-
Recent advances in the use of density functional theory to
design efficientsolar energy-based renewable systemsRamy Nashed,
Yehea Ismail, and Nageh K. Allam
Citation: J. Renewable Sustainable Energy 5, 022701 (2013); doi:
10.1063/1.4798483 View online: http://dx.doi.org/10.1063/1.4798483
View Table of Contents: http://jrse.aip.org/resource/1/JRSEBH/v5/i2
Published by the American Institute of Physics.
Additional information on J. Renewable Sustainable EnergyJournal
Homepage: http://jrse.aip.org/ Journal Information:
http://jrse.aip.org/about/about_the_journal Top downloads:
http://jrse.aip.org/features/most_downloaded Information for
Authors: http://jrse.aip.org/authors
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
Recent advances in the use of density functional theoryto design
efficient solar energy-based renewable systems
Ramy Nashed,1,2 Yehea Ismail,2 and Nageh K. Allam1,a)1Energy
Materials Laboratory, Physics Department, School of Sciences and
Engineering,The American University in Cairo, New Cairo 11835,
Egypt2Center of Nanoelectronics and Devices (CND), American
University in Cairo/Zewail Cityof Science and Technology, Cairo,
Egypt
(Received 18 January 2013; accepted 13 March 2013; published
online 27 March 2013)
This article reviews the use of Density Functional Theory (DFT)
to study the
electronic and optical properties of solar-active materials and
dyes used in solar
energy conversion applications (dye-sensitized solar cells and
water splitting). We
first give a brief overview of the DFT, its development,
advantages over ab-initiomethods, and the most commonly used
functionals and the differences between
them. We then discuss the use of DFT to design optimized dyes
for dye-sensitized
solar cells and compare between the accuracy of different
functionals in
determining the excitation energy of the dyes. Finally, we
examine the application
of DFT in understanding the performance of different photoanodes
and how it
could be used to screen different candidate materials for use in
photocatalysis in
general and water splitting in particular. VC 2013 American
Institute of Physics.[http://dx.doi.org/10.1063/1.4798483]
I. INTRODUCTION
The energy demand is currently increasing at an unprecedented
rate due to the high tech-
nology revolution. Humans used to depend on fossil fuels as
their primary source of energy.
However, fossil fuel reserves are limited1 and their production
is declining over the time. Also,
fossil fuels have serious adverse impacts on the environment
because the gases emitted from
burning the fuels trap the solar radiation leading to global
warming and threatening the lives of
humans and other creatures.2 For these reasons, scientists are
investigating alternative sources
of energy. Solar energy is one of the most attractive options as
the amount of solar energy
reaching the earth is four orders of magnitude greater than the
current worlds energy
consumption.3
Dye-Sensitized Solar Cells (DSSCs) represent one of the most
promising solar cell struc-
tures due to their low-cost, simple fabrication method, and
relatively high efficiency that
reached 11.4%.4 The main problem limiting the further
development of DSSCs is the fact that
dyes with high absorption coefficients have narrow bands and
vice-versa. Forster Resonance
Energy Transfer (FRET) mechanism proves to be an efficient
technique that combines high-
absorption dyes with wide-band ones in a donor-acceptor fashion
leading to systems with wide-
strong absorption.5,6 In order for this mechanism to work, the
absorption spectrum of the
acceptor should have large overlap with the emission spectrum of
the donor and the two materi-
als should be within one Forster radius apart.5 However, the
combination of dyes is still not
optimized and is based on guess-and-check procedures. A more
systematic approach isrequired in order to find an optimum
combination of dyes.
A similar approach to DSSCs is the solar-hydrogen production by
the photoelectrolysis of
water. Here, water acts as the electrolyte medium. Hydrogen is
particularly chosen since it
a)Author to whom correspondence should be addressed. Electronic
mail: [email protected]. Fax: 202 27957565.
1941-7012/2013/5(2)/022701/27/$30.00 VC 2013 American Institute
of Physics5, 022701-1
JOURNAL OF RENEWABLE AND SUSTAINABLE ENERGY 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
represents the richest fuel in terms of energy per unit mass.7
Briefly, the system for solar hydro-
gen production consists of two electrodes: a working electrode,
which is made of a semiconduc-
tor, and a counter electrode, which is mainly platinum, dipped
in water. When the working
electrode is illuminated, electrons are excited from the valence
band to the conduction band of
the semiconductor. The holes in the valence band diffuse to the
semiconductor/water interface
where they oxidize water to oxygen gas and hydrogen ions. The
excited electrons will flow
through the wire to the cathode where they reduce the hydrogen
ions forming hydrogen fuel.
The main challenge to achieve efficient solar hydrogen
production systems is to find a cheap
and stable material with wide absorption spectrum to be used as
the working electrode. Some
oxides, such as TiO2, tend to be stable due to their large
energy gap. Unfortunately, this large
band gap limits the absorption capabilities of the material. On
the other hand, low band materi-
als, such as Fe2O3 and Cu2O, have wide absorption spectrum but
tend to be unstable. This
incites researchers to use mixed metal oxides in which a high
band gap oxide is mixed with a
low band gap one hoping to arrive at a stable, yet widely
absorbing material. Different mixed
metal oxides such as Ti-Fe-O,8 Ti-Cu-O,9 Ti-W-O,10 Ti-Nb-Zr,11
and Ti-Pd12 have been tried
and showed very promising results for photocatalytic fuel
production. However, as in the case
of DSSCs, the mixing between oxides is highly combinatorial13
and still needs to be optimized.
A systematic approach still needs to be devised to understand
the effect of atomic composition
of metal oxides on the electronic and photocatalytic properties
of the material.14
Computational science is considered a good approach to solve the
guess-and-check prob-
lem involved in the design of DSSCs as well as solar hydrogen
production systems.
Specifically, it can be used to study the changes in the
electronic structures of the fabricated
systems. This would give a better understanding of the systems
behavior and hence helps to
pinpoint the positions of weaknesses in the design, which would
assist to arriving at more
optimized solutions. Density Functional Theory (DFT) is regarded
as one of the most efficient
computational tools to be used in this domain because it is
computationally inexpensive albeit
accurate. DFT is used in wide range of research topics such as
structural materials, catalysis
and surface science, magnetism, semiconductors and
nanotechnology, and biomaterials. Ikehata
et al.15 used DFT calculations to design a low-Youngs modulus
high-strength titanium alloy.Ford motor company makes use of DFT to
enhance the properties of aluminum cast alloys.16,17
Tripkovic et al.18 studied the oxygen reduction reaction
mechanism on platinum surface forPEM fuel cell design using DFT
calculations. DFT-based calculations have also been used to
suggest Fe-Co alloys as a good candidate for high-density
magnetic storage.19 In semiconductor
nanotechnology, DFT has been used to gain insight into the
electronic structure of carbon nano-
tubes, quantum dots, and semiconducting nanoparticles.20
It is clear that DFT has wide range of applications; however, we
will limit our discussion
to solar energy applications. In the next section, we will
discuss the development of DFT and
assess the accuracy of its different variations. Section III
discusses the use of DFT in the design
of DSSCs and Sec. IV discusses its use in designing the working
electrode for solar hydrogen
production.
II. OVERVIEW OF DFT
A. Pre-DFT attempts
The physical and chemical properties of any system can be
determined exactly by solving
the many-body Schrodinger equation,
H^Wir;R EiWir;R; (1)
where Wi is the wave function of the system, Ei is the
Eigen-values, which are the allowedenergy states produced by
solving Eq. (1), and H^ is the Hamiltonian operator. For
interactingatoms, H^ is defined as21
022701-2 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
H^ XPI1
h2
2MIr2I
XNi1
h2
2mr2i
e2
2
XPI1
XPJ 6I
ZIZJjRI RJj
e2
2
XNi1
XNj 6i
1
jrI rJj e2XPI1
XNi1
ZIjRI rij; (2)
where R {RI}, I 1,, P, is a set of P nuclear coordinates, and r
{ri}, i 1,, N, is a setof N electronic coordinates. ZI and MI are
the P nuclear charges and masses, respectively.
When interpreted physically, the first term on the right hand
side of Eq. (2) is the kinetic
energy of the P nuclei, the second term is the kinetic energy of
the N electrons, the third term
is the Coulomb repulsive potential between each pair of nuclei,
the fourth term is the Coulomb
repulsive potential between each pair of electrons, and the
fifth term is the Coulomb attraction
potential between the electrons and the nuclei in the
system.
It is obvious that the Hamiltonian for such systems is very
complicated and requires large
computational effort especially for large atoms and molecules.
Also, the analytical expression
for the many-electron Hamiltonian is not known. For these
reasons, various simplifications have
been introduced to Eq. (2). The first approximation is the
Born-Oppenheimer approximation22
which is based on neglecting the kinetic energy of nuclei and
treating their repulsive potential
as a constant. The plausibility of this approximation is due to
the fact that the mass of the
nuclei is much greater than that of the electrons and thus the
nuclei can be assumed stationary
with respect to the electrons. This gives rise to the so-called
electronic Hamiltonian,
H^elec XNi1
h2
2mr2i
e2
2
XNi1
XNj 6i
1
jrI rJj e2XPI1
XNi1
ZIjRI rij: (3)
The electronic energy, Eelec, can be found by substituting Eq.
(3) in Eq. (1). The total energycan then be calculated by adding
Eelec to the constant nuclear repulsion term Enuc,
Enuc e2
2
XPI1
XPJ 6I
ZIZJjRI RJj: (4)
Although the Hamiltonian was greatly simplified by the
Born-Oppenheimer approximation,
the second term in Eq. (3) still represents a computational
problem as it involves pair-wise
Coulombic correlation between electrons and hence it is required
to consider the contribution
of two electrons every time we write the wave function. This
renders the wave function compli-
cated and the solution of Schrodinger equation hard. Hartree
proposed a solution to this prob-
lem by assuming that each electron in the system feels an
average potential energy due to the
other electrons.23 This allows for treating a single electron at
a time and consequently to
express the wave function as a product of one-electron wave
functions. It uses separation of
variables to solve Schrodinger equation, which greatly
simplifies the calculations. To determine
the expressions for the one-electron wave functions, Hartree and
Fock (HF) introduced a
method which took into account Pauli exclusion principle where
the many-electron wave func-
tion is approximated by a product of anti-symmetrical
one-electron wave functions in the form
of a Slater determinant,21
WHF w1r1; r1 w1r2; r2 w1rN; rNw2r1; r1 w2r2; r2 w2rN; rN
. ..
wNr1; r1 wNr2; r2 wNrN; rN
; (5)
022701-3 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
where ri signifies the spatial position of the electron i and r
signifies its spin. Wi(ri, ri) areexpressed as a Linear
Combinations of Atomic Orbitals (LCAOs) to form Molecular
Orbitals
(MOs). Using this approximation, the energy of the system can be
calculated as
EHF XNi1
Hi 12
XNi;j1
Jij Kij; (6)
where Hi represents the kinetic energy and the electron-nucleus
Coulomb attraction, Jij are thecoulomb integrals, which represent
the repulsive potential that the electron feels due to an aver-
age distribution of the rest of the electrons, and Kij is the
exchange integrals that are a quantummechanical effect occurring
due to the overlapping of orbitals, which combines all possible
per-
mutations of electron energy distribution in the system. This
approximation is called HF or
Self-Consistent Field (SCF) approximation and it includes
particle exchange in an exact man-
ner.24,25 The main drawback of this method is that the
computational effort needed to compute
Eq. (5) scales by M3, where M is the number of atomic
orbitals.
B. Development of DFT
Despite the different approximations that have been applied to
the Hamiltonian and the
wave function, solving Schrodinger equation remains very hard
and nearly impossible for large
atoms and molecules since the wave function is a function of 3N
variables, where N is the
number of electrons in the system. Density functional theory
solved this problem by reducing
the number of variables to three variables only.26 This is
because DFT is based on using the
electron density, which is a function of the three spatial
coordinates, to calculate the energy of
the system. This considerably reduced the computational cost and
allowed for determining the
physical and chemical properties of large atoms and
molecules.
The efforts of Thomas27 and Fermi,28 which date back to 1927,
represent the seed of the
DFT. In Thomas-Fermi model, the energy of the system is
calculated in terms of the electron
density as
ETFqr 310
3p22=3q5=3rdr Z
qrr
dr 12
qr1qr2
r12dr1dr2; (7)
where q is the electron density. The first term in Eq. (7)
represents the kinetic energy of elec-trons, the second term is the
nuclear attraction between nuclei and electrons, and the third
term
is the Coulomb repulsion between electrons. The kinetic energy
term is found by solving a par-
ticle in a box problem assuming a constant electron density.
This is a very crude approximation
since the electron density is non-uniform and is actually
rapidly changing near the nuclei. Also,
the exchange and correlation effects are neglected.21 In 1930,
Dirac used the uniform electron
density approximation to introduce an expression for the
exchange energy,29 which gave rise to
Thomas-Fermi-Dirac theory.30 Weizsacker31,32 was the first to
target the non-uniform electrondensity problem in 1935 by providing
an expression for the kinetic energy of electrons that
depends on the gradient of the electron density in the
neighborhood. Considering the gradient
of electron density allowed for adding information about how the
electron density changes in
the vicinity of each point in space. This led to two refinements
to the Thomas-Fermi theory: (1)
Thomas-Fermi-Weizsacker theory,32 which corrects the kinetic
energy term in Thomas-Fermitheory by considering non-uniform
electron density but did not consider exchange correlation
energy, (2) Thomas-Fermi-Dirac-Weizsacker32 which not only
corrects the kinetic energy termin Thomas-Fermi but also includes
the exchange energy term using Dirac approximation.
However, this theory is still not accurate as it is based on
Dirac approximation.
DFT started to attract great attention after the work done by
Hohenberg and Kohn in 1964
who proved that the potential is a unique functional of electron
density.26 This is a marvelous
achievement because it means that for each electron density
distribution, there is one and only
one expression for the energy of the system. The proof of this
theorem comes from the fact
022701-4 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
that, in order to determine the Hamiltonian operator, one needs
to determine the number of
electrons in the system as well as the positions of the nuclei.
The electron density is very
powerful in this aspect as the integration of the electron
density over the whole space gives the
number of electrons and the positions of the cusps found when
plotting the electron density ver-
sus position represent the locations of the nuclei. According to
Hohenberg-Kohn theorem, the
energy of the system can be expressed as26
E vrqrdr 1
2
qrqr0jr r0j drdr
0 Gq; (8)
where v(r) represents the nuclear potential. The first term
represents the nuclear Coulombattraction, the second term is the
electron Coulomb repulsion and G[q] is the sum of the elec-tron
kinetic energy, T[q], and the exchange and correlation energy,26
Exc[q],
Gq TqExcq: (9)
In their original paper, Hohenberg and Kohn did not propose
explicit forms to the kinetic
energy and exchange and correlation energies. Kohn and Sham
addressed this problem in
1965,33 shortly after the publication of the original
Hohenberg-Kohn theorem. Kohn and Sham
provided an exact expression for T[q] as well as a semi-exact
expression for Exc[q]. The calcu-lation of the Exc[q] term depends
on splitting it into two terms: Exchange term, Ex, andCorrelation
term, Ec, where Ex is calculated exactly from Hartree-Fock
equations and Ec isapproximated under the assumption of a uniform
electron density. Although the calculation of
Exc[q] is very accurate, it requires large computational power
since the calculation of Ex isbased on Hartree-Fock equations,
which involve wave functions instead of electron density. For
this reason, a simpler expression for Exc[q] is suggested by
Kohn and Sham assuming uniformelectron density for the whole
expression of Exc[q]. From the above discussion, it is obviousthat
the main challenge in DFT is to find the proper expression for T[q]
and Exc[q].
C. Basis sets
The Kohn-Sham equation, which is analogous to Schrodinger
equation, can be written as33
12r2 vr
qr0jr r0j dr
0
lxcqr
wir eiwir; (10)
where v(r) is the attractive Coulomb potential between the
electron and the nuclei, lxc(r) is thedensity of Exc with respect
to q, and wi(r) is the Kohn-Sham orbitals which are analogous
towave functions in Schrodinger equation. The numerical solution of
Eq. (10) requires expanding
Kohn-Sham orbitals in a set of pseudopotentials (PPs).34
The main types of basis functions are the Slater-Type Orbitals
(STOs), Gaussian-Type
Orbitals (GTOs), Contracted Gaussian Functions (CGFs), and PPs.
Slater-Type Orbitals35 are
functions which decay exponentially far from the origin. They
closely resemble the true behav-
ior of atomic wave functions as they have cusps at the nuclei
positions. However, they require
large computational efforts. On the other hand, Gaussian-Type
Orbitals36 are not as accurate as
STOs but they are easier to handle numerically since the product
of two GTOs located at differ-
ent atoms is another GTO located between the atoms, whereas the
product of two STOs does
not lead to an STO.34 Contracted Gaussian functions37 represent
a compromise between the ac-
curacy of STO and the simplicity of GTO where CGF is constructed
by approximating STO by
a small number of GTOs. Pseudopotentials represent the most
attractive basis functions for sys-
tems with large number of electrons.34 The idea of using
pseudopotentials is based on the fact
that the binding energy of solids and molecules is dominated by
the valence electrons of each
atom and hence only the valence electrons need to be considered
in Eq. (10), which
022701-5 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
tremendously reduces the number of electrons treated explicitly.
This allows for performing
DFT calculations on large systems.
D. DFT functionals
As mentioned above, the main challenge in DFT is to find the
proper expression for T[q]and Exc[q]. Several expressions have been
proposed which are briefly described in this section.
1. Local density approximation
Local Density Approximation (LDA) is the first and simplest
approximation in DFT. It is
based on decomposing the real problem of a non-uniform
interacting system into two simpler
components: a spatially non-uniform non-interacting system to
calculate T[q], and a uniforminteracting system to calculate
Exc[q].
34 The expression of Exc[q] follows that proposed byKohn and
Sham:33
Excq qrexcqrdr; (11)
where exc[q] is the exchange and correlation energy per electron
of a uniform electron gas. InLDA, exc[q] is decomposed, like in
Kohn-Sham, into two functionals: exchange functional (ex)and
correlation functional (ec). The exchange functional is calculated
from Diracs form
38 while
the correlation function is unknown and has been simulated in
numerical quantum Monte Carlo
calculations assuming uniform electron density and yielded
nearly exact results.39 In LDA,
Exc[q] is very-well approximated since the errors in ec tend to
be cancelled by ex.34
2. Generalized gradient approximation
Generalized gradient approximation (GGA) builds on LDA by
considering non-uniform dis-
tribution of electrons. In GGA, exc is a functional of electron
density as well as its gradientwhich helps to take into account the
way by which the electron density changes in the vicinity
of the point of interest. This is very crucial when considering
points near the nuclei in which
the electron density is strongly changing. Nowadays, the most
popular GGA in physics is PBE
which was proposed by Perdew et al.,40 whereas BLYP, which is a
combination of Beckesexchange energy41 with Lee et al.s correlation
energy,42 is the most popular GGA inchemistry.34
3. Meta-GGA
Although GAA has shown great improvements in calculations
compared to LDA, the
chemical accuracy, which requires that the errors in
calculations should exceed 1 kcal/mol, was
not reached yet.34 For this reason, several beyond-GGA
functionals were introduced. Meta-
GGA43,44 is an example of beyond-GGA development in which the
exchange energy depends
on the Laplacian of the spin density, r2q, or the local kinetic
energy density, s. The incorpora-tion of Meta-GGA helped to solve
some problems of the previous functional such as self-
interaction problem in the correlation functional, increasing
the accuracy of calculating the
exchange functional by recovering the fourth order gradient
expansion for slowly varying den-
sities, and obtaining a finite exchange potential at the
nucleus.45
4. Hybrid exchange functionals
Although LDA and GGA give good approximations for many
calculations, they tend to
underestimate the transition energy. This is because they do not
contain the correct 1/R depend-
ence (where R is the distance between charges) in the exchange
functional expression. Hybrid
functionals can remedy this problem through the incorporation of
the exact Hartree-Fock
exchange functional.
022701-6 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
Hybrid functionals are based on the exact adiabatic approach,46
which allows for the exact
representation of the exchange and correlation energy functional
as
Excq 12
drdr0
1
k0
dkke2
jr r0j hqrqr0iq;k qrdr r0; (12)
where k is a coupling constant with k 0 corresponding to
non-interacting system and k 1corresponding to fully interacting
system. A non-interacting system is well-represented by
Hartree-Fock equations while GGA is a good representation for a
fully interacting system with
a uniform electron density. A logical approximation to the
integral in Eq. (12) is to consider
the extreme cases with k 0 and k 1 and use a weighted average to
approximate Exc[q].Beckes hybrid functional,47 B3, employs this
idea and is considered the most successful
exchange functional for chemical applications, especially when
combined with LYP GGA42
functional for Ec to form B3LYP functional which is the most
popular functional in quantumchemistry.34
5. Long-range corrected functionals
For hybrid functional, the asymptotic behavior in the
exchange-correlation expression
decays as 0.2/R. This is accurate enough for small molecules.
However, as the molecules
become more complicated, the discrepancy between the exact value
and that obtained by hybrid
functional increases. The idea of long-range-corrected (LC)
functionals, similar to hybrid func-
tionals, is based on using exact Hartree-Fock exchange
functional. However, the fraction of
exact exchange increases with increasing the separation
distances, whereas this fraction is kept
constant for hybrid functionals. This represents an optimization
of both the computational cost
and accuracy. At small separation distances, where there is
negligible difference between
Hartree-Fock and DFT calculations, the fraction of Hartree-Fock
is kept low. On the other
hand, a large fraction of Hartree-Fock exchange functional is
employed at longer distances. The
performance of LC functionals depends on the range-separation
parameter, l, whose optimumvalue was found to be
system-dependent.48 For example, the l-dependence of small
moleculeexcitation energies (CO, H2CO, and CH3CO) is markedly
different from that observed in larger
molecules (anthracene, indole, pyridazine, benzocyclobutendione,
benzaldehyde, and pyrrole).48
This difference was attributed to the fact that the valence
excited-state densities for these larger
molecules sample both the long- and short-range parts of the
Coulomb potential, whereas the
small molecules fit more or less within the length scale
described using normal TD-DFT.48
These observations stress the importance of having
large-molecule excitation energies in the
training set of any LRC functional that is intended for use in
TD-DFT.4850 In fact, the value
of l was found to be inversely proportional to the size of the
molecules under study.49
6. Time-dependent density functional theory
This theory is considered as the Hohenberg-Kohn analog in time
dependent systems in
which the electron density is a function of time and position.
The theory was developed by
Runge and Gross in 1984 to study the excited state properties of
the materials such as atomic
and nuclear processes, photoabsorption in atoms, and the dynamic
response of systems.51 The
main advantages of TD-DFT include the balance of accuracy and
efficiency as well as the wide
applicability range.51 However, TD-DFT must use an approximation
for the exchange-
correlation energy. The main drawback of TD-DFT is the
underestimation results for large
molecules even when used with hybrid functionals.52 However,
combining TD-DFT with LC
functionals represents a promising approach towards an accurate
study of large systems.4850
7. Performance of various functionals
The performance of LDA, GGA, and Meta-GGA functionals has been
assessed by Kurth
et al.53 The study was made on several molecules and the mean
relative error was calculated
022701-7 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
for each approximation. The GGA functionals considered were
BLYP, PBE, and HCTH,
whereas the Meta-GGA functionals considered were VS98 and PKZB.
Table I illustrates the
mean relative error in the atomization energy, unit cell volume,
and bulk modulus for each
functional. For atomization energy, GGA offers a great
improvement to LDA reducing the error
from 22% down to 3%. Meta-GGA does not provide a significant
improvement in this aspect.
On the other hand, LDA performs very well in approximating the
unit cell volume producing
results comparable to that of GGA. The errors in bulk modulus
tend to be large with the excep-
tion of PBE and PKZB, which show relatively small errors.
III. OPTIMIZING DYE-SENSITIZED SOLAR CELLS USING DENSITY
FUNCTIONAL
THEORY
Figure 1 illustrates the schematic diagram of DSSCs. The
operation of DSSCs can be sum-
marized as follows: Upon illumination, electrons are excited
from the HOMO levels to the
LUMO of the dye. The excited electrons are then transported to
the conduction band of the
semiconductor. The dye is regenerated with the help of the redox
electrolyte, which compen-
sates the electrons that the dye loses to the semiconductor.
In order to have an efficient system, certain criteria should be
met: (1) the absorption of
the dye should cover a wide range of the solar spectrum, (2) the
energy of the dyes LUMO
should be less negative than the conduction band edge of the
semiconductor, (3) the dye should
be well anchored to the semiconductor surface to allow for
efficient charge transfer, (4) the
conduction band of the semiconductor should be located well
above the HOMO of the dye to
minimize recombination, and (5) the Redox potential of the
electrolyte should lie above the
HOMO of the dye to permit dye regeneration. Having these
criteria in mind, it should be sim-
ple to design an efficient system if we can predict the band
structure of each component of the
system.
TABLE I. Mean relative error for atomization energy, unit cell
volume, and bulk modulus for different functionals.
Adapted from Ref. 55.
Functional Atomization energy (%) Unit cell volume (%) Bulk
modulus (%)
LDA 22 5 19
BLYP 5 8 22
PBE 7 4 10
HCTH 3 6 20
VS98 2 8 29
PKZB 3 3 9
FIG. 1. Schematic diagram of dye-sensitized solar cell.
022701-8 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
A lot of experimental work4,5465 has been done to obtain
efficient systems. However, most
of that work focused on optimizing a certain criterion and
neglecting the others. Therefore an
ultimately efficient system has not been realized yet, as the
criteria mentioned above are very
interdependent. Only recently, the computational methods have
been combined with experimen-
tal work in DSSCs to explain the behavior of current systems
aiming at the design of more
optimized systems. Density functional theory plays a very
important role in this aspect as it
provides information about the electronic and optical
characteristics of each component in the
system as well as how would a change in a certain component
reflect on the performance of
others.
For example, Kusama et al.66 studied the effect of adsorption of
N-containing heterocycles,namely 4-t-butylpyridine (TBP) and
imidazole, on the bandgap of TiO2. They used DFT to cal-
culate the density of states (DOSs) of the system before and
after adsorption of the hetero-
cycles. They found that both the valence and conduction band of
TiO2 move upwards with the
same value keeping the bandgap of TiO2 unchanged (Figure 2 (Ref.
66)). They also found that
the shift in case of imidazole is higher than that of TBP which
led to a higher open circuit volt-
age in case of imidazole as the distance between the Fermi level
of the semiconductor and the
Redox potential of the electrolyte increases. Also, assuming the
position of the TBPs and imi-
dazoles LUMO is the same, this shift led to a lower short
circuit current in case of imidazole
due to decreased injection rate. However, the position of the
LUMO for different dyes is gener-
ally different as asserted by Hagberg et al.52 Therefore, it is
more desirable to calculate the rel-ative shift of the conduction
band edge with respect to the heterocycle in order to assess
the
injection rate of electrons. Furthermore, the bandgap of TiO2
calculated from Figure 2 is about
2 eV which is quite far from the experimental value of 3.2 eV.
This discrepancy is due to two
reasons: (a) using the conventional PBE functional,40 which
suffers from errors due to elec-
tronic self-interaction energy,67 and (b) using DFT, which is
limited to ground state calcula-
tions, to calculate excited state. The first problem can be
alleviated by using a more exact
exchange correlation functionals such as hybrid functionals
whereas the second problem can be
overcome by implementing time-dependent density functional
theory (TD-DFT) proposed by
Runge and Gross51 who extended the Hohenberg-Kohn theorem to
arbitrary time dependent
systems, which allowed for studying the dynamics of the
systems.
FIG. 2. Calculated total density of states. Reprinted with
permission from H. Kusama et al., Sol. Energy Mater. Sol.
Cells92(1), 8487 (2008). Copyright 2008 Elsevier.
022701-9 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
The ability of TD-DFT to accurately calculate the electronic
properties of DSSCs was
proved by Mete et al.68 who calculated the HOMO-LUMO gap of
ruthenium bipyridine dyeusing the conventional PBE functional in
DFT as well as the hybrid B3LYP in TD-DFT. DFT
calculations led to a bandgap of 1.85 eV whereas TD-DFT
calculations gave a result of 2.64 eV
which is much closer to the experimental value of 2.7 eV.68
Following this conclusion, Mete
et al. adopted TD-DFT to study the effect of attaching different
ligands and halogen atoms onthe electronic properties of perylene
diimide (PDI) dye molecules. They pointed out that the
energy gap depends on the size of the halogen atom as well as
the geometric structure of the
molecule. As the size of the halogen increases, the bandgap
decreases and the bond length
increases due to increasing the delocalization of the density of
states68 (Table II). On the other
hand, the introduction of more carboxylic groups from ligands
does not affect the energy gap
since they introduce energy levels below the HOMO (shown in
Figure 3).68 However, introduc-
ing more of these levels by adding more carboxylic groups
increases photon harvesting, which
widens the absorption spectrum of the dye. Also, increasing the
number of carboxylic groups
improves the electron transfer from the dye to the semiconductor
due to better anchoring.
Although B3LYP tends to be in close agreement with experimental
values, the calculated
values of excitation energy are underestimated (2.64 eV versus
2.7 eV as reported in Ref. 68).
This underestimation was confirmed by Hagberg et al.52 who found
that the extent of underesti-mation increases by increasing the
distance between the donor and acceptor moieties, i.e., size
of the dye. This is a severe drawback of B3LYP since the use of
long-chain dyes is inevitable
due to their high extinction coefficient.52 Miao et al.69 tried
to find a better agreement withexperiments by trying other hybrid
functionals. They compared the accuracy of B3LYP and
PBE070 in calculating the excitation energy of 39N-substituted
1,8-naphthalimides dyes and
found that PBE0 provided a slightly better accuracy with a mean
absolute error of 0.21 eV
FIG. 3. Calculated TDDFT molecular orbital levels of PDI based
dye molecules. Adapted from Ref. 68.
TABLE II. The effect of halogen on the bandgap and carboxylic
group-PDI bond length. Adapted from Ref. 68.
Chromophore Bandgap (eV) Bond length (A)
F-PDI 2.50 1.38
Cl-PDI 2.44 1.74
Br-PDI 2.39 1.88
I-PDI 2.31 2.12
022701-10 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
compared to 0.28 eV obtained from B3LYP with the error in the
excitation energy depending
on the solvent.69 This has been asserted by Zhang et al.71 who
compared B3LYP, PBE0, andMPW1K72 in calculating the electronic
absorption of TA-St-CA dye sensitizer and found that
PBE0 and MPW1K are somewhat better than B3LYP with maximum
absorption occurring at
378 nm for PBE0 and MPW1K and at 395 nm for B3LYP as shown in
Figure 4.71 This is com-
parable to the experimental result of 386 nm.73
Another interesting observation from Figure 4 is the red shift
of the dye absorption spec-
trum when being in a solvent compared to that in vacuum. The
same phenomenon was
observed by Kurashige et al.74 as well as by Pastore et al.75
and was attributed to the polariza-tion of the solvent, which led
to electrostatic interaction with the charge-separated excited
state
leading to its stabilization.74,75 To model this interaction, a
Polarizable Continuum Model
(PCM) is introduced while doing the calculations. When ignoring
this phenomenon in calcula-
tions, Pastore et al. had an error of about 0.2 to 0.3 eV in the
excitation energy compared toabout 0.1 eV when PCM is included.75
However, Rocca et al.76 ignored this phenomenon dur-ing their study
on squaraine dyes and their calculations were within 0.14 eV of the
experimental
results, which suggests that this phenomenon is very
system-dependent and that the strength of
electrostatic interaction between the dye and the solvent should
be considered first before decid-
ing to include or exclude PCM in the calculations.
The adsorption of the dye on the semiconductor surface
represents one more parameter that
affects its absorption spectrum. It was shown experimentally7780
that the adsorption of dyes on
TiO2 leads to a blue shift of the dyes absorption spectrum due
to the deprotonation of the dye
where the proton is replaced by the metal ion leading to an
upward shifting of the LUMOs
energy.80 Two TD-DFT recent works by Agrawal et al.81 and
Sanchez-de-Armas et al.82 wereconducted in parallel on coumarin
derivatives and both studies asserted the upward shifting of
the LUMO, which led to blue shifting of the dyes absorption
spectrum. Deprotonation of these
dyes when adsorbed on TiO2 was confirmed by FTIR indicating the
presence of carboxylate
ion after adsorption.83 In their analysis, Sanchez-de-Armas et
al. ignored long range interactionsassuming the considered systems
are not very large and hence having negligible long range
interaction.82 Although this is true for the smallest dyes
considered, the error becomes
FIG. 4. Calculated electronic absorption spectra of TA-St-CA.
Reprinted with permission from Zhang et al., Curr. Appl.Phys.
10(1), 77-83 (2010). Copyright 2010 Elsevier.
022701-11 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
sufficiently large for larger dyes. The error increases from
0.15 eV for the smallest dye to
0.47 eV for the largest dye considered.82 This is in agreement
with the results of Agrawal et al.,where the error was 0.01 eV for
the smaller dye and 0.37 eV for the larger one.81 They attrib-
uted this error to using B3LYP functional.
Almost all of the previously mentioned works depend on utilizing
hybrid functionals in cal-
culating the electronic properties of the system. These
functionals are not very accurate as they
ignore long range interaction, which has a pronounced effect in
large systems. In other words,
these functionals provide a very fast decay of the potential and
hence fail to keep the 1/r de-
pendence of the potential especially at large distances. This
limits these methods to qualitative
descriptions and trends of the system only. Unfortunately,
quantitative description is still
required in order to design an optimized DSSC. Several efforts
have been made to find more
accurate methods. Kurashige et al.74 resorted to
wavefunction-based methods, namely configu-ration interaction
single (CIS) method and approximate coupled cluster singles and
doubles
(CC2) method, to investigate the excited states of coumarin
dyes. The main advantage of
wavefunction-based methods is that they do not suffer from
self-interaction error, which is con-
sidered the main drawback of DFT. They compared these methods
with TD-DFT using the pop-
ular B3LYP functional. As expected, they found that B3LYP
provides good results for small
dyes but introduces relatively large errors as the size of the
dye increases. Also, CIS is likely to
overestimate the excitation energy and oscillation strengths due
to the lack of electronic correla-
tion. CC2 seems to be the best candidate especially for large
dyes. For example, for C343 dye,
the experimental value for the excitation energy is 2.81 eV. The
calculated value from B3LYP
is 3.09 eV leading to an error of 0.28 eV whereas CC2 gives 3.19
eV leading to an error of
0.38 eV. CIS overestimates these values by giving 4.15 eV
leading to an error of 1.34 eV. On
the other hand, B3LYP does not seem to be a good choice for
large NKX-2586 dye underesti-
mating the excitation energy value by 0.24 eV. CC2 overestimates
the experimental value by
only 0.07 eV and again CIS overestimates the experimental value
by 0.79 eV.74 Table III lists
the experimental together with the calculated excitation
energies for different coumarin dyes
using different functionals.74
Although CC2 seems appealing from the accuracy point of view,
being a wavefunction-
based method renders it computationally expensive. To solve this
problem, Wong and
Cordaro50 used a LC functional, namely LC-BLYP, which was
proposed by Hirao et al.49,84,85
in TD-DFT calculations and compared the results with that of
Kurashige and co-workers.74 The
main idea of LC-TDDFT functional, which was initially proposed
by Savin,86 is to provide a
higher percentage of the exact Hartree-Fock exchange energy as
the distance increases to
recover the 1/r dependence of the potential. Wong and Cordaro
studied the same dyes of
Kurashige et al. and compared the results obtained by LC-BLYP
with that of CC2. Table IVcompares the values of the calculated
excitation energy of the dyes in the gas phase by LC-
BLYP method with those of CC2, CIS, and B3LYP obtained from
Kurashige et al.50,74 fromwhich we can observe the exceptional
agreement between LC-BLYP and CC2.
This agreement between LC-BLYP and CC2 is considered a
remarkable achievement
owing to the relatively modest computational cost of electron
density-based LC-BLYP com-
pared to wavefunction-based CC2. However, this achievement is
questionable since it was
shown that sometimes CC2 gives underestimated values as pointed
out by Schreiber et al.87
This was confirmed by Pastore et al.75 who compared several
TD-DFT functionals with
TABLE III. Vertical excitation energy for different coumarin
dyes in methanol. Adapted from Ref. 74.
CC2 CIS B3LYP Experimental
C343 3.19 4.15 3.09 2.81
NKX-2388 2.77 3.72 2.70 2.51
NKX-2311 2.63 3.47 2.43 2.46
NKX-2586 2.52 3.24 2.21 2.45
022701-12 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
wavefunction methods on a number of organic dyes. The long-range
corrected (LC) functional,
CAM-B3LYP,88 was specifically attractive as compared to B3LYP
and MPW1K hybrid func-
tionals as it always gave values higher than that of CC2 and
hence compensating the underesti-
mating nature of CC2.75 Also, CAM-B3LYP was preferred to other
wavefunction methods dis-
cussed as these methods tend to be computationally expensive and
unpractical for large
systems. However, the performance of this functional was not
compared to experiment.
From the above analysis, it seems that LC functionals are
promising methods to quantita-
tively describe the electronic behavior of the system. However,
more experiments should be
carried out to assess the performance of these functionals.
Also, LC functionals suffer from a
practical problem which is the choice of the range-separation
parameter, l, which was found tobe system and property
dependent89,90 and therefore there should be a systematic way to
obtain
the optimum value of l.
IV. DESIGNING EFFICIENT PHOTOANODES FOR SOLAR HYDROGEN
PRODUCTION
Hydrogen is considered as one of the most promising energy
carriers to replace fossil fuels
mainly because it does not produce any environmental unfriendly
emissions91 besides being the
richest fuel in energy per unit mass.7 Solar hydrogen production
is the most appealing method
to produce hydrogen92,93 as it depends on clean and abundant
sources of energy, that is, the so-
lar energy.
To produce hydrogen efficiently, certain criteria should be met
in the photocatalyst,
namely, (i) the conduction band minimum (CBM) should be located
above the hydrogen evolu-
tion potential (0 eV vs. NHE) and the valence band maximum (VBM)
should be located below
the oxygen evolution potential (1.23 eV vs. NHE), therefore, the
bandgap should be greater
than 1.23 eV, (ii) The bandgap should be low enough to absorb
light efficiently, (iii) the charge
carriers should have low effective mass to allow for good charge
separation and decrease the
probability of recombination, and (iv) the photocatalyst should
be stable in aqueous solutions.
In 1972, Fujishima and Honda94 used TiO2 as a photocatalyst to
produce hydrogen using
solar energy. TiO2 is well-known for its high catalytic activity
and stability in aqueous solution.
However, its bandgap is about 3.2 eV,95 which limits its
absorption capabilities to the ultraviolet
region only. Also, the conduction band edge should be raised to
allow for the hydrogen evolu-
tion.96 A lot of efforts have been made to reduce the bandgap of
TiO2 by doping with nonme-
tallic elements, such as N and C, or metallic elements, such as
Cr and V.97104 Adding small
amounts of these elements leads to marginal improvements.
Although it reduces the bandgap, it
causes other detrimental effects due to introduction of
recombination centers. Passivated co-
doping with isovalent donor-acceptor elements can provide better
engineering of the bandgap
and at the same time suppress the recombination centers.96,105
It can also provide better solubil-
ity for the alloying elements, which permits higher alloying
concentrations.106109
In this regard, DFT is considered a powerful tool that can give
a deep understanding about
the changes in the electronic and optical characteristics of the
material upon incorporating the
TABLE IV. Cap Comparison between the excitation energy
calculated by different methods for coumarin dyes in the gas
state. Adapted from Refs. 50 and 74.
Dye LC-BLYP (eV) CC2 (eV) CIS (eV) B3LYP (eV)
C343 3.36 3.44 4.40 3.32
NKX-2388 (s-trans) 3.01 2.99 3.94 2.90
NKX-2388 (s-cis) 2.85 2.80 3.75 2.78
NKX-2311 (s-trans) 2.91 2.89 3.73 2.70
NKX-2311 (s-cis) 2.73 2.71 3.50 2.56
NKX-2586 (s-trans) 2.81 2.81 3.53 2.50
NKX-2586 (s-cis) 2.66 2.66 3.34 2.40
NKX-2677 2.67 2.71 3.12 2.23
022701-13 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
co-dopants and thus help in selecting the best dopant
candidates. However, the accuracy by
which the bandgap is determined is very sensitive to the
functional used. A recent study on
Ta2O5 showed that the error in calculating the bandgap can be
reduced from 95% to only 5%
when using PEB0 hybrid functional.110 In this study, Allam et
al.110 compared between GGA-PBE and three hybrid functionals,
namely HSE06, B3LYP, and PBE0. The hybrid functionals
show more accurate results because they partially incorporate
the exact Hartree-Fock exchange.
PBE0 gave the most accurate result since it incorporates the
highest Hartree-Fock percentage.
PBE0 and HSE06 are essentially similar. The main difference
between them is that HSE
divides the exchange energy into short range and long range with
the short range only incorpo-
rating Hartree-Fock, whereas the long range uses pure PBE
functional. In PBE0, this split does
not occur which makes it more accurate especially in systems
with large atoms where there is
long range interaction. For systems with small atoms, the two
methods would give very similar
results since the long range interaction can be neglected.
Although this paper does not study
the effect of doping on the band structure of metal oxides, it
gives a good insight about how
one can choose a suitable functional for the system under test.
The effect of doping on the
band structure of metal oxides was studied by Yin et al.96 who
studied the codoping of TiO2 toremedy its limitations. The idea was
based on replacing Ti with 4d and 5d cations, which have
higher energy than the 3d of Ti and hence the CBM will be raised
above the hydrogen evolu-
tion potential. Also, O was replaced by 2p and 3p anions to
raise the VBM and reduce the
bandgap. Figure 5 shows the resulting bandgap for high and low
codoping concentration.96 For
low concentration, the symbol (X, Y) denotes that 3.1% of Ti was
replaced by X and 3.1% of
O was replaced by Y. (X, 2Y) denotes that 3.1% of Ti was
replaced by X and 6.25% of O was
replaced by Y and so on. For high concentration, the 3.1% is
replaced by 12.5% and 6.25% is
replaced by 25%. At low concentration, (Ta, P) and (Nb, P) tend
to give the lowest energy gap
as well as a slight upshift of the CBM. However, there is a
large mismatch between the atomic
size of P and that of O which would affect the charge transfer
negatively. (Mo, C) and (W, C)
have good bandgap but the CBM is downshifted which makes them
unappealing. (2Nb, C),
(2Ta, C), (Mo, 2N), and (W, 2N) have suitable bandgaps as well
as upshift of the CBM.
However, (Mo, 2N) and (W, 2N) provide better carrier mobility
due to more dispersive bands
and hence they represent the best candidates for low doping
concentration. For high concentra-
tion, (Mo, 2N) and (W, 2N) suffer from CBM lowering and hence
they are discarded. (2Ta, C)
and (2Nb, C) have suitable bandgaps but the VBM is shifted above
the oxygen evolution poten-
tial and so they are discarded too. The best candidates for high
doping concentration are
(Nb, N) and (Ta, N). These calculations were based on General
Gradient Approximation
FIG. 5. Calculated GGA band offsets (at the C point) for TiO2
and TiO2 alloyed with various passivated donor-acceptorcombinations
in the (a) low-concentration regime, (b) high-concentration regime.
The CBM of pure TiO2 is set to zero as
the reference and the band gap is corrected using a scissor
operator. Reprinted with permission from Yin et al., Phys.Rev. B
82, 045106 (2010). Copyright 2010 American Physical Society.
022701-14 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
(GGA) functional of the DFT which is well-known for its
underestimation of bandgaps. To test
the accuracy of these calculations, the authors performed the
calculations using the hybrid func-
tional HSE111 with 22% of the exact Hartree-Fock exchange and
found that the position of the
CBM agrees well whereas the VBM is underestimated by 0.3 eV in
the GGA calculations
developed by Perdew and Wang,112 which means that the bandgap
reduction shown in Figure 5
is underestimated by 0.3 eV. To study the effect of doping
concentration on the carrier mobility,
Yin et al.96 studied the effect of N concentration in (Ta, N)
codopant system of TiO2 and foundthat increasing the concentration
of N leads to higher hole mobility while leaving the electron
mobility nearly intact. This is illustrated from the density of
states given in Figure 6 which
shows higher curvature at the VBM as the N concentration
increases.96
Zhang et al.113 studied the effect of Ag-La codoping on the
electronic and optical proper-ties of CaTiO3, known as perovskite.
Although it has a wide bandgap of 3.5 eV, perovskite is
an attractive material due to its low cost, ease of synthesis,
and high stability.113 Zhang et al.based their DFT calculations on
GGA functional of Perdew et al.40 The band structures ofundoped
CaTiO3, Ag-doped CaTiO3, and Ag-La codoped CaTiO3 are shown in
Figure 7.
113
The calculated bandgap of undoped CaTiO3 is 2.45 eV, which is
about 1 eV lower than the ex-
perimental value. This is due to the use of the inaccurate GGA
functional. However, qualitative
conclusions can still be drawn from these values. Adding Ag to
CaTiO3 shifts the VBM by
0.23 eV while keeping the CBM nearly unchanged leading to a
reduction in the bandgap by
0.23 eV. Further codoping with Ag and La further decreases the
bandgap by 0.55 eV. This cor-
responds to about 23% reduction in the bandgap, which is
considered a significant bandgap nar-
rowing. However, the VBM is nearly flat indicating a small hole
mobility which will in-turn
decrease the rate of water oxidation and increase the
probability of carrier recombination.
As the reduction in bandgap comes from the upward shifting of
the VBM, it is suggested
that the dopant materials introduce energy states around the
valence band. This is illustrated in
Figure 8, which shows the DOS for each element in the undoped as
well as the doped
CaTiO3.113 It is clear that for both undoped and doped CaTiO3,
the conduction band is domi-
nated by Ti-3d orbital, which is the reason why the position of
the CBM was not changed in
FIG. 6. Calculated GGA band structures for (a) pure TiO2;(b)(d)]
TiO2 coincorporated with (Ta, N) with different concen-
trations in which 3.1% O, 12.5% O, and 25% O were replaced by N,
respectively. Band offsets are taken into account in
these plots. Reprinted with permission from Yin et al., Phys.
Rev. B 82, 045106 (2010). Copyright 2010 AmericanPhysical
Society.
FIG. 7. Band structures of (a) CaTiO3, (b) CaTiO3 doped with Ag,
and (c) CaTiO3 codoped with AgLa. Reprinted with
permission from Zhang et al., J. Alloys Compd. 516, 9195 (2012).
Copyright 2012 Elsevier.
022701-15 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
the different plots of Figure 7. On the other hand, for Ag-doped
CaTiO3, Ag-5s extended above
the valence band and mix with O-2p orbital leading to a shift in
the VBM. The same argument
applies for Ag-La codoped CaTiO3 in which La-4s extends the
valence band.
Cu delafossites, CuMO2 (M group III-A and III-B elements)
represent another promisingclass of photocatalysts for hydrogen
production due to their excellent hole mobility and high
stability in aqueous solutions.114120 Nie et al.121 studied the
electronic structure of CuMO2with M from group III-A using (LDA) of
DFT. All studied compounds had indirect bandgap of
1.97 eV, 0.95 eV, and 0.41 eV for CuAlO2, CuGaO2, and CuInO2,
respectively. The direct
bandgap at C was found to decrease from 2.93 eV for CuAlO2, to
1.63 eV for CuGaO2, and to0.73 eV for CuInO2 as shown in Figure
9.
121 The LDA calculations are expected to introduce
an underestimation of 0.8 eV.121 However, even with the
compensation for this error, the trend
in the calculated bandgap does not follow the experimental
results, which show an increase in
the bandgap from 3.5 eV in CuAlO2, to 3.6 eV in CuGaO2, and to
3.9 eV in CuInO2. The reason
for this discrepancy is that the absorption near the fundamental
gap at C is negligible due tosame (even) parity of the valence and
conduction band states. This low absorption is depicted
in Figures 9(d)9(f).121
FIG. 8. DOS of (a) CaTiO3, (b) CaTiO3 doped with Ag, and (c)
CaTiO3 codoped with AgLa. Reprinted with permissions
from Zhang et al., J. Alloys Compd. 516, 9195 (2012). Copyright
2012 Elsevier.
FIG. 9. (a) to (c) The calculated LDA band structures for
CuAlO2, CuGaO2, and CuInO2, respectively. Energy zero is at
the highest valence band at F. The VBMs appeared off F are
marked by the black circles. (d) to (f) The corresponding tran-
sition matrix elements between the band edge states. Reprinted
with permissions from Nie et al., Phys. Rev. Lett. 88,066405
(2002). Copyright 2002 American Physical Society.
022701-16 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
A similar observation was reported by Huda et al.105 for group
III-B delafossites in whichthe transition probability at the C
point for CuLaO2 is nearly zero (Figure 10(a)) due to thesame
parity violation observed for group III-A delafossites. The reason
for this parity violation
stems from the inversion symmetry of their crystal structure.120
To break this symmetry, Huda
and coworkers coincorporated group III-A and group III-B
delafossites105 Figure 10(a) shows
that the probability of absorption at the C point is no longer
zero when La is mixed with Gaand Figure 10(b) illustrates the
enhancement in the optical absorption which is around 1 eV.104
The calculations were carried out with GGA of the DFT and hence
might suffer from underesti-
mation of errors. However, it gives a qualitative insight about
the effect of mixing group III-A
with group III-B delafossites. Furthermore, Huda et al. tried
substitutional alloying of Mn andCr with Y in CuYO2; however, the
material suffered from a large electron effective mass as
well as unwanted defect levels due to the multiple ionizations
of these transition metals.105
Although mixing group III-A with group III-B delafossites
provided local symmetry break-
ing, it failed to provide global symmetry breaking.122 To
effectively break the symmetry, an
element with large size as well as chemical potential mismatch
should be incorporated.122 Huda
et al.122 proposed Bi to be incorporated in group III-B
delafossites. Group III-B delafossites arepreferred to group III-A
delafossites because they are direct bandgap materials113 and
hence
will provide better absorption. The rationale behind choosing Bi
was based on: (i) Bi is isova-
lent to group III elements and so it will not create any
unwanted recombination centers, (ii) Bi
has large size and potential mismatch with group III elements,
and (iii) Bi is expected to reduce
the hole effective mass by introducing delocalized states at the
valence band maxima due to Bi
6 s lone pair electron.123125 Figure 11 (Ref. 122) shows the
partial density of states for differ-
ent Bi-alloyed group III-B delafossites, namely, Cu(Sc, Bi)O2,
Cu(Y, Bi)O2, and Cu(La, Bi)O2.
It is clear that the conduction band minimum is dominated by
Bi-p states, which is the reason
for reducing the bandgap. For CuScO2, the fundamental bandgap is
reduced from 3.05 eV to
FIG. 10. (a) Transition probability for CuLaO2 (blue) and
Cu(La,Ga)O2 (red) at different symmetry points to show the
effect of mixing group III-A and group III-B on the transition
probability of delafossites. (b) Calculated optical absorption
coefficients for Cu(La,Ga)O2. The arrow in the x-axis shows the
band gap for pristine CuLaO2. The reduction of optical
band gap due to this isovalent alloying is clear from the
figure. Reprinted with permissions from Huda et al., Sol.
HydrogenNanotechnol. V, 77700F (2010). Copyright 2010 SPIE.
FIG. 11. Partial density of state (p-DOS) of Bi alloyed (a)
CuScO2, (b) CuYO2, and (c) CuLaO2. Reprinted with permission
from Huda et al., J. Appl. Phys. 109, 113710 (2011). Copyright
2011 American Institute of Physics.
022701-17 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
1.44 eV after doping with Bi whereas for CuYO2, the bandgap is
reduced from 2.95 to
1.24 eV.122 These results would have been a real breakthrough if
the material could sustain its
direct bandgap property. However, incorporation of Bi changes
group III-B delafossites to indi-
rect bandgap materials.122 The minimum calculated direct bandgap
is 2.543 eV for Cu(Sc,
Bi)O2 and 2.348 eV for Cu(Y, Bi)O2 which is still large for
efficient visible light absorption.122
The band structures of pristine delafossites as well as the
alloyed ones are shown in Figure
12.121 Another observation that can be drawn from the figure,
besides the reduction in the
bandgap, is the carriers effective masses. It is clear that the
electrons as well as the holes
effective masses are reduced, especially for Cu(Y, Bi)O2. The
data given in Figures 11 and 12
represent high Bi concentration in which the ratio between Bi
and group III-B element is 1:1.
To study the effect of concentration, low Bi concentration is
considered for Cu(Y, Bi)O2 in
which Bi:Y is 11:1. For this low concentration, the reduction in
the bandgap was negligible
because the width of Bi-p, which dominates the CBM, is very
narrow compared to the high
concentration case.122 To decrease the bandgap at this low Bi
concentration, Ga has been co-
incorporated with Bi since Ga was found to decrease the bandgap
of CuYO2.104 However, the
bandgap remained indirect with the minimum direct bandgap being
around 2.73 eV.122 The
reduction in the bandgap was not significant since the Ga-s
states are at higher energy than that
of Bi-p.122 The calculations were done using GGA functional with
U-correction125,126 to correct
for the underestimation in the calculated energy.
MCu2O2, where M is a group II-A element, is another Cu-based
promising candidate for
solar water splitting. Although they have been used as
transparent metal oxides, they can also
be used as photocatalysts owing to their direct bandgap. Nie et
al.127 studied these materialsand found that the bandgap increases
with increasing the size of incorporated group II-A ele-
ment, which suggests using small-size elements, namely Mg or Ca.
The calculated bandgap for
CaCu2O2 and MgCu2O2 are 3.01 eV and 2.45 eV,128 respectively,
that is comparable to the fun-
damental direct bandgap of Bi-doped group III-B delafossites.122
Figure 13 shows the bandgap
structure for the different MCu2O2 materials.128 The
calculations were based on LDA and as a
result the calculated conduction bands were shifted by 1.5 eV to
account for the underestimation
of LDA. Nie and coworkers showed that the contribution to the
CBM comes mainly from the
group II-A element as seen in Figure 14.128 Incorporation of Bi
might further reduce the
bandgap as it was the case for group III-B delafossites.122
However, the position of Bi-p states
relative to the states of group II-A element should be
calculated to determine whether Bi would
introduce new states near the CBM. Also, the effect of Bi on
changing the material into an
indirect bandgap still needs to be investigated.
The strong effect that Bi has on the host materials has
intrigued researchers to think of Bi-
based compounds as efficient photocatalysts. BiVO4 is one of the
recent compounds that were
studied intensively as a photocatalyst129133 owing to its direct
bandgap of 2.4-2.5 eV (Refs.
125, 134, and 135) good optical properties, its suitable bandgap
alignment for H2 and O2
FIG. 12. Band structures of (a) CuScO2, (b) Cu(Sc,Bi)O2, (c)
CuYO2, and (d) Cu(Y,Bi)O2. Reprinted with permission from
Huda et al., J. Appl. Phys. 109, 113710 (2011). Copyright 2011
American Institute of Physics.
022701-18 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
evolution131,134 and its capability to be doped with a wide
range of dopants with 10 atomic per-
cent or more,136,137 which allows for its bandgap engineering.
Liu et al.138 studied the photoca-talytic properties of
Bi0.5M0.5VO4 (MLa, Eu, Sm, and Y). The bandgap of BiVO4 wasreduced
from 2.82 eV in the undoped case to 2.58 eV in Bi0.5Y0.5VO4 (see
Figure 15).
138
However, co-doping of Bi and Y resulted in an indirect bandgap.
The fundamental direct
bandgap for Bi0.5Y0.5VO4 was 2.90 eV.138 These results were
higher than the experimental val-
ues of 2.42.5 eV for pristine BiVO4 because the studied crystal
is of zircon type whereas that
discussed in experiments has a monoclinic crystal structure. The
presence of Y does not affect
the bandgap significantly because the VBM is dominated by O-2p
whereas CBM is dominated
by V-3d leaving Bi and Y with little contribution as shown in
Figure 15. To determine the ori-
gin of photocatalytic activity, Liu and co-workers considered
different VO4 compounds,
namely, Bi0.5Y0.5VO4, La0.5Y0.5VO4, and Ce0.5Y0.5VO4 and found
that only Bi0.5Y0.5VO4showed photocatalytic activity concluding
that Bi ion and not VO4 was responsible for enhanc-
ing the photocatalytic response.138 This is mainly due to
reducing the carriers effective masses
as reported by Huda et al. in their study of Bi-doped group
III-B delafossites.121
Yin et al.139 studied monoclinic BiVO4 (which is more active and
hence a better photocata-lyst than zircon BiVO4
128) and calculated the bandgap to be ca. 2.06 eV which is
underesti-
mated by about 0.44 eV owing to the use of GGA functional.
Figure 16(a) illustrates the calcu-
lated band structure whereas Figure 16(b) shows the density of
states.139 In agreement with Liu
et al.,138 the VBM is dominated by O-2p states whereas V-3d
dominates the CBM.To optimize the electron and hole conductivities,
Yin et al. considered doping BiVO4 with
different group I-A, group II-A, group II-B, group IV-B, and
group VI-B elements.139 They
also considered substituting O with C or N or F. Table V lists
the electron ionization energies
of different donors, whereas the transition energy levels of the
different acceptors are listed in
Table VI.139 (Xvac) denotes an intrinsic defect occurring due to
a vacancy position of X, (Xy)
denotes a substitutional defect due to replacing X with Y, and
(Xint) is an interstitial defect of
element X. Tables IV and V show that intrinsic defects (Ovac,
VBi, Vint, Bivac, Vvac, BiV, Oint)
FIG. 13. Calculated semi relativistic electronic band structure
for (a) MgCu2O2, (b) CaCu2O2, (c) SrCu2O2, and (d)
BaCu2O2. Energy zero is at VBM. The energy of the conduction
bands are shifted upwards by 1.5 eV to correct the LDA
band gap error. Reprinted with permission from Nie et al., Phys.
Rev. B 65, 075111 (2002). Copyright 2002 AmericanPhysical
Society.
022701-19 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
can only provide moderate n- and p- conductivity, respectively,
if compared to the extrinsic
defects.139
Interstitial Li, Na, and K have very low ionization energy.
However, due to the compensa-
tion between intrinsic and extrinsic defects, the Fermi level is
pinned at 0.6, 1.2, and 0.9 eV
below the CBM for Li, Na, and K, respectively, which makes them
poor electron conductors.
The same effect is noticed for group II-A and group II-B
elements. For group IV-B, it is
FIG. 14. Calculated total and local density of states (DOS) for
SrCu2O2. Reprinted with permission from Nie et al., Phys.Rev. B 65,
075111 (2002). Copyright 2002 American Physical Society.
FIG. 15. DOS plots of zircon BiVO, YVO, and BYV solid solution.
Reprinted with permission from J. Solid State Chem.
186, 7075 (2012), Copyright 2012 Elsevier.
022701-20 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
relatively small with only 0.2 eV pinning of Fermi level below
CBM for Zr and Hf. Group
VI-B represents the best n-type conductors, especially MoV whose
Fermi level is limited by its
ionization energy and does not suffer donor-acceptor
compensation.139
For p-type conductivity, MgBi and ZnBi have the lowest
transition energy of 0.02 eV.
However, they have high formation energy (1 eV), which makes
CaBi and SrBi more attractive
as they have lower formation energy and their Fermi level
position is determined by the
FIG. 16. (a) Band structure of monoclinic BiVO4, (b) Partial
density of states of bulk monoclinic BiVO4. Reprinted with
permissions from Yin et al., Phys. Rev. B 83, (2011), 155102.
Copyright 2011 American Physical Society.
TABLE V. Electron ionization energies of donors in intrinsic and
different dopings of BiVO4. Reprinted with permission
from Yin et al., Phys. Rev. B 83, 155102 (2011). Copyright 2011
American Physical Society.
Ei(0/1) Ei(0/2) Ei(0/3) Ei(0/4) Ei(0/5) Ei(0/6)
Ovac 0.22 0.17VBi 0.52 0.68
Vint 0.31 0.46 0.56 0.76 0.94
Liint 0.02
Naint 0.03
Kint 0.04
Mgint 0.06 0.12
Caint 0.04 0.11
Srint 0.04 0.10
Znint 0.08 0.15
TiBi 0.20
Tiint 0.20 0.27 0.39 0.50
ZrBi 0.04
Hfint 0.05 0.18 0.30 0.40
CrBi 1.08 0.85 1.16
CrV 0.41
Crint 0.70 0.77 1.03 1.08 1.30 1.40
MoBi 0.21 0.42 0.57
MoV 0.04
Moint 0.67 0.74 0.90 0.95 1.14 1.22
WBi 0.09 0.16 0.23
WV 0.01
Wint 0.50 0.55 0.68 0.72 0.88 0.93
022701-21 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
transition energy (0.1 eV above VBM). Also, NaBi and KBi show
similar values of 0.1 and0.2 eV above the VBM, respectively. For
substitutional alloying of O, the results did not give
good n- or p-conductivity due to strong compensation from
intrinsic defects.139
Although Table VI suggests that WV would provide shallower donor
level than MoV and
hence provide better electron conduction, Yin et al. suggested
that incorporation of Mo wouldgive the most efficient electron
conduction due to donor-acceptor compensation, which pins the
Fermi level at 0.2 eV below the CBM.138 However, Park et al.140
proved experimentally thatincorporation of W gave higher
photocurrent than Mo. Incorporation of W increases the photo-
current by 6 times than that of pristine BiVO4 whereas
incorporation of Mo increases the cur-
rent by only 3.5 times.139 This is because the difference in
ionization energy between W and
Mo is about 0.02 eV,140 which is in close agreement with the
0.03 eV shown in Table V.140
Consequently, W can give electrons more efficiently to the host
material than Mo. There seems
to be a controversy between the experimental results of Park et
al. and the theoretical calcula-tions of Yin et al., which requires
more work to resolve. Coincorporation of W and Mo leadsto a better
improvement compared to incorporation of W alone as shown in Figure
17(a).139
This is due to increasing the carrier density which in turn
increases the electric field in the
depletion region and hence provides better separation of charge
carriers.140 The charge density
for W/Mo-doped BiVO4 is nearly twice that of W-doped BiVO4 as
depicted from the slope of
the Mott-Schottky curve in Figures 17(b) and 17(c).140 DFT
calculations show that the shape of
the band structure as well as the bandgap is nearly equal for
pristine BiVO4, W-doped BiVO4,
and Mo-doped BiVO4 asserting that the improvement of the
photocatalytic activity is due to
better charge carrier separation (see Figure 18).140
V. CONCLUSION
DFT represents a very effective means that can provide a
systematic approach towards the
design of efficient solar energy conversion systems such as
sensitizers for dye-sensitized solar
TABLE VI. Transition energy levels of acceptors in intrinsic and
different dopings of BiVO4. Reprinted with permission
from Yin et al., Phys. Rev. B 83, 155102 (2011) Copyright 2011
American Physical Society.
ei(0/-1) ei(0/-2) ei(0/-3) ei(0/-4) ei(0/-5)
Bivac 0.14 0.16 0.18
Vvac 0.23 0.25 0.27 0.30 0.26
BiV 1.23 1.15
Oint 0.74 0.49
LiBi 0.09 0.11
LiV 0.22 0.24 0.27 0.29
NaBi 0.08 0.11
NaV 0.23 0.25 0.28 0.30
KBi 0.16 0.18
KV 0.22 0.24 0.27 0.28
MgBi 0.02
MgV 0.20 0.22 0.25
CaBi 0.08
CaV 0.21 0.24 0.25
SrBi 0.09
SrV 0.21 0.25 0.26
ZnBi 0.02
ZnV 0.21 0.24 0.26
TiV 0.15
ZrV 0.17
HfV 0.16
022701-22 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
cells and photoanode materials for water splitting. DFT is
preferred over ab-initio methods dueto its considerably reduced
computational cost as it is based on calculating the electron
density
instead of the wavefunction. Different functionals have been
reported to investigate the best
candidate dyes for use in dye-sensitized solar cells. LDA- and
GGA-based functionals tend to
underestimate the excitation energy. Implementing hybrid
functionals in TD-DFT give good
results for small dyes but tend to underestimate the results for
large ones. LC-based functionals
incorporated in TD-DFT are the best candidates for larger dyes
allowing one to reproduce the
electronic absorption spectra of organic dyes. Using TD-DFT, the
LC formalism can be used to
correctly predict any increase in the excited-state electric
dipole moment of the dyes. As a
result, the LC/TD-DFT formalism will help provide the insight
needed to guide the design of
efficient solar cell dyes. In particular, one promising approach
will be to investigate light-
harvesting organic sensitizers, which selectively absorb photons
with specific wavelengths. To
this end, the understanding of critical features such as
excitation energies and charge-transfer
states using the LC/TD-DFT technique can provide a step towards
this goal
For solar-driven hydrogen production, metal oxides are the most
promising candidates for
robust photoanodes. However, metal oxides should be modified to
satisfy the requirements for
efficient photoanodes. DFT is a very effective means to screen
materials and indentify the most
efficient mixed metal oxides. Hybrid functionals seem to give
very accurate results for such
systems, reducing the discrepancy between theoretical
calculations and experiments to only 5%,
provided the right functional is chosen. In order to be able to
select the right functional for the
system, one should look closely at the expression of the
exchange-correlation energy of the
functional and have a rough understanding about the correlation
between the electrons in the
system under study. For example, for systems like Ta2O5,
functionals that implement the exact
FIG. 17. (a) Linear sweep voltammograms of undoped BiVO4 (blue),
W-doped BiVO4 (red), and W/Mo-doped BiVO4
(black) with chopped light under visible irradiation in the 0.1M
Na2SO4 aqueous solution (pH 7, 0.2M sodium phosphatebuffered). Beam
intensity was about 120 mW cm2 from a full xenon lamp, and the scan
rate was 20mV s1; (b) Mott-Schottky plots of W-doped BiVO4, (c)
Mott-Schottky plots of W/Mo-doped BiVO4. AC amplitude of 10mV was
applied
for each potential, and three different AC frequencies were used
for the measurements: 1000Hz (blue), 500Hz (red), and
200Hz (black). Tangent lines of the M-S plots are drawn to
obtain the flat band potential. Adapted from Ref. 140.
022701-23 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
Hartree-Fock on longer range are more desirable since the Ta
atoms are quite large and so they
are expected to exert a long range force on the nearby nuclei
and electrons. For this reason,
PBE0 gives a better approximation to the bandgap of Ta2O5 than
HSE06. In fact, this is the
main source of discrepancy that is usually found between
theoretical and experimental data.
This opens the vista for further work to resolve these
controversies. In this regard, DFT can be
considered as a fast computational screening method, with
respect to stability and bandgap, to
discover new light harvesting materials for water splitting.
FIG. 18. Density of states projected onto the Bi 6s (red), Bi 6p
(pink), O 2p (blue), and V 3d (green) states for: (a) pristine
BiVO4, (b) W-doped BiVO4, (c) Mo-doped BiVO4. Adapted from Ref.
140.
022701-24 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
ACKNOWLEDGMENTS
This research was partially funded by The American University in
Cairo, Zewail City of
Science and Technology, the STDF, Intel, Mentor Graphics,
andMCIT.
1P. D. Wurfel, Physics of Solar Cells: From Principles to New
Concepts (Wiley-VCH, Weinheim, 2005)2S. J. Hassol, Arctic Climate
Impact Assessment, Arctic Monitoring and Assessment Programme,
Program for theConservation of Arctic Flora and Fauna, and
International Arctic Science Committee, Impacts of a Warming
Arctic:Arctic Climate Impact Assessment (Cambridge University
Press, Cambridge, UK, New York, 2004).3N. S. Lewis, Mrs Bull.
32(10), 808820 (2007).4L. Y. Han, A. Islam, H. Chen, C. Malapaka,
B. Chiranjeevi, S. F. Zhang, X. D. Yang, and M. Yanagida,
EnergyEnviron. Sci. 5(3), 60576060 (2012).5J. I. Basham, G. K. Mor,
and C. A. Grimes, ACS Nano 4(3), 12531258 (2010).6G. K. Mor, J.
Basham, M. Paulose, S. Kim, O. K. Varghese, A. Vaish, S. Yoriya,
and C. A. Grimes, Nano Lett. 10(7),23872394 (2010).7I. P. Jain,
Int. J. Hydrogen Energy 34(17), 73687378 (2009).8G. K. Mor, H. E.
Prakasam, O. K. Varghese, K. Shankar, and C. A. Grimes, Nano Lett.
7(8), 23562364 (2007).9G. K. Mor, O. K. Varghese, R. H. T. Wilke,
S. Sharma, K. Shankar, T. J. Latempa, K. S. Choi, and C. A. Grimes,
NanoLett. 8(10), 3555 (2008).
10M. Fernandez-Garcia, A. Martinez-Arias, A. Fuerte, and J. C.
Conesa, J. Phys. Chem. B 109(13), 60756083 (2005).11N. K. Allam, F.
Alamgir, and M. A. El-Sayed, ACS Nano 4(10), 58195826 (2010).12N.
K. Allam, A. J. Poncheri, and M. A. El-Sayed, ACS Nano 5(6),
50565066 (2011).13M. Woodhouse and B. A. Parkinson, Chem. Mater.
20(7), 24952502 (2008).14E. Thimsen, S. Biswas, C. S. Lo, and P.
Biswas, J. Phys. Chem. C 113(5), 20142021 (2009).15H. Ikehata, N.
Nagasako, S. Kuramoto, and T. Saito, MRS Bull. 31(9), 688692
(2006).16C. Wolverton, X. Y. Yan, R. Vijayaraghavan, and V.
Ozolins, Acta Mater. 50(9), 21872197 (2002).17V. Vaithyanathan, C.
Wolverton, and L. Q. Chen, Acta Mater. 52(10), 29732987 (2004).18V.
Tripkovic, E. Skulason, S. Siahrostami, J. K. Norskov, and J.
Rossmeisl, Electrochim. Acta 55(27), 79757981(2010).
19T. Burkert, L. Nordstrom, O. Eriksson, and O. Heinonen, Phys.
Rev. Lett. 93(2), 027203 (2004).20N. Marzari, MRS Bull. 31(9),
681687 (2006).21J. Kohanoff and N. I. Gidopoulos, Density
functional theory: Basics, new trends and applications, in Handbook
ofMolecular Physics and Quantum Chemistry, edited by S. Wilson
(John Wiley and Sons, Chichester, 2003), pp. 532568.
22J. M. Ziman, Electrons and Phonons: The Theory of Transport
Phenomena in Solids (Clarendon, Oxford, 1960).23D. R. Hartree, Math
Proc. Cambridge Philos Soc. 24(01), 89110 (1928).24V. Fock, Z.
Physik 61(12), 126148 (1930).25J. C. Slater, Phys. Rev. 35(2),
02100211 (1930).26P. Hohenberg and W. Kohn, Phys. Rev. 136(3B),
B864B871 (1964).27L. H. Thomas, Proc. Cambridge Philos. Soc. 23,
542548 (1927).28E. Fermi, Z. Physik 48 (12), 7379 (1928).29P. A. M.
Dirac, Math. Proc. Cambridge Philos. Soc. 26, 376385 (1930).30E. H.
Lieb, Rev. Mod. Phys. 53 (4), 603641 (1981).31C. F. v. Weizsacker,
Z. Physik 96 (78), 431458 (1935).32J. L. Gazquez and E. V. Ludena,
Chem. Phys. Lett. 83(1), 145148 (1981).33W. Kohn and L. J. Sham,
Phys. Rev. 140(4A), A1133A1138 (1965).34K. Capelle, Braz. J. Phys.
36(4A), 13181343 (2006).35J. C. Slater, Phys. Rev. 36(1), 00570064
(1930).36P. M. W. Gill, Adv. Quantum. Chem. 25, 141205 (1994).37E.
Clementi, IBM J. Res. Dev. 9(1), 2 (1965).38W. Koch and M. C.
Holthausen, A Chemists Guide to Density Functional Theory
(Wiley-VCH, Weinheim, Chichester,2000).
39D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45(7), 566569
(1980).40J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett.
77(18), 38653868 (1996).41A. D. Becke, Phys. Rev. A 38(6), 30983100
(1988).42C. T. Lee, W. T. Yang, and R. G. Parr, Phys. Rev. B 37(2),
785789 (1988).43J. P. Perdew, S. Kurth, A. Zupan, and P. Blaha,
Phys. Rev. Lett. 82(12), 25442547 (1999).44A. D. Becke, J. Chem.
Phys. 104(3), 10401046 (1996).45J. M. Tao, J. P. Perdew, V. N.
Staroverov, and G. E. Scuseria, Phys. Rev. Lett. 91(14), 146401
(2003).46O. Gunnarsson and B. I. Lundqvist, Phys. Rev. B 13(10),
42744298 (1976).47A. D. Becke, J. Chem. Phys. 98(7), 56485652
(1993).48M. A. Rohrdanz and J. M. Herbert, J. Chem. Phys. 129(3),
034107 (2008).49H. Iikura, T. Tsuneda, T. Yanai, and K. Hirao, J.
Chem. Phys. 115(8), 35403544 (2001).50B. M. Wong and J. G. Cordaro,
J. Chem. Phys. 129(21), 214703 (2008).51E. Runge and E. K. U.
Gross, Phys. Rev. Lett. 52(12), 9971000 (1984).52D. P. Hagberg, T.
Marinado, K. M. Karlsson, K. Nonomura, P. Qin, G. Boschloo, T.
Brinck, A. Hagfeldt, and L. Sun,J. Org. Chem. 72(25), 95509556
(2007).
53S. Kurth, J. P. Perdew, and P. Blaha, Int. J. Quantum Chem.
75(45), 889909 (1999).54K. Tennakone, G. R. R. A. Kumara, I. R. M.
Kottegoda, and V. P. S. Perera, Chem. Commun. 1999(1), 1516.55S.
Ito, S. M. Zakeeruddin, R. Humphry-Baker, P. Liska, R. Charvet, P.
Comte, M. K. Nazeeruddin, P. Pechy, M. Takata,H. Miura, S. Uchida,
and M. Gratzel, Adv. Mater. 18(9), 12021205 (2006).
022701-25 Nashed, Ismail, and Allam J. Renewable Sustainable
Energy 5, 022701 (2013)
Downloaded 20 Jun 2013 to 171.67.34.205. This article is
copyrighted as indicated in the abstract. Reuse of AIP content is
subject to the terms at:
http://jrse.aip.org/about/rights_and_permissions
-
56Y. Chiba, A. Islam, Y. Watanabe, R. Komiya, N. Koide, and L.
Y. Han, Jpn. J. Appl. Phys., Part 2 45(2428),L638L640 (2006).
57K. S. Kim, H. Song, S. H. Nam, S. M. Kim, H. Jeong, W. B. Kim,
and G. Y. Jung, Adv. Mater. 24(6), 792 (2012).58M. Wang, P. Chen,
R. Humphry-Baker, S. M. Zakeeruddin, and M. Gratzel, Chemphyschem
10(1), 290299 (2009).59I. K. Ding, J. Melas-Kyriazi, N. L.
Cevey-Ha, K. G. Chittibabu, S. M. Zakeeruddin, M. Gratzel, and M.
D. McGehee,Org. Electron. 11(7), 12171222 (2010).
60H. N. Tian and L. C. Sun, J. Mater. Chem. 21(29), 1059210601
(2011).61Z. Yu, H. N. Tian, E. Gabrielsson, G. Boschloo, M. Gorlov,
L. C. Sun, and L. Kloo, RSC Adv. 2(3), 10831087 (2012).62D. Song,
M. S. Kang, Y. G. Lee, W. Cho, J. H. Lee, T. Son, K. J. Lee, S.
Nagarajan, P. Sudhagar, J. H. Yum, and Y. S.Kang, Phys. Chem. Chem.
Phys. 14(2), 469472 (2012).
63Y. Hao, X. C. Yang, J. Y. Cong, A. Hagfeldt, and L. C. Sun,
Tetrahedron 68(2), 552558 (2012).64S. Y. Qu, B. Wang, F