ARTICLE

pubs.acs.org/cm

Phosphates as Lithium-Ion Battery Cathodes: An Evaluation Based on

High-Throughput ab Initio Calculations
Geoffroy Hautier, Anubhav Jain, Shyue Ping Ong, Byoungwoo Kang, Charles Moore, Robert Doe, and
Gerbrand Ceder*
Massachusetts Institute of Technology, Department of Materials Science and Engineering, 77 Massachusetts Avenue,
Cambridge, MA 02139

bS Supporting Information
ABSTRACT: Phosphate materials are being extensively studied as
lithium-ion battery electrodes. In this work, we present a high-
Downloaded from pubs.acs.org by LA TROBE UNIV on 01/05/19. For personal use only.

throughput ab initio analysis of phosphates as cathode materials.

Capacity, voltage, specific energy, energy density, and thermal stability
are evaluated computationally on thousands of compounds. The limits
in terms of gravimetric and volumetric capacity inherent to the
phosphate chemistry are determined. Voltage ranges for all redox
couples in phosphates are provided, and the structural factors influen-
cing the voltages are analyzed. We reinvestigate whether phosphate
materials are inherently safe and find that, for the same oxidation state,
Chem. Mater. 2011.23:3495-3508.

oxygen release happens thermodynamically at lower temperature for

phosphates than for oxides. These findings are used to recommend
specific chemistries within the phosphate class and to show the
intrinsic limits of certain materials of current interest (e.g., LiCoPO4 and LiNiPO4).
KEYWORDS: Li-ion battery, cathode, phosphates, ab initio, DFT, high-throughput, thermal stability, safety

1. INTRODUCTION The high scalability of computing furthermore offers the possi-

The first commercial lithium-ion batteries, appearing in the bility to search for new cathode materials using a computational
market in the early 1990s, were based on transition metal oxide high-throughput approach by computing properties on thou-
cathodes such as layered LiCoO2.1 In 1997, lithium iron phos- sands of potential battery materials. This approach can be used to
phate (LiFePO4) in the olivine structure was demonstrated as a screen and discover new or overlooked compounds (see for
viable cathode material.2 Since then, the exceptional safety and instance Kim et al.19) but also to analyze, using a large data set,
the limits and factors that control electrochemical properties
improvements in rate capabilities of lithium iron phosphate have
across an entire chemical class.
led to a strong focus from the battery community on polyanionic
In this work, we compute the voltage, capacity (gravimetric
chemistries and especially on phosphates.3,4
and volumetric), specific energy, energy density, stability, and
Despite the important amount of work on phosphate materi-
safety of thousands of phosphate compounds using such a
als, few phosphates have been precisely characterized and tested
computational high-throughput approach. From this large data-
electrochemically: LiMnPO4,5 LiVOPO4,6 Li3Fe2(PO4),7 Li3V2-
base of calculated properties on phosphates, important conclu-
(PO4)3.8 The limited size of this data set makes it difficult to
sions can be made on the limits and opportunities for phosphates
understand the trends and limits of phosphates as battery elec-
as cathode materials.
trodes. Moreover, comparison between experimentally tested
materials is sometimes challenging as electrochemical measure-
ment are the result of many different intrinsic and extrinsic 2. METHODS
materials properties (e.g., rate can be controlled by limitations 2.1. Ab Initio High-Throughput Methodology. All ab initio
from the intrinsic materials transport or by electrode fabrication computations were performed in the density functional theory (DFT)
and structure). framework using a generalized gradient approximation (GGA) func-
Ab initio computations in the density functional theory (DFT) tional parametrized by Perdew-Burke and Ernzerhof (PBE).20 The
framework have been used for almost 20 years in the battery field transition metals, Cu, Fe, Cr, Co, Mn, Ni, V, Nb, and Mo, have been
to provide insight into the fundamental properties of electrode
materials .9
18 Ab initio computations are nowadays accurate Received: April 2, 2011
enough to understand and even predict many important battery Revised: June 19, 2011
properties (e.g., voltage, stability, safety, and lithium diffusion). Published: July 13, 2011

r 2011 American Chemical Society 3495 dx.doi.org/10.1021/cm200949v | Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

assigned a U parameter to correct for the self-interaction error present in Entropic contributions were neglected. When the delithiated state still
GGA.21,22 This U parameter was fitted to experimental binary metal contained lithium in the structure (e.g., LiV2(PO4)3), the lithium
formation energies using the Kubaschewski tables,23 following the ordering over the available sites was determined by enumerating possible
approach of Wang et al.24 The only exception is the U for Co, set at ordered states of lithium and lithium vacancies using Hart’s algorithm
5.7 eV, following Zhou et al.25 We followed Wang et al. and did not use a and choosing the state leading to the lowest electrostatic energy
U parameter on Ti. While hybrid functional, for instance in the Heyd- computed by an Ewald sum on formal valence point charges.34
Scuseria-Ernzerhof (HSE) scheme,26 can also correct for the self-inter- Thermodynamic stability at zero K and zero pressure was evaluated
action error,27 the high-throughput nature of our work favors the use of using ab initio computed total energies. The stability of any phase was
the less computationally expensive GGA+U framework. evaluated by comparing it to other phases or linear combination of
All compounds were run in their ferromagnetic states with a k-point phases leading to the same composition using the convex hull construc-
density of at least 500/(number of atom in unit cell) k-points. The tion. GGA and GGA+U computations were combined using Jain et al.’s
Vienna ab initio software package (VASP)28 was used with the plane- methodology.41 The stability of any compound was quantified by
augmented wave (PAW) pseudopotentials provided.29 The computa- evaluating the energy above the hull, which is the negative energy of the
tions are expected to be converged within a few meV/at. More details on reaction that decomposes a compound into the most stable combination
the high-throughput ab initio methodology can be found in Jain et al.30 of other phases. When the compound is stable at zero K, its energy above
2.2. Compounds Data Set. Our list of redox active elements and the hull is 0 meV/at. An energy above the hull is always positive, and its
allowed oxidation state consists of Ti (+2 to +4), V (+2 to +5), Cr (+2 to +6), magnitude indicates the degree of instability of the compound.
Mn (+2 to +4), Fe (+2 to +4), Co (+2 to +4), Ni (+2 to +4), Cu (+1 Safety or thermal stability was computed as in Ong et al. by evaluating
to +3), Nb (+3 to +5), Mo (+3 to +6), Sn (+2 to +4), Sb (+3 to +5) and the oxygen chemical potential necessary for the compound to decom-
Bi (+3 to +5). pose at equilibrium through oxygen gas evolution.42 This approach
Our set of known compounds was obtained from the 2006 ICSD assumes an equilibrium process and an entropic contribution to the
database.31 All known compounds containing lithium, phosphorus, reaction solely from the oxygen gas. The oxygen chemical potential
oxygen, and a redox active metal were considered. reference (μO2 = 0 meV) is chosen to be air at 298 K, according to the
Duplicate crystal structures were identified and removed using an tabulated entropy of oxygen in the JANAF tables and the fitted oxygen
affine mapping technique. Two structures are considered equivalent if an molecule energy from Wang et al.24,43 Oxygen chemical potential ranges
affine mapping exists transforming one into another within reasonable (with respect to this reference) can be found for typical binary oxides in
tolerances.32 The algorithm used to perform this comparison was based the Supporting Information of Hautier et al.’s paper.37
on Hundt et al.33 This algorithm was also used to assign a crystal 2.5. Phosphates Classification. In this work, any compound
structure prototype to any of the generated compounds. containing phosphorus atoms tetrahedrally coordinated by oxygen will
The experimental crystallographic data of some known compounds be considered as a phosphate. We will follow the classification of
contain partial occupancies. For those, we used an ordering procedure to phosphates from Durif.44 The first criteria used to classify phosphates
enumerate unique supercells using an algorithm proposed by Hart.34 is based on the oxygen to phosphorus ratio. Three categories can then be
Among all the enumerated structures, the closest to the structure’s distinguished:
stoichiometry, as listed in the ICSD, and lowest in electrostatic energy • O/P = 4 orthophosphates or monophosphates
(according to an Ewald summation on formal valence point charges) was • O/P > 4 oxyphosphates
chosen. • 5/2 < O/P < 4 condensed phosphates
In addition to known compounds from the ICSD, we generated The condensed phosphates have the particularity that some PO4
“virtual” compounds by substituting ions in known ICSD compounds to groups share corners with other PO4 groups. The condensed phosphates
form new possible compounds containing lithium, phosphorus, oxygen, can be subdivided between:
and a redox active metal . This substitution process was driven by a data- • Linear polyphosphates. They contain groups of PO4 tetrahedron
mined substitution model presented in Hautier et al.35 sharing one or two corners with other PO4 tetrahedron. The general
To evaluate stability of the computed compounds, all competing formula for these is (PnO3n+1)(n+2)
. Pyrophosphates (n = 2)
phases known in the ICSD were also computed. For instance, to make (P2O7)4
belong to this group. Metaphosphates (PO3)
are
the stability study of a Li
M
P
O system possible, we computed the formed if the chain is infinite.
compounds present in the ICSD in all the related binary (i.e., Li
O, • Cyclophosphates. They present corner sharing rings of tetrahedron
M
O, Li
M) and ternary (i.e., Li
M
O, Li
P
O, M
P
O, PO4 and are of formula (PnO3n)n
. Among those, the (cyclic)
Li
M
P) systems.36 In addition, a data-mined ternary oxide structure metaphosphates are common (PO3)
.
prediction algorithm has been used to propose any likely new compound • Ultraphosphates. They present PO4 tetrahedron sharing three of
candidate in the M
P
O systems.37 In total, our database contains their oxygen with other PO4 groups. (P2m+nO5m+3n)n

4074 computed compounds containing containing lithium, a redox
active element, phosphorus. and oxygen.
2.3. Local Environment Assignment. The local environment 3. RESULTS
for a given cation was assigned automatically through a polyhedron
matching algorithm. The cation’s neighboring oxygen atoms were 3.1. Thermodynamical Stability of Phosphates. We analyze
detected through a Voronoi construction.38 In a second step, the poly- the thermodynamic stability at zero K and zero pressure of the
hedron formed by this set of neighbors was compared to a library of known ICSD compounds. For each of those known compounds,
perfect local environments (containing environments such as octahe- the energy above the hull was computed. It should be noted that
dral, square planar, tetrahedral, square pyramidal, trigonal bipyramidal, it is possible for a known compound to be unstable in this analysis
trigonal prismatic square face monocapped, trigonal, and square anti- (i.e., having a nonzero energy above the hull). Besides possible
prismatic). The comparison was performed using the continuous errors from DFT, some phases present in the ICSD are prepared
symmetry algorithm developed by Pinsky et al.39 at high temperature (and sometimes at high pressure con-
2.4. Battery Properties Computations. Voltages were com- ditions), stabilizing a compound that would be unstable at zero
puted directly from a difference in total energy between the lithiated and K and zero pressure. Moreover, this analysis is purely thermo-
delithiated state following the methodology presented in Aydinol et al.40 dynamical, and compounds can be synthesized while metastable.
3496 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

Table 1. Predicted New (with respect to the ICSD database) Stable Phases at Zero Ka
formula space group based on based on ICSD nb exp. evidence

LiSn2(PO3)5 Pc(7) LiPb2(PO3)5 61207 none

LiSb(PO3)4 C2/c(15) LiYb(PO3)4 67705 none
LiCr4(PO4)3 Pnma(62) Cd4Na(PO4)3 8115 none
Li2Cr(P2O7) C2/c(15) Li2Cu(P2O7) 72485 none
Li3Mo2(PO4)3 R3(148) Li3Fe2(PO4)3 95973 yes53
LiCo(P2O7) P 21(4) LiSc(P2O7) 91496 none
LiCo(PO4) Pn21a(33) LiZn(PO4) 79352 none
LiBi(PO3)4 C2/c(15) LiYb(PO3)4 67705 none
LiCr(P2O7) P 21(4) LiSc(P2O7) 91496 ICSD: 240965
LiMn(P2O7) P 21(4) LiFe(P2O7) 63509 ICSD: 415153
LiMo2(PO4)3 P21/c(14) Li3Fe2(AsO4)3 90880 yes53
LiV2(PO4)3 R3c(167) LiTi2(PO4)3 95979 none
LiCo(PO3)4 Pbcn(60) LiAl(PO3)4 74860 none
LiFe(PO3)4 Pbcn(60) LiAl(PO3)4 74860 yes52
LiCr6(P2O7)2(P3O10) P21/m(11) NaMn6(P2O7)2(P3O10) 280928 none
LiTi(PO3)5 P21/m(11) Na2Mn(PO3)5 91544 none
LiV(PO3)4 Pbcn(60) LiAl(PO3)4 74860 none
LiTi(PO3)4 Pbcn(60) LiAl(PO3)4 74860 none
Li3Cr2(PO4)3 R3(148) Li3Fe2(PO4)3 95973 yes53,54
Li9Cr3(P2O7)3(PO4)2 P3c1(165) Li9Fe3(P2O7)3(PO4)2 50958 ICSD:417755
Li2Fe3(P2O7)2 P21/c(14) Li2Ni3(P2O7)2 83662 yes51
LiMn(PO3)4 Pbcn(60) LiAl(PO3)4 74860 none
LiMn2P3O10 P21/m(11) LiCo2P3O10 83582 none
LiCo(PO3)3 P 212121(19) LiMn(PO3)3 51632 ICSD:153515
LiCr(PO3)4 Pbcn(60) LiAl(PO3)4 74860 yes52
LiMo(PO3)4 Pbcn(60) LiAl(PO3)4 74860 none
LiMo2(PO4)3 R3(148) LiSn2(PO4)3 83832 yes53
a
We indicated the Hermann-Maugin notation and the space group number of the phase along with the formula and number of the ICSD entry on which
the new compound is based (e.g., through ionic substitution). Previous experimental evidence refers to that mentioned in the literature without a precise
crystal structure assignment and to entries entered after 2006 in the ICSD.

However, it is expected that for existing compounds, any instability A similar thermodynamical analysis has been performed for
from the zero K convex hull should be in a reasonable energy the compounds proposed by the substitution algorithm.35
range. Among the 37 compounds containing at least Li, P, O, Twenty-seven novel phases were found to be on the convex hull
and a redox active metal, and which are reported with no partial and thus computationally stable at zero K. Table 1 shows the
occupancies in the ICSD, only six have a distance to the hull thermodynamic data for those compounds. A few compounds
larger than 10 meV/at (see Table 2 of the Supporting In- have been inserted in the ICSD after 2006 and were not in our
formation). Obvious reasons for metastability at zero K could working version of this database (LiCo(PO3)3, LiCrP2O7,
be identified for four of those six compounds. The NASICON LiMnP2O7, and Li9Cr3(P2O7)3(PO4)2). Li2Fe3(P2O7) has re-
LiV2(PO4)3 is formed by electrochemical delithiation and was cently been synthesized but is not yet entered in the ICSD.50,51 It
never synthesized directly.45 The LiCu(PO3 )3 P1 structure is is encouraging that in each case the correct crystal structure was
probably also a metastable compound as it is known to found by our compound prediction scheme. A few compounds of
transform to the LiCu(PO3)3 orthorhombic form upon formula LiM(PO3)4 based on the LiAl(PO3)4 crystal structure
heating.46,47 The LiFePO4-β is a high-pressure phase higher are also predicted. Evidence for their synthesis with M = Fe and
in energy than the olivine structure.48 The energy above the Cr are presented in Murashova et al.52 We also predict the
hull of Li2Mn2P6O18.10H2O is extremely high (400 meV/at), existence of two molybdenum compounds with the NASICON-
and while it is possible that this lack of stability results from like structure (Li3Mo2(PO4)3 and LiMo2(PO4)3). Although not
inaccuracy in the representation of water in our database, it is referenced in the ICSD, synthesis of those compounds has been
more likely due to a missasignment of the hydrogen positions claimed in the literature.53 Similarly, the Li3Cr2(PO4)3 NASI-
in the ICSD structure, as is hinted by the large ionic relaxa- CON is absent in the ICSD but referenced in several papers.53,54
tions observed during the DFT computations.49 With more This demonstrates the ability of a substitution algorithm coupled
accurate hydrogen positions, this structure is likely to have a with a DFT phase stability analysis to discover new phases
much lower calculated energy. Surprisingly, LiCoPO4 olivine beyond the ICSD 2006 data set.
is found to be 24 meV/at above a predicted LISICON 3.2. Preferred Local Environments. Our large data set of
structure in which the transition metals are tetrahedrally computed phosphates can be used to study the local environ-
coordinated. ments that each ion favors. Figure 1 plots for each ion in different
3497 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

Figure 2. Best achievable gravimetric capacity for different M(n
1)+/Mn+

Figure 1. Lowest distance to the convex hull for different coordination couples as a function of the mass-to-charge ratio of the (poly)anion.
environments of each ionic species in our phosphate data set. Each Oxides and different phosphate polyanions are indicated on the mass-to-
symbol refers to a different local environment: O6, octahedral; S4, charge ratio axis. For the sake of comparison, this plot assumes the mass
square planar; T4, tetrahedral; S5, square pyramidal; T5, trigonal of iron for the M element.
bipyramidal; T7, trigonal prismatic square face monocapped; T6,
trigonal; and S8, square antiprismatic. An energy above the hull of 0 knowledge (even in low temperature synthesis conditions).58 To
meV/at indicates that the environment is found in a compound verify the accuracy of this result, we assessed the LiCoPO4
thermodynamically stable at 0 K. phase stability with GGA with different U values, with the
Heyd
Scuseria
Ernzerhof (HSE) hybrid functional and in an
coordination environments the lowest energy above the hull that antiferromagnetic state (Supporting Information).26 Even though
is found in our data set. Each symbol indicates a different local these additional computations come with some different physical
environment. Overlap between symbols can happen when two approximations, they all led to the same conclusion placing the
local environments are present in the same compounds or in LISICON structure lower in energy than the olivine. This
compounds with similar stability. Zero energy above the hull theoretical result suggests that a tetrahedral LiCoPO4 might be
indicates that at least one stable compound exists with this synthesizable. If direct synthesis is not successful, an alternative
coordination environment. synthesis route could be to form NaCoPO4, which is known to
The vast majority of transition metals are stable in octahedral form under certain conditions with tetrahedrally coordinated
environments (blue circles). For some +2 ions such Fe, Mn, and cobalt and perform a Li
Na ion-exchange.59
Co, trigonal bipyramidal environments (which can also be seen 3.3. Capacity Limits in Phosphates. The amount of charge
as distorted square pyramidal environments) exist as well (red extractable from an electrode material per unit weight is the
crosses). The four ICSD compounds: Fe2P2O7, Ni2P2O7, gravimetric capacity. It is possible to place general bounds on the
Co2P2O7, and Co3(PO4)2, as well as its predicted isostructural gravimetric capacity of compounds of formula LixMyXz, where M
Mn3(PO4)2, are all compounds with mixed octahedral and trigonal is a redox active element and X is a (poly)anion. The maximum
bipyramidal environments. This type of mixed environment achievable capacity for a one-electron delithiation process de-
compounds are also observed in lithium-containing phosphates pends on the one-electron couple used (M1+/M2+, M2+/M3+,...),
in the recently discovered Li2FeP2O7 and Li2MnP2O7.55,56 the mass of the element M, and the charge to mass ratio for X. For
Both Cu2+ and Cr2+ are stable in square planar environments instance, assuming an orthophosphates compound (X =
such as in the ICSD Li2CuP2O7 compound and the isostructural (PO4)3
) and a Fe2+/Fe3+ couple, the best capacity achievable
predicted Li2CrP2O7. Mixed octahedral and square planar are is around 170 mAh/g (this corresponds to the LiFePO4
also possible such as in Cu3(PO4)2. stoichiometry). Figure 2 plots the maximum capacity for differ-
Larger ions such as Sb3+, Bi3+ ,and Sn2+ show higher coordina- ent one-electron redox couples (the different colored lines) as
tion environments (8- or 7-fold coordination). On the contrary, function of the charge-to-mass ratio of the (poly)anion X. For the
when stable, their oxidized equivalent (e.g., Sb5+ and Sn4+) form sake of comparison, this plot assumes the mass of iron for the M
in octahedral environments because of their smaller size. element, which is a reasonable average for the 3d transition
Few tetrahedrally coordinated redox active elements are found metals. Figure 2 highlights the importance of the cation charge
to be stable in phosphates. Only the trigonal phase of FePO4 and the charge-to-mass ratio on the anion. For a high valent
contains a Fe3+ in a tetrahedrally coordinated environment. cation redox couple (e.g., M5+/M6+), a large amount of anion
No stable tetrahedral redox active element is found in a lithium mass is required in the compound to compensate, thereby
containing phosphate. decreasing the specific capacity. Hence, from a specific energy
However, tetrahedrally coordinated structures such as the perspective, low valent redox couples, with reasonably high
LISICON-like structures formed by LiZnPO4 can be very voltage and combined with (poly)anions with high negative
energetically competitive for some ions such as Fe2+ and Mn2+ charge-to-mass ratio, are optimal. Different classes of phosphates
(only 5 to 9 meV/at above the stable olivine structure). For Co2+, are represented in Figure 2 at different charge-to-mass ratio:
we predict the cobalt LISICON LiCoPO4 to be significantly orthophosphates (X = (PO4)3
), pyrophosphates (X =
more stable than olivine by 24 meV/at. While it has been shown (P2O7)4
), and metaphosphates (X = (PO3)
) (see Methods
that other polymorphs can be stabilized under high pressure,57 section for the classification). For comparison, the pure oxides
they all have octahedrally coordinated Co, and no report of a are also indicated on the graph (X = O2
). All the phosphates
phase with tetrahedrally coordinated Co exists to the best of our show lower maximal capacities than the oxides. For instance, for
3498 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

the couple M2+/M3+, the best achievable capacity for oxides is

around 300 mAh/g, while that for the orthophosphates is around
170 mAh/g. Moreover, condensation of the PO4 groups tends to
increase the mass-to-charge ratio and to decrease dramatically
the maximal capacity. For the couple M2+/M3+, the capacity is
reduced to 100 mAh/g for metaphosphates (X = (PO3)
).
Oxyphosphates are situated between the oxides and orthopho-
sphates in terms of mass-to-charge ratio and have therefore a
higher achievable maximal capacity than ortho or condensed
phosphates.
3.4. Voltage Data Mining. 3.4.1. Voltages Range. The voltage
delivered by cathode materials with respect to a certain anode
(here lithium metal) is a critical battery property. The voltage in-
fluences directly the energy stored per unit mass and volume
(i.e., the specific energy and energy density). However, a limit to
this voltage is imposed by the electrolyte. Current commercial
electrolytes are not stable on long-term cycling above a voltage
threshold situated around 4.5 V (versus lithium metal). In this
work, all voltages are referenced to lithium metal.
We computed the delithiation voltage for all of the quaternary
lithium-containing phosphates in our data set. Only compounds
reasonably stable in their lithiated state (energy above hull
<50 meV/at) were considered. The amount of lithium removed
corresponds to a one-electron oxidation of the redox active
element. The delithiation and computational procedure is pre-
sented in the Methods section. In total, our data set contains
651 voltage computations, on which we performed statistical
analysis.
Figure 3. Voltage distribution for different redox couples in phosphates.
The first parameter influencing the voltage is the active redox Red circles indicate the average voltage calculated for all ternary oxides in
couple.60 A cathode working on the Ni2+/Ni3+ couple is expected the ICSD database. The blue bar indicates the range that captures 90% of
to deliver a much higher voltage than one operating on Ti3+/Ti4+. all phosphate compounds with that redox couple. The red dashed line at
We assumed a normal distribution of the voltages around a 4.5 V indicates the practical electrolyte limit. The couples are sorted by
different mean for each redox couple. The standard deviation the mass of their element.
takes into account the effect of the specific crystal structure and is
considered common to all couples. The mathematical expression
for the fitted normal distribution is voltage for the M(n
1)+/Mn+ couple. Therefore, the presence of
the phosphate group has the effect of increasing the voltage
VMðn
1Þþ =Mnnþ ¼ NðμMðn
1Þþ =Mnþ , σÞ ð1Þ (above that of a “pure” oxide).
Interestingly, according to our data, this inductive effect seems
where N represents a normal distribution and μM(n
1)+/Mn+ is the nonuniform across couples. Couples such as Ti3+/Ti4+ and Fe2+/
mean value for the M(n
1)+/Mn+ redox couple. A maximum Fe3+ have small inductive effects, whereas couples such as Mn2+/
likelihood fit to the data determined a standard deviation σ of Mn3+ and Mn3+/Mn4+ have higher inductive effect. The lack of
0.49 V and the mean value for each couple. Figure 3 represents anion effect observed in the case of titanium Ti3+/Ti4+ is con-
for each redox couple the mean voltage with a bar indicating the sistent with early calculations comparing oxides, sulfides, and
range that captures 90% of all compounds with that redox couple. selenides40 but somewhat in contradiction with the phosphate
The mean voltage for the ternary oxides present in the ICSD are literature. Indeed, the important difference in voltage between
indicated with red dots for comparison. The 4.5 V voltage limit lithium insertion in titanium NASICON (2.5 V) and lithium
for current commercial electrolytes stability is indicated by the insertion into the spinel-related Li4Ti5O12 (1.6 V) has been
red dashed line. reported as an evidence for the inductive effect.62 Our data shows
The average voltage for each element increases with oxidation to the contrary that on average the inductive effect is not very
state (e.g., the average voltage for Co3+/Co4+ is higher than for pronounced for the Ti3+/Ti4+ couple. Our average voltage for
Co2+/Co3+). The only exception is tin, for which the average Sn2+/ oxides with titanium is higher than 1.6 V and lies closer to 2.3 V.
Sn3+ voltage is higher than Sn3+/Sn4+, indicating that Sn3+ is Lithium insertion in spinel-based structures tends to be low in
not stable and will disproportionate into Sn2+ and Sn4+. voltage for structural reasons. Lithiating Li4Ti5O12 creates very
Figure 3 shows that phosphates exhibit higher voltages than strong electrostatic Li+
Li+ interactions, which reduces the
oxides for a similar redox couple. This fact is well-known in the intercalation voltage, and the focus of the comparison on
battery field and is explained by the inductive effect.61 The insertion in Li4Ti5O12 has somewhat biased the analysis in the
phosphorus ion sharing a common oxygen with the transition literature. Lithiation of TiO2 anastase proceeds at a higher
metal M, through a M
O
P link, influences the covalency of voltage (slightly lower than 2 V),63 and ramsdellite LiTi2O4
the M
O bond. The electronegative phosphorus attracts elec- has recently been reported experimentally as deintercalating at a
trons from the oxygen and weakens the covalent metal to oxygen even higher potential (2 to 2.5 V).64 Both those results are much
bond. The weaker the M
O covalent bond, the higher the closer to our computed average voltage.
3499 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

The inductive effect is one of the motivations for the interest in

polyanionic systems. Couples often too low in voltage in oxides,
such as Mn2+/Mn3+, V3+/V4+, or Fe2+/Fe3+, can be higher than
3 V in phosphates. However, this effect can be detrimental for
certain couples. For instance, Mn3+/Mn4+, Co3+/Co4+, Ni2+/
Ni3+, and Ni3+/Ni4+, while used in commercial oxide-based
lithium ion batteries, are too high in voltage to be of interest in
phosphates with the current electrolyte technology.
Before comparing our results to the experimental literature, we
should note that our data set consists only of voltages obtained
from the delithiation of rather stable lithiated compounds. No
voltage from lithium insertion in stable compounds is used.
There is a technological justification for this choice. As the
standard commercial anode (carbon) is made without lithium,
there is a need for the cathode to be the lithium source of the
battery. This choice influences somewhat the obtained voltage Figure 4. Mean voltage in phosphates versus maximum gravimetric
range. Voltages obtained from lithium insertion into stable capacity achievable. Specific energy curves at 600 and 800 Wh/kg are
drawn on the figure (blue dashed lines). The red dashed line indicates
compounds are usually lower than voltages obtained by delithia-
the upper voltage which we consider safe against decomposition of the
tion of stable compounds (in similar chemistries and activating electrolyte. Different colors and markers have been used for different
the same redox couple). Indeed, as the voltage is the difference elements.
between the energy of the charged and discharged state, inserting
Li in stable compounds will lead in general to lower voltage than side. The V3+/V4+ couple has been studied experimentally in
delithiating stable compounds. For instance, the Fe2+/Fe3+ redox many different structures from the Li3V2(PO4)3 monoclinic
couple can be activated in pyrophosphate by inserting lithium (3.8 V)7,45 and rhombohedral (3.8 V)71 NASICON to the
into the stable LiFeP2O7 or by removing lithium from the stable pyrophosphate LiVP2O7 (4.1 V).62 All those experimental results
Li2FeP2O7. The crystal structures for those two compounds are are in agreement with our predicted voltage range. An outlier is
different, and the lithium insertion into the stable charged state in found in Li4VO(PO4)2 inserting at 1.5 V, indicating most likely a
LiFeP2O7 is measured to be 2.9 V,61 while the delithiation
significant polarization.72 The position of the V4+/V5+ couple, on
voltage from the stable discharged state Li2FeP2O7 is higher at
the other hand, has been more polemical in the experimental
3.5 V.55
literature. Our computations locate the couple on average at
The Fe2+/Fe3+ couple has been well studied in phosphates.
4.2 V. However, extraction of the last lithium in the NASICON
This couple is reported at 2.8 V for insertion in LiFe2(PO4)3
NASICON, at 2.9 V for insertion into LiFeP2O7, at 3.1 V in (Li1V2(PO4)3 f V2(PO4)3), which requires the oxidation of
amorphous Fe4(P2O7)3.4H2O,61 and at 3 V in the high-pressure vanadium to +5 was believed early on to occur at a potential
form of FePO4.65 The higher voltage of 3.5 V in olivine LiFePO4 higher than 4.5 V.73 More recent experimental results show a
seems to be unusual among this data set but is close to our significant hysteresis between the charge and discharge of the last
computed average. More recent experimental results on delithia- lithium in NASICON with a charge around 4.6 V but a discharge
tion compounds working on the Fe2+/Fe3+ couple such as much lower at 4.1 V.7 A mechanism based on charge-ordering
Li2FeP2O7 at 3.5 V,55 Li2Fe3(P2O7)2 at 3.4 V,50 and LiFe2- asymmetry has been proposed by Yin et al. to explain this
(P3O10) at 3.2 V66 indeed show higher voltages. behavior.45 The activity of V4+/V5+ in other compounds such as
The Ti3+/Ti4+ couple has been identified around 2.5 V when in the three polymorphs of LiVOPO4 oxyphosphates, R (3.7 V),74
inserting into the NASICON structure: LiTi2(PO4)3.67,68 A β (4 V),75 ε (4 V),76 and LiVOP2O7 (4.1 V)77 are all in
potential at 2.6 V is found for insertion into TiP2O7.62 Hydro- agreement with our computational voltage range.
genated phosphates as Ti(HPO4) have also been inserted at a There is less data for high voltage couples as issues with
voltage around 2.5 V.69 All those experimental results fit well with electrolyte decomposition often obscure delithiation. The high
our computed average of 2.6 V for Ti3+/Ti4+. voltage electrochemical tests reported for the nickel and cobalt
The Mn2+/Mn3+ couple is less studied with a strong focus on olivines confirm the high voltage associated with the two Co2+/
LiMnPO4 measured at 4.1 V.5 Recent results on Li2Mn(P2O7) Co3+ and Ni2+/Ni3+ couples. LiNiPO4 has been measured at
showed a voltage around 4.5 V.56 Both voltages lie in our pre- 5.2 V78 and LiCoPO4 at 5.0 V and 4.8 V.79,80 Some experimental
dicted voltage range. results have recently appeared on the Mn3+/Mn4+ couple. Those
Experimental measurements of redox active chromium in experimental results based on LiMnPO4 doped with Mg (4.8 V),81
phosphates is scarce. Patoux et al. mention the inactivity of the and on Li2Mn(P2O7) (5.3 V)56 confirm the high voltage com-
Cr2+/Cr3+ couple in the Li3Cr2(PO4)3 and Li1TiCr(PO4)3 puted for this couple. Our computations can provide an explana-
NASICONs.54 A claimed measurement through cyclic voltame- tion for some unsuccessful electrochemical delithiation attempts.
try of the Cr3+/Cr4+ couple in LiCrP2O7 between 3.1 and 3.5 V For instance, LiCo(PO3)3 was tested electrochemically up to 5 V
seems extremely low but would correspond with our voltage but did not show any activity as expected for the high voltage of
range for Cr2+/Cr3+.70 The data might have been wrongly inter- the Co3+/Co4+.82 Similarly, the failed attempt to extract lithium
preted as a measurement of the voltage during delithiation, while from LiFeP2O7 can be attributed to the high voltage of the Fe3+/
the actual process was a lithium insertion into LiCrP2O7. Fe4+ couple.83
Vanadium is an extensively studied element in phosphates. We predict the average voltage for the Mo3+/Mo4+, Mo4+/
The V2+/V3+ couple is reported in NASICON at 1.7 V60 and at Mo5+, and Mo5+/Mo6+ couples to lie exactly in the 3
4.5 V
2 V for LiVP2O7.62 Both lie in our voltage range but on the lower voltage window. It is therefore surprising that very few experimental
3500 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

reports on molybdenum electrochemistry in phosphates have

been published. To our knowledge, only lithium insertion in
(MoO2)2P2O7 has been reported at 2.6 V.84 Unfortunately, the
experimental results indicate amorphization during lithiation and
are difficult to compare to computations which assume topotac-
tic insertion.
3.4.2. Specific Energy Analysis. The data mining of the voltage
(Figure 3) can be used in conjunction with our previous analysis
on best gravimetric capacities achievable (Figure 2) to detect
couples that may lead to interesting specific energies. Figure 4
plots the mean voltage for each couple with respect to the best
achievable capacity in a phosphate (excluding oxyphosphates) as
computed in section 3.3 and in Figure 2. The specific energy is
the product of the voltage and the capacity; specific energy curves
at 600 and 800 Wh/kg are drawn on the figure (blue dashed
lines). The red dashed line indicates the upper voltage that we
consider safe against decomposition of the electrolyte. For
comparison, the theoretical specific energy of LiFePO4 olivine Figure 5. Computed voltage for different redox couples in six different
is 600 Wh/kg. crystal structure prototypes. Each color represents a different crystal
Using this plot, couples such as Sn3+/Sn4+, Mo3+/Mo4+, Mo4+/ structure prototype. The prototypes are labeled by one of the com-
Mo5+, Mo5+/Mo6+, Sb3+/Sb4+, and Sb4+/Sb5+, while of interest pounds that forms this crystal structure prototype as indicated in the
ICSD. The rank in voltage is preserved across different redox couples.
for their voltages, have to be excluded (as one-electron systems)
High or low voltages crystal structures are the same for all redox couples.
because of their high atomic weights. The one-electron redox
couples that are likely to improve the gravimetric specific energy
3.4.3. Factors Determining the Voltage. The considerable
over LiFePO4 (600 Wh/kg) and are below 4.5 V are the Fe2+/
spread of the voltage around its average for a given redox couple
Fe3+, Mn2+/Mn3+, and Cu1+/Cu2+ couples. While there is
can be related to the effects of the crystal structure. Padhi et al.
extensive work on the two first couples in LiFePO4 and LiMnPO4,
previously studied the influence of the crystal structure on four
to our knowledge there has been no experimental work on
different Fe2+/Fe3+ phosphate cathode materials.2
phosphate Cu1+/Cu2+ compounds. The Cu1+/Cu2+ couple not
To investigate the effect of structure, we calculated many of the
only exhibits a voltage likely to fit in our targeted voltage window,
couples in seven different crystal structures. Figure 5 plots the
but also benefits from a higher achievable capacity because of its
voltage obtained for different redox couples (all +3/+4 couples)
low oxidation state (Figure 2).
and for seven different crystal structure prototypes. Each colored
It is challenging to find a one electron phosphate compound
line corresponds to a different structure prototype. Each proto-
competitive in terms of specific energy and located in the 3
4.5 V
type is labeled by a compound that forms this crystal structure as
voltage window. The remaining strategy to increase the specific
given in the ICSD database. Across the different couples, the
energy is to consider two-electron couples. Both of the couples
voltage ranking between crystal structures is conserved: the high
present in the two-electron transition should both lie in the
voltage and low voltage prototypes are similar for all redox
3
4.5 V window. From Figure 4, molybdenum (with three
couples. For instance, the purple line in the figure shows the
couples in the voltage window: Mo3+/Mo4+, Mo4+/Mo5+, and
voltages in the structure type of LiAl(PO3)4. Those compounds
Mo5+/Mo6+) and vanadium (two couples: V3+/V4+ and V4+/V5+)
tend to be very high in voltage. On the other hand, the light blue
are the only two elements with acceptable two-electron voltages.
line, related to compounds isostructural with NaAl(P2O7), is
The heavier weight and lower voltage for molybdenum com-
very low in voltage. Moreover, the spread between the highest
pounds compared to those of vanadium argues against them. For
voltage structure and the lowest one is rather equivalent between
instance, while we predict a molybdenum Li3Mo2(PO4)3 NASI-
redox couples confirming our previous assumption of a common
CON to be stable, its specific energy (532 Wh/kg) compared
standard deviation among couples.
to the theoretical specific energy for the vanadium NASICON
A statistical view on the influence of prototype and redox
(722 Wh/kg) is 25% lower.
couple on the voltage versus lithium anode can be obtained using
Among the nontransition metals, we notice that Sb3+/Sb5+ is a
an analysis of variance (ANOVA).85 The ANOVA analysis is a
two-electron redox couple that should be considered, even though
statistical tool used to determine if one or several given factors are
the weight of antimony mitigates the benefit of using a two-
statistically influencing a variable and how much of the variable
electron system.
variance can be explained by this factor.
Experimental effort is ongoing in finding phosphates capable
The ANOVA analysis relies on the assumption that a voltage
of activating two electrons using a M2+/M4+ couple. For instance,
k for a given redox couple i and a given crystal structure prototype
Li2M(P2O7) (M = Fe, Mn, V) has recently been targeted as a
j can be modeled as
two-electron system with the promise of high capacity.55,56 Our
voltage analysis indicates, however, that it will be extremely Vijk ¼ μ þ Ri þ βj þ εijk ð2Þ
difficult to find a two electron M2+/M4+ phosphate compound
with the current electrolyte technology. From all the possible where μ is a constant, Ri represents the deviation to μ due to the
couples, either the M2+/M3+ couple is too low (e.g., V2+/V3+), redox couple, βj indicates the deviation to μ due to the crystal
the M3+/M4+ couple is too high (e.g., Mn3+/Mn4+, Fe3+/Fe4+, structure prototype, and εijk is a normally distributed disturbance.
Cr3+/Cr4+), or both couples are too high (e.g., Ni2+/Ni3+ and We used the ANOVA analysis to evaluate if the chemistry and the
Ni3+/Ni4+). crystal structure are statistically significant in determining the
3501 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

Figure 6. Ab initio computed voltage as a function of the voltage

computed by electrostatics alone. The electrostatic term was computed
using an Ewald sum on the ab initio relaxed structure. The different
colors indicate different redox couples. While redox couples span Figure 7. Voltage Z-score in function of the phosphorus-to-oxygen
different range of voltages, the highest electrostatic voltages correspond ratio (P/O). The different regions of the phosphates classification,
to the highest ab initio voltages. oxyphosphates, orthophosphates, and condensed phosphates, are in-
dicated. Higher phosphorus-to-oxygen ratio are correlated with higher
voltage and how much of the variance observed in the computed voltages. The blue dashed line is to aid the eye.
voltages these two factors can explain.
Not surprisingly, performing this ANOVA analysis shows that
both redox couple (chemistry + oxidation state) and crystal The lower voltage observed for V4+/V5+ during delithiation of
structure have a significant statistical influence on voltage (both LiVOPO4 compared to that of the NASICON delithiation was
with a high statistical significance of P < 0.0001, see Supporting explained by the smaller number of PO4 groups connecting the
Information). The total variance can be estimated to be 47
66% vanadium octahedron for the oxyphosphate (four links) com-
from the redox couple and 21
33% from the crystal structure. pared to the NASICON (six links).6 The analysis of our data set
Hence, the majority of the observed variance can be explained by did not show any influence of the numbers of PO4 groups linked
a combination of the redox couple and the crystal structure to the redox active element (Supporting Information).
prototype, as previously observed computationally by Arroyo-de Padhi et al., on the other hand, used the length of the P
O
Dompablo et al.86 bond connected to the redox active element as an indicator of a
It is of interest to not only relate the voltage to structure more or less pronounced inductive effect and used this parameter
prototype but also to understand what particular structural factors to rationalize the difference in voltages between four phosphate
influence the voltage. In ionic compounds such as phosphates, cathodes working on the Fe2+/Fe3+ couple.61 We found that a
the electrostatic energy should influence the energetics and relation between the voltage and the P
O bond length does
therefore the voltage.61 To test this hypothesis, we computed indeed exist (Supporting Information).
an “electrostatic voltage” by computing the voltage obtained by The inductive effect can be also estimated by the phosphorus-
the difference in electrostatics through a simple Ewald summa- to-oxygen ratio. More phosphorus compared to oxygen should
tion on the formal charges. Figure 6 plots the ab initio computed provide a higher inductive effect. Figure 7 plots the Z-score in
voltage versus this electrostatic voltage for a few redox couples. function of the phosphorus-to-oxygen ratio. The orthopho-
While there is significant scattering, a relation can be observed: sphates, condensed phosphates, and oxyphosphates are indicated
the higher the electrostatic voltage, the higher the ab initio on the P/O axis. The amount of phosphorus present in the
voltage. compound influences the voltage, with higher voltages observed
To discover what other factors influence the voltage besides for higher P/O ratio, but the scatter is significant. Condensed
the redox couple and the electrostatics, we used the Z-score phosphates have slightly higher voltages than orthophosphates,
(or standard score) defined as which are themselves higher in voltages than oxyphosphates.
Using an ANOVA model combining redox couple, electro-
V
μMðn
1Þþ =Mnþ statics, and phosphorus-to-oxygen ratio, the three factors were
Z¼ ð3Þ shown to be statistically significant. This analysis attributes 76%
σ
of the variance for the redox couple, 6% for the electrostatics, and
where V is a voltage, μM(n
1)+/Mn+ is the mean of the active redox 4% for the phosphorus-to-oxygen ratio (Supporting Information).
couple, and σ is the voltage standard deviation. Converting to the 3.5. Phosphates Volumetric Capacity and Energy Density.
Z-score allows the different redox couples to be compared on the Volumetric quantities are as important as gravimetric ones for
same footing. To investigate whether other factors have statisti- certain battery applications (e.g., portable electronics). In addi-
cally significant effects on the voltage, one can plot the Z-score tion, volumetric energy densities are linked more directly to the
versus this factor. price per energy stored than specific energy.87
A factor often used to rationalize differences in voltages for Using our data set, we plotted in Figure 8 the volumetric
the same redox couple in phosphates is the inductive effect. capacity versus the gravimetric capacity for phosphates (red circles).
Gaubicher et al., for instance, used the number of PO4 groups A linear regression curve can be fitted to this data (red line),
linked to the redox active element to rationalize differences in showing an average density for phosphates around 3.1 kg/l. For
voltages between different crystal structures at the same couple. comparison, the theoretical energy densities of the oxides present
3502 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

Figure 9. Critical oxygen chemical potential of the charged cathode as a

function of the transition metal valence state for the phosphates
investigated. The color of each circle represent the voltage. When
multiple charged compounds have the same transition metal valence
and critical oxygen chemical potential, the voltage was averaged over
them in order to keep the graph legible. The red color represents higher
voltages, while the blue color represents lower voltages. The ions are
sorted by their critical oxygen chemical potential starting with the safest
ions. All values at μO2 = 4 eV represent values higher or equal to 4 eV.

of LiFePO4 (2061 Wh/l) due to the higher voltage and compar-

Figure 8. Volumetric capacity and energy density as a function of the ably (but slightly lower) density.
gravimetric capacity for oxides (blue diamonds) and phosphates (red While comparing energy density of oxides and phosphates, we
circles). Linear regression curves have been plotted in blue for the oxides should take into account the higher average voltage in phos-
and red for the phosphates. Only one-electron systems with voltage phates due to the inductive effect. Panel (b) of Figure 8 plots the
lower than 4.5 V and relatively stable in the lithiated state (<50 meV/at energy density versus the gravimetric capacity for our data set. As
above hull) have been considered in the analysis. The theoretical expected, the higher voltage of phosphates makes the phosphate
volumetric and gravimetric capacities for three current cathode materials, and oxide curves closer than when comparing capacities. It also
LiFePO4, LiMn2O4 spinel, and LiCoO2 layered, are indicated in increases the spread. Some cathode phosphates get competitive
the graph.
in terms of volumetric energy densities. For instance, a LiCoPO4
in the Cmcm structure (based on the high-pressure crystal
in ICSD (obtained by DFT) are plotted in blue diamond with a structure of LiNiPO4 and LiFePO4) exhibits a voltage just
linear fit (blue curve). The theoretical volumetric and gravimetric below the cutoff at 4.4 V and a rather good volumetric capacity
capacities for three current cathode materials, LiFePO 4 , (624 mAh/cc). This would lead to a very high volumetric
LiMn2O4 spinel, and LiCoO 2 layered, are also indicated in capacity of 2765 Wh/l. This β-LiCoPO4 phase has been synthe-
the graph. Only one-electron systems were considered in sized by high-pressure methods,57 and it is not surprising to find
this plot. higher density in high-pressure phases. Unfortunately, currently
As discussed previously, one-electron oxides can reach higher known high-pressure phosphate phases have suffered so far from
gravimetric capacities than phosphates (up to 300 mAh/g versus low lithium mobility (e.g., in β-LiFePO4).65 However, on
170 mAh/g for phosphates). In addition, for a similar gravimetric average, the voltage increase is not strong enough (especially if
capacity, oxides tend to exhibit higher energy density. This is due only voltages lower than 4.5 V are taken into account) to
to a higher density for oxide materials in comparison to phos- compensate for the difference in densities between oxides and
phates. This difference in density is linked to the difference in phosphates. For the same gravimetric capacity, an oxide has
crystal structures formed by phosphates and oxides. Oxides typically an energy density 30% higher than a phosphate.
generally form close-packed structures, whereas phosphates tend 3.6. Safety and Thermal Stability. Safety is one of the key
to be stable in open frameworks such as NASICONs. In our data considerations in the design of lithium-ion battery cathodes.
set, for the same gravimetric capacity, the oxides will have on Charged cathodes can be a safety hazard. A typical runaway
average a 37% higher volumetric capacity. Among the phos- reaction in a Li battery starts by an overheating event (e.g., caused
phates, the olivine structure is actually very dense, leading to by an internal or external short) that causes a chain reaction of
volumetric capacities around 600 mAh/cm3. It would be difficult exothermic processes, typically initiated on the anode side, but a
to obtain denser phosphate materials than the olivine com- critical process is the ability of the charged cathode to release
pounds. The only option to improve the energy density in oxygen and combust the flammable organic electrolyte, ulti-
one-electron phosphates compared to LiFePO4 would be to mately leading to fire.88,89 Safe battery materials require a
use an equivalently dense structure but to increase the voltage. If thermally stable cathode, as well as a nonflammable electrolyte
LiMnPO4 olivine full capacity could be realized, its energy that is electrochemically stable within the operating range of the
density (2224 Wh/l) would indeed be slightly higher than that battery.
3503 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

In this work, we focus on the contribution of a phosphate

cathode to the thermal stability. We analyzed the safety of the
phosphates by calculating the critical oxygen chemical potential
at which the charged (delithiated) cathode begins to evolve
oxygen gas (henceforth referred to as μO2) as predicted by first
principles calculations. This methodology was described in
greater details in our previous work that investigated the thermal
stability of LiMnPO4 and LiFePO4.42 It should be noted that in
the present work, we only studied the equilibrium critical μO2.
No consideration has been given to possible kinetic barriers to
decomposition, which could be important for some compounds.
Figure 9 shows the critical chemical potential at which oxygen
would be released from a charged cathode, as a function of the
redox active ion in all the phosphates investigated. The lower the
critical μO2, the more stable the charged cathode is, as a more
reducing environment (i.e., higher temperature for a given
oxygen partial pressure) is required to cause the cathode to evolve
oxygen gas. For reference, the critical μO2 for the charged cathodes
of existing lithium battery cathodes in commercial production,
CoO2 (fully charged LiCoO2) and Mn2O4 (charged LiMn2O4),
are labeled, as well as some of the promising olivine phosphates
with significant research interest. The thermal stability of charged
LiCoO2 is well known to be a significant safety concern.90 The
plot also shows the average voltage associated with the ion. Figure 10. Critical oxygen chemical potential (μO2) for oxides versus
A significant spread of the critical oxygen chemical potential phosphates with the same oxidation state of active ion. The data points
exists for a given ion. Indeed, depending on the exact position of correspond to an average of our data set, and the error bar shows the
the phase in the phase diagram and of its competing phases, the minimum and maximum values observed. Different colors and markers
oxygen chemical potential can be quite different. General con- have been used for different elements.
clusions can however be made for each of the ions and, if we
consider a material with an oxygen chemical potential lower than computed prediction (Co3(PO4)2, Co(PO3)2 and oxygen gas).
Mn2O4 delithiated spinel to be safe, the safe ions are Ti4+, V3+, However, the computed Co2P2O7 is only 1 meV/at less stable
Cr3+, Nb4+, Sn3+, Nb4+, Cu2+, Sn4+, Fe3+, Mo4+, Sb4+, Mo5+, V4+, than a combination of Co3(PO4)2 and Co(PO3)2, which is well
V5+, Mn3+, Mo6+, and Sb5+. within the error tolerance we expect from our DFT calculations.]
Figure 9 shows also that high voltages are associated with This result is of technological significance as those Co and Ni
unsafe, high oxygen chemical potential materials. Huggins al- olivines are sometimes considered as possible high voltage, high
ready suggested this relation between safety and voltage for energy density cathode materials, and our work identifies that
oxides and conversion reactions.91,92 they are associated with major intrinsic safety concerns.87
Our computations are in agreement with the well-known Phosphates in the olivine structure and phosphates in general
exceptional thermal stability of the Fe3+ in delithiated LiFePO4.93 have been often claimed to be intrinsically safer materials than
In addition, DSC experiments on LiV2(PO4)3 showed a better oxides.66,99
103 The rationalization behind the intrinsic phos-
thermal stability for this V4+ phosphate than that for Mn2O4 phate thermal stability invokes the strength of the phosphorus

delithiated spinel, in agreement with our computations.8 oxygen bond. As this P
O bond is strong, it is rationalized that it
Of the three alternative olivine compounds being heavily should be very difficult to remove oxygen from a phosphate
researched LiMPO4 (where M = Mn, Co, Ni), all are computed compound to form the reduced product(s). To test this state-
to be significantly less thermally stable than LiFePO4. The ment computationally, we compared in Figure 10 the critical
thermal instability of charged LiMnPO4 has already been dis- oxygen chemical potential for oxides and phosphates as a
cussed at length in our previous work and experimental works by function of the oxidation state of their redox active ion.
Kim et al. and Richardson et al.42,94,95 Our analysis shows that The prevalence of points below the diagonal in Figure 10
both LiNiPO4 and LiCoPO4 have even worse thermal stability indicates that most of the phosphates do actually have a similar or
than that of LiMnPO4. Bramnik et al. had previously found worse thermodynamic thermal stability than oxides for the same
experimentally that delithiated CoPO4 released oxygen at a oxidation state.
relatively low temperature of 100
200 °C.96 On the other hand, At equilibrium, the argument for the stability of phosphates
Okada et al. found that Li0.17CoPO4 decomposes at 280 °C, based on the difficulty to break a P
O bond is not valid. Indeed,
which is higher than the 200 °C for charged LiCoO2.97 Regard- this argument would only hold if some dangling P bond would be
less of the disagreement, temperatures found in both works are present in the reduced products. However, the most common
significantly lower than for delithiated LiFePO4. Preliminary oxygen evolution reaction from phosphates involves forming
experimental results by Wang et al. show that LiNiPO4 decom- oxygen gas and accommodating the loss of oxygen in the phosphate
poses immediately to Ni2P2O7 and oxygen gas upon delithiation structure by forming condensed phosphates (e.g., pyrophosphates
at room temperature.98 The predictions from our phase diagrams or metaphosphates). This is, for instance, what happens in many
support both of these experimental observations. [Bramnik et al. olivine compounds: 2 MPO4 f M2P2O7 + 1/2O2
reported Co2P2O7 and oxygen gas to be the decomposition One of the ions with the largest difference between its critical
product for the reduction of CoPO4, which differs from our oxygen chemical potential in oxides and phosphates is Mn4+,
3504 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

The maximum achievable gravimetric capacity for a phosphate

is smaller than for an oxide because of the lower negative charge
per unit wight for the (PO4)3
anion. The higher the phosphorus
content and the degree of condensation of the phosphate groups,
the lower the achievable capacity becomes.
The lower densities from phosphate compounds lead to on
average 37% lower volumetric capacity compared to that of
oxides. The higher voltage achievable in phosphates only slightly
mitigates the impact on the energy density, which is on average
30% below that of oxides with the same capacity.
Our work identifies statistically V3+/V4+, V4+/V5+, Cr2+/Cr3+,
Mn2+/Mn3+, Fe2+/Fe3+, Cu1+/Cu2+, Mo3+/Mo4+, Mo4+/Mo5+,
Mo5+/Mo6+, Sb3+/Sb4+, and Sb4+/Sb5+ as the redox couples of
interest for phosphate-based cathodes in current electrolyte
Figure 11. Voltage versus oxygen critical chemical potential (μO2) for technology (above 4.5 V). The inductive effect is observed in
charged ICSD oxides (blue diamonds) and charged phosphates from our data set with higher voltages identified for phosphates
our data set (red circles). Linear regressions have been added to the plot compared to oxides. This inductive effect makes several couples
to outline the general trends.
that are attractive in oxides, such as Mn3+/Mn4+, Fe3+/Fe4+, Cr3+/
Cr4+, Cu2+/Cu3+ Co2+/Co3+, Co3+/Co4+, Ni2+/Ni3+, and Ni3+/
which is disproportionally more stable in an oxide. For instance, Ni4+, too high in voltage for commercial electrolytes, even
comparing the critical oxygen chemical potential of MnP2O7 and though they might be of interest with advanced electrolyte
Li2MnO3, the phosphate releases oxygen at μO2 = 0.5 eV (i.e., will technology. We should note, however, that this statement is
be unstable in air at 298 K) and the oxide releases oxygen at a statistical, and some of those couples, if present in the right low
much lower μO2 =
5 eV. This difference is striking as Mn4+ is voltage crystal structure, could show activity at voltages lower
used as a stabilizer improving thermal stability in some oxides than 4.5 V. This is especially true for couples with an average
layered compounds.104 voltage very close to the 4.5 V limit such as Mn3+/Mn4+. Besides
The thermal stability of a compound is more driven by the the inductive effect, we find that the electrostatic and the P/O
oxidation state of the transition metal than by the presence of a ratio also influence the voltage.
P
O bond. For instance, Fe3+ is a very thermally stable ion in In terms of safety, the common statement that olivine-based
both oxides and phosphates (in average, μO2 =
5 eV in oxides compounds and phosphates in general are intrinsically safer than
and μO2 =
4 eV in phosphates). However, in phosphates, the oxides is challenged by our results. Thermodynamically, we
Fe2+/Fe3+ couple benefits from the inductive effect and is compute that most phosphate compounds evolve oxygen at
significantly higher in voltage than in oxides. This tendency is lower temperature than oxides for the same oxidation state of
general and while for the same oxidation state phosphates are not the active redox metal. Phosphates can, however, provide in most
more thermally stable than oxides, phosphates can provide a situations cathode materials with higher voltages than oxides for a
higher voltage for an equivalent safety. This is illustrated in similar thermal stability. Hence, the reason that LiFePO4 is so
Figure 11, where the voltage is plotted versus the critical oxygen stable against reduction by the electrolyte should at least in part
chemical potential of the charged cathode. The red circles are be attributed to its low voltage. This is consistent with our
phosphates, and the blue diamonds are ICSD oxides. Two computational and recent experimental evidence that indicates
regression lines have been added to indicate the general trend. that the higher voltages olivine (LiMnPO4, LiCoPO4, and
As observed by Huggins, the general tendency for a higher LiNiPO4) do not share the excellent thermal stability of LiFe-
voltage to be associated with higher critical chemical potential PO4. The inherent safety of phosphates is often extended to
is present for both oxides and phosphates.91,92 However, phos- polyanionic systems in general (e.g., silicates103) and this calls for
phates present in general a higher voltage for a similar thermal a future similar computational investigation of those alternative
stability. On average, we found phosphates to be around 0.5 V polyanionic chemistries.
higher in voltage for a similar thermal stability. If one generalizes to other polyanions the observation made
Our computations only measure the thermodynamic driving on phosphates, high inductive effect polyanions such as sulfates,
force for oxygen release and different chemistries may have fluoro-phosphates, and fluoro-sulfates might provide the highest
somewhat different kinetics. While kinetics probably controls voltage for a given thermal stability.
the oxygen release of the materials that decompose at very low On the basis of our data, the only one-electron couples
temperature (i.e., the very unstable materials) to be viable, presenting acceptable safety combined with voltages in the
cathodes have to be stable up to 200 °C and higher.89 At such 3
4.5 V window and that are lightweight are Fe2+/Fe3+, Mn2+/
temperatures, thermodynamic control is more likely. Mn3+, and Cu1+/Cu2+. The two first couples have been exten-
sively studied in olivine LiMPO4. While iron olivine is definitely a
successful cathode material, the manganese version shows much
4. DISCUSSION poorer electrochemical performances.105 Our work identifies
In this work, we used high-throughout ab initio computations some new opportunities for these redox couples. We find that
to explore the properties of phosphate compounds as lithium-ion energetically competitive tetrahedral structures based on LISI-
battery cathode materials. The intrinsic limits and expected range CON exist. These alternative structures, if they are successfully
of important battery properties, stability, capacity, voltage, synthesized, might provide similar energy densities than the
and safety, have been analyzed as well as some of the factors olivines. The Cu1+/Cu2+ couple is not commonly used in Li
influencing them. batteries. Our analysis shows that the combination of the higher
3505 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

maximal capacity for +1/+2 compounds with the acceptable calculations predict that molybdenum also forms a Li3Mo2-
average voltage around 3.2 V makes it theoretically possible to (PO4)3 NASICON structure. This molybdenum compound
surpass the iron olivine’s 600 Wh/kg specific energy limit. is not very competitive by itself in terms of energy densities
Unfortunately, no stable lithium-containing Cu1+ phosphate is (526 Wh/kg and 1732Wh/l). However, by replacing one-quarter
known, and none was predicted to be stable in this work. The of the nonactive vanadium in Li3V2(PO4)3 by Mo3+, it might be
stability analysis showed a Cu1+ phosphate in tetrahedral co- possible to extract this last lithium with lower hysteresis. This
ordination at composition Li5Cu(PO4)2 in a LISICON structure strategy suggests a Li3Mo0.5V1.5(PO4)3 compound with a com-
as closest to the hull. This compound lies 20 meV/at above its petitive theoretical specific energy of 654 Wh/kg (3.5 V average
decomposition products: Cu, Cu2O, Li3PO4, and Li2CuP2O7. voltage with a capacity of 187 mAh/g). The energy density is also
The Li3Cu3(PO4)2 would be more interesting in terms of attractive at 1900 Wh/l. Both specific energy and energy density
gravimetric capacity and is computed to be 54 meV/at above are competitive with LiFePO4 (600 Wh/kg and 1900 Wh/l). In
the hull in the LISICON structure. Two potential issues with a addition, computations predict that molybdenum would mix
battery working on the Cu1+/Cu2+ couple would be the very easily in the vanadium NASICON structure. The best ordering
different preferred local environment for the Cu1+ and Cu2+ ions for Li3Mo0.5V1.5(PO4)3 is 4 meV/at (or 40 meV/transition metal
(tetrahedral and octahedral or square planar) and the high sites) above the hull and within the range where entropy effects
mobility of Cu1+. Both effects could make it difficult to retain will stabilize the mixing. This example illustrates how a high-
the crystal structure upon cycling. throughput computational study can be used to help the rational
Because of the limitations on the gravimetric and volumetric design of new battery compounds.
capacity of phosphates, two-electron systems are of strong It is remarkable that for all properties investigated, except the
interest with this anion chemistry. Our analysis shows that few voltage, LiFePO4 is close to the optimal of what could be
elements could provide two-electron activity in the 3
4.5 V expected from a one-electron phosphate. The 170 mAh/g
voltage window. No M2+/M4+ redox couple was found to satisfy capacity is the highest achievable (except for oxyphosphates).
those requirements. Only molybdenum (Mo3+ to Mo6+) and Olivines are among the most dense phosphate compounds with
vanadium (V3+ to V5+) show redox couples in the targeted volumetric capacities around 590 mAh/cm3, and the safety of the
voltage window with high enough achievable gravimetric capa- delithiated iron olivine is extremely good for a material working
cities. In addition, those two elements are stable in phosphates in at 3.5 V. Hence, unless one can increase the voltage of olivines
an octahedral environment for all oxidation states and are in without sacrificing other properties, not much improvement should
general relatively thermally stable. Our compound prediction be expected for one-electron phosphates.
algorithm identified a novel two-electron vanadium compound, All battery properties cannot be directly computed ab initio on
Li9V3(P2O7)3(PO4)2, to be very close to stability at zero K a large scale, and our investigations only focused on some
(7 meV/at above the hull). This mixed ortho- and pyrophosphate, necessary but not sufficient conditions that new phosphate
isostructural to Li9Fe3(P2O7)3(PO4)2,106 presents attractive the- cathodes should meet. We believe the developments in accurate
oretical specific energy (726 Wh/kg) and has been synthesized modeling of lithium diffusion110 and polaron migration111,112 will
and tested electrochemically in our research group.107 Very recently, in the future provide the ability to perform the same kind of
Kuang et al. reported independently on the synthesis and elec- analysis on a larger set of properties such as ionic and electronic
trochemical testing of this compound.108 conductivity. Beside the difficulties in accurately modeling those
The Sb3+/Sb5+ redox couple is also of interest, but the weight properties, many challenges lies in the scaling up of their
of antimony combined with the strong preference for different computations without human intervention in a high-throughput
local environment for Sb3+ and Sb5+ (8-fold coordinated and framework.
octahedral) are detrimental.
Oxyphosphates are an attractive subclass of phosphates with
their possible higher capacity because of their lower phosphorus 5. CONCLUSION
content, while still benefiting from the inductive effect. In addition, We have performed a large scale computational evaluation of
the connectivity of their transition metal octahedra might be phosphates as cathode materials. A computational database of
favorable for electronic conductivity. Only a few oxyphosphates thousands of known and virtual phosphate materials has been
are known. Most of them exist in vanadium, molybdenum, and built. Essential battery properties such as stability, capacity
titanium chemistries. The lithium vanadium oxyphosphate Li- (volumetric and gravimetric), voltage, specific energy, energy
VOPO4 has been already well studied and shows good specific density, and safety have been computed using this data. Limits
energy but poor rate capability.75 The two-electron redox couples and opportunities in phosphate chemistries have been evaluated.
available for vanadium and molybdenum combined with their Such a high-throughput approach can help the experimental
tendency to form oxyphosphates call for more investigations of process of new cathode materials design focus on the most
these chemistries. promising directions and will be extended in the future to other
The data provided in this work can be used to propose new chemistries.
chemistries to explore but also to suggest incremental modifica-
tions to known compounds. For instance, the vanadium NASI-
CON Li3V2(PO4)3 is a promising battery cathode with a ’ ASSOCIATED CONTENT
computed theoretical energy density of 724 Wh/kg and 2099
Wh/l.8,109 Unfortunately, the last lithium extraction process bS Supporting Information. Stability results for lithium-
shows an important hysteresis, and only the capacity related to containing ICSD phosphates, relative stability of LiCoPO4 phases
the V3+/V4+ couple can be used practically. Without this last with different functionals, and details on the ANOVA analysis of
lithium, the theoretical energy density (456Wh/kg and 1330 Wh/l) the factors determining the voltage. This information is available
is much lower than that of LiFePO4. On the other hand, our free of charge via the Internet at http://pubs.acs.org/
3506 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508
Chemistry of Materials ARTICLE

’ AUTHOR INFORMATION (27) Chevrier, V. L.; Ong, S. P.; Armiento, R.; Chan, M. K. Y.; Ceder,
G. Phys. Rev. B 2010, 82, 075122.
Corresponding Author (28) Kresse, G.; Furthmuller, J. Comput. Mater. Sci. 1996, 6,
*E-mail: gceder@mit.edu. 15–50.
(29) Bl€ochl, P. Phys. Rev. B 1994, 50, 17953–17979.
(30) Jain, A.; Hautier, G.; Moore, C. J.; Ong, S. P.; Fischer, C. C.;
’ ACKNOWLEDGMENT Mueller, T.; Persson, K. A.; Ceder, G. Comput. Mater. Sci. 2011,
50, 2295–2310.
The authors would like to acknowledge Bosch and Umicore (31) Inorganic Crystal Structure Database; FIZ Karlsruhe: Germany,
for funding. Anubhav Jain acknowledges funding from the U.S. 2006.
Department of Energy, Department of Energy Computational (32) Burzlaff, H.; Malinovsky, Y. Acta Crystallogr., Sect. A: Found.
Science Graduate Fellowship (DOE CSGF) (under Grant DE- Crystallogr. 1997, 53, 217–224.
FG02-97ER25308). (33) Hundt, R.; Sch€on, J. C.; Jansen, M. J. Appl. Crystallogr. 2006,
39, 6–16.
(34) Hart, G. L. W.; Forcade, R. W. Phys. Rev. B 2008, 77, 1–12.
’ REFERENCES (35) Hautier, G.; Fischer, C.; Ehrlacher, V.; Jain, A.; Ceder, G. Inorg.
(1) Whittingham, M. S. Chem. Rev. 2004, 104, 4271–4302. Chem. 2011, 50, 656–663.
(2) Padhi, A.; Nanjundaswamy, K.; Goodenough, J. B. J. Electrochem. (36) Ong, S. P.; Wang, L.; Kang, B.; Ceder, G. Chem. Mater. 2008,
Soc. 1997, 144, 1188. 20, 1798–1807.
(3) Masquelier, C.; Patoux, S.; Wurm, C.; Morcrette, M. In Poly- (37) Hautier, G.; Fischer, C. C.; Jain, A.; Mueller, T.; Ceder, G.
anion-Based Positive Electrode Materials; Nazri, G.-A., Pistoia, G., Eds.; Chem. Mater. 2010, 22, 3762–3767.
Springer: New York, 2003; pp 445
474. (38) O’Keefe, M. Acta Crystallogr., Sect. A 1979, 35, 772–775.
(4) Whittingham, M. S.; Song, Y.; Lutta, S.; Zavalij, P. Y.; Chernova, (39) Pinsky, M.; Avnir, D. Inorg. Chem. 1998, 37, 5575–5582.
N. A. J. Mater. Chem. 2005, 15, 3362. (40) Aydinol, M.; Kohan, A.; Ceder, G.; Cho, K.; Joannopoulos, J.
(5) Delacourt, C.; Poizot, P.; Morcrette, M.; Tarascon, J.-M.; Phys. Rev. B 1997, 56, 1354–1365.
Masquelier, C. Chem. Mater. 2004, 16, 93–99. (41) Jain, A.; Hautier, G.; Ong, S. P.; Moore, C. J.; Fischer C. C.;
(6) Gaubicher, J.; Mercier, T. L.; Chabre, Y.; Angenault, J.; Quarton, Persson, K. A.; Ceder, G. Phys. Rev. B 2011, accepted.
M. J. Electrochem. Soc. 1999, 146, 4375–4379. (42) Ong, S. P.; Jain, A.; Hautier, G.; Kang, B.; Ceder, G. Electrochem.
(7) Patoux, S.; Wurm, C.; Morcrette, M.; Rousse, G.; Masquelier, C. Commun. 2010, 4, 1–4.
J. Power Sources 2003, 119
121, 278–284. (43) Chase, M. W. NIST-JANAF Thermochemical Tables; American
(8) Saïdi, M.; Barker, J.; Huang, H.; Swoyer, J. L.; Adamson, G. Institute of Physics: Woodbury, NY, 1998.
J. Power Sources 2003, 119
121, 266–272. (44) Durif, A. Crystal Chemistry of Condensed Phosphates; Plenum
(9) Ceder, G.; Chiang, Y.-M.; Sadoway, D. R.; Aydinol, M.; Jang, Press, New York, 1995.
Y. I.; Huang, B. Nature 1998, 392, 694–696. (45) Yin, S.-C.; Grondey, H.; Strobel, P.; Anne, M.; Nazar, L. F.
(10) Van Der Ven, A.; Aydinol, M. K.; Ceder, G. J. Electrochem. Soc. J. Am. Chem. Soc. 2003, 125, 10402–11.
1998, 145, 2149. (46) La€ugt, M.; Tordjman, I.; Guitel, J. C.; Roudaut, R. Acta
(11) Kang, K.; Meng, Y. S.; Br
eger, J.; Grey, C. P.; Ceder, G. Science Crystallogr., Sect. B: Struct. Crystallogr. Cryst. Chem. 1972, 28, 2352–2358.
2006, 311, 977–980. (47) La€ugt, M.; Durif, A.; Martin, C. J. Solid State Chem. 1974, 7, 448.
(12) Kang, B.; Ceder, G. Nature 2009, 458, 190–193. (48) García-Moreno, O.; Alvarez-Vega, M.; García-Alvarado, F.;
(13) Meng, Y. S.; Wu, Y. W.; Hwang, B. J.; Li, Y.; Ceder, G. García-Jaca, J.; Gallardo-Amores, J. M.; Sanju
an, M. L.; Amador, U.
J. Electrochem. Soc. 2004, 151, A1134. Chem. Mater. 2001, 13, 2455–2455.
(14) Arroyo-deDompablo, M. E.; Dominko, R.; Gallardo-Amores, (49) Averbuch-Pouchot, M. T. Acta Crystallogr., Sect. C: Cryst. Struct.
J. M.; Dupont, L.; Mali, G.; Ehrenberg, H.; Jamnik, J.; Mor
an, E. Chem. Commun. 1989, 45, 1856–1858.
Mater. 2008, 20, 5574–5584. (50) Ramana, C. V.; Ait-Salah, A.; Utsunomiya, S.; Mauger, A.;
(15) Arroyo Y De Dompablo, M.; Amador, U.; Tarascon, J.-M. Gendron, F.; Julien, C. M. Chem. Mater. 2007, 19, 5319–5324.
J. Power Sources 2007, 174, 1251–1257. (51) Ramana, C. V.; Kopec, M.; Mauger, A.; Gendron, F.; Julien,
(16) Ceder, G.; Hautier, G.; Jain, A.; Ong, S. P. MRS Bull. 2011, C. M. J. Appl. Phys. 2009, 106, 064304.
36, 185–191. (52) Murashova, E. V.; Chudinova, N. N. Inorg. Mater. 2000, 36,
(17) Meng, Y. S.; Arroyo-de Dompablo, M. E. Energy Environ. Sci. 1277–1281.
2009, 2, 589. (53) Lesnyak, V. V.; Slobodyanik, N. S. Theor. Exp. Chem. 1999,
(18) Armstrong, a. R.; Arrouvel, C.; Gentili, V.; Parker, S. C.; Islam, 35, 338–342.
M. S.; Bruce, P. G. Chem. Mater. 2010, 2, 101117124944079. (54) Patoux, S.; Rousse, G.; Leriche, J.; Masquelier, C. Solid State Sci.
(19) Kim, J. C.; Moore, C. J.; Kang, B.; Hautier, G.; Jain, A.; Ceder, 2004, 6, 1113–1120.
G. J. Electrochem. Soc. 2011, 158, A309. (55) Nishimura, S.-i.; Nakamura, M.; Natsui, R.; Yamada, A. J. Am.
(20) Perdew, J.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, Chem. Soc. 2010, 132, 13596–7.
77, 3865–3868. (56) Zhou, H.; Upreti, S.; Chernova, N. A.; Hautier, G.; Ceder, G.;
(21) Anisimov, V.; Zaanen, J.; Andersen, O. Phys. Rev. B 1991, Whittingham, M. S. Chem. Mater. 2011, 23, 293–300.
44, 943–954. (57) Amador, U.; Gallardo-Amores, J.; Heymann, G.; Huppertz, H.;
(22) Anisimov, V. I.; Aryasetiawan, F.; Lichtenstein, A. I. J. Phys.: Mor
an, E.; Arroyo Y De Dompablo, M. Solid State Sci. 2009, 11, 343–348.
Condens. Matter 1997, 9, 767–808. (58) Huang, X.; Ma, J.; Wu, P.; Hu, Y.; Dai, J.; Zhu, Z. Mater. Lett.
(23) Kubaschewski, O.; Alcock, C. B.; Spencer, P. J. In Thermo- 2005, 59, 578–582.
chemical Data, 6th ed.; Pergamon Press:Elmsford, NY, 1993; Chapter 5, (59) Chippindale, A. M.; Cowley, A. R.; Chen, J.; Gao, Q.; Xu, R.
pp 257
323. Acta Crystallogr., Sect. C: Cryst. Struct. Commun. 1999, 55, 845–847.
(24) Wang, L.; Maxisch, T.; Ceder, G. Phys. Rev. B 2006, 73, 1–6. (60) Padhi, A.; Manivannan, V.; Goodenough, J. B. J. Electrochem.
(25) Zhou, F.; Cococcioni, M.; Kang, K.; Ceder, G. Electrochem. Soc. 1998, 145, 1518.
Commun. 2004, 6, 1144–1148. (61) Padhi, A.; Nanjundaswamy, K.; Masquelier, C.; Okada, S.;
(26) Heyd, J.; Scuseria, G. E.; Ernzerhof, M. J. Chem. Phys. 2003, Goodenough, J. B. J. Electrochem. Soc. 1997, 144, 1609–1613.
118, 8207. (62) Patoux, S.; Masquelier, C. Chem. Mater. 2002, 14, 5057–5068.

3507 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508

Chemistry of Materials ARTICLE

(63) Yang, Z.; Choi, D.; Kerisit, S.; Rosso, K. M.; Wang, D.; Zhang, (100) Yamada, A.; Kudo, Y.; Liu, K.-Y. J. Electrochem. Soc. 2001,
J.; Graff, G.; Liu, J. J. Power Sources 2009, 192, 588–598. 148, A747.
(64) Kuhn, A.; Martín, M.; Garcia-Alvarado, F. J. Solid State Chem. (101) Shanmukaraj, D.; Murugan, R. Ionics 2004, 10, 88–92.
2010, 183, 20–26. (102) Lyness, C.; Delobel, B.; Armstrong, A. R.; Bruce, P. G. Chem.
(65) Arroyo-de Dompablo, M. E.; Gallardo-Amores, J. M.; Amador, Commun. 2007, 4890.
U. Electrochem. Solid-State Lett. 2005, 8, A564. (103) Nyten, A.; Abouimrane, A.; Armand, M.; Gustafsson, T.;
(66) Ait-Salah, A.; Ramana, C. V.; Gendron, F.; Morhange, J.; Mauger, Thomas, J. Electrochem. Commun. 2005, 7, 156–160.
A.; Selmane, M.; Julien, C. M. Mater. Res. Soc. Symp. Proc. 2007, 972. (104) Jouanneau, S.; MacNeil, D. D.; Lu, Z.; Beattie, S. D.; Murphy,
(67) Delmas, C.; Nadiri, A.; Soubeyroux, J. L. Solid State Ionics G.; Dahn, J. R. J. Electrochem. Soc. 2003, 150, A1299.
1988, 419–423. (105) Kang, B.; Ceder, G. J. Electrochem. Soc. 2010, 157, A808.
(68) Aatiq, A.; Menetrier, M.; Croguennec, L.; Suard, E.; Delmas, C. (106) Poisson, S.; D’Yvoire, F.; Nguyen Huy, D.; Bretey, E.; Berthet,
J. Mater. Chem. 2002, 12, 2971–2978. P. J. Solid State Chem. 1998, 138, 32–40.
(69) Kishore, M.; Pralong, V.; Caignaert, V.; Varadaraju, U.; Raveau, (107) Ceder, G.; Jain, A.; Hautier, G.; Kim, J. C.; Kang, B.; Daniel,
B. J. Power Sources 2007, 169, 355–360. R. Mixed Phosphate-Diphosphate Electrode Materials and Methods of
(70) Bhuvaneswari, G. D.; Kalaiselvi, N. Appl. Phys. A: Mater. Sci. Manufacturing Same. Patent PCT application PCT/US11/031946, 2011.
Process. 2009, 96, 489–493. (108) Kuang, Q.; Xu, J.; Zhao, Y.; Chen, X.; Chen, L. Electrochim.
(71) Gaubicher, J.; Wurm, C.; Goward, G.; Masquelier, C.; Nazar, L. Acta 2011, 56, 2201–2205.
Chem. Mater. 2000, 12, 3240–3242. (109) Barker, J.; Huang, H.; Swoyer, J. L.; Adamson, G. Electrochem.
(72) Satya Kishore, M.; Pralong, V.; Caignaert, V.; Malo, S. Solid-State Lett. 2002, 5, A149–A151.
Hebert, S.; Varadaraju, U.; Raveau, B. J. Solid State Chem. 2008, 181, (110) Kang, K.; Morgan, D.; Ceder, G. Phys. Rev. B 2009, 79, 1–4.
976–982. (111) Maxisch, T.; Zhou, F.; Ceder, G. Phys. Rev. B 2006, 73, 1–6.
(73) Nanjundaswamy, K.; Padhi, A.; Goodenough, J. B.; Okada, S.; (112) Ong, S.; Chevrier, V.; Ceder, G. Phys. Rev. B 2011, 83, 1–7.
Otsuka, H.; Arai, H.; Yamaki, J. Solid State Ionics 1996, 92, 1–10.
(74) Dupr
e, N.; Gaubicher, J.; Le Mercier, T.; Wallez, G.; Angenault,
J.; Quarton, M. Solid State Ionics 2001, 140, 209–221.
(75) Barker, J.; Saidi, M. Y.; Swoyer, J. L. J. Electrochem. Soc. 2004,
151, A796.
(76) Kerr, T. A.; Gaubicher, J.; Nazar, L. F. Electrochem. Solid-State
Lett. 2000, 3, 460–462.
(77) Kishore, M. S.; Pralong, V.; Caignaert, V.; Varadaraju, U.;
Raveau, B. Solid State Sci. 2008, 10, 1285–1291.
(78) Wolfenstine, J.; Allen, J. J. Power Sources 2005, 142, 389–390.
(79) Wolfenstine, J.; Allen, J. J. Power Sources 2004, 136, 150–153.
(80) Amine, K. Electrochem. Solid-State Lett. 1999, 3, 178.
(81) Wohlfahrt-Mehrens, M. Lithium Ion Batteries: Battery Materials
and Ageing Processes. Conference on Advanced Battery Technologies for
Automobiles and Their Electric Power Grid Integration, 2010.
(82) Choi, S.; Hong, S.-T. Mater. Res. Bull. 2005, 40, 1787–1795.
(83) Uebou, Y. Solid State Ionics 2002, 148, 323–328.
(84) Uebou, Y.; Okada, S.; Yamaki, J.-i. J. Power Sources 2003,
115, 119–124.
(85) Sahai, H.; Ageel, M. I. Analysis of Variance: Fixed, Random and
Mixed Models; Birkh€auser: Boston, 2000; p 784.
(86) Arroyo-de Dompablo, M.; Armand, M.; Tarascon, J.; Amador,
U. Electrochem. Commun. 2006, 8, 1292–1298.
(87) Howard, W.; Spotnitz, R. J. Power Sources 2007, 165, 887–891.
(88) Dahn, J.; Fuller, E.; Obrovac, M.; Vonsacken, U. Solid State
Ionics 1994, 69, 265–270.
(89) Xu, K. Chem. Rev. 2004, 104, 4303–4418.
(90) MacNeil, D. D.; Zhonghua, L.; Zhaohui, C.; Dahn, J. R. J. Power
Sources 2002, 108, 8–14.
(91) Godshall, N. A.; Raistrick, I. D.; Huggins, R. A. J. Electrochem.
Soc. 1984, 131, 543.
(92) Huggins, R. A. ECS Trans. 2009, 16, 37–47.
(93) Delacourt, C.; Poizot, P.; Tarascon, J.-M.; Masquelier, C. Nat.
Mater. 2005, 4, 254–260.
(94) Kim, S.-W.; Kim, J.; Gwon, H.; Kang, K. J. Electrochem. Soc.
2009, 156, A635.
(95) Chen, G.; Richardson, T. J. J. Power Sources 2010, 195, 1221–1224.
(96) Bramnik, N. N.; Nikolowski, K.; Trots, D. M.; Ehrenberg, H.
Electrochem. Solid-State Lett. 2008, 11, A89.
(97) Okada, S.; Ueno, M.; Uebou, Y.; Yamaki, J. J. Power Sources
2005, 146, 565–569.
(98) Wang, D.; Xiao, J.; Zu, W.; Zhang, J.-G. Investigation of
LiNiPO4 as a Cathode Material for Lithium Ion Battery; International
Meeting on Lithium Batteries, 2010.
(99) Huang, H.; Faulkner, T.; Barker, J.; Saidi, M. J. Power Sources
2009, 189, 748–751.

3508 dx.doi.org/10.1021/cm200949v |Chem. Mater. 2011, 23, 3495–3508