Supplementary Information:
Materials and Methods
Figures S1-S7
Tables S1-S3
References (1-41)
Supplementary Information for
Solar fuels photoanode materials discovery by integrating high-throughput
theory and experiment
Authors:
Qimin Yan
1,2
*, Jie Yu
3,4,5
, Santosh K. Suram
3
, Lan Zhou
3
, Aniketa Shinde
3
, Paul F.
Newhouse
3
, Wei Chen
4
, Guo Li
1,2,5
, Kristin A. Persson
4,6
, John M. Gregoire
3
*, Jeffrey B.
Neaton
1,2,7
*
Affiliations:
1
Molecular Foundry, Lawrence Berkeley National Laboratory, Berkeley, California 94720,
United States
2
Department of Physics, University of California, Berkeley, CA 94720, United States
3
Joint Center for Artificial Photosynthesis, California Institute of Technology, Pasadena, CA
91125, United States
4
Environmental Energy Technologies Division, Lawrence Berkeley National Laboratory,
Berkeley, CA 94720, United States
5
Joint Center for Artificial Photosynthesis, Lawrence Berkeley National Laboratory, Berkeley,
CA 94720, United States
6
Department of Materials Science and Engineering, University of California, Berkeley,
California 94720, United States
7
Kavli Energy NanoSciences Institute at Berkeley, Berkeley, CA 94720, United States
*Correspondence to: qiminyan@temple.edu, gregoire@caltech.edu, jbneaton@lbl.gov
Literature phases and screening criteria
Our literature survey to identify previously-reported phases was aimed at identifying metal oxide
compounds that meet the following criteria:
1.
Measured band gap (direct or indirect) in the range 1.2-2.8 eV
2.
Photoelectrocatalytic (PEC) experiment demonstrating anodic photocurrent in an aqueous
electrolyte (regardless of the pH) without the addition of a sacrificial reagent
These criteria are commensurate with those of the screening pipeline, which is reflected in the
high throughput screening results. All 4 previously-reported photoanodes in the search space
(pure VO
4
motif phases in Supplementary Table 1) passed all computational screening criteria
and the 3 of those phases that were experimentally synthesized passed the remainder of the
experimental screening criteria. Supplementary Table 1 includes 6 metal vanadates that contain
V-O structures beyond that of the present focus (VO
4
in d
0
configuration), and it is worth noting
that the set of cations in these phases (Fe, Cu, Ni, Bi) is a subset of the cations that appear in the
pipeline-identified phases, indicating that the VO
4
-scaffold hypothesis may be generalized to
include structures with a partial VO
4
motif and motivating further research to this effect.
Concerning the maximum solar water splitting efficiencies attainable with these metal oxide
photoanodes, we consider the model recently presented by Fountaine et al. (1) that provides
“realistic limiting efficiencies.” For their “Earth abundant photodiodes” device, the global
maximum efficiency is 16.2% and the lowering of the photoanode band gap from 2.4 eV to 1.8
eV provides an approximately 2-fold increase in device efficiency. For their “high performance”
device the global maximum efficiency is 28.3% and the lowering of the photoanode band gap
from 2.4 eV to 1.8 eV increases the limiting device efficiency from approximately 7.3% to 22%,
a 3-fold increase.
Table S1: The 16 previously-reported photoanode phases noted in the manuscript along with the
reference demonstrating photoelectrocatalytic activity.
phase
Band gap
(eV)
Pure VO
4
motif
Ref.
Fe
2
WO
6
1.5
(2)
α-Fe
2
O
3
1.9
(3)
Fe
2
VO
4
1.9
(4)
FeV
2
O
4
1.9
(4)
FeVO
4
1.9
Y
(4-6)
ZnFe
2
O
4
1.923
(7, 8)
α-CuV
2
O
6
1.95
(9)
β-Cu
3
V
2
O
8
2.05
Y
(10)
BiFeO
3
2.1
(11)
Bi
4
V
2
O
11
2.15
(12)
NiV
2
O
6
2.16
(13)
Fe
2
TiO
5
2.2
(14)
α-Ag
3
VO
4
2.2
Y
(15)
Fe
2
V
4
O
13
2.25
(16)
CuWO
4
2.34
(17)
BiVO
4
2.4
Y
(18)
Choice of computational method for band gaps
The focus of this work is to identify new ternary oxides that exhibit efficient optoelectronic
function including band gaps and band edges. Given the number of ternary oxides and their
chemical complexity, a high-throughput survey is a challenging task, and an exhaustive search
through the whole of material space is not feasible. We require a well-defined search space and a
screening pipeline with multiple layers. DFT is the formalism of choice for estimating properties
related to optoelectronic functionality. However, static DFT is a ground-state theory; and
although Kohn-Sham orbital energies generate an approximate band structure and spectrum,
standard semilocal functionals underestimate the band gaps of semiconductors and lead to
(sometimes considerable) band gap and self-interaction errors, the latter a particular challenge
for localized d-states in TMOs (19), such as those of interest here. DFT+U reduces self-
interaction errors for localized states by including an on-site repulsion on transition metal d
electrons, but there are still limitations associated with the use of DFT for band gaps and optical
properties. To overcome the well-known band-gap underestimation problem induced by local or
semi-local DFT methods, different levels of electronic structure theory have been proposed for
the high-throughput computational screening of band gaps, including
-sol (19), GLLB-SC (20),
and ab initio many-body perturbation theory within the GW approximation and the Bethe-
Salpeter equation (BSE) approach (21, 22). The so-called ab initio GW-BSE method is currently
the formalism of choice for accurate band structures and optical properties. A rigorous excited-
state method, GW-BSE can provide accurate band gaps and optical spectra for diseperate classes
of materials. However, the computational cost of GW-BSE is significantly greater than DFT; and
moreover, such calculations rely on a careful choice of starting point and are rather sensitive to
convergence (23); and therefore GW-BSE calculations for a large number of materials are not
feasible for this study. Previous studies (19, 24) have shown that DFT spectra obtained with the
HSE functional – a hybrid functional featuring local fractional exact exchange – can be
approximately predictive for band gaps for a range of materials. DFT-HSE has also been proven
to be an accurate tool to explore the fundamental chemistry and physics of strongly correlated
oxides (24). In this study, we use DFT with the HSE functional and a modified mixing parameter
(
=0.17) as a compromise between accuracy and efficiency. In screening for band gaps across
121 compounds, we neglect exchange and correlation effects absent in DFT-HSE generalized
Kohn-Sham single-particle spectra; and we neglect electron-hole interactions. Additionally, here
we evaluate direct and indirect gaps as differences in DFT-HSE generalized Kohn-Sham
eigenvalues, fully considering the materials space independent of whether lowest-energy
transitions are symmetry-allowed or symmetry-disallowed.
Figure S1. (A) Calculated band gaps with both HSE(
=0.17) and PBE+U for 116 ternary metal
vanadates with PBE+U band gaps below 3.0 eV. (B) The distribution of band gaps for both HSE
and PBE+U methods. The plot data are included in the SI Appendix, Table S3.
Details of high-throughput computations
All the high-throughput DFT computations are performed using the Vienna software package
(VASP) (25),(26) with the PAW pseudopotentials (27), the generalized gradient approximation
(GGA) as implemented by Perdew, Burke and Ernzerhoff (PBE) (28), and the screened hybrid
functional of Heyd, Scuseria, and Ernzerhof (HSE) (29, 30). The mixing parameter for the
Hartree-Fock exchange potential is reduced from 25% to 17%, based on the success of the latter
value for predicting gaps of known metal vanadates (including BiVO
4
and CuO-V
2
O
5
systems).
We refer to this functional as HSE(
=0.17). A uniform reduction factor for the q-point grid
representation of the exact exchange potential (NKRED = 2) is applied to accelerate the HSE
calculations. An even-number Gamma-centered k-point mesh for the integrations over the
Brillouin zone is used with k-point densities at or larger than 1000 k-points per atom (kppa).
Spin-polarization is included in all calculations. The lattice parameters employed in this work are
obtained from the Materials Project database which are calculated using the PBE+U functional
with the high-throughput computation parameters described by Jain
et al.
(31). The Hubbard U
values used in the PBE+U calculations are adopted from the previous work (32). We use an
energy cutoff of 400 eV for the static HSE(
=0.17) calculations and surface slab calculations.
All data analysis is performed using the Pymatgen package (33).
Band gap energies
DFT-HSE (with a mixing parameter of
=0.17, modified from the standard value of 0.25) and
DFT-PBE+U band gaps for the 116 metal vanadates are shown in Supplementary Fig. 1a. It is
clear that the correlation between the HSE(
=0.17) and PBE+U band gaps is not simple enough
to be described by a “scissor” shift, motivating the use of HSE(
=0.17) in this study. This fact is
also manifested by the different energy distribution of band gaps obtained by PBE+U and
HSE(
=0.17) [Supplementary Fig. 1b]. Note that the band gap difference is dependent on the
choice of Hubbard U in PBE+U or the choice of screening parameter in HSE. The discrepancy
between HSE and PBE+U results becomes especially large for the band gaps that are relevant to
photocatalysis.
For oxides with a HSE(
=0.17) band gap between 1.2 eV and 2.8 eV, its PBE+U
band gap is within a range between 0 eV and 2.5 eV. It indicates that our choice of the PBE+U
band gap window (< 3 eV) for the first screen layer is appropriately restrictive to avoid false-
negative results that would erroneously exclude a promising phase for experimental study.
Band edge energies relative to vacuum
In order to predict the absolute positions of band edge energies, several methods based on
different assumptions have been employed in the literature. For example, an empirical method
based on atomic Mulliken electro-negativities from bulk calculations has been used to predict the
band edge positions (34, 35). However, as pointed out in Ref. (36), this method can lead to errors
for some transition metal oxides of more than 1 eV. Also, this approach is not able to correctly
predict the difference in band edge positions for materials with the same formula unit but
different crystal symmetries. For instance, rutile and anatase TiO
2
are known to have similar
band gaps, while the band offset between these two phases is as large as 0.4 eV (37).
Previous work shows that a reliable prediction of the position of the semiconductor band edges
relative to the two reaction potentials in the presence of aqueous solution can be realized via the
knowledge of the band edge energies relative to the vacuum (36). Band edge energies with
respect to vacuum can provide a good estimate of the alignment of the semiconductor band edges
with water redox potential without computationally-demanding explicit calculations of
semiconductor/water interfaces. Also, it has been shown that the lowest energy surface
orientations and terminations are the most relevant for predicting band edge energies (36). It
indicates that in complex systems without enough experimental information the search for the
lowest energy surface orientation and/or termination becomes essential.
Figure S2. (A) The averaged electrostatic potential (in black) and macroscopic electrostatic
potential (in green) of a sample surface slab system. (B) The band edge energies for mon-CrVO
4
with different surface orientations.
In this work, we employ a standard two-step scheme to calculate the band edge energies relative
to the vacuum level. In the bulk the electronic eigenvalues are referenced to the average
electrostatic potential (ionic and Hartree). From calculations of a supercell with a surface slab
region and a vacuum region, the vacuum level can be defined by the potential far away from the
surface slab, well in the vacuum region; the macroscopically averaged electrostatic potential
taken from deep inside the slab is defined as the “bulk” average electrostatic potential. In this
manner, the potential step Δ
V
between the vacuum and the bulk is established and the bulk
electronic eigenvalues can be referenced to the vacuum. As an example, the average electrostatic
potential of the (
001
) surface slab of a candidate oxide is shown in Supplementary Fig. 2a. Note
that Δ
V
is surface orientation and termination dependent.
Recently, combining the quasiparticle perturbation theory GW and DFT (+U) methods, this
methodology has been used for calculating the absolute band edge positions of 14 transition
metal oxides and obtained a reasonably good agreement with experiment(36). The rationale
behind this choice is that the potential distribution in the slab model depends primarily on the
charge density which is the ground state quantity and hence could be relatively well described by
the traditional DFT(+U) method. In this work, we evaluate the absolute band edge positions by
combing DFT-HSE(
=0.17) bulk calculations and PBE+U surface slab calculations. This
combination has been used to evaluate the band alignment in traditional nitride semiconductors
(38) and good agreement with experiment have been achieved. Supplementary Figure 2b shows
the absolute band edge energies of mon-CrVO
4
for several different surface orientations with
different surface energies. Clearly, the band edge energies are sensitive to surface
orientations/terminations. This observation is consistent with the previous work(36) and reveals
the need for explicit calculations of band edge positions for different surface orientations.
The surface slabs are constructed using an automated workflow developed in a previous work
(39). We only consider low Miller index (
h
,
k
,
l)
[smaller than (
111
)] and non-polar surfaces as
those surfaces are typically of lowest energy and likely not to reconstruct. Different terminations
are considered for each specific surface orientation. We generate around 600 surface slab
systems for 43 metal vanadates, carry out structural relaxations, and compute the surface
energies which are defined as the total energy difference between the slab and the bulk systems
per unit of surface area:
ܧ
௦
=
ா
ೞೌ್
ିா
್ೠೖ
ଶ
, where A is the surface area,
E
slab
is the total energy of
slab, and
E
bulk
is the total energy of bulk with equivalent number of atoms. We compute band
edge energies from surfaces with lowest surface energies. Since the E
VBM
screening criterion
(tier 4) is particularly lenient, we note that the down-selection of phases in the pipeline is
somewhat insensitive to the approximation of E
VBM
using the lowest-energy surface since the
higher energy surfaces also typically pass the screening criterion. This approximation is most
pertinent for the results of Fig. 3 where the E
VBM
exhibits strong anti-correlation with the valence
band character parameter, and we note that this general trend is also robust to the selection of
band edge energy among the low index surfaces.
Crystal structures and comparison with experiment
To avoid expensive structural relaxations with the HSE functional, the structures under
investigation are obtained from the Materials Project database relaxed with PBE+U. To validate
this choice, we perform the structural relaxation with both HSE and PBE+U for 41 transition
metal oxides in the ICSD database. The comparisons for the calculated equilibrium volume with
experimental data are shown in Supplementary Fig. 3.
0 100 200 300 400 500 600
0
100
200
300
400
500
600
HSE
PBE+U
Theory volume (A
3
)
Expt. volume (A
3
)
0
200
400
0.80
0.85
0.90
0.95
1.00
1.05
V
PBE+U
(Angstrom
3
)
V
HSE
/ V
PBE+U
Figure S3 (A) Comparison between the calculated equilibrium volumes for 41 transition metal
oxides obtained from both HSE(α=0.25) and PBE+U methods and the experimental data
obtained from the ICSD database. (B) The volume ratio between the equilibrium lattice
structures relaxed with PBE+U and these relaxed with HSE(α=0.25).
Clearly, the equilibrium volume data obtained using HSE achieve an excellent agreement with
experimental data with a mean error of 0.2% difference. On the other hand, as expected, PBE+U
tends to overestimate the equilibrium volume of materials. In spite of that, the differences in the
unit cell volume between PBE+U and HSE are within a reasonable range between 3% and 8%
(Supplementary Fig. 3b). The HSE band gaps calculated at PBE+U structures are in general only
slightly smaller than those obtained at HSE structures with an average difference of less than 0.2
eV. Given the fact that the changes in lattice parameters and band gaps scale almost linearly with
the change in mixing parameter, we expect that the difference between PBE and HSE(α=0.17)
should be smaller than that between PBE and HSE(α=0.25). We thus conclude that it is feasible
to perform theHSE(α=0.17) electronic structure calculations using the PBE+U relaxed structures
without introducing additional significant errors in the estimated band gaps.
Magnetic ordering
Although all PBE+U calculations in the MP database are spin-polarized, only ferromagnetic
(FM) spin configurations are considered. In transition metal oxides, the exchange interaction can
be strong enough to make the antiferromagnetic (AF) configurations more favorable even at
room temperature. For instance, NiO, MnO, and Fe
2
O
3
are known to be stable in AF spin ground
states at room temperature.
Figure S4. (A) Differences in the total energies per atom between the FM phases and AFM/FR
phases for 55 transition metal oxides. (B) Differences in the band gaps between the FM and
AFM/FR phases for 55 transition metal oxides.
To validate our choice, we study the energetics and electronic structures of some selected
transition metal oxides in three simple magnetic structures including FM, AF, and ferrimagnetic
(FR) spin configurations. We develop an automatic scheme to generate the initial
antiferromagnetic magnetic orderings. Firstly, we create the supercells which are automatically
expanded from the primitive unit cells and contain at least 8 atoms for each magnetic elements.
We then treat the spin up and spin down species for the same element as two different elements,
and choose the AF spin configuration with highest symmetries (largest space group numbers) as
an initial guess of the stable AF (or FR) configurations. The initial magnetic moment on each
magnetic atom is set to be the highest possible value according to the number of occupied d-
electrons.
With the relaxation of both the structural and the magnetic degree of freedom using the PBE+U
method, we obtain the total energies and the band gaps of these magnetic phases. As shown in
Supplementary Fig. 4a, FM phases are more energetically favorable for most of the selected
transition metal oxides. Furthermore, as shown in Supplementary Fig. 4b, the difference in band
gaps between the two magnetic configurations is smaller than 0.2 eV for most compounds. We
therefore choose FM phase as the default magnetic configuration in our data-driven high-
throughput computational screening process.
Definition of band characters at band edges
The band character of V 3d at the CBM is defined as
ܹ
,
ଷௗ
=
ைௌ
(
,
ଷௗ
)
ைௌ
(
௧௧
)
ா
ಳಾ
ା
.
ହ
ா
ಳಾ
ܧ݀
, where
E
CBM
is the energy of the CBM,
DOS(total)
is the total density of states, and
DOS(V,3d)
is the
projected density of states for V 3d orbitals. The band character of O 2p at the VBM is defined as
ܹ
ை
,
ଶ
=
ைௌ
(
ை
,
ଶ
)
ைௌ
(
௧௧
)
ா
ೇಳಾ
ா
ೇಳಾ
ି
.
ହ
ܧ݀
, where
E
VBM
is the energy of the VBM and
DOS(O,2p)
is the
projected density of states for O 2p orbitals.
A summary of the computational screening results is provided in Supplementary Table 3.
AFM favorable
FM favorable
Compound
-0.01
0.00
0.01
E
FM
total
-
E
AFM
total
(eV/atom)
-1.0
-0.5
0.0
0.5
1.0
E
g
FM
-
E
g
AFM
(eV)
PVD and XRD experiments
The metal-vanadium (M-V) oxide composition libraries were fabricated using RF magnetron co-
sputtering onto 100 mm-diameter, 2.2 mm-thick glass substrates with a Fluorine doped Tin oxide
(FTO) coating (Tec7, Hartford Glass Company) in a sputter deposition system (Kurt J. Lesker,
CMS24) with 10
-5
Pa base pressure. Co-sputtering was performed using a V target and an
additional metal target (M) with the exception of the Ag-containing libraries, which used an
Ag
2
O target. The composition libraries were either deposited as “metal” or “oxide” thin films,
referring to the absence or presence of reactive O
2
in the chamber. The working atmosphere was
composed of inert sputtering gas Ar (0.72 Pa) and
reactive gas O
2
(0.08 Pa) for oxide depositions, and
0.80 Pa Ar for metal deposition. The composition
gradients in the co-sputtered continuous composition
spreads were attained by positioning the deposition
sources in a non-confocal geometry. The power
applied on each source was adjusted according to the
deposition rate calibration from the sputter source.
The film thickness was not measured for each
composition library but was estimated to be 200 nm
based on the deposition rate calibration assuming the
average molar density of the elemental oxides. The
as-deposited composition libraries were subsequently
placed flat on a quartz support and annealed in a
Thermo Scientific box oven in flowing air at various
temperatures and durations. The annealing was
preceded by a 2 h temperature ramp and was
followed by free-cooling to near ambient
temperature. Supplementary Table 2 summarizes the
deposition and annealing conditions for the 15 phases identified by the screening pipeline.
The crystal structures and phase distribution of the composition libraries were determined
through XRD measurements using a Bruker DISCOVER D8 diffractometer with Cu K
α
radiation
from a Bruker I
S source. The x-ray spot size was limited to a 1 mm length scale, over which the
composition is constant to within approximately 1%. The XRD measurements were performed
on a series of evenly-spaced positions along the composition gradient. Diffraction images were
collected using a two-dimensional VÅNTEC-500 detector and integrated into one-dimensional
patterns using DIFFRAC.SUITE™ EVA software. For patterns in which multiple crystalline
phases were identified, the relative phase fraction of each phase was calculated using the
measured intensity and relative sensitivity factor of the most distinguishing peak of each phase.
Partial XRD patterns for representative phases are shown in Supplementary Figure 5 to illustrate
phase identification.
Table S2: The deposition and annealing conditions are listed for the 15 phases identified by the
screening pipeline along with Materials Project identification numbers (mp-id). All the vanadate
Figure S5: Partial XRD patterns for 3
representative phases where the reference
patterns are shown as black lines and regions
with strong substrate signal are noted with
gray boxes.
depositions used metal targets (M and V), except a silver oxide (Ag
2
O) target was used for the
Ag-V libraries.
mp-id
Phase
Deposition condition
Annealing condition
Metal or oxide
deposition
V power
(W)
M power
(W)
temperature
(°C)
Duration
(hrs)
mp-851269
Cr
2
V
4
O
13
metal
150
70
610
1
mp-19418
orth-CrVO
4
metal
150
70
610
1
mp-19688
mon-CrVO
4
oxide
150
165
550
3
mp-540630
tri-FeVO
4
metal
150
43.5
610
1
mp-773310
α-CoV
2
O
6
oxide
160
55
550
3
mp-540833
Co
3
V
2
O
8
oxide
160
55
550
3
mp-557404
Ni
2
V
2
O
7
oxide
150
52
550
3
mp-542151
Ni
3
V
2
O
8
oxide
150
52
550
3
mp-505508
α-Cu
2
V
2
O
7
oxide
150
6
610
1
mp-559660
β-Cu
2
V
2
O
7
oxide
150
11
550
10
mp-540833
γ-Cu
3
V
2
O
8
oxide
150
6
610
1
mp-505456
Cu
11
V
6
O
26
oxide
150
6
610
1
mp-18889
α-Ag
3
VO
4
oxide
150
Ag
2
O, 47
550
3
mp-19412
β-Ag
3
VO
4
oxide
180
Ag
2
O, 30
300
10
mp-504878
mon-BiVO
4
oxide
175
14
550
3
For each M-V system, multiple composition libraries were typically synthesized using both
metal and oxide depositions and different annealing conditions. In instances where multiple
synthesis conditions yielded the same target phase in sufficiently high purity to pass tier 5
criteria, all such libraries were passed to tier 6 and 7 screening, and the sample exhibiting the
cleanest UV-vis and PEC data was chosen to represent the target phase. In total, ~50 M-V oxide
libraries were synthesized in tier 5 screening, highlighting the need for high throughput
experiments to sufficiently evaluate computational predictions.
UV-vis experiments
Ultraviolet-visible (UV-vis) absorption spectroscopy was performed using a custom scanning
spectroscopy instrument described in detail previously.(40) Briefly, the dual integrating sphere
system measured both the fractional transmittance (T) and reflectance (R) at 1521 locations
across the entire 100 mm substrate (sample pitch = 2 mm) using illumination from a 200 W
(Hg)Xe broadband source (Newport/Oriel Apex) and Spectral Products, Inc. model SM303
spectrometers. The T and R signals were used to calculate the spectral absorption coefficient (α)
up to a factor of film thickness (τ): α × τ = -ln[T × (1-R)
-1
] from which direct and indirect Tauc
plots were generated. Band gap energies were estimated from the Tauc plots using a constrained
piece-wise linear fitting based algorithm. While a direct band gap energy was obtained from each
direct-allowed Tauc signal, several indirect-allowed Tauc signals did not exhibit a clear
transition, resulting in the use of an inequality expressing the upper limit of the indirect band gap
energy. Representative Tauc signals and band gap extractions are shown in Supplementary Fig.
6. Typically each target phase existed over a sufficiently large area of the film that dozens of
UV-vis spectra were acquired on samples with high phase purity, and the reported band gap
values were validated through manual inspection of many corresponding Tauc spectra.
PEC experiments
A scanning drop electrochemical cell (SDC) was utilized to perform photoelectrochemical (PEC)
experiments on the PVD libraries as described in our previous work.(41) An aqueous electrolyte
made up of 0.1 M boric acid, 0.05 M potassium hydroxide, and 0.25 M sodium sulfate (pH 9)
provided continuous flow into the 3-electrode cell while a 385 nm light emitting diode (LED,
Thor Labs, M385F1, 0.7 A current limit, 3 mW illumination power) provided illumination. Short
circuit photocurrent density (0 V vs. O
2
/H
2
O Nernstian potential; 1.23 V vs. RHE)
was
calculated from toggled-illumination (0.5 s on, 0.5 s off) chronoamperometry (CA)
measurements over 44 s, a portion of which is shown in Supplementary Figure 7 for three
representative samples. The average current from a portion of the illuminated period was
subtracted by the average current from a portion of the previous dark period and divided by the
circular illumination area (1.5 mm diameter) to provide values for J
O2/H2O
in Table 1 and Figure
2. All 16 phases in tier 6, Figure 1, were measured for PEC activity, 15 of which resulted in
measurable photocurrent. Linear sweep voltammetry (LSV) starting at 1.23 V vs RHE was
performed on each sample after the CA measurement, and while the LSV data was not used to
generate screening criteria, manual inspection of the LSV was performed to confirm the
photoactivity of each phase. For example, the CA characterization of β-Ag
3
VO
4
reveals a
substantial dark current indicating the presence of a non-PEC reaction. Inspection of both the CA
and LSV reveal that the photocurrent remains nearly constant despite strong variation in the dark
current. This example includes the lowest-purity phase to pass tier 5 screening, for which the thin
film samples contains fcc-Ag as a minority phase. Correspondingly, the anodic dark current in
the CA is likely due to electrooxidation of Ag, and the onset of cathodic dark current in the LSV
due to its electroreduction, both of which proceed without impacting the photoactivity of β-
Ag
3
VO
4
.While Faradaic efficiency measurements are not available in the high throughput PEC
experiments, this type of detailed analysis is applied to each tier-7 phase to confirm that the
observed photoactivity is due to OER photoelectrocatalysis by the majority phase. While PEC
stability is a primary focus of ongoing work, we note that for photoactive phases for which the
photocurrent is primarily due to photoanodic corrosion, the photocurrent typically vanishes over
Figure S6: Tauc signals (normalized by maximum value) for direct (red) and indirect (cyan) transitions with
band gap energy determination from piecewise linear modelling (intersection of dashed-purple Tauc line and
dashed-black baseline). The indirect band gap energy of β-Ag
3
VO
4
is reported as an inequality due to the lack
of a clear transition in the indirect Tauc signal.
Figure S7: Toggled-illumination chronoamperometry (A, 1.23 V vs RHE) and linear sweep voltammetry (B, -
0.02 V s
-1
) in pH 9.2 electrolyte. Of these 3 phases, orth-CrVO
4
exhibits the highest photovoltage but lowest
photocurrent, β-Ag
3
VO
4
exhibits a varying dark current signal (believed to be from a minority phase) that does
not strongly impact the photoactivity, and β-Cu
2
V
2
O
7
exhibits high, stable photocurrent.
the 44 s CA measurement, prompting our use of the photocurrent at the end of the 44 s CA as the
screening criterion.
Table S3. Materials information for 174 tier 1 compounds including the farthest tier achieved by
each phase.
Materials
Project ID
Formula
E above
Hull
(eV/atom)
PBE+U
band
gap
(eV)
F
arthest
tier
HSE
indirect
band gap
(eV)
V 3d
character
at CBM
O 2p
character
at VBM
VBM
energy
(eV)
mp-18812 NdVO
4
0.0000
3.02
Tier1
mp-18734 HoVO
4
0.0000
3.03
Tier1
mp-19169
PrVO
4
0.0000
3.03
Tier1
mp-18815
KVO
3
0.0000
3.04
Tier1
mp-18827
AlVO
4
0.0000
3.04
Tier1
mp-25726 V
2
Pb
3
O
8
0.0000
3.05
Tier1
mp-19031 RbVO
3
0.0000
3.05
Tier1
mp-556791 K
5
V
3
O
10
0.0000
3.06
Tier1
mp-540778 CsVO
3
0.0000
3.11
Tier1
mp-19162
LaVO
4
0.0000
3.12
Tier1
mp-19083 NaVO
3
0.0000
3.23
Tier1
mp-19034 Mg
3
V
2
O
8
0.0000
3.31
Tier1
mp-565574 K
4
V
2
O
7
0.0000
3.44
Tier1
mp-19660 Sr
2
V
2
O
7
0.0000
3.45
Tier1
mp-19474 Ba
2
V
2
O
7
0.0000
3.46
Tier1
mp-19386 Sr
3
V
2
O
8
0.0000
3.68
Tier1
mp-19365 Ba
3
V
2
O
8
0.0000
3.74
Tier1
mp-780545 Na
3
VO
4
0.0000
4.03
Tier1
mp-19219 Li
3
VO
4
0.0000
4.03
Tier1
mp-639787 K
3
VO
4
0.0000
4.04
Tier1
mp-566195 Mg
2
V
2
O
7
0.0013
3.16
Tier1
mp-19368 Sr
2
V
2
O
7
0.0014
3.39
Tier1
mp-583094 Li
3
VO
4
0.0028
3.94
Tier1
mp-648893 Na
4
V
2
O
7
0.0039
3.38
Tier1
mp-542076 Ca
3
V
2
O
8
0.0189
3.51
Tier1
mp-764673 Na
3
VO
4
0.0190
3.90
Tier1
mp-19373
LiVO
3
0.0201
3.03
Tier1
mp-639402 K
3
VO
4
0.0215
3.63
Tier1
mp-25110 V
2
Pb
3
O
8
0.0327
3.05
Tier1
mp-18989
LaVO
4
0.0357
3.51
Tier1
mp-19052
K
3
VO
4
0.0361
3.67
Tier1
mp-763901 NaVO
3
0.0425
3.21
Tier1
mp-779358 Na
3
VO
4
0.0477
4.02
Tier1
mp-770094 Y
2
V
2
O
7
0.0816
0.49
Tier1
mp-771790 V
3
NiO
8
0.0825
1.19
Tier1
mp-761301
VIO
4
0.0847
1.06
Tier1
mp-510657 VCu
3
O
4
0.0848
0.10
Tier1
mp-783902 Dy
2
V
2
O
7
0.0855
0.47
Tier1
mp-779376 Na
2
V
2
O
5
0.0872
0.97
Tier1
mp-775570 Na
3
VO
3
0.0881
0.22
Tier1
mp-772127 V
2
CrO
6
0.0898
0.00
Tier1
mp-764595 Na
2
V
2
O
5
0.0924
1.27
Tier1
mp-772672 Sm
2
V
2
O
7
0.0936
1.35
Tier1
mp-853245 KV
12
O
30
0.0937
0.37
Tier1
mp-769865 VCr
2
O
4
0.0938
0.08
Tier1
mp-763248 V4FeO
12
0.0953
0.01
Tier1
mp-780306 Na
2
V
2
O
5
0.0987
1.17
Tier1
mp-32432 Mg
2
VO
4
0.0989
0.00
Tier1
mp-743557 V
2
Bi
4
O
11
0.1009
2.50
Tier1
mp-769665 VCoO
4
0.1146
0.37
Tier1
mp-649492 V
2
PbO
6
0.1208
2.34
Tier1
mp-771556 VCoO
4
0.1413
0.26
Tier1
mp-705670 V
2
Cu
2
O
7
0.2036
0.00
Tier1
mp-18740 V
2
Cd
2
O
7
0.0000
2.56
Tier2
3.24
0.70
0.69
mp-18784 DyVO
4
0.0000
2.98
Tier2
-
-
-
mp-18799 YbVO
4
0.0022
0.00
Tier2
0.00
-
-
mp-18807 V
2
Zn
2
O
7
0.0325
2.34
Tier2
2.99
0.71
0.66
mp-18880
ErVO
4
0.0358
2.72
Tier2
0.19
0.03
0.72
mp-18929 BaV
2
O
6
0.0000
2.86
Tier2
3.58
0.72
0.70
mp-18960
ErVO
4
0.0000
2.97
Tier2
0.00
-
-
mp-18993
LuVO
4
0.0000
2.94
Tier2
3.67
0.70
0.71
mp-19068 TmVO
4
0.0000
2.96
Tier2
0.00
-
-
mp-19121
TbVO
4
0.0000
2.98
Tier2
-
-
-
mp-19133
YVO
4
0.0000
2.97
Tier2
3.80
0.68
0.72
mp-19142 Mn
2
V
2
O
7
0.0000
1.20
Tier2
0.95
0.66
0.29
mp-19214
CeVO
4
0.0485
0.00
Tier2
1.08
0.67
0.05
mp-19247
ScVO
4
0.0000
2.64
Tier2
3.38
0.65
0.73
mp-19295
LuVO
4
0.0341
2.60
Tier2
3.28
0.72
0.72
mp-19323 SmVO
4
0.0000
3.00
Tier2
-
-
-
mp-19440
LiVO
3
0.0000
2.82
Tier2
3.86
0.71
0.70
mp-19582 V
2
Zn
3
O
8
0.0303
2.70
Tier2
4.21
0.70
0.65
mp-19707 V
2
Zn
2
O
7
0.0000
2.55
Tier2
3.30
0.71
0.67
mp-25113
VInO
4
0.0000
2.97
Tier2
3.96
0.61
0.71
mp-25140 GdVO
4
0.0000
2.97
Tier2
3.77
0.68
0.71
mp-25142 V
2
Pb
3
O
8
0.0032
2.98
Tier2
3.59
0.62
0.56
mp-25153
TlVO
3
0.0000
2.97
Tier2
3.66
0.68
0.51
mp-25160 Na
5
VO
5
0.0024
2.24
Tier2
2.91
0.69
0.58
mp-25559 MnVO
4
0.0235
0.19
Tier2
0.35
0.09
0.29
mp-25796 V
2
Pb
2
O
7
0.0000
2.87
Tier2
3.43
0.69
0.54
mp-32406 Ta
9
VO
25
0.0000
2.61
Tier2
3.55
0.23
0.74
mp-32407
TaVO
5
0.0000
2.17
Tier2
2.86
0.52
0.74
mp-32479 Tl
3
VO
4
0.0000
2.60
Tier2
3.19
0.35
0.45
mp-32500 Mg
2
V
2
O
7
0.0000
2.71
Tier2
3.65
0.72
0.71
mp-504820 VCl
3
O
0.0502
2.89
Tier2
3.55
0.66
0.08
mp-504923 V
2
Zn
4
O
9
0.0186
2.55
Tier2
3.39
0.70
0.63
mp-505253 U
2
V
2
O
11
0.0361
1.99
Tier2
3.34
0.33
0.55
mp-505265 La
11
V
4
O
26
0.0066
1.02
Tier2
0.85
0.69
0.19
mp-505290 EuVO
4
0.0000
2.78
Tier2
0.00
-
-
mp-505392 Ba
3
V
4
O
13
0.0000
2.95
Tier2
3.70
0.70
0.69
mp-505679 HfV
2
O
7
0.0000
2.65
Tier2
3.34
0.68
0.71
mp-541368 Tl
4
V
2
O
7
0.0000
2.61
Tier2
2.94
0.63
0.34
mp-541501 VInO
4
0.0205
2.86
Tier2
3.70
0.63
0.72
mp-559090 VBiO
4
0.0209
2.53
Tier2
3.12
0.66
0.60
mp-559440 V
2
(CuO
2
)
5
0.0288
0.02
Tier2
1.17
0.12
0.48
mp-561207 UV
2
O
8
0.0000
2.16
Tier2
3.10
0.16
0.69
mp-565447 ThV
4
O
12
0.0000
2.57
Tier2
3.56
0.71
0.71
mp-565725 ZrV
2
O
7
0.0000
2.59
Tier2
3.28
0.65
0.71
mp-572632 Mn
2
V
2
O
7
0.0040
1.46
Tier2
1.12
0.66
0.26
mp-613172 VBiO
4
0.0000
2.81
Tier2
3.23
0.69
0.57
mp-634381 V
2
Zn
2
O
7
0.0306
2.02
Tier2
3.11
0.71
0.66
mp-647265 V
2
Bi
7
O
15
0.0197
1.25
Tier2
-
-
-
mp-647385 V
2
Pb
4
O
9
0.0000
2.91
Tier2
3.46
0.62
0.50
mp-690568 V
2
Bi
4
O
11
0.0754
2.77
Tier2
3.37
0.61
0.59
mp-698685 V
2
Bi
4
O
11
0.0518
2.30
Tier2
2.89
0.63
0.59
mp-763984 Na
3
VO
3
0.0665
0.19
Tier2
0.00
-
-
mp-766784 V
3
CoO
8
0.0368
1.12
Tier2
0.89
0.73
0.18
mp-769777 NbVO
4
0.0805
1.35
Tier2
1.16
0.08
0.15
mp-769888 Ho
2
V
2
O
7
0.0771
0.51
Tier2
-
-
-
mp-771386 VBO
4
0.0681
2.63
Tier2
3.32
0.72
0.74
mp-771484 VCoO
4
0.0516
0.60
Tier2
1.07
0.13
0.54
mp-771685 V
4
SnO
12
0.0405
2.51
Tier2
3.37
0.70
0.71
mp-771856 MnVO
4
0.0763
0.50
Tier2
1.10
0.09
0.48
mp-771872 V
2
NiO
6
0.0000
2.69
Tier2
2.95
0.66
0.40
mp-772238 V
2
SiO
7
0.0522
2.44
Tier2
3.15
0.71
0.72
mp-773218 V
2
FeO
6
0.0247
1.72
Tier2
1.04
0.71
0.04
mp-773430 NbV
2
O
7
0.0669
0.00
Tier2
0.00
-
-
mp-773455 V
2
SnO
7
0.0000
2.61
Tier2
3.57
0.68
0.72
mp-773503 V
2
CrO
7
0.0347
0.62
Tier2
0.97
0.06
0.51
mp-773930 V
3
(WO
6
)
2
0.0783
0.18
Tier2
0.00
-
-
mp-774246 Nb
9
VO
25
0.0015
2.00
Tier2
2.83
0.07
0.74
mp-774376 Na
5
VO
4
0.0560
0.81
Tier2
0.00
-
-
mp-774963 Na
5
VO
4
0.0766
0.31
Tier2
0.00
-
-
mp-775001 V
3
FeO
8
0.0802
1.26
Tier2
0.00
-
-
mp-776985 MnV
3
O
8
0.0573
1.08
Tier2
0.83
0.72
0.21
mp-778022 Na
3
VO
3
0.0652
0.40
Tier2
0.06
0.10
0.15
mp-778222 Na
3
VO
3
0.0730
0.59
Tier2
0.50
0.08
0.15
mp-778356 Na
8
V
2
O
7
0.0604
0.65
Tier2
0.00
-
-
mp-778780 VBO
4
0.0000
2.90
Tier2
3.61
0.72
0.73
mp-780893 Na
6
V
2
O
7
0.0756
0.91
Tier2
1.01
0.49
0.19
mp-850239 Mn
2
V
3
O
12
0.0770
0.00
Tier2
0.31
0.02
0.68
mp-891955 Ca
10
V
6
O
25
0.0288
2.32
Tier2
2.97
0.69
0.57
mp-891956 Ca
10
V
6
O
25
0.0000
2.35
Tier2
2.95
0.70
0.59
mp-19096 Ba
2
VO
4
0.0000
1.52
Tier3
1.72
0.68
0.20
-2.13
mp-554926 Sr
2
VO
4
0.0396
1.53
Tier3
1.46
0.68
0.20
-4.30
mp-565780 Sr
2
VO
4
0.0000
1.53
Tier3
1.45
0.72
0.21
-2.06
mp-566989 NpV
4
O
12
0.0283
0.15
Tier3
2.06
0.46
0.09
--
mp-766869 V
2
Bi
24
O
41
0.0068
1.66
Tier3
2.35
0.14
0.60
-4.56
mp-767265 V(Bi
5
O
8
)
5
0.0118
1.83
Tier3
2.47
0.18
0.59
-4.95
mp-849942 Na
4
VO
4
0.0000
1.14
Tier3
1.31
0.27
0.20
-2.09
mp-853167 Na
2
VO
3
0.0473
1.74
Tier3
1.92
0.73
0.15
-2.82
mp-18889 VAg
3
O
4
0.0000
0.92
Tier7
1.70
0.19
0.34
-6.63
mp-18949
VFeO
4
0.0146
1.82
Tier4
1.81
0.14
0.61
-5.78
mp-19412 VAg
3
O
4
0.0301
0.60
Tier7
1.61
0.05
0.36
-7.64
mp-19418
VCrO
4
0.0015
1.91
Tier7
2.21
0.44
0.28
-6.41
mp-19688
VCrO
4
0.0127
2.21
Tier7
2.48
0.48
0.33
-6.82
mp-19692 Mn
3
V
2
O
8
0.0000
1.92
Tier4
2.10
0.68
0.32
-6.17
mp-25122
VBiO
4
0.0161
2.26
Tier4
2.71
0.67
0.61
-7.62
mp-504747 V
2
Cu
3
O
8
0.0163
0.00
Tier7
1.73
0.18
0.51
-7.81
mp-504878 VBiO
4
0.0161
2.30
Tier7
2.72
0.67
0.61
-7.58
mp-505456 V
6
Cu
11
O
26
0.0292
0.00
Tier7
1.38
0.13
0.50
-7.41
mp-505508 V
2
Cu
2
O
7
0.0064
0.00
Tier7
1.84
0.23
0.53
-8.04
mp-540630 VFeO
4
0.0792
2.05
Tier7
2.10
0.17
0.55
-7.38
mp-540833 V
2
Co
3
O
8
0.0000
1.74
Tier7
2.03
0.55
0.11
-6.27
mp-542151 V
2
Ni
3
O
8
0.0000
2.57
Tier7
2.54
0.47
0.44
-7.39
mp-547693 V
2
Co
2
O
7
0.0000
1.83
Tier6
2.00
0.63
0.13
-6.54
mp-557404 V
2
Ni
2
O
7
0.0000
2.67
Tier7
2.72
0.54
0.38
-8.18
mp-559660 V
2
Cu
2
O
7
0.0000
0.00
Tier7
1.84
0.27
0.52
-8.02
mp-565529 V
4
Fe
2
O
13
0.0000
2.31
Tier4
2.54
0.19
0.66
-7.99
mp-600273 V
2
Cu
3
O
8
0.0240
0.33
Tier4
1.47
0.24
0.53
-8.08
mp-614005 VCuO
3
0.0770
0.80
Tier4
1.37
0.69
0.12
-7.18
mp-624691 VAgO
3
0.0097
1.92
Tier4
2.65
0.70
0.25
-6.06
mp-687096 V
2
Cu
2
O
7
0.0230
0.03
Tier4
1.85
0.22
0.52
-8.17
mp-761307 TiV
2
O
7
0.0000
2.03
Tier4
2.73
0.50
0.72
-8.06
mp-763634 VCrO
4
0.0606
1.83
Tier4
2.12
0.42
0.31
-6.52
mp-765013 Zr
9
VO
20
0.0610
1.71
Tier4
1.96
0.73
0.21
-7.11
mp-766904 MnV
4
O
12
0.0608
0.92
Tier4
1.81
0.28
0.70
-7.04
mp-767805 Zr
11
VO
24
0.0517
1.77
Tier4
1.99
0.72
0.19
-6.90
mp-769887 NbVO
5
0.0292
1.79
Tier4
2.45
0.50
0.74
-7.58
mp-769890 NbVO
5
0.0000
1.89
Tier5
2.60
0.34
0.73
-9.12
mp-772351 NbVO
5
0.0025
2.04
Tier4
2.76
0.36
0.73
-8.87
mp-773310 V
2
CoO
6
0.0000
2.09
Tier7
2.16
0.67
0.08
-6.05
mp-777555 MnV
2
O
6
0.0000
1.97
Tier4
2.18
0.71
0.24
-6.58
mp-850978 VFeO
4
0.0715
1.92
Tier4
1.75
0.14
0.62
-7.60
mp-851269 V
4
Cr
2
O
13
0.0000
2.33
Tier7
2.55
0.60
0.28
-7.40
References
1.
Fountaine KT, Lewerenz HJ, & Atwater HA (2016) Efficiency limits for photoelectrochemical water-
splitting.
Nat Commun
7:13706.
2.
Khader MM, Saleh MM, & El-Naggar EM (1998) Photoelectrochemical characteristics of ferric tungstate.
Journal of Solid State Electrochemistry
2(3):170-175.
3.
Hardee KL & Bard AJ (1976) Semiconductor Electrodes: V . The Application of Chemically Vapor
Deposited Iron Oxide Films to Photosensitized Electrolysis.
Journal of The Electrochemical Society
123(7):1024-1026.
4.
Mandal H
, et al.
(2016) Development of ternary iron vanadium oxide semiconductors for applications in
photoelectrochemical water oxidation.
RSC Advances
6(6):4992-4999.
5.
Arai T, Konishi Y, Iwasaki Y, Sugihara H, & Sayama K (2007) High-throughput screening using porous
photoelectrode for the development of visible-light-responsive semiconductors.
J Combinatorial Chem
9(4):574-581.
6.
Morton CD, Slipper IJ, Thomas MJK, & Alexander BD (2010) Synthesis and characterisation of Fe–V–O
thin film photoanodes.
J Photochem Photobiol, A
216(2–3):209-214.
7.
Valenzuela MA, Bosch P, Jiménez-Becerrill J, Quiroz O, & Páez AI (2002) Preparation, characterization
and photocatalytic activity of ZnO, Fe2O3 and ZnFe2O4.
J Photochem Photobiol, A
148(1–3):177-182.
8.
De Haart LGJ & Blasse G (1985) Photoelectrochemical properties of ferrites with the spinel structure.
Solid
State Ionics
16:137-139.
9.
Guo W
, et al.
(2015) Synthesis and Characterization of CuV2O6 and Cu2V2O7: Two Photoanode
Candidates for Photoelectrochemical Water Oxidation.
The J Phys Chem C
119(49):27220-27227.
10. Seabold JA & Neale NR (2015) All 1st Row Transition Metal Oxide Photoanode for Water Splitting Based
on Cu
3
V
2
O
8
.
Chem. Mater
27(3):1005-1013.
11. Chen XY
, et al.
(2007) Application of weak ferromagnetic BiFeO3 films as the photoelectrode material
under visible-light irradiation.
Appl Phys Lett
91(2):022114.
12. Jiang Z
, et al.
(2016) One-Pot Solvothermal Synthesis of Bi4V2O11 as A New Solar Water Oxidation
Photocatalyst.
Sci Rep
6:22727.
13. Dang HX, Rettie AJE, & Mullins CB (2015) Visible-Light-Active NiV 2 O 6 Films for
Photoelectrochemical Water Oxidation.
J Phys Chem C
119(26):14524-14531.
14. Courtin E
, et al.
(2014) New Fe2TiO5-based nanoheterostructured mesoporous photoanodes with improved
visible light photoresponses.
J Mater Chem A
2(18):6567-6577.
15. Chemelewski WD, Mabayoje O, & Mullins CB (2015) SILAR Growth of Ag3VO4 and Characterization
for Photoelectrochemical Water Oxidation.
J Phys Chem C
119(48):26803-26808.
16. Tang D
, et al.
(2015) Facile growth of porous Fe2V4O13 films for photoelectrochemical water oxidation.
J
Mater Chem A
00:1-9.
17. Doumerc JP, Hejtmanek J, Chaminade JP, Pouchard M, & Krussanova M (1984) A photoelectrochemical
study of CuWO4 single crystals.
physica status solidi (a)
82(1):285-294.
18. Sayama K
, et al.
(2003) Photoelectrochemical decomposition of water on nanocrystalline BiVO4 film
electrodes under visible light.
Chem Commun
(23):2908-2909.
19. Chan MKY & Ceder G (2010) Efficient Band Gap Prediction for Solids.
Phys Rev Lett
105(19).
20. Gritsenko O, Vanleeuwen R, Vanlenthe E, & Baerends EJ (1995) Self-Consistent Approximation to the
Kohn-Sham Exchange Potential.
Phys Rev A
51(3):1944-1954.
21. Onida G, Reining L, & Rubio A (2002) Electronic excitations: density-functional versus many-body
Green's-function approaches.
Rev Mod Phys
74(2):601-659.
22. Hybertsen MS & Louie SG (1986) Electron correlation in semiconductors and insulators: Band gaps and
quasiparticle energies.
Phys Rev B
34(8):5390-5413.
23. Shih BC, Xue Y, Zhang PH, Cohen ML, & Louie SG (2010) Quasiparticle Band Gap of ZnO: High
Accuracy from the Conventional G(0)W(0) Approach.
Phys Rev Lett
105(14):146401.
24. Wen XD, Martin RL, Henderson TM, & Scuseria GE (2013) Density Functional Theory Studies of the
Electronic Structure of Solid State Actinide Oxides.
Chem Rev
113(2):1063-1096.
25. Kresse G & Furthmuller J (1996) Efficiency of ab-initio total energy calculations for metals and
semiconductors using a plane-wave basis set.
Comp Mater Sci
6(1):15-50.
26. Yan Q
, et al.
(2015) Mn2V2O7: An Earth Abundant Light Absorber for Solar Water Splitting.
Adv Energy
Mater
5(8):1401840.