Composition-Designed High Entropy Perovskite Oxides for Oxygen Evolution Catalysis

doi:10.21203/rs.3.rs-1944245/v1

Download PDF

Article

Composition-Designed High Entropy Perovskite Oxides for Oxygen Evolution Catalysis

https://doi.org/10.21203/rs.3.rs-1944245/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Oxygen evolution reaction (OER) catalysts play an essential role in energy-conversion electrochemical reactions. High entropy oxides (HEOs) were recently investigated as promising candidates to realize highly active and cost-effective OER catalysts. The vast composition space for HEOs needs considerable efforts to find possible catalysts, which disturbs the further development beyond simple chemical compositions like equimolar ones. In this study, we conducted the fast and efficient design of the perovskite HEOs of La(Cr, Mn, Fe, Co, Ni)O₃ with high OER catalytic activity using Bayesian optimization and elucidated the relationship between chemical compositions and OER catalytic activities. The HEOs with optimized compositions exhibited much higher activities than equimolar LaCr_1/5Mn_1/5Fe_1/5Co_1/5Ni_1/5O₃, which was previously reported as an active catalyst. Bayesian optimization adjusted the concentrations of OER active elements of Fe, Co, and Ni to enhance the catalytic activities and provided the insight that inactive elements (Cr and Mn) in HEOs even promoted the OER activities. These findings suggest the solution of data-based predictions to improve catalytic performances in multi-element transition metal oxides.

Oxygen evolution reaction (OER: 4OH^– → O₂ + 2H₂O + 4e^– in alkaline conditions) is one of the crucial electrochemical reactions, such as water-splitting for hydrogen production and metal-air battery¹. However, sluggish kinetics of OER force us to induce excess voltages over equilibrium potential, 1.23 V vs. reversible hydrogen electrode (RHE), to proceed with OER. Platinum group oxide catalysts, e.g., IrO₂ and RuO₂, are utilized to promote OER, but their high cost and scarcity hinder large-scale applications. Perovskite oxides represented by the general formula of AMO₃ (A: alkaline earth and rare earth metals, M: 3d transition metals) have been investigated as cost-effective catalysts because of their great varieties of chemical compositions². Especially, Fe-, Co-, and Ni-based perovskites such as LaNiO_3–δ³, PrBa_0.5Sr_0.5Co₂O₅⁴, and Ba_0.5Sr_0.5Fe_0.5Co_0.5O_3–δ² were reported as active catalysts. The OER catalysis in perovskites is generally limited by charge-transfer kinetics between the active site of M metals and the adsorbats^5,6. The electronic structures involving 3d band of transition metals and 2p band of oxygen are associated with OER kinetics, which can be altered by valence states and oxygen deficiency⁷. Electronic structures of perovskites have been adjusted to promote the charge transfers under the OER condition by mixing multiple elements^7–11. Indeed, OER activities in complex metal oxides, e.g., LaMn_0.3Co_0.7O₃⁸ and LaFe_0.1Co_0.9O₃⁹, are generally higher than those of perovskite oxides containing a single transition metal element, which are supported by experimental and theoretical electronic-state studies. Ternary 3d metal perovskite oxides of La(Mn,Co,Ni)O₃¹⁰ and LaNi_0.5Co_0.25Fe_0.25O₃¹¹ are more active than unary or binary perovskites.

High entropy oxides (HEOs) containing at least five transition metal elements in a single crystallographic site are recently investigated as promising materials such as active oxides of electrochemical capacitors¹², ionic conductors¹³, and OER catalysts¹⁴. The change in mixing entropy ΔS_mix = ) categorizes multi-metal oxides into HEO with ΔS_mix > 1.5R and middle entropy oxide with ΔS_mix = R–1.5R, where R is the gas constant (= 8.314 J K^–1 mol^–1) and c_n is the atomic concentration for each metal. The multi-metal oxides (N 5) are classified into HEOs with ΔS_mix 1.61R. Rock-salt HEOs (Co, Ni, Mn, Zn, Fe)₃O_3.2¹⁵ and spinel HEOs (Co_0.2Mn_0.2Ni_0.2Fe_0.2Zn_0.2)₃O₄¹⁶ exhibited higher OER activities than single 3d transition metal oxides, where ΔS_mix was considered to be a crucial parameter to contribute to the activity^16–18. Recently, perovskite HEOs were also studied as active OER catalysts. Nguyen et al. reported that perovskite HEO of LaCr_1/5Mn_1/5Fe_1/5Co_1/5Ni_1/5O₃ (M_1/5) displayed higher activities than LaMO₃ (M = Cr, Mn, Fe, Co, and Ni)¹⁹. They discovered a much more active catalyst with Co-rich composition LaCr_1/6Mn_1/6Fe_1/6Co_1/3Ni_1/6O₃ (M_1/6Co_1/3) than M_1/5, which was derived from both oxygen deficiency and high valence Co ions. In addition, Tang et al. reported that the catalytic activities were improved according to the increase in ΔS_mix from binary (La_0.6Sr_0.4)(Co_0.2Mn_0.8)O₃ (ΔS_mix = 1.27R) to quinary oxides (La_0.6Sr_0.4)(Co_0.2[Fe, Mn, Ni, Mg]_0.8)O₃ (ΔS_mix = 1.83R)¹⁸, claiming that the enhancement came from the lattice oxygen-participated mechanism reported in perovskites²⁰. However, it is significantly hard to estimate catalytic activities in enormous composition spaces of HEOs, e.g., 4.4 million samples are expected in a quinary system at 1 at% interval. Selected samples with nearly equimolar compositions were investigated in these reports. Since it is impossible to discover much active catalysts from experiments all over the compositions in the HEO system, rational strategies for HEO catalysts with high activities are desired. In this study, the data-driven method by Bayesian optimizations was adopted for perovskite HEOs of La(Cr, Mn, Fe, Co, Ni)O₃ to tailor chemical compositions for active catalysts effectively. We synthesized 28 composition-distinct perovskite HEO samples and investigated OER catalytic activities. The catalysts more active than the reported oxide were effectively discovered from the limited numbers of samples. The relationship between OER activities and chemical compositions revealed that both total and individual atomic concentrations of OER active elements (Fe, Co, and Ni) play an essential role in electronic structures of HEO to develop active catalysts including multiple transition metal ions.

We conducted three cycles of Bayesian optimization and experiment for HEO catalysts: syntheses, phase identifications by X-ray powder diffraction (XRD), electrochemical measurements, and Bayesian optimizations (Fig. 1). The samples whose compositions are determined using the Bayesian optimization and random numbers are denoted as BO-HEOs and R-HEOs, respectively. The first dataset was constructed by chemical compositions and catalytic activities of R-HEOs (15 samples), single-transition-metal-element oxides LaMO₃ with M = Cr, Mn, Fe, Co, and Ni (five samples), and LaFe_1/3Co_1/3Ni_1/3O₃ (denoted by M_1/3), as displayed in “1st cycle” of Fig. 1. The R-HEOs served to construct the low-biased random data in the first dataset. The first generation of optimization using the above-mentioned first dataset (21 samples in total) suggested five composition candidates of active catalysts as the BO-HEOs. The data for the BO-HEOs were added to the first dataset to form the second dataset with 26 samples. The second generation of Bayesian optimization provided five more compositions of BO-HEOs, adding up to 31 samples. The third generation of optimization using the second dataset generated three candidates of BO-HEOs, and then the sequence of the Bayesian optimization was terminated. The active HEO references of LaCr_1/5Mn_1/5Fe_1/5Co_1/5Ni_1/5O₃ (M_1/5) and LaCr_1/6Mn_1/6Fe_1/6Co_1/3Ni_1/6O₃ (M_1/6Co_1/3) were prepared as well as the previous work¹⁹.

Table S1 lists the concentrations of transition metal elements and mixing entropies (ΔS_mix) of transition metal ions for R-HEOs and BO-HEOs. Each sample of R-HEOs and BO-HEOs is denoted as Rl (l = 1–15) and BOn-m (n = 1–3, m: a, b, c, d, e), respectively. The n represents the generation, and the m gives the order of the hyperparameter λ determining the coefficient of variance for Bayesian optimization. All the Bayesian optimizations estimated the upper confidence bound (UCB) of achieved functions to generate candidates of chemical compositions. The functions of UCB were represented as the linear combination of mean µ(x) and variance σ(x) as follows: UCB(x) = µ(x) + λσ(x), where λ is hyperparameter. When the m changed from “a” to “e” in the same generation, the λ values were increased in our program (see “Bayesian Optimization” in Method); BOn-a has the most exploitive composition based on input data, whereas BOn-e is most exploratory with high confidences. It is apparent that all the compositions of BO-HEOs and R-HEOs were dispersed unlike the uniform compositions of M_1/5 and M_1/6Co_1/3. Especially, the compositions of BO-HEOs were biased toward Fe, Co, and Ni. Only the five samples (R4, R12, R13, R14, and BO1-e) were categorized into high entropy oxides satisfying the definition of ΔS_mix > 1.5R, whereas most of the others were categorized into the medium entropy oxides (R < ΔS_mix < 1.5R). While most of the samples did not satisfy the strict definition of high entropy, we kept the notation of HEO in this study.

The single-phase perovskite oxides were obtained for all the HEOs. Figure 2a displays the powder X-ray diffraction (XRD) patterns of LaMO₃ (M = Cr, Mn, Fe, Co, Ni) and references (M_1/3 and M_1/5). The XRD patterns of LaCrO₃ and LaFeO₃ were assigned as the GdFeO₃-type orthorhombic perovskite structures (space group: Pnma), whereas those of LaMnO₃, LaCoO₃, and M_1/3 were assigned as the rhombohedral perovskite structures (space group: R\(\stackrel{-}{3}\)c), referred from the simulated patterns of the inorganic crystal structure database (ICSD). For M = Ni (denoted as LaNiO_y), the stoichiometric perovskite phase could not be obtained because the strong oxidizing atmosphere is necessary to achieve the perovskite with Ni³⁺. The M_1/5 crystallized in the orthorhombic perovskite, which was consistent with the previous reports^19,21.

Figure 2b shows the XRD profiles for R-HEOs. Some samples (R1, R2, R3, R4, R5, R7, R9, and R12) crystallized in the orthorhombic perovskite structure, whereas the others (R6, R8, R10, R11, R13, R14, and R15) rhombohedral one. No impurity phase was observed for all. Figure 2c represents the XRD patterns for BO-HEOs. All the BO-HEOs crystallized in the rhombohedral perovskite with no impurity. Consequently, perovskite phases were successfully obtained in non-equimolar various compositions except for LaNiO₃, which needs strong oxidizing conditions. Then, this quinary system La(Cr, Mn, Fe, Co, Ni)O₃ was suitable for the investigation of vast chemical compositions to OER catalytic activities. The lattice constants of HEOs were refined by the Rietveld analysis using the XRD data (profiles in Fig. S1, parameters in Table S2). Figure 3 compares the lattice volume per formula unit (V_f.u) for all the R- and BO-HEOs. The average V_f.u. of BO-HEOs [= 57.3(4) Å³] was about 2% smaller than that of R-HEOs [= 58.5(8) Å³] (see dotted lines in Fig. 3). The lattice shrinkage in the BO-HEOs is attributed to the increase in the atomic contents of relatively small ions, Co³⁺ and Ni³⁺ in six-fold coordination.

The absence of a substantial amount of oxygen deficiencies in the BO-HEOs was confirmed by the Rietveld refinement using synchrotron X-ray diffraction data because of the significantly low solubility limit of the BO-HEO samples into the chloric acid in titration analysis. The occupancy factors at oxygen sites [g(O)] were retained near the unity [g(O) > 0.980]. Since the g(O) values refined in the same manner were consistent with the iodometric titration^4,22,23, we concluded that the BO-HEO samples contained negligible amounts of oxygen deficiencies.

Figure S2 shows the scanning electron microscopy (SEM) images for the BO-HEOs, in addition to the R11, which was the most active sample within the R-HEOs as described below. The particle sizes of BO-HEOs ranged between 0.1 and 1 µm, in contrast to the fine grains of about 0.1 µm for R11. Hence, we expected no superiority in morphology for BO-HEOs, which may contribute to catalytic activity.

Figure 4 displays the linear sweep voltammograms (LSVs) for the LaMO₃ (M = Cr, Mn, Fe, Co), LaNiO_y, references (M_1/3, M_1/5, and M_1/6Co_1/3), R-HEOs, and BO-HEOs. The current density was normalized by the disk area. The LSV curves and Tafel slopes of all the HEOs are shown in Fig. S3 and S4, respectively. The OER activity (A_OER) was determined by the current density at 1.6 V vs. RHE (see dashed lines in Fig. 4). In the single-transition-metal oxides, the late-3d transition metal oxides of LaFeO₃, LaCoO₃, and LaNiO_y were more active than the early series of LaCrO₃ and LaMnO₃ (Fig. 4a), as reported previously²⁴. If the activities of multielement oxides are dominated by the weighted averages of native activities of constituent elements as in the simple-mixing rule, the poorly active elements of Cr and Mn should be excluded. We examined M_1/3 composing of only three active elements of Fe, Co, and Ni, but the catalytic activity of M_1/3 was comparable to that of M_1/5 (Fig. 4a). This fact implies that the high catalytic activities of the HEOs originate beyond the simple mixing rule. M_1/6Co_1/3 displayed higher activity (0.12 mA cm^− 2) than M_1/5 as reported in the previous paper¹⁹. The R-HEOs demonstrated wide-ranging activities between 0.01 and 0.09 mA cm^− 2 (Fig. 4b), indicating that the OER catalytic activities of HEOs are significantly dispersed when no restrictions and biases are imposed in the composition. The activities of only two samples (R8 and R11) were comparable to those of M_1/3 and M_1/5, whereas the 13 remaining R-HEOs were not as active as M_1/3 and M_1/5. Consequently, we conclude that it is hard for the HEOs with vast composition space to find catalysts exceeding M_1/5 and M_1/6Co_1/3.

We conducted the Bayesian optimizations to adjust the chemical compositions by using the composition-activity dataset for LaMO₃ (M = Cr, Mn, Fe, and Co), LaNiO_y, M_1/3, and R1–R15 HEOs (see Fig. 1 and “Bayesian optimizations” in Method). Figure 4c displays the LSV curves for the BO1-m series. BO1-a and BO1-b were much more active (A_OER = 0.20–0.23 mA cm^− 2) than BO1-c, BO1-d, and BO1-e (A_OER = 0.08–0.13 mA cm^− 2). It is notable that BO1-a and BO1-b exceeded M_1/6Co_1/6, which was reported as the best catalyst in the previous study¹⁹. Even BO1-c, BO1-d, and BO1-e were more active than M_1/5. The second and third BO generations (Fig. S3g–3h) firmly generated high activities exceeding that of M_1/6Co_1/3 with high probability as well as the first BO generation. As a result, the activities of all the BO-HEOs were higher than or comparable to that of the M_1/5 reference. The present study reasonably demonstrates that the high probability (12/13 = 92%) of active BO-HEOs over M_1/5 is in contrast with the low probabilities (2/15 = 13%) of R-HEOs. Figure 4d compares LSVs between the most active catalysts in each generation (BO1-a, BO2-b, and BO3-a) and references of M_1/5, M_1/3, and M_1/6Co_1/3. All BO1-a, BO2-b, and BO3-a were more active than M_1/6Co_1/3. Especially, the latest generation of BO3-a exhibited the highest activity (A_OER = 0.26 mA cm^− 2), confirming that the evolution in the Bayesian optimization serves to refine the chemical compositions for more active HEO catalysts. Consequently, we succeeded in the Bayesian-optimization-based adjustments of compositions of highly active HEO of La(Cr, Mn, Fe, Co, Ni)O₃ in the limited number of trials.

We investigated the detailed contributions of each transition metal element to catalytic activities in the HEOs. Figure 5a-5b plots the A_OER and atomic contents for the HEOs. BO-HEOs exhibit higher activities than any R-HEOs and single transition metal oxide references. Comparing the activities in the HEOs, the 26-fold difference in A_OER was observed between the most active BO1-a and the least active R5 (Fig. 5a). The average A_OER in BO-HEOs [0.14(6) mA cm^− 2] was much larger than that of R-HEOs [0.04(2) mA cm^− 2], which is apparently related to the chemical composition. BO-HEOs contain larger amounts of OER active elements (Fe, Co, and Ni) than R-HEOs with diverse individual atomic contents (Fig. 5b), as the average value of the total concentration of OER active elements (Fe, Co, and Ni) in BO-HEO (90.1 at%) was much larger than that of R-HEO (57.5 at%). The correlation coefficients of total concentrations of combinations of three transition metals were calculated against the A_OER values (Fig. S5) to prove the tendency of concentrations of OER active elements to A_OER. The (Fe, Co, Ni)-content showed the highest coefficient (0.91) in the values of the other combinations. We concluded that the total atomic contents of Fe, Co, and Ni elements in HEOs play a crucial role in increasing the OER activities in La(Cr,Mn,Fe,Co,Ni)O₃ HEOs.

We studied the relationship between catalytic activities A_OER and total ratios of OER active elements (Fe, Co, and Ni) for HEOs as shown in Fig. 6. A positive linear correlation was found, indicating that the catalytic activities are dominated by (Fe, Co, Ni)-content in HEOs. The improvement in activities from M_1/5 to M_1/6Co_1/3 reported previously¹⁹ is consistent with the positive tendency of activities of HEOs according to the increase in (Fe, Co, Ni)-content. M_1/6Co_1/3 was exceptionally positioned out of the distribution. Although most of the data points are included near the linear fitting result within ± 60% (shaded area in Fig. 6), significant differences in activity were observed in similar (Fe, Co, Ni)-content. For instance, comparing the samples consisting of almost only Fe, Co, and Ni, the activity of BO1-a in the upper end of shaded area is about three times larger than that of M_1/3 in the bottom end of the shaded area, suggesting the superiority of the unbalanced composition between Fe, Co, and Ni [e.g., Fe/Co/Ni = 34/51/14 (at%) for BO1-a] to the equivalent compositions [Fe/Co/Ni = 33/33/33 (at%) for M_1/3]. Consequently, the appropriate adjustment of individual concentrations of Fe, Co, and Ni is the second essential factor to promote catalytic activities of La(Cr,Mn,Fe,Co,Ni)O₃.

We examined the effects of ΔS_mix on OER catalytic activity for HEOs. Figure 7 illustrates the relationship between A_OER and ΔS_mix for HEOs. There is a rough positive correlation between ΔS_mix and A_OER for R-HEOs (the green-shaded area), in which the M_1/6Co_1/3 is located at the top of this category. This trend is similar to the previous reports on other systems of perovskite HEOs (La_0.6Sr_0.4)(Co_0.2[Fe, Mn, Ni, Mg]_0.8)O₃¹⁸ and spinel HEOs (Cr_0.2Mn_0.2Fe_0.2Co_0.2Ni_0.2)₃O₄¹⁷, where the catalytic activities are increased as the ΔS_mix increases. In contrast, no clear correlation was observed for most BO-HEOs in the red-shaded area. This indicates that ΔS_mix unnecessarily affects the catalytic activity probably because the BO-HEOs are distributed in the limited range of chemical compositions around M_1/3 with relatively small ΔS_mix. The ΔS_mix dependence of catalytic activity may emerge in a narrow range of chemical compositions with high ΔS_mix. Considering that the OER activities were depended on chemical compositions displayed in Fig. 6, the modifications of individual concentrations of transition metals were more suitable than the ΔS_mix values to enhance OER catalytic activities in this system La(Cr, Mn, Fe, Co, Ni)O₃.

We collected the X-ray absorption spectra to investigate the valence states in the selected BO-HEOs with 90 at% of (Fe,Co,Ni) contents. Figure 8 shows the X-ray absorption near edge structure (XANES) spectra of Fe, Co, and Ni K-edges. The absorption energy E₀(M) was determined as the value exceeding 0.5 of the normalized absorption (M = Cr, Mn, Fe, Co, and Ni) listed in Table S4. M_1/5, M_1/3, M_1/6Co_1/3, BO3-a, and BO1-a exhibited similar absorption energy (~ 7122 eV) to the Fe³⁺ reference of LaFeO₃ rather than the Fe⁴⁺ reference of CaFeO₃ in Fe K-edge (Fig. 8a), indicating that the trivalent states are predominant for all of M_1/5, M_1/3, BO3-a, and BO1-a. BO-HEOs with 90 at% of (Fe,Co,Ni) contents contained Fe³⁺ ions, confirmed by XANES spectra (Fig. S6). The Cr ions in M_1/5 and M_1/6Co_1/3 are also considered to be the trivalent state, as determined from Cr K-edge spectra (Fig. S6). The Mn valence in M_1/5 and M_1/6Co_1/3, was estimated to be between + 3 and + 4 as shown in Fig. S6.

The Co K-edge absorption energy for M_1/5 and M_1/6Co_1/3 was ~ 7719 eV (Fig. 8b), which is between those of TeCo²⁺O₃ (7717.6 eV) and LaCo³⁺O₃ (7720.4 eV). This observation suggests the intermediate valence (tentatively, Co^~ 2.8+) between + 2 and + 3. Considering the Mn valence (+ 3.4) calculated from E₀(Mn) in Fig. S6, the charge transfer between Co and Mn ions in M_1/5 and M_1/6Co_1/3 reduced the Co valences from the trivalent state expected from the single-transition-metal oxides La³⁺M³⁺O₃ as follows: Mn³⁺ + Co³⁺ → Mn⁴⁺ + Co^2 + 25. In contrast, M_1/3, BO1-a, and BO3-a exhibited higher E₀(Co) of ~ 7720 eV than M_1/5, retaining the trivalent states. Because of the substantial decrease in Mn content from M_1/5, the charge transfer is suppressed for M_1/3 and BO1-a, contributing to the activity with Co³⁺ species. BO-HEOs with 90 at% of (Fe,Co,Ni) contents contained Co³⁺ ions, confirmed by XANES spectra (Fig. S6).

Figure 8c compares the Ni K-edge spectra for M_1/3, M_1/5, BO1-a, and BO3-a, together with references of TeNi²⁺O₃ and LaNi³⁺O₃. Unlike LaNiO_y synthesized under ambient conditions, LaNiO₃ was prepared in a strong oxidization atmosphere using KClO₄²⁴ at high-pressure and high-temperature, being adopted as the pure Ni³⁺ reference. The Ni K-edge spectra of M_1/3, M_1/5, BO1-a, and BO3-a were positioned between TeNiO₃ [E₀(Ni) ~ 8339.6 eV] and LaNiO₃ (8342.2 eV), i.e., their Ni valence states are between + 2 and + 3. The absorption energy [E₀(Ni) ~ 8340.8 eV] of M_1/5 represented the approximate Ni valence of + 2.5. The Ni K-edge absorptions for M_1/3 and BO1-a were close to LaNiO₃, calculating the valences of + 2.9. The slight decrease in Ni valence for M_1/5 and BO3-a are derived from the increase in the Mn⁴⁺ contents, as well as the charge transfer between Co and Mn. From the above XANES analyses, the order of Ni valences is estimated as follows: TeNiO₃ (Ni²⁺) < M_1/5 (Ni^2.5+) < BO3-a (Ni^2.7+) < M_1/3, BO1-a (Ni^2.9+) < LaNiO₃ (Ni³⁺). The Ni valences in BO-HEOs with 90 at% of (Fe, Co, Ni) contents changed between divalent and trivalent according to chemical compositions more sensitively than Fe and Co valences (Fig. S6).

We attempt to understand the catalytic activities of HEO using simple electronic structure models speculated from valences of 3d transition metals and chemical compositions. Figure 9 illustrates the simple schematics of electronic band structures for the selected samples, focusing on the unoccupied 3d (M_3d,unocc) bands of transition metals. The OER activities of various 3d transition metal oxides are associated with the M_3d,unocc bands²⁶ as well as oxygen 2p band²⁰. In other words, M_3d,unocc bands in lower energy levels and oxygen 2p band (O_2p,occ) moderately close to the Fermi energy are favorable, consequently proposing the charge-transfer energies between M_3d,unocc and O_2p,occ as the rational descriptor⁷. In the comparison between M_1/5 with M_1/3, the latter apparently satisfies more suitable conditions: the absence of the inactive elements (Cr, Mn) and higher valence states in the late 3d metals (Co, Ni) lead to the smaller charge-transfer energies (Fig. 9). On the other hand, considering the fact that M_1/5 with M_1/3 possessed almost the same activities, unexpected contributions of Cr and Mn ions to complement the apparent factors should exist. BO1-a with the tiny amount of Cr and Mn (< 1 at%) is considered to be an almost a ternary system with Fe, Co, and Ni in the deviated concentration (Fe/Co/Ni = 34/51/14) compared to the equal content for M_1/3 (Fe/Co/Ni = 33/33/33). The drastic improvement in catalytic activity from M_1/3 to BO1-a would originate from the adjustment of chemical compositions within the three constituent elements of Fe, Co, and Ni (see the density of states and band depths of “M_1/3” and “BO1-a” in Fig. 9), as observed in the improvement in the quinary system from M_1/5 to M_1/6Co_1/3. Finally, BO3-a, the most active catalyst in the present study, is regarded to be further optimized by adding the fourth element of Mn (the highest band of BO3-a in Fig. 9). Although the inactive element of Mn in BO3-a may not directly contribute to the activity and rather the Ni valence is slightly decreased from BO1-a to BO3-a (see Fig. 8c), the catalytic activity of BO3-a was unexpectedly increased from that of BO1-a by 13%. Together with the retained activity of M_1/5, the unrecognized roles of the inactive elements emerge in the multielement systems. Although detailed analyses are needed to elucidate the mechanistic aspect, it is beyond the scope of this study. One of the possibilities is oxygen deficiencies on the surface, which was suggested in the previous study on HEOs^18,19.

The above discussion proposes that the Bayesian optimizations adjust the chemical compositions among the active elements (Fe, Co, and Ni) and attempts the addition of inactive elements (Cr and Mn) to excavate their hidden roles, which cannot be discovered by only empirical study. These findings suggest that the data-driven approaches shed light on novel design principle of chemical compositions in perovskite oxides with multiple transition metals for active OER catalysts.

In summary, we investigate the OER activity of high entropy oxides (HEOs) in La(Cr, Mn, Fe, Co, Ni)O₃ system. Bayesian optimizations using the compositions-activities dataset of references and HEOs with random compositions, successfully obtained the highly active catalysts with optimal chemical compositions. All the optimized HEOs exhibited higher activities than any other oxides in single transition metal perovskites, HEOs with random compositions, and LaCr_1/5Mn_1/5Fe_1/5Co_1/5Ni_1/5O₃. Moreover, we obtained the several HEOs exceeding LaCr_1/6Mn_1/6Fe_1/6Co_1/3Ni_1/6O₃, the ever reported most active catalyst in this system. The high probabilities of active catalysts for the optimization elucidated that the Bayesian method is effective in boosting catalytic activities in high-entropy systems. Total atomic contents of OER active elements, Fe, Co, and Ni, were revealed to dominate catalytic activities in HEOs by the correlation between catalytic activities and (Fe, Co, Ni)-content. Additionally, modifications of individual concentrations of 3d transition metals enhanced OER catalytic activities. We concluded that the Bayesian approach has capabilities to discover unreported chemical compositions of highly active catalysts. The findings would promote the Bayesian-optimization-based design of chemical compositions in multiple 3d metal oxide catalysts.

Procedure in Experiments and Optimizations

Multiple cycles of experiments were conducted to construct the database of chemical compositions and OER activities (Fig. 1) when a single cycle was defined as experiments of synthesis, phase identifications using XRD, and electrochemical measurements as one cycle. The first dataset of 21 samples (15 samples of R-HEO and 6 references) was built by the first cycle experiment. The first generation of Bayesian optimization was conducted using the first dataset to generate 5 compositions. The second cycle experiment was conducted in five samples with optimized compositions. The second generation of Bayesian optimization was performed to generate five samples by using the second dataset, which is composed of the first dataset and additional data of optimized five samples. The third optimization was conducted to generate three samples by using 26 samples and added data of optimized 5 samples to the second dataset. Synthesis and electrochemical measurements were conducted in final cycles.

Synthesis

All the samples were synthesized by the polymerization method using metal nitride solutions. Stoichiometric metal nitrates: La(NO₃)₃·6H₂O (99.9%), Cr(NO₃)₃·9H₂O (99%), Mn(NO₃)₃·6H₂O (99.9%), Fe(NO₃)₃·9H₂O (99.9%), Co(NO₃)₂·6H₂O (98%), and Ni(NO₃)₂·6H₂O (99.9%), were dissolved in deionized water to prepare 1 mol L^–1 concentration solutions with a typical volume of 50 mL. 0.5 mL of La nitrate solution and 0.5 mL of solutions of 3d transition metal nitrates were mixed in the stoichiometric ratio, La:(Cr,Mn,Fe,Co,Ni) = 1:1, listed in Table S1 using automatic aspirating/dispensing robot, OT-2 (Opentrons Inc.). The individual concentrations of less than 1at% were ignored in the mixing stage because of the volume resolution (5 µL) of the automatic pipette. A 10 folds excess of citric acid and 1 g of 1,2-ethanediol were added to 1 mL mixed solutions (5*10^–4 mol of ions of La and total 3d transition metal). The mixtures were dried at 423 K overnight and heated at 673 K for 1 h to remove water and organic compounds. The powders from grinding of dried mixtures were fired at 1273 K for 12 h in air to achieve the samples.

Characterization

XRD data were collected by a diffractometer (Ultima IV, Rigaku, Japan) with Cu Kα radiation. Lattice parameters were refined by Rietveld method using of program RITAN-FP with orthorhombic Pnma and rhombohedral R\(\stackrel{-}{3}\)c space groups. Oxygen deficiencies were analyzed by Synchrotron XRD data collected in BL02B2 beamline of SPring-8. XANES spectra of Fe, Co, and Ni K-edge for references and BO-HEOs with more than 90 at% of (Fe, Co, Ni) contents were collected using transmission mode in BL14B2 beamline of SPring-8. All spectra were normalized to the unit edge jump. The morphologies of all BO-HEO and R11 were confirmed by SEM images (S-4800, Hitachi High-Tech, Japan).

Electrochemical characterization

Working electrodes were prepared using the drop-casting method of inks containing catalysts on glassy carbon electrode, referred to previous papers^2,4. A 5 wt% proton-type Nafion suspension (Sigma-Aldrich, USA), 0.1 M KOH aqueous solution (Nacalai Tesque, Inc., Japan), and tetrahydrofuran (THF, Sigma-Aldrich, USA) 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 µL 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-b325, BAS Inc., Japan). We used a Pt wire electrode and a Hg/HgO electrode (International Chemistry Co., Ltd., Japan) filled 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 O₂saturation at room temperature. This fixed the equilibrium potential of the O₂/H₂O redox couple to 0.304 V 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^–1. The disk potential was represented in those vs. reversible hydrogen electrode (RHE), with IR-compensation (R = 43 Ω). The capacitive effect was compensated by averaging the cathodic and anodic scans. All currents were normalized by the disk area (0.126 mm²).

Bayesian optimization

Bayesian optimizations were conducted for maximizing activities of HEO La(Cr, Mn, Fe, Co, Ni)O₃. The used dataset was composed of atomic contents of transition metals as explanatory variables and OER activities as objective variables. All optimizations were conducted by upper confidence bounds method with the Gaussian kernel using the “mlrMBO” package in R studio²⁷. The optimization of the first generation was calculated from 21 samples of five single-transition-metal-oxides (LaCrO₃, LaMnO₃, LaFeO₃, LaCoO₃, and LaNiO_y), M_1/3, and 15 samples of R-HEO to generate five samples. The Bayesian optimizations in the second and third generations were respectively conducted by using datasets of 26 and 31 samples, which were added by novel samples generated in previous generations. The second and third generation optimized five and three compositions, respectively.

Acknowledgements

The synchrotron radiation experiments were performed at SPring-8 under the approval of the Synchrotron Radiation Research Institute (Proposal Nos. 2021B1902, 2022A1493, and 2022A1828). This work was supported by JSPS KAKENHI (Grant Nos. JP19H02438, JP20H02825), JST SPRING (Grant Number JPMJSP2139), Takahashi Industrial and Economic Research Foundation, and JFE 21st Century Foundation.

Author contributions

I.Y. conceived, designed, and co-wrote the paper. Y.O. co-wrote the paper, conducted the Rietveld analysis, and X-ray absorption spectroscopic experiments. Y.F. synthesized the samples and conducted the electrochemical measurements. Y.O. and Y.F. contributed equally to this work. H.M. designed the code for Bayesian optimizations. S.Y. discussed the results of electrochemical measurements. H.M., N.M., Y.N., and H.I. discussed the results of Bayesian optimizations. All the authors discussed the paper.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary information The online version contains supplementary material available at https://...

Data availability

The data in this manuscript is available from the authors upon reasonable requests.

Hong, W. T. et al. Toward the rational design of non-precious transition metal oxides for oxygen electrocatalysis. Energy Environ. Sci. 8, 1404–1427 (2015).
Suntivich, J., May, K. J., Gasteiger, H. A., Goodenough, J. B. & Shao-Horn, Y. A perovskite oxide optimized for oxygen evolution catalysis from molecular orbital principles. Science. 334, 1383–1385 (2011).
Yang, C. et al. Revealing pH-Dependent Activities and Surface Instabilities for Ni-Based Electrocatalysts during the Oxygen Evolution Reaction. ACS Energy Lett. 3, 2884–2890 (2018).
Grimaud, A. et al. Double perovskites as a family of highly active catalysts for oxygen evolution in alkaline solution. Nat. Commun. 4, 2439 (2013).
Man, I. C. et al. Universality in Oxygen Evolution Electrocatalysis on Oxide Surfaces. ChemCatChem 3, 1159–1165 (2011).
Bockris, J. O. M. & Otagawa, T. Mechanism of oxygen evolution on perovskites. J. Phys. Chem. 87, 2960–2971 (1983).
Hong, W. T. et al. Charge-transfer-energy-dependent oxygen evolution reaction mechanisms for perovskite oxides. Energy Environ. Sci. 10, 2190–2200 (2017).
Sun, J. et al. Bifunctional LaMn_0.3Co_0.7O₃ Perovskite Oxide Catalyst for Oxygen Reduction and Evolution Reactions: The Optimized e_g Electronic Structures by Manganese Dopant. ACS Appl. Mater. Interfaces 12, 24717–24725 (2020).
Duan, Y. et al. Tailoring the Co 3d-O 2p Covalency in LaCoO₃ by Fe Substitution to Promote Oxygen Evolution Reaction. Chem. Mater. 29, 10534–10541 (2017).
Sun, J. et al. A bifunctional perovskite oxide catalyst: The triggered oxygen reduction/evolution electrocatalysis by moderated Mn-Ni co-doping. J. Energy Chem. 54, 217–224 (2021).
Sheelam, A. et al. Improved Oxygen Redox Activity by High-Valent Fe and Co³⁺ Sites in the Perovskite LaNi_{1 – x}Fe_0.5xCo_0.5xO₃. ACS Appl. Energy Mater. 5, 343–354 (2022).
Li, H. et al. High-entropy oxides: Advanced research on electrical properties. Coatings 11, 628 (2021).
Bérardan, D., Franger, S., Meena, A. K. & Dragoe, N. Room temperature lithium superionic conductivity in high entropy oxides. J. Mater. Chem. A 4, 9536–9541 (2016).
Sun, Y. & Dai, S. High-entropy materials for catalysis: A new frontier. Sci. Adv. 7, 1–24 (2021).
Zhang, Y., Dai, W., Zhang, P., Lu, T. & Pan, Y. In-situ electrochemical tuning of (CoNiMnZnFe)₃O_3.2 high-entropy oxide for efficient oxygen evolution reactions. J. Alloys Compd. 868, 159064 (2021).
Zhang, Y. et al. Stabilizing Oxygen Vacancy in Entropy-Engineered CoFe₂O₄-Type Catalysts for Co-prosperity of Efficiency and Stability in an Oxygen Evolution Reaction. ACS Appl. Mater. Interfaces 12, 32548–32555 (2020).
Sun, Z. et al. High entropy spinel-structure oxide for electrochemical application. Chem. Eng. J. 431, 133448 (2022).
Tang, L. et al. High Configuration Entropy Activated Lattice Oxygen for O₂ Formation on Perovskite Electrocatalyst. Adv. Funct. Mater. 2112157, 2112157 (2022).
Nguyen, T. X., Liao, Y. C., Lin, C. C., Su, Y. H. & Ting, J. M. Advanced High Entropy Perovskite Oxide Electrocatalyst for Oxygen Evolution Reaction. Adv. Funct. Mater. 31, 2101632 (2021).
Grimaud, A. et al. Activating lattice oxygen redox reactions in metal oxides to catalyse oxygen evolution. Nat. Chem. 9, 457–465 (2017).
Guo, M. et al. Inactive Al³⁺-doped La(CoCrFeMnNiAl_x)_1/(5+x)O₃ high-entropy perovskite oxides as high performance supercapacitor electrodes. J. Adv. Ceram. 11, 742–753 (2022).
Yamada, I. et al. High-pressure synthesis of highly oxidized Ba_0.5Sr_0.5Co_0.8Fe_0.2O_3–δ cubic perovskite. Mater. Chem. Front. 3, 1209–1217 (2019).
Togano, H. et al. Highly active hydrogen evolution catalysis on oxygen-deficient double-perovskite oxide PrBaCo₂O_6–δ. Mater. Chem. Front. 4, 1519–1529 (2020).
Yamada, I. et al. Systematic Study of Descriptors for Oxygen Evolution Reaction Catalysis in Perovskite Oxides. J. Phys. Chem. C 122, 27885–27892 (2018).
Dass, R. I. & Goodenough, J. B. Multiple magnetic phases of La₂CoMnO_6–δ (0 ≤ δ ≤ 0.05). Phys. Rev. B - Condens. Matter Mater. Phys. 67, 144011–144019 (2003).
Cheng, X. et al. Oxygen Evolution Reaction on La_{1 – x}Sr_xCoO₃ Perovskites: A Combined Experimental and Theoretical Study of Their Structural, Electronic, and Electrochemical Properties. Chem. Mater. 27, 7662–7672 (2015).
Bischl, B. et al. mlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions. (2017).

There is NO Competing Interest.

SupplementaryInfo.pdf
Supplementary Information

Download PDF

Version 1

posted

You are reading this latest preprint version

Composition-Designed High Entropy Perovskite Oxides for Oxygen Evolution Catalysis

Status:

Version 1

Abstract

Figures

Full Text

Results And Discussion

Conclusion

Method

Procedure in Experiments and Optimizations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1