Article

pubs.acs.org/accounts

Oxygen Transport in Perovskite-Type Solid Oxide Fuel Cell Materials:

Insights from Quantum Mechanics
Ana B. Muñoz-García,† Andrew M. Ritzmann,‡ Michele Pavone,† John A. Keith,§ and Emily A. Carter*,∥
†
Department of Chemical Sciences, University of Naples Federico II, Naples 80126, Italy
‡
Department of Chemical and Biological Engineering, Princeton University, Princeton, New Jersey 08544-5263, United States
§
Department of Chemical and Petroleum Engineering, University of Pittsburgh, Pittsburgh, Pennsylvania 15261, United States
∥
Deparment of Mechanical and Aerospace Engineering, Program in Applied and Computational Mathematics, and Andlinger Center
for Energy and the Environment, Princeton University, Princeton, New Jersey 08544-5263, United States

CONSPECTUS: Global advances in industrialization are precipitating increasingly

rapid consumption of fossil fuel resources and heightened levels of atmospheric CO2.
World sustainability requires viable sources of renewable energy and its efficient use.
First-principles quantum mechanics (QM) studies can help guide developments in
energy technologies by characterizing complex material properties and predicting
reaction mechanisms at the atomic scale. QM can provide unbiased, qualitative
guidelines for experimentally tailoring materials for energy applications.
This Account primarily reviews our recent QM studies of electrode materials for solid
oxide fuel cells (SOFCs), a promising technology for clean, efficient power generation.
SOFCs presently must operate at very high temperatures to allow transport of oxygen
ions and electrons through solid-state electrolytes and electrodes. High temperatures,
however, engender slow startup times and accelerate material degradation. SOFC
technologies need cathode and anode materials that function well at lower temperatures,
which have been realized with mixed ion−electron conductor (MIEC) materials.
Unfortunately, the complexity of MIECs has inhibited the rational tailoring of improved SOFC materials. Here, we gather theoretically
obtained insights into oxygen ion conductivity in two classes of perovskite-type materials for SOFC applications: the conventional
La1−xSrxMO3 family (M = Cr, Mn, Fe, Co) and the new, promising class of Sr2Fe2−xMoxO6 materials. Using density functional theory
+ U (DFT+U) with U−J values obtained from ab initio theory, we have characterized the accompanying electronic structures for the
two processes that govern ionic diffusion in these materials: (i) oxygen vacancy formation and (ii) vacancy-mediated oxygen migration.
We show how the corresponding macroscopic oxygen diffusion coefficient can be accurately obtained in terms of microscopic
quantities calculated with first-principles QM.
We find that the oxygen vacancy formation energy is a robust descriptor for evaluating oxide ion transport properties. We also find it
has a direct relationship with (i) the transition metal−oxygen bond strength and (ii) the extent to which electrons left behind by the
departing oxygen delocalize onto the oxygen sublattice. Design principles from our QM results may guide further development of
perovskite-based MIEC materials for SOFC applications.

1. INTRODUCTION Mixed ion-electron conductor (MIEC) cathode materials permit

Fossil-fuel-driven industrialization hastens a daunting two-sided lower Top by increasing the active area for ORR via facile bulk ion
problem: nonrenewable natural resources are depleted while transport (Figure 1). MIEC cathodes therefore can provide lower
atmospheric CO2 concentrations alarmingly increase. Efficient, overpotentials and improved ORR kinetics2 in intermediate
renewable, and CO2-neutral energy technologies are needed. temperature (IT-)SOFCs (Top ≈ 600−800 °C). Although MIEC
Solid oxide fuel cells (SOFCs) can potentially deliver clean and cathode materials are promising, further improvements are
highly efficient power via electrochemistry. SOFC electrolytes, needed before SOFCs would be deployed widely.
anodes, and cathodes can be manufactured from low-cost Understanding oxygen ion mobility is critical to improving
ceramic materials. However, cathode and anode materials carry MIEC materials.3−5 Common MIECs are based on perovskite-
high intrinsic activation energies for ionic conductivity. That type transition metal (TM) oxides with the stoichiometric formula
inevitably leads to high operating temperatures (Top = 800− AMO3 (A = La and/or Sr; M = Mn, Fe, Co, and/or Cr). Oxygen
1000 °C) that cause deleterious thermal stresses, high start-up ion transport through these lattices occurs via vacancy-mediated
energy inputs, and short cell lifetimes.1 Simply reducing SOFC
operation temperatures would curb oxygen diffusion and confine
Special Issue: DFT Elucidation of Materials Properties
the oxygen reduction reaction (ORR) to occur only at the triple
phase boundary between air, the cathode, and the electrolyte; Received: December 26, 2013
therefore, lower temperatures engender higher overpotentials.

© XXXX American Chemical Society A dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

2. DFT+U FOR PHYSICALLY CORRECT DESCRIPTIONS

OF STRONGLY CORRELATED MATERIALS
KS-DFT is widely used in first-principles QM studies of
extended materials because of its relative accuracy and
computational efficiency. Exchange-correlation (XC) functionals
utilizing local (LDA) and semilocal (GGA) approximations often
sufficiently predict many different material properties. However,
the inexact exchange energy approximation in these XC
functionals introduces self-interaction errors (SIEs) that can
result in overly delocalized electron densities and other
unphysical behaviors. SIEs become most significant when
multiple open-shell electrons reside within tightly contracted
orbitals, such as in d or f electrons in some TM oxides.9
Out of several SIE correcting approaches,10 two are
summarized here. DFT+U theory applies an averaged intra-
atomic electron−electron potential to the electrons in the
subset of atomic orbitals that are highly susceptible to SIE,
providing a more physically correct description of Coulomb
(U) and exchange (J) interactions within this subset.11 This
Figure 1. Two possible oxygen incorporation paths: surface path computationally efficient treatment uses a single U−J parameter,
showing the triple-phase boundary (TPB) and bulk path highlight- but deriving this parameter from ad hoc experimental data
ing in red the increased active area for the oxygen reduction introduces empiricism that impairs transferability for other
reaction. predictions. Obtaining this parameter from first-principles theory
is possible but not trivial. Alternatively, hybrid DFT adds to all
electrons (rather than just a subset) a partial amount of exact,
“hopping” mechanisms. The oxide ion diffusion coefficient, D,
nonlocal exchange energy from Hartree−Fock theory to offset
has an Arrhenius form: D = A·exp[−Q/(kBT)], where the
overall activation energy Q comprises two terms: (1) the the SIE.12 While accounting for SIEs from all electrons, hybrid
oxygen vacancy formation free energy and (2) the activation DFT solved with plane-wave basis sets and periodic boundary
enthalpy to transport an O2− to its nearest vacant site. conditions is computationally expensive due to the nonlocal
Consequently, high bulk conductivity requires low oxygen exchange integral evaluations in real space. Since DFT+U and
vacancy formation energies and low migration barrier heights. hybrid DFT often provide qualitatively similar results, we employ
Experimental measurements can determine D for existing DFT+U whenever it has been confirmed to describe the correct
materials, but they usually cannot reveal design principles that physics of the material. We wish to predict properties of materials
improve D. Thus, materials optimization usually proceeds by with oxygen vacancies, so we use the GGA functional of Perdew,
trial and error, requiring expensive, time-consuming syntheses. Burke, and Ernzerhof (PBE),13 which better describes bond
Rational materials optimization can benefit from fundamental, energies and reaction energetics (such as the loss of O2) than
atomic-scale insights that explain how and why structural and LDA.14
electronic factors affect oxygen bulk conductivity. Using theory We derive U−J parameters using the ab initio scheme of
to understand oxide ion transport in SOFCs helps direct Mosey, Liao, and Carter.15,16 Here, a unique U−J value for each
experimentalists toward promising new materials while also TM element is obtained from unrestricted Hartree−Fock
creating a feedback loop to improve theory. This Account (UHF) calculations on electrostatically embedded clusters. This
summarizes our recent computational efforts to provide these unbiased approach requires no experimental data apart from
insights. crystal and magnetic structures, allowing DFT+U calculations
Some oxides (including the perovskites discussed herein) are to be maximally independent of experiment with transferability
strongly correlated materials, since highly localized electrons at in predictions. We furthermore contend that this approach is
the TM sites govern their material properties. Standard Kohn− closer to the physical meaning of the U and J parameters than
Sham density functional theory6 (KS-DFT) often fails to first-principles approaches that derive these from DFT
accurately predict physical properties of these materials. We calculations using linear-response theory.17 We previously
outline procedures to overcome these failures. We then showed that this ab initio DFT+U scheme succeeds for iron,
summarize insights from explicitly modeling oxygen vacancy chromium, nickel, and manganese oxides,16,18,19 materials that
formation and oxide ion diffusion in two classes of materials: standard DFT describes poorly. Very recently, a comparison of
the conventional Sr-doped LaMO3 (M = Cr, Mn, Fe, Co) the linear-response versus UHF approaches revealed serious
perovskites and the new, promising Sr2Fe2−xMoxO6 mater- limitations for the former but not the latter.20
ials. The first class represents the state-of-the-art for both Indeed, conventional DFT also incorrectly predicts electronic
applications in high-temperature SOFCs and MIEC-based and magnetic features of LaMO3 perovskites. For example,
IT-SOFCs.7 The second class with x = 0.5, Sr2Fe1.5Mo0.5O6 LDA and GGA predict LaMnO3 to have a ferromagnetic
(SFMO0.5), has recently demonstrated outstanding electro- metallic ground state, while in reality it is an antiferromagnetic
catalytic performance in symmetric SOFCs.8 We conclude by (AFM) insulator.21 Instead, properties from ab initio-derived
rationalizing experimentally observed material properties DFT+U theory of LaFeO3, LaCoO3, and Sr-doped LaMO3
through their electronic, magnetic, and structural features, qualitatively agree with measurements. Likewise, the SIE in
showing how quantum mechanics (QM) may help guide next- PBE causes an incorrect description of the parent material
generation IT-SOFC development. Sr2Fe2−xMoxO6 with x = 1 (SFMO1.0), which is predicted to be
B dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

a nonmagnetic semiconductor. By contrast, the experimentally

observed half-metallic AFM character of SFMO1.022 is correctly
described by PBE+U (Figure 2).

Figure 3. (top) Oxygen vacancy formation energies, ΔEf,vac, for

LaMO3 (M = Cr, Mn, Fe, Co). PBE+U (solid line) and experimental
(dashed line) values are shown. Adapted from ref 25. (bottom)
Magnetic moment variations of the nearby M ion upon formation of
an oxygen vacancy. Schematic of the d orbital occupation of M ions
before (M3+, black half-arrows) and after (M2+, extra red half-arrow
representing the electron added upon reduction) formation of an
oxygen vacancy.
Figure 2. Projected densities of states (PDOS) on Fe d, Mo d, and O
p orbitals of SFMO1.0 predicted by PBE (top panel) and PBE+U consistent with experimental data from thermogravimetric
(bottom panel). Positive/negative PDOS refer to α-/β-spin electrons. analyses. Calculated ΔEf,vac values do not monotonically
Fermi energies set to be zero. Adapted from ref 26.
decrease with increasing atomic number. This trend matches
experimental bulk oxide ion diffusion rates for these four
3. OXYGEN VACANCY FORMATION ENERGY materials as well.33 Notably, the consistent errors between
theory and experiment are approximately halved when
Oxygen vacancy formation can be expressed according to eq 1 accounting for the error associated with the known overbinding
(in Kröger−Vink notation):23 of O2 by DFT-GGA.
1 With QM calculations, we can elucidate the origins of this
2MM X
+ OOX → 2M′M + VO•• + O2 (g)
2 (1) trend by analyzing how the electronic structure changes upon
V••
O formation.
25
After removal of a neutral oxygen atom from
The reaction enthalpy can be obtained by calculating energies
the ideal lattice by breaking two TM−O bonds, the two
for the perfect lattice (with an oxygen OxO between two TM
electrons formerly associated with the formally O2− ion must
ions MxM), the defective lattice with an oxygen vacancy (V••O) redistribute within the defective crystal. Figure 4 shows electron
and its two adjacent reduced metal sites (M′M), and the oxygen
density difference plots illustrating how this remaining charge
molecule in gas phase in its triplet ground state.
localizes on the two TMs closest to the vacancy. This localized
Creating a V•• O by removing a neutral oxygen atom leads to reduction is consistent with a decrease of atomic Bader charge34
defective AMO3−δ/Sr2Fe2−xMoxO6−δ lattices, with a non-
(by ∼0.4e) of each TM nearest the vacancy. Bader charges of
stoichiometry degree (δ) that depends on the periodic supercell
La remain unaltered, so La ions are merely spectators during
model size. We obtain the electronic energies and crystal
this process. The thermodynamics of V•• O formation therefore
vibrations of structurally relaxed perfect/defective structures
depend on M−O−M bond strengths and the electron affinity
(Eperfect/Edefective) from spin-polarized DFT+U calculations
of the two adjacent TM ions that reduce from M3+ to M2+.
using the Vienna Ab Initio Simulation Package (VASP).24
Lower ΔEf,vac values are obtained with weaker TM−O bonds
Further computational details (structural models, convergence
and/or M3+ TMs with higher electron affinities. Regarding
parameters, U−J values, magnetic arrangements of M
TM−O bond strengths, early TMs having low d occupation form
sublattices) can be found elsewhere.25−32 The simplest
strong polar−covalent, donor−acceptor bonds with oxygen
approximation of the vacancy formation free energy for eq 1
whereas late TMs with doubly occupied d orbitals create
can be directly obtained from electronic energies (eq 2): repulsion with oxygen lone pairs that weaken TM−O bonds.35
1 This is well-understood by inorganic chemists in terms of the
ΔEf,vac = Edefective − Eperfect + EO2
2 (2) heuristically observed “oxo wall”.36 To understand electron
affinities of M3+ ions, we analyze the configuration of their dn
We previously showed that additional zero-point and thermal
shells before and after oxygen vacancy formation. Measured and
energy contributions affect the reaction enthalpy by less than
computed magnetic moments show that Cr3+ and Mn3+ ions are
0.1 eV in LSF (xSr = 0, 0.25, 0.50).29 Thus, we use ΔEf,vac to
high spin (HS) whereas Co3+ can be considered intermediate spin
analyze trends for perovskite materials.
(IS) at SOFC operating temperatures (vide inf ra). In these cases,
3.1. LaMO3 the magnetic moment increases when the oxygen is removed
Calculated values for ΔEf,vac for LaMO3 (M = Cr, Mn, Fe, Co) (due to the electrons left behind entering the majority spin
are shown in Figure 3 (top). PBE+U ΔEf,vac values are largely manifold). The increased magnetic moment brings increased
C dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

Figure 4. Electron density differences (e−/bohr3) upon oxygen vacancy formation in LaCrO3: (a) 3D isosurface. Yellow/cyan surface refers to an
accumulation/loss of electrons. (b) 2D slice along the (100) plane. Red contours (accumulated electronic density) indicate localized reduction of
neighboring TM ions. Figure from ref 25. Reproduced by permission of The Royal Society of Chemistry.

favorable exchange interaction energy and causes a lower ΔEf,vac.

By contrast, when the extra electrons are added necessarily to the
minority spin manifold in Fe3+, reduction produces Fe2+ with a
decreased magnetic moment. This provides no extra favorable
exchange and leads to extra Coulomb repulsion within a doubly
occupied d orbital. For these reasons, the increase of magnetic
moment on the TM ions correlates (inversely) with the ΔEf,vac
trend (Figure 3, bottom).
Such a correlation has consequences for LaCoO3. Motivated
by the controversy around the actual spin state of LaCoO3
(invoked to be a mixture of Co3+ low spin (LS), IS, and HS states
to explain some anomalous properties37), we have explicitly
analyzed oxygen vacancy formation for several LaCoO3 spin states:
LS, IS, HS, and a 1:1 mixture of HS and LS. We find that LS Co3+
has the lowest ΔEf,vac among the spin states surveyed, which
again can be rationalized in terms of higher exchange stabilization
upon reduction. The LS Co3+ ion (t2g6 eg0) gains four favorable
exchange interactions upon reduction to HS Co2+ (t2g5 eg2), which
offsets the added Coulomb repulsion from the additional electron,
the cost of promoting an electron from the t2g orbitals to the eg
orbitals, and the extra energy required to break the stronger Figure 5. (top) Oxygen vacancy formation energies, ΔEf,vac, for
(LS Co3+)−O bond.38 We then predict that LaCoO3 oxygen La1−xSrxMO3 (M = Mn, Fe). (bottom) Percentage of electrons
delocalized over the oxygen sublattice after oxygen vacancy formation.
vacancy concentrations would be higher in materials with more
LS Co3+.32
3.2. La1−xSrxMO3 ion bulk diffusion.40 The effect of xSr on ΔEf,vac is less for LSM
than for LSF. Our electronic structure analysis does not support a
Large divalent cations (e.g., Sr2+, Ca2+, Ba2+) can serve as A-site rationalization based on localized reduction; a subtler picture
substitutions for La3+ in LaMO3-based SOFC cathodes to arises, as discussed next.
increase charge carrier (hole) concentrations and enhance p- As noted earlier, the nearest TM ions receive most of the
type conductivity. Adding Sr2+ induces the formation of a extra charge in both LaMnO3 and LaFeO3 (La1−xSrxMO3 with
(formally) TM4+ to maintain electroneutrality. These holes xSr = 0). Only ∼30% and ∼20% of this charge goes into the
affect the material valence band edge and qualitatively change oxygen sublattice for M = Mn and Fe, respectively. Hybrid-
the electronic structure, resulting in completely new material ization of the eg M 3d and O 2p orbitals partially delocalizes the
properties. charge over the oxygen atoms. Sr doping enhances this effect,39
To understand this phenomenon, we analyzed La1−xSrxMnO3 which further affects ΔEf,vac. In fact, we find that ΔEf,vac(xSr)
(LSM) and La1−xSrxFeO3 (LSF) with xSr = 0.25 and 0.5. LSM is directly (inversely) correlates with the percentage of extra
a common high-temperature SOFC cathode material, a good charge that delocalizes onto the oxygen sublattice (Figure 5,
electro-catalyst but a poor oxide ion conductor. LSF exhibits bottom). This percentage is still low for Mn (∼40−50%)
good bulk oxide ion transport and is the base material for the because a partial local reduction of Mn cations is retained for
prototypical MIEC cathode, La0.75Sr0.25Co0.2Fe0.8O3.39 We both xSr = 0.25 and 0.5.31 Surprisingly, almost no local reduction
recently evaluated how Sr doping affects the electronic structures occurs for Fe ions in LSF; the charge left by the oxygen vacancy
of these two materials.29,31 Figure 5 (top) displays ΔEf,vac versus formation is mostly accommodated by the oxygen sublattice
Sr content (xSr) in both materials. ΔEf,vac(xSr) of LSM and LSF (∼80%). The very low ΔEf,vac for LSF are therefore due to
follow the same trend in which the presence of Sr lowers the favorable delocalization of this charge onto oxygen atoms, thus
vacancy formation energy due to favorable charge compensation minimizing repulsion due to the defect. In general, larger
(excess Sr-derived holes compensated by excess O-vacancy- amounts of extra charge deposited onto oxygen correspond to
derived electrons). This trend is consistent with measurements lower vacancy formation energies. As shown below, we find that
showing that Sr-doped LaMO3 materials exhibit improved oxide ΔEf,vac in SFMO-based materials also feature this behavior.
D dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

3.3. Sr2Fe2−xMoxO6−δ (x = 1.0, 0.5)

Antisite defects, naturally present in SFMO1.0 perovskites, allow
us to analyze oxygen vacancy formation within different local
environments, that is, those occurring along Fe−O−Mo, Fe−
O−Fe, and Mo−O−Mo bonds.26 The average ΔEf,vac of these
three types of oxygen vacancies are shown in Figure 6 (left).

Figure 6. (left) Oxygen vacancy formation energies, ΔEf,vac, along A

(Fe−O−Mo), B (Fe−O−Fe), and C (Mo−O−Mo) bonds in
SFMO1.0. VO•• shown in yellow. Adapted from ref 26. (right)
Corresponding distribution among the Fe, Mo, and O sublattices of
the extra charge left behind upon oxygen vacancy formation.

Figure 7. (top) Oxygen vacancy formation energies, ΔEf,vac, as a

The 3−5 eV range in energy is similar to that for AMO3 function of oxygen vacancy concentration δ along Fe−O−Mo and
materials. However, ΔEf,vac along the Fe−O−Fe bonds are Fe−O−Fe bonds in SFMO0.5. Color code as in Figure 6. Experimental
significantly lower than those along Fe−O−Mo and Mo−O− δ = 0.10(2) indicated [ref 27]. (middle and bottom) Corresponding
Mo bonds by ∼1 and 2 eV, respectively. As in the AMO3 series, distribution of the extra charge remaining within the Fe, Mo, and O
the trend ΔEf,vac(Fe−VO••−Fe) < ΔEf,vac(Fe−VO••−Mo) < sublattices upon oxygen vacancy formation.
ΔEf,vac(Mo−V•• O −Mo) follows the bond strength rule, since
Fe is a late TM and Mo is an early TM. Moreover, analyzing
than ΔEf,vac(Fe−V•• O −Mo) at all δ, explaining the observed
electron density differences upon oxygen vacancy formation
selectivity in oxygen sites exhibiting vacancies. As was found in
shows that most of the extra charge remains localized on Mo
La1−xSrxFeO3, the facile electron delocalization onto the oxygen
sites in the Mo−V•• O −Mo case, whereas Fe and O sublattices
sublattice upon reduction causes the extraordinarily low
share the extra charge in the Fe−V•• O −Fe (Figure 6, right).
ΔEf,vac(Fe−V•• O −Fe) (Figure 7, bottom). The resulting inherent
Thus, excess Fe (instead of Mo) should provide SFMO-based
oxygen vacancy concentration likely boosts oxygen mobility, since
materials with higher concentrations of oxygen vacancies that
the diffusion process will thus depend on only the migration barrier.
facilitate oxide ion diffusion.
Indeed, the Fe-rich SFMO0.5 performs extremely efficiently 4. DIFFUSION COEFFICIENTS
in symmetric IT-SOFCs. Its ionic conductivity rivals that
of Co-containing cathodes such as La 0.6Sr0.4CoO3 and DFT+U accurately predicts energetic, structural, and electronic
La0.6Sr0.4Fe0.8Co0.2O3. In joint experimental and theoretical properties of many strongly correlated materials at different
work,27 we explained why the SFMO0.5 structure yields temperatures. Calculation of kinetic coefficients, such as
excellent MIEC properties. First, SFMO0.5 is a nonstoichio- diffusivity, can be more challenging. The self-diffusion
metric solid. Powder neutron diffraction data determined an coefficient, DO, for a vacancy-mediated hopping mechanism
oxygen-deficient composition of Sr2Fe1.5Mo0.5O6−δ (δexp = can be expressed in terms of the vacancy diffusion coefficient,
0.10(2)). No Fe/Mo ordering is found, in agreement with Dv, within transition state theory30,32,41,42 as
the theoretically predicted lower energy of the cell without Mo ⎛ ⎞
Cv Cv
Dv ≈ ⎜ eΔSf,vac / kB e−ΔHf,vac /(kBT )⎟
aggregation, that is, that with Mo in the two opposite vertices of 1
DO = Dv =
CO 3 − Cv ⎜ p ⎟
the cube formed by TM ions (Figure 7, inset). Moreover, ⎝ O2 ⎠
oxygen vacancies can arise from two possible types, those from ⎛ 1 2 ΔS⧧/ kB −ΔHmigr /(kBT )⎞
Fe−O−Fe bonds or from Fe−O−Mo bonds (no Mo−O−Mo × ⎜ v a e e ⎟
⎝6 0 ⎠ (5)
are expected in SFMO0.5 with homogeneously distributed Fe/Mo,
and as Figure 6 shows, these are the least favorable). Figure 7 where CO and Cv are the oxygen and oxygen vacancy
(top) shows calculated ΔEf,vac(Fe−V•• ••
O −Fe) and ΔEf,vac(Fe−VO − concentrations, respectively, expressed in moles per mole of
Mo) in SFMO0.5 for different concentrations of oxygen vacancies ABO3 and we consider dilute vacancies. The apparent activation
(δtheo = 0.25, 0.125, and 0.0625). ΔEf,vac(Fe−V••
O −Fe) appears to energy for diffusion is the sum of the enthalpies of formation and
be nearly zero for δtheo ≈ δexp, confirming SFMO0.5’s non- migration of an oxygen vacancy. The temperature-dependent
stoichiometry. As expected, ΔEf,vac(Fe−V•• O −Fe) is much lower entropic contributions account for the entropy change upon
E dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

Figure 8. (a) Minimum energy path, (b) energy profile, and (c) magnetic moment profile for oxide ion migration in LaFeO3. Reprinted with
permission from ref 29. Copyright 2013 American Chemical Society.

defect formation and the thermal expansion effects at high SOFC oxygen vacancies in SFMO0.5 are those along Fe−O−Fe bonds
operating temperatures.29,41 DO is related to the tracer diffusion (vide supra), so we considered the migration pathway from a
coefficient, DO*, through the correlation factor, f (f ≈ 0.69 for Fe−O−Fe occupied site to the nearest Fe−V•• O −Fe. As with
perovskites43), and to the chemical diffusion coefficient, Dchem, other perovskites, oxide ion migration follows a curved path in
through the thermodynamic factor, γ. Calculating ΔHmigr, v0, and a four-atom plane. This plane contains three iron atoms
ΔS⧧ requires knowledge of the transition state (TS) structure. (involved in the migration) and a Mo atom at the opposite,
We identify the minimum energy path and its TS using the fourth corner. The calculated migration energy for this jump
climbing image nudged elastic band method.44 We include (Figure 9, A → B) is ∼0.3 eV, much lower than typical values
vibrational entropic effects via calculation of the real phonon
frequencies (νi) of the initial minimum state and the TS,
obtained at the Γ point from the eigenvalues of the Hessian
matrix. We neglect configurational entropy terms because of the
low concentration of vacancies in our materials.
For the prototypical LaMO3 series, experiments indicate that
O2− migration energies are very similar (∼0.8−0.9 eV) and not
strongly dependent on Sr-doping.45 We chose LaFeO3 as a
representative case to study O2− migration in La1−xSrxMO3.
Our calculations confirm the curved two-dimensional pathway,
a well-known feature in perovskite-type materials.46 The
migrating oxide ion moves in a plane that contains four TM
ions (Fe in this case) framed by two La atoms above and below
its center (Figure 8, left). Our predicted migration barrier,
ΔHmigr = 0.79 eV, matches the value from oxygen tracer
experiments (ΔHmigr =0.77 eV).43 Localized reduction adjacent
to the vacant site couples electron transfer with ion motion, as
shown by the abrupt change in Fe magnetic moments during the
oxide ion migration (Figure 8, right). We cannot expect analogous Figure 9. (top) Diffusion planes showing three different (A → B, C →
coupled electron transfer for La1−xSrxMO3 (xSr > 0) materials with D, and E → F) Fe−O−Fe → Fe−V•• O −Fe jumps in SFMO0.5. Color
delocalized extra charge. However, this electronic feature should code as in Figure 6. (bottom) Corresponding energy profiles. Adapted
have little effect on the migration barrier, which we have shown to from ref 28.
be largely governed by steric and electrostatic effects.29
Calculating diffusion coefficients that are directly comparable (∼0.8−0.9 eV) measured and calculated for La,Sr-based
with experiment requires accounting for oxygen vacancy perovskites. The calculated self-diffusion coefficient (at 750 °C)
concentrations (Cv in eq 5). An artificially high ΔEf,vac will with this migration barrier (taking ΔHf,vac = 0 due to SFMO’s
lead to artificially low Cvs in perfect LaFeO3, resulting in nonstoichiometry) is 5.1 × 10−6 cm2 s−1.28 Although this is
diffusion coefficients DO several orders of magnitude in error. very close to the Dchem reported by Chen and co-workers (5.0 ×
We recently showed that accounting for the slight La sub- 10−6 cm2 s−1),47 the γ factor mentioned above brings uncertainty
stoichiometry observed in experiments with LaFeO3 corrects of approximately two orders of magnitude.
these discrepancies.30 We found that even a small concentration Interestingly, the migration barrier for the parent material
of La vacancies, V‴ La, lead to dramatically higher oxygen vacancy without Mo, namely, SrFeO3, is ∼0.9 eV.48 Thus, Mo plays an
concentrations. VLa ‴ introduces holes that compensate the extra important role in oxygen diffusion. We analyzed this role by
charge from a V•• O , just as Sr did in previous examples. Thus, computing the migration barrier heights also for Fe−O−Fe →
promoting cation vacancies is a route to increasing oxygen vacancy Fe−V••O −Fe jumps in a layered cell with varying Mo configurations.
concentrations. Having no Mo cation in the migration plane (Figure 9, C → D)
Besides being nonstoichiometric, SFMO0.5 also exhibits increases the migration energy by 0.4 eV, and having a layered
unusual migration features compared with LaMO3. Preferred Mo distribution (Figure 9, E → F) increases it further by 0.2 eV
F dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX
Accounts of Chemical Research Article

(Figure 9, bottom). We observed no change in atomic Bader formation, they stabilize TS structures and lower the
charges, magnetic moments, or PDOS along the migration migration barrier to ∼0.3 eV. Our predicted oxide ion
pathway in any of the three jumps described above. The similar diffusion coefficient is very high and agrees with recent
Shannon ionic radii49 of Fe(III) and Fe(IV) in 6-fold coordination experimental estimates.
(0.64 and 0.58 Å, respectively) and that of Mo(VI) (0.59 Å) rule In summary, ΔEf,vac is a robust descriptor for evaluating the
out steric effects determining the diffusion behavior. Thus, we oxide ion transport properties of perovskite oxides used as
conclude that the differences in migration barriers arise only from SOFC cathodes. There is also a direct relationship between the
Mo-based electrostatic effects. These results suggest that attraction ΔEf,vac values and (i) the transition metal−oxygen bond strength
between the highly charged Mo5+δ cation and the negatively and (ii) the degree to which the O sublattice accommodates the
charged oxygen ion produces a lower migration energy when the vacancy’s extra charge. We derived QM-inspired design principles
Mo’s are closer to the migration path. Furthermore, we find that for improved IT-SOFC cathodes: (1) mid-to-late transition metal
the Madelung potential of the crystal stabilizes the TS more for ions with weak M−O bonds and high electron affinities should be
the homogeneous Fe/Mo case than for the layered one.28 incorporated, (2) dopants and defect engineering should exploit
hybridization of M 3d and O 2p states to enhance charge
5. CONCLUSIONS delocalization, and (3) the Fe:Mo ratio in SFMO-based materials
This Account details the successful application of the DFT+U should be optimized to include more iron to enhance oxygen
method with ab initio-derived U−J values to explain oxide nonstoichiometry. Further studies on ORR electrocatalysis and
ion diffusion in two classes of cathode materials for SOFC oxygen surface exchange rates are ongoing.
applications: La1−xSrxMxO3 and Sr2Fe2−xMoxO6. QM calcu-
lations elucidated the microscopic effects governing the two
processes that determine diffusivity in these materials, that is,
■ AUTHOR INFORMATION
Funding
oxygen vacancy formation and oxygen migration. We then This work was supported as part of HeteroFoaM, an Energy
showed how the oxide ion diffusion coefficients measured in Frontier Research Center funded by the U.S. Department of
experiments can be accurately predicted from first-principles Energy, Office of Science, Office of Basic Energy Sciences
methods. Our insights clarify several points: under Award Number DE-SC0001061.
• In LaMO3 materials, the availability of vacant oxygen sites Notes
that allow oxide ion transport governs oxide ion diffusion.
The authors declare no competing financial interest.
Thus, the oxygen vacancy formation energy, ΔEf,vac, is a
critical descriptor for this process. The electronic rearrange- Biographies
ment upon oxygen vacancy formation involves breaking two Ana B. Muñoz-Garcı ́a graduated (Valedictorian) in Chemistry in 2006
TM−oxygen bonds and reducing two TMs closest to the and obtained her Ph.D. in Theoretical Chemistry and Computational
vacancy. Weaker M−O bonds and M3+ ions with greater Modeling in 2011 from the Universidad Autónoma de Madrid. After
electron affinities facilitate oxygen vacancy formation. one year as a Postdoctoral Research Associate at Princeton University,
• In Sr-doped LaMO3 materials, aliovalent substitution at she became an Assistant Professor at the Università di Napoli Federico II.
the La site forms holes that affect electronic states at the
valence band edge and consequently the defect Andrew M. Ritzmann obtained a B.S. in Chemical Engineering
chemistry. These holes facilitate the reduction produced (magna cum laude) from Cornell University in 2005. Since 2008, he
by the electrons left behind when the oxygen vacancy has been a graduate student at Princeton University.
forms, via charge compensation. The p-type character Michele Pavone graduated (cum laude) in Chemistry and obtained his
also increases the hybridization of the eg transition metal Ph.D. in Chemical Sciences from the Università di Napoli Federico II
3d and oxygen 2p states, allowing delocalization of the where he is now an Assistant Professor. From 2010 to 2012, he was a
extra charge across the M−O sublattice. These two Postdoctoral Research Associate at Princeton University.
factors significantly lower oxygen vacancy formation John A. Keith obtained his B.A. with high honors in chemistry from
energies of La1−xSrxMxO3 compared to LaMO3. Wesleyan University and his Ph.D. in chemistry from Caltech. After an
• For LaFeO3-based materials, calculated migration barriers Alexander von Humboldt postdoctoral fellowship at the University of
(∼0.8 eV) are in excellent agreement with experiments. By Ulm, he was an Associate Research Scholar in the Carter group at
accounting for La vacancies, a quantitative match with the Princeton. In 2013, he became the inaugural R.K. Mellon Faculty
experimental DO is achieved. As found with Sr-doping, Fellow in Energy at the University of Pittsburgh.
A-site vacancies introduce holes in the system lowering
Emily A. Carter is the Founding Director of the Andlinger Center for
ΔEf,vac and consequently increasing the concentration of
Energy and the Environment at Princeton University and the Gerhard
vacant sites on oxygen sublattice. Cation vacancies
R. Andlinger Professor in Energy and the Environment, as well as
indirectly promote oxide ion diffusion.
Professor of Mechanical and Aerospace Engineering and Applied and
• In Sr2Fe2−xMoxO6 materials, the electronic features and
Computational Mathematics.

■
oxide ion diffusion largely depend on the Fe:Mo ratio. Fe-
rich compositions exhibit lower oxygen vacancy formation
REFERENCES
energies and better oxide ion transport properties, because
Mo−O bonds are stronger and electron density is more (1) Boudghene-Stambouli, A.; Traversa, E. Solid oxide fuel Cells
(SOFCs): A Review of an Environmentally Clean and Efficient Source
localized at Mo centers upon reduction. of Energy. Renewable Sustainable Energy Rev. 2002, 6, 433−455.
• Theory and experiment both find that Sr2Fe1.5Mo0.5O6−δ (2) Adler, S. B. Factors Governing Oxygen Reduction in Solid Oxide
is inherently nonstoichiometric. Migration barrier heights Fuel Cell Cathodes. Chem. Rev. 2004, 104, 4791−4843.
govern oxide ion diffusion in this material. Even though (3) Chroneos, A.; Yildiz, B.; Tarancón, A.; Parfitt, D.; Kilner, J. A.
highly charged Mo6+ centers hinder oxygen vacancy Oxygen Diffusion in Solid Oxide Fuel Cell Cathode and Electrolyte

G dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX

Accounts of Chemical Research Article

Materials: Mechanistic Insights from Atomistic Simulations. Energy (26) Muñoz-García, A. B.; Pavone, M.; Carter, E. A. Effect of Antisite
Environ. Sci. 2011, 4, 2774−2789. Defects on the Formation of Oxygen Vacancies in Sr2FeMoO6:
(4) Yueh-Lin, L.; Kleis, J.; Rossmeisl, J.; Shao-Horn, Y.; Morgan, D. Implications for Ion and Electron Transport. Chem. Mater. 2011, 23,
Prediction of Solid Oxide Fuel Cell Cathode Activity with First- 4525−4536.
Principles Descriptors. Energy Environ. Sci. 2011, 4, 3966−3970. (27) Muñoz-García, A. B.; Bugaris, D. E.; Pavone, M.; Hodges, J. P.;
(5) Kuklja, M. M.; Kotomin, E. A.; Merkle, R.; Mastrikov, Y. A.; Huq, A.; Chen, F.; zur Loye, H.-C.; Carter, E. A. Unveiling Structure-
Maier, J. Combined Theoretical and Experimental Analysis of Property Relationships in Sr2Fe1.5M0.5O6−δ, an Electrode Material for
Processes Determining Cathode Performance in Solid Oxide Fuel Symmetric Solid Oxide Fuel Cells. J. Am. Chem. Soc. 2012, 134, 6826−
Cells. Phys. Chem. Chem. Phys. 2013, 15, 5443−5471. 6833.
(6) Kohn, W.; Sham, L. J. Self-Consistent Equations Including (28) Muñoz-García, A. B.; Pavone, M.; Ritzmann, A. M.; Carter, E. A.
Exchange and Correlation Effects. Phys. Rev. 1965, 140, A1133− Oxide Ion Transport in Sr2Fe1.5M0.5O6‑δ, a Mixed Ion-Electron
A1138. Conductor: New Insights from First-Principles Modeling. Phys.
(7) Wachsman, E. D.; Singhal, S. C. Solid Oxide Fuel Cell Chem. Chem. Phys. 2013, 15, 6250−6259.
Commercialization, Research, and Challenges. Am. Ceram. Soc. Bull. (29) Ritzmann, A. M.; Muñoz-García, A. B.; Pavone, M.; Keith, J. A.;
2010, 89, 22−32. Carter, E. A. Ab Initio DFT+U Analysis of Oxygen Vacancy Formation
(8) Liu, Q.; Dong, X.; Xiao, G.; Ma, S.; Zhao, F.; Chen, F. A Novel and Migration in La1−xSrxFeO3−δ (x = 0, 0.25, 0.50). Chem. Mater.
Electrode Material for Symmetrical SOFCs. Adv. Mater. 2010, 22, 2013, 25, 3011−3019.
5478−5482. (30) Ritzmann, A. M.; Muñoz-García, A. B.; Pavone, M.; Keith, J. A.;
(9) Carter, E. A. Challenges in Modeling Materials Properties Carter, E. A. Ab Initio Evaluation of Oxygen Diffusivity in LaFeO3:
without Experimental Input. Science 2008, 321, 800−803. The Role of La Vacancies. MRS Commun. 2013, 3, 161−166.
(10) Huang, P.; Carter, E. A. Advances in Correlated Electronic (31) Pavone, M.; Muñoz-García, A. B.; Ritzmann, A. M.; Carter, E. A.
Structure Methods for Solids, Surfaces, and Nanostructures. Annu. Rev. Ab Initio Study of Lanthanum Strontium Manganite: Insights into
Phys. Chem. 2008, 59, 261−290. Electronic Features and Oxygen Vacancy Formation. J. Phys. Chem. C
(11) Anisimov, V. I.; Zaanen, J.; Andersen, O. K. Band Theory and 2014, DOI: 10.1021/jp500352h.
Mott Insulators: Hubbard U Instead of Stoner I. Phys. Rev. B 1991, 44, (32) Ritzmann, A. M.; Pavone, M.; Muñoz-García, A. B.; Keith, J. A.;
943−954. Carter, E. A. Ab Initio DFT+U Analysis of Oxygen Transport in
(12) Becke, A. D. A New Mixing of Hartree-Fock and Local Density- LaCoO3: The Effect of Co3+ Spin States. J. Mater. Chem. A 2014, 2,
Functional Theories. J. Chem. Phys. 1993, 98, 1372−1377. 8060−8074.
(13) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient (33) Jiang, S. P.; Li, J. In Solid Oxide Fuel Cells: Materials Properties
Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865−3868. and Performances; Fergus, J.W., Hui, R., Li, X., Wilkinson, D.P., Zhang,
(14) Johnson, B. G.; Gill, P. M. W.; Pople, J. A. The Performance of a J., Eds.; CRC Press 2009; pp 131−137.
Family of Density Functional Methods. J. Chem. Phys. 1993, 98, 5612− (34) Tang, W.; Sanville, E.; Henkelman, G. A Grid-based Bader
5626. Analysis Algorithm without Lattice Bias. J. Phys.: Condens. Matter 2009,
(15) Mosey, N. J.; Carter, E. A. Ab initio Evaluation of Coulomb and 21, No. 084204.
(35) Carter, E. A.; Goddard, W. A., III Early- versus Late-Transition-
Exchange Parameters for DFT+U Calculations. Phys. Rev. B 2007, 76,
Metal-Oxo Bonds: The Electronic Structure of Oxovanadium(1+) and
No. 155123.
Oxoruthenium(1+). J. Phys. Chem. 1988, 92, 2109−2115.
(16) Mosey, N. J.; Liao, P.; Carter, E. A. Rotationally Invariant Ab
(36) Winkler, J. R.; Gray, H. B. Electronic Structures of Oxo-Metal
Initio Evaluation of Coulomb and Exchange Parameters for DFT+U
Ions. In Molecular Electronic Structures of Transition Metal Complexes I;
Calculations. J. Chem. Phys. 2008, 129, No. 014103.
Mingos, D. M. P., Day, P., Dahl, J. P., Eds.; Structure and Bonding;
(17) Kulik, H. J.; Cococcioni, M.; Scherlis, D. A.; Marzari, N. Density
Springer: Berlin/Heidelberg, 2012; pp 17−28.
Functional Theory in Transition-Metal Chemistry: A Self-Consistent (37) Radaelli, P. G.; Cheong, S.-W. Structural Phenomena Associated
Hubbard U Approach. Phys. Rev. Lett. 2006, 97, No. 103001. with the Spin-state Transition in LaCoO3. Phys. Rev. B 2002, 66,
(18) Toroker, M. C.; Kanan, D. K.; Alidoust, N.; Isseroff, L. Y.; Liao, No. 094408.
P.; Carter, E. A. First Principles Scheme to Evaluate Band Edge (38) Hong, W. T.; Gadre, M.; Lee, Y.-L.; Biegalski, M. D.; Christen,
Positions in Potential Transition Metal Oxide Photocatalysts and H. M.; Morgan, D.; Shao-Horn, Y. Tuning the Spin State in LaCoO3
Photoelectrodes. Phys. Chem. Chem. Phys. 2011, 13, 16644−16654. Thin Films for Enhanced High-Temperature Oxygen Electrocatalysis.
(19) Kanan, D. K.; Carter, E. A. Band Gap Engineering of MnO via J. Phys. Chem. Lett. 2013, 4, 2493−2499.
ZnO Alloying: A Potential New Visible-Light Photocatalyst. J. Phys. (39) Kilner, J. A.; Berenov, A.; Rossiny, J. In Perovskite Oxide for Solid
Chem. C 2012, 116, 9876−9887. Oxide Fuel Cells; Ishihara, T., Ed.; Springer: Dordrecht, the
(20) Yu, K.; Carter, E. A. Communication: Comparing ab Initio Netherlands, 2009; pp 95−116.
Methods of Obtaining Effective U Parameters for Closed-Shell (40) Mizusaki, J.; Sasamoto, T.; Cannon, W. R.; Bowen, H. K.
Materials. J. Chem. Phys. 2014, 140, No. 121105. Electronic Conductivity, Seebeck Coefficient, and Defect Structure of
(21) Hashimoto, T.; Ishibashi, S.; Terakura, K. Jahn-Teller Distortion La1‑xSrxFeO3 (x = 0.l, 0.25). J. Am. Ceram. Soc. 1983, 66, 247−252.
and Magnetic Structure in LaMnO3: A First-principles Theoretical (41) Vineyard, G. H. Frequency Factors and Isotope Effects in Solid-
Study with Full Structure Optimizations. Phys. Rev. B 2010, 82, State Rate Processes. J. Phys. Chem. Solids 1957, 3, 121−127.
No. 045124. (42) Marino, K. A.; Carter, E. A. First-principles Characterization of
(22) Kobayashi, K. I.; Kimura, T.; Sawada, H.; Terakura, K.; Tokura, Ni Diffusion Kinetics in β-NiAl. Phys. Rev. B 2008, 78, No. 184105.
Y. Room-Temperature Magnetoresistance in an Oxide Material with (43) Ishigaki, T.; Yamauchi, S.; Kishio, K.; Mizusaki, J.; Fueki, K.
an Ordered Double-Perovskite Structure. Nature 1998, 395, 677−680. Diffusion of Oxide Ions in Perovskite-type Oxides. J. Solid State Chem.
(23) Kröger, F. A.; Vink, H. J. Relations between the Concentrations 1988, 73, 179−187.
of Imperfections in Crystalline Solids. Solid State Phys. 1956, 3, 307− (44) Henkelman, G.; Uberuaga, B. P.; Jónsson, H. A Climbing Image
435. Nudged Elastic Band Method for Finding Saddle Points and Minimum
(24) Kresse, G.; Furthmüller, J. Efficient Iterative Schemes for Ab Energy Paths. J. Chem. Phys. 2000, 113, 9901−9904.
Initio Total-Energy Calculations Using a Plane-Wave Basis Set. Phys. (45) Mastrikov, Y. A.; Merkle, R.; Kotomin, E. A.; Kuklja, M. M.;
Rev. B 1996, 54, 11169−11186. Maier, J. Formation and Migration of Oxygen Vacancies in
(25) Pavone, M.; Ritzmann, A. M.; Carter, E. A. Quantum- La1‑xSrxCo1‑yFeyO3‑δ Perovskites: Insight from ab Initio Calculations
Mechanics-Based Design Principles for Solid Oxide Fuel Cell Cathode and Comparison with Ba1‑xSrxCo1‑yFeyO3‑δ. Phys. Chem. Chem. Phys.
Materials. Energy Environ. Sci. 2011, 4, 4933−4937. 2013, 15, 911−918.

H dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX

Accounts of Chemical Research Article

(46) Islam, M. S. Ionic Transport in ABO3 Perovskite Oxides: A

Computer Modeling Tour. J. Mater. Chem. 2000, 10, 1027−1038.
(47) Xiao, G.; Liu, Q.; Zhao, F.; Zhang, L.; Xia, C.; Chen, F.
Sr2Fe1.5Mo0.5 O6 as Cathodes for Intermediate-Temperature Solid
Oxide Fuel Cells with La0.8Sr0.2Ga0.87Mg0.13O3 Electrolyte. J. Electro-
chem. Soc. 2011, 158, B455−B460.
(48) Wang, L.; Merkle, R.; Maier, J. Surface Kinetics and Mechanism
of Oxygen Incorporation into Ba1‑xSrxCoyFe1‑yO3‑δ SOFC Micro-
electrodes. J. Electrochem. Soc. 2010, 157, B1802−B1808.
(49) Shannon, R. D. Revised Effective Ionic Radii and Systematic
Studies of Interatomic Distances in Halides and Chalcogenides. Acta
Crystrallogr. 1976, A32, 751−767.

I dx.doi.org/10.1021/ar4003174 | Acc. Chem. Res. XXXX, XXX, XXX−XXX