LOM-O BRI and AEM were categorized as transforming steps from *OH to *O, whereas the PDSs of AEM-O BRI were assigned to adsorption processes of the reactant OH À . We conclude that the LOM-O BRI with the lowest h th (0.85 V) examined in the present DFT calculation is the most probable mechanism to explain the high OER activity of postspinelstructured oxides.

Conclusion
In summary, we investigated the OER catalytic activity in the postspinel-structured oxides CaB 2 O 4 (B ¼ Cr, Mn, and Fe), revealing higher OER activities and smaller charge-transfer resistances than the perovskite-and spinel-structured counterparts. The DFT calculation on the surface of CaFe 2 O 4 elucidates that a novel reaction mechanism with the lowest theoretical overpotential, where O BRI and O ad are combined to generate oxygen, is more probable than the 3-step reaction mechanism with simultaneous adsorption of OH À proposed in the previous study. Consequently, the geometric congurations around adsorption sites tolerating additional bonding are another factor to activate OER beyond the conventional single-site OER mechanism.

Conflicts of interest
There are no conicts to declare.