Identifying Stable Electrocatalysts Initialized by Data Mining: Sb2WO6 for Oxygen Reduction

Abstract Data mining from computational materials database has become a popular strategy to identify unexplored catalysts. Herein, the opportunities and challenges of this strategy are analyzed by investigating a discrepancy between data mining and experiments in identifying low‐cost metal oxide (MO) electrocatalysts. Based on a search engine capable of identifying stable MOs at the pH and potentials of interest, a series of MO electrocatalysts is identified as potential candidates for various reactions. Sb2WO6 attracted the attention among the identified stable MOs in acid. Based on the aqueous stability diagram, Sb2WO6 is stable under oxygen reduction reaction (ORR) in acidic media but rather unstable under high‐pH ORR conditions. However, this contradicts to the subsequent experimental observation in alkaline ORR conditions. Based on the post‐catalysis characterizations, surface state analysis, and an advanced pH‐field coupled microkinetic modeling, it is found that the Sb2WO6 surface will undergo electrochemical passivation under ORR potentials and form a stable and 4e‐ORR active surface. The results presented here suggest that though data mining is promising for exploring electrocatalysts, a refined strategy needs to be further developed by considering the electrochemistry‐induced surface stability and activity.


How to define the stable materials based on phase species
After confirming that ΔG pbx is lower than 0.5 eV/atom, we should consider the existence of solid phase species in the bulk Pourbaix diagram to ensure the stability of this material.Based on previous studies [1], we summarized the standard how to define the stable materials based on phase species: ⅰ) in a defined pH and potential range, if the phase species aligns with the composition of the compound or encompasses it, this compound is deemed to exhibit stability; ⅱ) in a defined pH and potential range, if the phase species does not encompass the compound but nevertheless represent the solid phases across the entire potential range or within a broad potential range, this compound also can be considered to possess potential stability.

Details of pH-dependent microkinetic modeling methods
We used the Quantum Espresso code [2] to calculate the electric field effects.
Electric fields were applied using a saw-tooth potential that corresponds to fields ranging from −0.8 to 0.8 V/Å.At each applied field, adsorbates were allowed to relax with a force convergence threshold of 0.05 eV/Å.We used the lowest energy conformation to predict the adsorbate energy under that field.
To describe the potential and pH dependence, we related the electric fields to both the standard hydrogen electrode (SHE) and reversible hydrogen electrode (RHE) potential using a parallel-plate capacitor model.The model is described by Equation (1): where σ refers to charge density,  0 refers to the vacuum permittivity (8.85 × 10 −12 F m −1 ),  refers to the dielectric constant (unitless), C H refers to the Helmholtz capacitance (μF cm −2 ), V SHE refers to the potential vs. SHE, and V PZC refers to the potential at the point of zero charges (PZCs) vs. SHE.
Fumagalli et al. [3] demonstrated that the dielectric constant of water near a surface is 2. The C H can vary with the surface and potential but typically ranges between 20 and 30 over the majority of the potential range, with more elevated values near the PZC.For simplicity, we assumed a constant C H of 25 μF cm −2 across all surfaces [4].
To measure an adsorbate's response to the field, we fitted a second-order polynomial to the calculations for each adsorbate across the range of fields.We then used Equation (2) to determine the values of the intrinsic dipole moment (μ) and polarizability (α).
where G ads PZC refers to the binding energy of adsorbate at the PZC, which corresponds to the energy calculated with no applied field.These fits, along with values for μ and α for each adsorbate on each electrode, are displayed in Table S6.
Our method for calculating binding energy dependence on the SHE potential differs from the methods employed by Ref. [5].Our method requires the input of fixed PZC and C H , whereas their method allows PZC and C H to vary with adsorption but relies on the ability of implicit solvent methods to accurately predict these values.However, the accuracy of these previous predictions has been called into question by several studies [6].
In this study, we utilized the implicit method VASPsol to determine the computational PZCs for Sb 2 WO 6 .We set the VASPsol parameters to default values, including a bulk dielectric constant ε k = 78.4,width of dielectric cavity σ = 0.6, cutoff charge density ρ cut = 0.0025 Å −3 , and a surface tension parameter of 0.525 meV/Å 2 .
The C HE was used to correct the binding energies for RHE dependence using Equation (3) as follows: where n refers to the number of electrons (relative to water), e refers to the charge of an electron, and V RHE refers to the potential versus RHE.
Ultimately, the free energy of adsorbate at the given V RHE and V SHE is shown by Equation ( 4): The Cutoff charge density ρ cut = 0.0025 Å −3 , and a surface tension parameter of 0.525 meV/Å 2 .
For the relaxation freedom, Nørskov and colleagues [4] found that performing single-point calculations without a structural relaxation on transition metals may lead to misleading conclusions in the pH-field coupled analysis.In this work, all the electric field effect simulations and field-induced energetic calculations have considered full structural relaxations.
Solvation effects play a crucial role in electrochemical reactions occurring at the electrode-electrolyte interface [7].Earlier studies have highlighted the pronounced solvent stabilization of HO* on Pt(111), presenting values ranging from 0.1 to 0.3 eV, contingent upon coverage [8].[7]."

Low-index surfaces of Sb2WO6 were selected for the analysis
We calculated the surface energies of set of high-index crystal facets (110), (011), and (101), and low-index crystal facets (100), (001), and (010) in Table S5 and found that the energies of high-index crystal facets are higher than those of the low-index crystal facets.Therefore, we focused on the more stable (100), (001), and (010) crystal facets for further analysis, among which (100) has lowest surface energy value of -0.267 eV/Å 2 .Next, we calculated surface Pourbaix diagram for (100), (001) and (010), as shown in Figs.5g and S11-S12.In the ORR potential window, we observed that (001) is partly covered by O* at the high-potential region of ORR (Fig. S12a-b), while the catalyst surface of (010) was partly covered by HO* after the potential is higher than 0.75 V (vs.RHE) (Fig. S12c-d).
These conclusions are similar to the (100) surface (Fig. 5g).Furthermore, after developing the volcano model plotted with (100), (001), and (010) in Fig. 5h, (100) exhibited the highest activity.Because catalysis is site-specific (i.e., the most active site/facet may predominant the overall activity of a catalyst) [11], we consider that the (100) facet is the origin of the high alkaline ORR activity of the Sb-W-O material.Considering the lowest surface energy and the highest identified activity, the (100) surface was mainly selected for the theoretical analysis.Besides, the lattices observed under TEM (Fig. 5d-e) are all orthogonal to (100), further suggesting that the exposed active surface are correctly modelled.The CB loading is about 0.05 mg cm -2 , which is the same as that of CB loading during the test of Sb 2 WO 6 sample (0.2 mg cm -2 of metal oxide + 0.05 mg cm -2 CB).

Fig. S1 .
Fig. S1.The number of various elements present in our dataset.

Fig. S3 .
Fig. S3.Stable MOs for OER across a pH range from 0 to 14 screened by the search engine developed in this work.

Fig. S4 .
Fig. S4.Stable MOs for HER across a pH range from 0 to 14 screened by the search engine developed in this work.

Fig. S9. a
Fig. S9. a Raman spectra of the catalysts before and after stability test.b Fitting of the broad peak at 874 cm -1 with two Lorenzt peaks.c Relative intensity between the two fitted peaks.

Fig. S10 .
Fig. S10.ECSA measurement of the catalysts before and after stability test in different electrolytes.a CV curves obtained at different scan rates in a 0.1 M HClO 4 electrolyte.b Fitting current differences and scan rates of different catalysts.Inset compares the calculated C dl of different catalysts.
[10]kovic et al. [9]identified a stabilization of HOO* by 0.5 eV, achieved through a semi-dissociated water layer on Pt(111).Meanwhile, Liu and colleagues[10]observed that O* and O 2 * undergo negligible stabilization.In contrast, HO* and HOO* were found to be stabilized at approximately 0.6 eV and 0.7 eV, respectively.The Sb 2 WO 6 synthesized in this study is categorized under metal oxides.Consequently, we utilized solvation correction energies of 0.15 eV for HO* and 0.4 eV for HOO*, as earlier determined on IrO 2

Table S1
Statistics of the metal elements from the defined stable MOs at a pH = 0.The normalized number is calculated by dividing the number of each element in the defined stable MOs by the number of various elements in our dataset.Table S2Statistics of the metal elements from the defined stable MOs at a pH = 7.The normalized number is calculated by dividing the number of each element in the defined stable MOs by the number of various elements in our dataset.

Table S3
Statistics of the metal elements from the defined stable MOs at a pH = 14.The normalized number is calculated by dividing the number of each element in the defined stable MOs by the number of various elements in our dataset.

Table S4 .
XPS peak position assignment of different elements.

Table S6 .
Field effects on the Sb2WO6