2
E. K. Hamal and M. C. Toroker
extends beyond previous studies by adding a detailed analysis of iron composition
and location dependence under strain. We find that both iron concentration and strain
affect the local bond distances near the active site and have a direct consequence on
efficiency.
2 Methods and Calculation Details
The VASP program was used for spin-polarized density functional theory (DFT)
calculations [22, 23]. The chosen functional is the Perdew–Burke–Ernzerhof (PBE)
[24] functional with the DFT + U formalism of Duradev et al. [25] at an effective
U-J term of 5.5 and 3.3 eV for Ni and Fe as previously done for doped NiOOH [17,
26–29]. We found in our previous work [30], that these values for U-J are essential for
capturing the qualitative experimental result that Fe doping reduces the overpotential
required for water oxidation. In this study, we also provide results showing that using
a different functional such as Heyd–Scuseria–Ernzerhof (HSE06) [31] fails to agree
with experiment on the favorable effect of Fe doping on catalysis. We also show
that van der Waals D2 corrections of Grimme [32] have negligible contribution (see
Tables 5, 6, 7 and 8). Projected-augmented wave (PAW) potentials replaced the
electrons of Ni 1s2s2p3s3p, Fe 1s2s2p3s3p, and O 1s [33, 34].
The unit cell of β-NiOOH [17, 35–38] was cleaved at the (0 1 5) plane since this
surface was studied in previous literature and we wanted to simplify the analysis and
compare to previous results [17]. The facet (001) has been recently suggested [16,
39] due to its surface stability, but according to our band edge calculations then this
facet should be less active [28]. Hence, the same slab sizes were built as in Refs.
[17, 28, 40]. The iron content was considered to be with at% of: 8, 25, 33, and 42%
at various locations of nickel atomic substitution. We considered 8% as in previous
studies [18] and higher concentrations below 50% where the material is known to be
catalytically more efficient. The Fe content in Ni 1−x Fe x OOH alloys was considered
at several representative locations in the surface unit cell, while the selection of the
models was based on varying number of iron atoms close and away from the active
site (see Fig. 1). At 8%, both iron and nickel were considered at the active site for
comparison, but at alloying concentration of 25% and above we considered iron as
the active site since this site was found chemically active in previous studies [10, 17].
The final configuration that was selected for applying strain was the one with the
lowest total energy and overpotential at preliminary calculation tests (Tables 9, 10,
11 and 12). Hence, for 8%, 25%, 33%, and 42% Fe contents, “location 2”, “location
2”, “location 3”, and “location 2” were selected, respectively. The energy cutoff of
600 eV and k-point Gamma-centered grid of 2 × 2 × 1 were converged to within
<1 meV. The ion positions were converged until the force components on all ions
were less than 0.03 eV/Å. Additional details are given in the supporting information,
and a data set collection of computational results is available in the ioChem-BD
repository [41].
E. K. Hamal and M. C. Toroker
extends beyond previous studies by adding a detailed analysis of iron composition
and location dependence under strain. We find that both iron concentration and strain
affect the local bond distances near the active site and have a direct consequence on
efficiency.
2 Methods and Calculation Details
The VASP program was used for spin-polarized density functional theory (DFT)
calculations [22, 23]. The chosen functional is the Perdew–Burke–Ernzerhof (PBE)
[24] functional with the DFT + U formalism of Duradev et al. [25] at an effective
U-J term of 5.5 and 3.3 eV for Ni and Fe as previously done for doped NiOOH [17,
26–29]. We found in our previous work [30], that these values for U-J are essential for
capturing the qualitative experimental result that Fe doping reduces the overpotential
required for water oxidation. In this study, we also provide results showing that using
a different functional such as Heyd–Scuseria–Ernzerhof (HSE06) [31] fails to agree
with experiment on the favorable effect of Fe doping on catalysis. We also show
that van der Waals D2 corrections of Grimme [32] have negligible contribution (see
Tables 5, 6, 7 and 8). Projected-augmented wave (PAW) potentials replaced the
electrons of Ni 1s2s2p3s3p, Fe 1s2s2p3s3p, and O 1s [33, 34].
The unit cell of β-NiOOH [17, 35–38] was cleaved at the (0 1 5) plane since this
surface was studied in previous literature and we wanted to simplify the analysis and
compare to previous results [17]. The facet (001) has been recently suggested [16,
39] due to its surface stability, but according to our band edge calculations then this
facet should be less active [28]. Hence, the same slab sizes were built as in Refs.
[17, 28, 40]. The iron content was considered to be with at% of: 8, 25, 33, and 42%
at various locations of nickel atomic substitution. We considered 8% as in previous
studies [18] and higher concentrations below 50% where the material is known to be
catalytically more efficient. The Fe content in Ni 1−x Fe x OOH alloys was considered
at several representative locations in the surface unit cell, while the selection of the
models was based on varying number of iron atoms close and away from the active
site (see Fig. 1). At 8%, both iron and nickel were considered at the active site for
comparison, but at alloying concentration of 25% and above we considered iron as
the active site since this site was found chemically active in previous studies [10, 17].
The final configuration that was selected for applying strain was the one with the
lowest total energy and overpotential at preliminary calculation tests (Tables 9, 10,
11 and 12). Hence, for 8%, 25%, 33%, and 42% Fe contents, “location 2”, “location
2”, “location 3”, and “location 2” were selected, respectively. The energy cutoff of
600 eV and k-point Gamma-centered grid of 2 × 2 × 1 were converged to within
<1 meV. The ion positions were converged until the force components on all ions
were less than 0.03 eV/Å. Additional details are given in the supporting information,
and a data set collection of computational results is available in the ioChem-BD
repository [41].