PHYSICAL REVIEW B VOLUME 42, NUMBER 10 1 OCTOBER 1990

Lattice dynamics and origin of ferroelectricity in BaTi03.

Linearized-augmented-plane-wave total-energy calculations

R. E. Cohen*
Complex Systems Theory Branch, Naval Research Laboratory, 8'ashington, D. C. 20375-5000

H. Krakauer
Department of Physics, College of William and Mary, Williamsburg, Virginia 23185
(Received 2 May 1990)

Phonon frequencies and eigenvectors have been computed from first principles for the three optic
F, „modes in BaTi03 using the full-potential linearized-augmented-plane-wave method. We find
that the ferroelectric instability in BaTi03 can be understood from calculations for a perfect crystal
with periodic boundary conditions. The energy wells for the soft-mode distortion are deeper for
rhombohedral [111]displacements relative to tetragonal [001] displacements, but they are relatively
shallow and comparable to the transition temperature. The nonrigid part of the charge-density dis-
tortion is centered around the Ti ion rather than the 0, and the Ti charge is closer to 2.9+ than
4+. There is significant hybridization between the Ti and 0, but the Ba is quite ionic and is well
described as a Ba'+ ion. The Ti-0 hybridization is essential to the ferroelectric instability.

I. INTRODUCTION At 393 K, with decreasing temperature, BaTi03 trans-

Forty years ago Slater began a paper by saying that the
structure of BaTi03 is so simple that there is good hope
of understanding its ferroelectric properties fairly com-
pletely. ' This goal has remained elusive, and it seems
that there are more questions than ever about the origin
of ferroelectricity in BaTi03 and the nature of its fer-
roelectric phase transitions. Interest remains high in
BaTi03 not only because of basic questions about the ori-
gin of its behavior, but also because it is the most
economically important ceramic material because of its
ferroelectric transition near room temperature. It
remains the classic and most studied ferroelectric, both
theoretically and experimentally.
Unresolved questions include the nature of its phase
transitions, the primary driving force for the transitions,
and the importance and effects of strain and impurities.
Particularly interesting is the strong anharmonicity and
order-disorder phenomena associated with the soft mode.
Both the experimental and theoretical studies are compli-
cated by the fact that both the displacements and the en-
ergy differences involved in the ferroelectric transition
are small. Thus impurities, defects, stresses, grain size,
domain formation, and even light can affect the transi-
tions. Almost aH of the theoretical studies have em-
ployed empirical or semiempirical models in which
parametrized interatomic forces and polarizabilities are
used in combination with lattice dynamics or molecular
dynamics. The only previous first principles study of Ba-
Ti03 was performed using the linearized muffin-tin orbit-
al (LMTO) method. A preliminary account of the
present calculations has been reported; the total energies
in Ref. 4 for the ferroelectric phase were in error and are
corrected here.
BaTi03 displays three ferroelectric phase transitions.
forms from a paraelectric cubic phase to a ferroelectric
tetragonal structure with a macroscopic polarization
along a cube face direction [001]. At 278 K, it trans-
forms to an orthorhombic structure with the polarization
along a cube edge direction [110], and at 183 K it be-
comes rhombohedral with the polarization along a cube
diagonal direction [111]. The rhombohedral phase ap-
pears to be a well-ordered structure with the Ti and 0
displaced in opposite directions along [111]. There is
some controversy about the higher-temperature struc-
tures, as to whether they are well-ordered structures or
whether they are disordered. In the "eight-site model, "
for example, there are minima in the potential along the
eight [111] cube diagonals. According to the order-
disorder model, in the paraelectric phase the Ti atoms are
disordered among the eight sites, in the tetragonal state
they preferentially occupy four of the eight sites, giving a
net polarization along [001], in the orthorhombic phase
they preferentially occupy two of the eight sites, and
finally in the rhombohedral phase they are completely or-
dered. ' In the pure "soft-mode" picture, the atoms pri-
marily occupy the ideal cubic sites at high temperatures.
As the temperature is lowered, one of the transverse-
optic (TO) modes with F, „symmetry goes soft and unsta-
ble, and the crystal transforms to the tetragonal phase
with atomic displacements along the [001] direction.
Another TO mode goes unstable at each of the subse-
quent phase transitions. Muller et al. have pointed out
that the order-disorder and soft-mode pictures are not
mutually exclusive. As temperature is lowered, one of
the modes can soften as in the soft-mode picture, but as
the transition temperature T, is approached, there is a
crossover to order-disorder behavior as correlated clus-
ters grow. Most of the experimental data are consistent
with a complicated combination of soft-mode and order-

1990 The American Physical Society

42 LATTICE DYNAMICS AND ORIGIN OF FERROELECTRICITY. . .
disorder behavior. No attempt will be made here to re-
view the voluminous experimental, phenomenological,
and potential modeling literature on BaTi03, but excel-
lent reviews are given by Lines and Glass and Jona and
Shirane.
A fundamental question is: what is the underlying
static potential for the ferroelectric distortions? If this
were known molecular dynamics or Monte Carlo simula-
tions could be used to study the transition, or the statisti-
cal mechanics of the phase transition could be quantita-
tively modeled. One piece of information that is needed
is the relative well depths for displacements in the [001]
and [111]directions. Symmetry requires that the curva-
ture of the potential surface with respect to displacement
amplitude be the same in all directions, since the soft FI„
mode is three-dimensionally degenerate at q =0. Howev-
er, the well minima positions and depths can be quite
different in different directions. In order to study the
order-disorder versus displacive models, the correlation
energy between cells is required as well as the zone
center, q =0, potential surface. However, for the ground
state to be the rhombohedral phase, as is observed, the
[111]wells must be deeper than the [001] wells for q =0.
Of even more fundamental interest is the underlying
origin of the ferroelectric instability. Most atomistic
models of ferroelectricity in oxide perovskites such as
BaTi03 are ionic models with polarizable oxygen ions. '
Here we study whether hybridization between the oxygen
and titanium ions is an important feature for driving the
ferroelectric instability.
We investigate these questions using the full-potential
linearized-augmented-plane-wave (LAPW) method"
within the local-density approximation (LDA). ' We in-
vestigate the lattice dynamics of the F, „modes using a
frozen-phonon approach, in which the total energy is cal-
culated for fixed distortions of the crystal. The lowest-
frequency F, „mode, which is related to the structural
distortions in the ferroelectric phases, is studied as a
function of volume for both tetragonal [001] displace-
ments, and rhombohedral [111] displacements. The
charge densities of the cubic and ferroelectric phases are
studied and compared with ionic reference models.
II. METHODS
A. LAPW calculations
In the full-potential LAPW method, both the basis set
and the representation of the potential and charge density
are based on a dual representation which consists of
spherical harmonics inside spherical "muffin tins" around
each atom, k, which are joined onto a plane-wave expan-
sion in the interstitial region between the muffin tins.
The basis functions inside each muffin tin consist of the
solution to Schrodinger s equation at an energy parame-
ter EI" for each angular momentum I, and derivatives of
these solutions with respect to EI . Multiple windows can
be used, in which the solutions are linearized with respect
to different energy parameters EI in each window. There
are no shape approximations for either the potential or
the charge density in the interstitial, and the basis set can
6417
essentially be fully converged. Convergence of the basis
set is governed by a dimensionless parameter RK, „,
where R is the minimum muffin-tin radius and K
the maximum reciprocal space K for the plane-wave ex-
„ is
pansion in the interstitial. LAPW computations can be
systematically converged by increasing the parameter
RK, „and the maximum angular momentum I,„. The
sensitivity of the results to the incompleteness of the basis
set inside the muffin tins can be studied by checking the
stability of the results as the energy parameters are
varied.
We treated the 0 2s states, Ba Ss and 5p states and the
Ti 3p states in a lower, semicore window, and the valence
states in an upper window. A 4X4X4 special k-point
mesh' gave 10 k points in the irreducible wedge of the
Brillouin zone in the rhombohedral space group, 6 k
points in the tetragonal symmetry, and 4 k points in the
cubic case. Increasing the mesh did not change the total
energies by more than 0. 1 mRy. The muffin-tin radii
were 2.5166 bohr (1 bohr=0. 527 A) for the Ba, 1.9148
bohr for Ti, and 1.55 211 bohr for the O. The parameter
RE, „(for 0) was 7 for most of the calculations, but
tests with RK, „up to 8 were made, and the energy
differences are converged to 0. 1 mRy at RK,„=7. An
RK,„of 7 gives approximately 675 basis functions at
each k point at the experimental zero pressure lattice
constant of 7.57 bohr. The core states were treated fully
relativistically, and the band states were treated semirela-
tivistically. As a test, spin orbit was turned on for the
band states as a second variation, ' and no change in en-
ergy differences was found for the fully relativistic calcu-
lations compared with the results for the scalar relativis-
tic band states. Most of the calculations were done using
the Hedin-Lundqvist' parametrization of the local
exchange-correlation potential, but tests were made with
the Wigner' and Kohn-Sham (exchange only)' function-
als as well. Different local exchange-correlation function-
als only shifted the equations of state of both cubic and
distorted phases up or down in volume by a couple per-
cent.
B. Frozen-phonon calculations
In order to find the eigenvectors, as well as eigenfre-
quencies for the F, „modes, the harmonic dynamical ma-
trix for the F I „modes was computed at the experimental
zero pressure lattice constant for the rhombohedral
phase, 7.57 bohr. The frozen-phonon calculations were
all performed for a cubic lattice, and lattice strains were
investigated separately. All of the calculations were per-
formed for 5 atom units cells, so that the wave vector was
strictly equal to zero for the distortions. Thus only TO
frequencies were obtained, and each of the F, „modes is
then threefold degenerate. There are 4 F, „modes, one of
which is an acoustic distortion with zero frequency, that
is a rigid translation of the lattice. Rather than determin-
ing the full 4X4 block of the dynamical matrix, only a
3 X 3 block need be determined by projecting out the
translational mode. This is done by using center-of-mass
coordinates for the displacements, that is symmetry coor-
dinates that are orthogonal to a uniform displacement of
6418 R. E. COHEN AND H. KRAKAUER 42

the center of mass. Since it is convenient for the total en-
ergy calculations to be done for displacements of one
atom at a time when possible, symmetry coordinates were
constructed by taking displacements of one atom, and
then shifting the origin to keep the center-of-mass con-
stant. This gives a set of symmetry coordinates that are
not mutually orthogonal. Thus it is necessary to find the
so-called kinetic-energy matrix, or 6 matrix, that diago-
nalizes the kinetic energy. ' This procedure is given ex-
plicitly here since this is the first time the frozen-phonon
procedure has been used to determine eigenvectors and
eigenfrequencies for polar modes. These complexities do
not appear for nonpolar modes.
In the tetragonal space group for BaTi03 there is one
Ba at (0,0, 5, ), one Ti at (0.5,0.5,0.5 +5&}, one O(1) ion at
(0.5,0.5, 53), and two O(2) ions at (0,0.5,0.5+54) and
(0.5,0,0.5+ 54 }, and the 5's are zero for the cubic
perovskite structure. The amplitude for each symmetry
coordinate S is given by
S=BX,
where X is the displacement of the Ba, Ti, O(1), and two
O(2) ions, respectively, and 8 is given by
0 —0. 5 —0. 5
0 —0. 5 —0. 5
—0. 5 —0. 5
B= 0 1
m Bg m'pl mQ mQ mQ
0 0 0 1
where the fourth row is the center-of-mass coordinate,
and the fifth row is the Fz„mode. The force constant
matrix F is then simply the second derivative matrix of
the energy with respect to displacements of the Ba, Ti
and O(1) ions, keeping the O(2) ions fixed. The secular
equation is given by'
fF G'X( =0—,
where A=co are the eigenvalues, and the kinetic-energy
matrix G is given by
6 =DD
of displacement was fit to a quadratic with an rms error
of about 0.2 mRy or less, so that within the precision of
the calculations a higher order fit was not justified. The
off-diagonal terms were each determined with only one
additional total-energy calculation for displacements of
each pair of ions. The off-diagonal element for a pair of
ions is obtained from the difference of the total energy
from the undistorted crystal energy for the displaced pair
of ions, compared with that obtained for the same dis-
placements for each individual ion.
III. RESULTS AND DISCUSSION
The calculated eigenfrequencies and eigenvectors are
shown in Table I. At the experimental zero pressure lat-
tice constant, we do find an unstable mode. The relative
displacements of Ba:Ti:O(1) for the soft mode are
0.53:1:-0.29 if the unstable mode eigenvector is expressed
in terms of displacements relative to a unit Ti displace-
ment, with the O(2) atoms fixed. The experimental dis-
tortion expressed in this way is 0.53:1:— 0. 28, so the cal-
culated soft-mode eigenvector is in perfect agreement
with the experimentally observed ferroelectric distortion
for the tetragonal phase.
The stable calculated frequencies at a =7. 57 are not in
very good agreement with the experimental F, „ frequen-
cies for the cubic phase (at a temperature of just above
the tetragonal to cubic transition, 395 K, ' which is prob-
ably due to anharmonicity that affects the higher-
frequency modes through anharmonic coupling to the
soft mode. The finite difference procedure used here can
also introduce some errors when the potential surface is
nonquadratic.
A. Soft-mode potential surface
Figure 1 shows the energy versus Ti displacement for
the soft mode at a =7. 57 bohr and at a =7. 5 bohr, using
the same displacement pattern, for distortions in both the
[001] and [111] directions. The wells are found to be
deeper in the [111]direction than the [001] direction, as
expected. The potential surface is extremely sensitive to
the volume. At a lattice constant of 7.5 bohr the well
depths are extremely shallow, and at higher pressures

where
J.
D„=B, m ', and m is the atomic mass of atom
they presumably disappear. This is consistent with the
large negative dependence of T, on pressure. In fact,
there is an observed tricritical point at about 3.5 GPa and
The three diagonal elements of the force constant ma- 233 K, and the phase transition changes from first-order
trix were determined by performing several total-energy to second order as pressure is increased. ' This is con-
calculations as functions of the displacement of the Ba, sistent with the underlying potential wells becoming
Ti, and O(1) ions, respectively. The energy as a function much shallower as pressure is increased.

TABLE I. Calculated eigenfrequencies and eigenvectors for BaTi03 at a = 7. 57 bohr (4.006 A).
calc exp* Calculated eigenvectors
CID Ba T1 O(1) O(2)
—72 unstable —0.07 —0.63 0.58 0.36 0.36
161 180 —0.64 0.61 0.21 0.30 0.30
397 482 —0.03 0. 16 0.74 —0.46 —0.46
*Reference 18.
 LATTICE DYNAMICS AND ORIGIN OF FERROELECTRICITY. ..

C)
V= 43
l

O
I

1.01 1.02

a=3.97 A
C)
OQ

-0. a=7.5 bohr

1 0.0 0.1
Ti Displacement rel. to Ba (A)
FIG. 1. Energy vs displacement for the lowest-frequency
mode at two volumes. The solid curves are for the distortion in
the rhombohedral (111) direction, and the dashed line is for the Qo-
tetragonal (001) distortion. The arrows and dots indicate the
minimum energy for the distortion when the tetragonal e/a
strain is included.
~ IA
I

B. Effects of lattice strain a=7.57 bohr

Experimentally there is a 1% c/a strain in the tetrago-

nal BaTi03 structure, and the rhombohedral angle in the
rhombohedral phase is 89.87'. In order to study the im- I
89.7 89.9 90 90.1
portance of the soft-mode coupling with the lattice strain
we performed displacements along the [001] and [111]
a (deg)
directions for several different values of strain. Figure 2 FIG. 2. Energy vs strain for the tetragonal and rhom-
shows the energy versus strain for the experimental atom- bohedral structures of BaTi03 with experimental atomic dis-
ic displacements for the tetragonal and rhombohedral placements. The energies are given relative to the cubic phase
structures. The energies are given relative to the cubic at each volume. (a) Tetragonal c/a strain at constant volume.
structure at each volume. We find an almost negligible (b) Rhombohedral strain at constant lattice constant.
effect on the total energies for the rhombohedral strain,
but the calculated energies are consistent with the rhom-
bohedral being slightly smaller than 90, as is observed.
On the other hand, we find a very strong dependence of [001] ferroelectric distortion. The 1% lattice strain is of
the same order of magnitude as the atomic displace-
the energy on the tetragonal c/a strain, and the energy
ments, and the tetragonal lattice strain appears to be of
decreases markedly when the lattice is strained. The ar-
rows in Fig. 1 indicate the lowering of the energy when
great importance in stabilizing the tetragonal phase. Be-
cause of the strain, the atoms probably do spend a
the lattice is strained for the experimental distortion. We
significant amount of time in the [001] wells in the tetrag-
have not done a complete set of strain versus displace-
onal structure, rather than only hopping among the [111]
ment calculations, but it is clear that the tetragonal strain
wells.
lowers the energy so significantly that the strained tetrag-
onal structure is probably only slightly higher than the
rhombohedral phase, and may even be lower at elevated C. Equation-of-state and e8'ects of pressure
pressures.
The c/a lattice strain at fixed volume increases the Ti- Figure 3 shows the total energy as a function of volume
0 distance along the tetragonal axis, thus facilitating the for the cubic BaTi03 structure and for various distor-
6420 R. E. COHEN AND H. KRAKAUER
o rhornbohedr
+ tetragonal
(s
x tetragonal (u
cubic
I
58 60 62 64
v (A')
FIG. 3. Calculated total energy as a function of volume for
the cubic BaTi03 structure (solid curve), for the rhombohedral
[lllj distortion, and the minimum energy strained and un-
strained (i.e., cubic lattice) tetragonal structures. The absolute
energy scale is arbitrary.
tions. The Hedin-Lundqvist parametrization of the
LDA, in general, underestimates the zero pressure
volume. For BaTi03 we find that the minimum-energy
lattice constant is 7.45 bohr, rather than the experimental
value of 7.57 bohr, and the cubic structure is computed
to be the static ground-state structure. This may partly
be due to the underestimation of the volume, but the
pressure required to stabilize the rhombohedral ground
state is only about —3 GPa, which is close to what is ex-
pected for the zero-point pressure. The experimental
pressure dependence of T, extrapolates to 0 K at about 8
GPa, so that only a small change in the lattice constant is
required to destroy the ferroelectric instability. On the
other hand, the soft mode is observed to be overdamped
in the cubic phase, which suggests a double well depth
that is comparable to the transition temperature. This
suggests that LDA may be slightly in error regarding the
weil depth versus volume, and the well depth at zero
pressure may be slightly deeper than what we calculate at
a =7. 5 bohr.
Both increasing temperature and increasing pressure
can lead from the tetragonal to cubic phases of BaTi03,
but we predict that the ferroelectric-paraelectric transi-
tion with pressure is very different from the transition
with temperature. When pressure is increased, the fer-
roelectric instability is destroyed in the underlying static
potential, but when temperature is increased, the long-
range correlations between atomic displacements disap-
42
pear, even though the atoms still spend most time in the
off-center well positions. Thermal expansion deepens the
potential wells, but nevertheless temperature favors the
cubic phase. The phase transition with increasing tem-
perature arises from the loss of correlations from one cell
to the next.
D. Charge density
Figure 4 shows the calculated valence charge density
for tetragonal BaTi03. The valence charge on the Ba is
very small, but there is significant hybridization between
the Ti and the O. When the atoms are displaced from the
cubic sites, the valence charge on the Ti distorts appreci-
ably. The large amount of valence charge on the Ti arises
primarily from hybridization between the 0 p states and
Ti d states.
We find it useful to compare the total calculated
charge density with a reference charge density that is
constructed from overlapping spherical ions. The
difference between the self-consistent charge density and
the ionic reference charge density highlights the inAuence
of covalency and polarization. We use an ab initio model
charge density formed from overlapping spherical ions
generated according to the potential induced breathing
(PIB) model. The PIB model charge densities are in ex-
cellent agreement with self-consistent charge densities for
the alkaline earth oxides, ' and any deviations between
the self-consistent and PIB charge densities emphasize
the difference between the actual bonding and a prototyp-
ical ionic crystal. Figure 5 shows the difference between
the self-consistent LAPW charge density and the refer-
ence fully ionic PIB charge density for both cubic, tetrag-
onal, and rhombohedral BaTiO&. The Ba is very close to
a Ba + ion, but the 0 valence states are strongly hybri-
dized with the Ti. Also, when the ferroelectric distortion
is introduced, a large polarization of the valence charge
around the Ti ion is observed. This is contrary to the
traditional picture of oxygen polarizability that assumes a
dipolar distortion of charge centered around the oxygen
ion.
C)
D
V
&110&
FIG. 4. Calculated valence charge density for tetragonal
BaTi03. The contour interval is 0.05 e/bohr' outside the thick
curves, and 0. 1 e/bohr' inside.
42 LATTICE DYNAMICS AND ORIGIN OF FERROELECTRICITY . ~ ~ 6421

'(a) Cubic $~
(
(
O—

A
D
C)
v

r f7~ C
0
tragonal ~s
o o- l

I
CI

A
D
ov

~\
n
mbohedral
' -2
I
-1.9
I
-1.8 -1.7
1 l

-1.6 -1.5
Q ion charge (electrons)
FIG. 6. Differences between the integrated charge in each
A
atomic sphere for overlapping ionic charge densities and the
O self-consistent charge density for cubic BaTi03 as functions of
D
v the0 (and Ti) charge used in the overlapping ion density.

static charges in BaTi03 are greatly reduced from the full

ionic charges.
Figure 8 shows the difference between the total LAPW

FIG. 5. Differences
(110)
between the total self-consistent charge
charge density and overlapping O',
Ti +, and Ba +
ions. The complicated, largely dipolar deformation of the
densities and those formed from overlapping spherical Ba Ti is apparent, as is a bonding charge between the short
Ti +, and 0 ions. The contour interval is 0.005 0. 1 e/bohr'.
(a) cubic perovskite (b) tetragonal (c) rhombohedral. The
—
Ti 0 bond. Also seen are the differences between the
O(1) and O(2) ions, which include the lack of z-direction
dashed lines are negative contours.
polarization for the O(2) ion and the formation of

It appears that there is a significant amount of charge

transfer back to the Ti, and that the Ti static charge is
significantly less than 4+. In order to investigate the
static charges on the ions, we can use the PIB reference
densities for different ionic configurations. Figure 6
shows the differences in the total charge in each atomic
sphere as a function of the oxygen charge used in gen-
erating the reference ionic densities. The Ba ionic charge
is kept as 2+, and the Ti charge is varied to give charge
neutrality. The best-fit configuration has an 0
charge of
1.63 —and a Ti charge of 2. 89+. We can also compare
the band eigenvalues of the self-consistent potential with
those generated by the ionic charge densities (Fig. 7).
When the fully charged ionic configuration is used, a me-
tallic band structure is obtained from the non-self-
consistent potential, but as the charge is lowered the ei-
genvalues approach those of the self-consistent band
structure, with a best fit for a majority of the bands with
similar charges to those determined by examining the to-
tal charge in each atomic sphere. Thus we find that the
—
O(2) Ti bonding charge different than the O(1) Ti —
bond.
We cannot simply obtain the dynamic effective
charges, which are the quantities necessary to determine
the polarization, from the self-consistent charge densi-
ty. The dynamic charges depend on how the crystal
charge density distorts when each nucleus is displaced. If
the valence charge on the Ti moves when an 0
is dis-
placed, it will contribute to the oxygen effective charge,
so the issue of atomic polarizabilities is a complicated one
that cannot be obtained from the change in the self-
consistent charge density alone; either supercells or
linear response theory is needed to recover the dynamic
charges and polarizabilities.
K. Driving force for the transition
The ferroelectric transition in BaTi03 arises from a
delicate cancellation between long-range electrostatic and
short-range overlap forces. The Madelung (long-range
electrostatic) contribution to the energy greatly favors
6422 R. E. COHEN AND H. KRAKAUER

bic. This is clear from ionic total-energy calculations us-

ing the PIB mode, which gives no ferroelectric transi-
tion (except at greatly expanded volumes). The hybridi-
zation we find in the self-consistent calculations serves to
CV
C) reduce the overlap repulsion, softening the short-range
C)
I
forces so that the ferroelectric displacement can take
place. It is probably not an accident that most oxide
perovskite ferroelectrics have a d B cation such as Ti +
or Zr +; the low-lying d states allow for significant hy-
C)
I bridization with the oxygen, thus softening a ferroelectric
mode.
Q
To test the importance of the Ti d hybridization on the
O ferroelectric instability, we performed a set of calcula-
0
tions with the Ti d energy parameter raised up above the
4d asymptote, effectively removing the Ti 3d states from
OQ the variational basis. When this is done, the ferroelectric
C)
O instability disappears, and the formerly unstable distor-
tion becomes a high-frequency mode.
We find that the lattice strain is quite significant for the
C) tetragonal distortion, but rather unimportant for the
O
I
rhombohedral distortion. The c/a strain helps stabilize
the tetragonal phase, and complicates the character of
the phase transition as much as the order-disorder behav-
ior among the multiple wells. The phase transition arises
from a complicated coupling of a displacive soft mode,
I I 1 l
-1.9 -1.8 -1.7 —1.6 -1.5
order-disorder "hopping" (or extremely anharmonic
0 ion charge (electrons)
motion) between the wells, and lattice strain.
FIG. 7. Differences between valence eigenvalues of the non-
self-consistent potential generated by overlapping ionic charge
densities and the self-consistent LAPW eigenvalues at the k IV. CONCLUSIONS
points used for self-consistency. For an 0
charge of less than
—1.75, a metallic band structure is obtained from the ionic po-
In summary, first-principles LAPW calculations indi-
tential.
cate a ferroelectric instability in BaTi03 that arises large-
ly from covalent hybridization between the Ti and the 0
the ferroelectric distortion; the electrostatic energy for ions. We have mapped out part of the soft-mode poten-
the fully ionic configuration decreases by 760 meV for the tial surface using LAPW frozen-phonon calculations, and
rhombohedral distortion relative to the cubic structure. find that the potential surface is strongly dependent on
The short-range forces, however, greatly favor the undis- volume. Thus we expect the phase transition characteris-
torted structure. For example, in the tetragonal distor- tics to vary sharply with pressure, as is observed experi-
tion, the Ti and O(l) ions are displaced towards each oth- mentally. The small lattice strain in tetragonal BaTi03 is
er. If there were no covalency or polarizability the found to be a significant contribution to the stabilization
short-range forces would be su5cient to keep BaTi03 cu- of the tetragonal room-temperature structure.

A A
CO D
C) O
V V

/
r '~ ~
I
I

(110) (010)
FIG. 8. Differences between the LAPW charge density for tetragonal BaTi03 and the best-fit overlapping ion density. The con-
tour interval is 0.005 e/bohr'. (a) (110) plane; (b) (100) plane. The dashed contours are negative.
42 LATTICE DYNAMICS AND ORIGIN OF FERROELECTRICITY. . .
ACKNOWLEDGMENTS
We would like to thank L. L. Boyer, D. Singh, %. E.
Pickett, M. J. Mehl, K. M. Rabe, P. Edwardson, D. A.
Papaconstantopoulos, and B. M. Klein for many helpful
discussions. The computations were mostly performed
'Present address: Geophysical Laboratory, Carnegie Institution
of Washington, 5251 Broad Branch Road, N. W. ,
Washington, DC 20015-1305.
'J. C. Slater,
Phys. Rev. 78, 748 (1950).
zM. E. Lines and A. M. Glass, Principles and Applications of
Ferroelectrics and Related Materials (Clarendon, Oxford,
1977), pp. 680.
K.-H. Weyrich and R. Siems, Z. Phys. 61, 63 (1985); K.-H.
Weyrich, Ferroelectrics (to be published).
4R. E. Cohen, L. L. Boyer, M. J. Mehl, W. E. Pickett, and H.
Krakauer, in Perovskites: A Structure of Great Interest to
Geophysics and Materials Science, 1989, AGU, pp. 55-66.
5R. Comes, M. Lambert, and A. Guinier, Solid State Commun.
6, 715 (1968); R. Comes, M. Lambert, and A. Guinier, Acta
Crystallogr. A 26, 244 (1970); K. Itoh, L. Z. Zeng, E.
Nakamura, and N. Mishima, Ferroelectrics 63, 29 (1985).
J. Harada, J. D. Axe, and G. Shirane, Phys. Rev. B 4, 155
(1971).
7K. A. Muller, Y. Luspin, J. L. Servoin, and F. Gervais, J. Phys.
43, L-537 (1982).
8F. Jona and G. Shirane, Ferroelectric Crystals (Pergamon, New
York, 1962), p. 402.
K. M. Rabe and J D. Joannopoulos, Phys. Rev. Lett. 59, 570
~
(1987); Phys. Rev. B 36, 6631 (1987).
' H. Bilz, G. Benedek, and A. Bussmann-Holder, Phys. Rev. B
35, 4840 (1987).
"S.-H. Wei and H. Krakauer, Phys. Rev. Lett. 55, 1200 (1985).
' W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
6423
on the Cray 2 at the National Center for Supercomputing
Applications and were supported by the National Science
Foundation, and some calculations were performed on
the NRL Cray XMP with computing production grants
from NRL. This work was supported in part by the
OSce of Naval Research and by National Science Foun-
dation Grant No. DMR-8719535 to H. K.
H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976);
J. D. Pack and H. J. Monkhorst, Phys. Rev. B 16, 1748
(1977).
' D. D. Koelling and B. N. Harmon, J. Phys. C 10, 3107 (1977).
'5L. Hedin and B. I. Lundqvist, J. Phys. C 4, 2063 (1971).
6E. Wigner, Trans. Faraday Soc. 34, 678 (1938), and corrected
in D. Pines, Solid State Phys. 1, 375 (1955).
' E. B. Wilson, Jr. J. C. Decius, and P. C. Cross, Molecular Vi-
,
brations: The Theory of Infrared and Raman Vibrational
Spectra (Dover, New York, 1955), p. 388.
'8Y. Luspin, J. L. Servoin, and F. Gervais, J. Phys. C 13, 3761
(1980).
' D. L. Decker and Y. X. Zhao, Phys. Rev. B 39, 2432 (1989).
L. L. Boyer et al. , Phys. Rev. Lett. 54, 1940 (1985); M. J.
Mehl, R. J. Hemley, and L. L. Boyer, Phys. Rev. B 23, 900
(1986).
M. J. Mehl, R. E. Cohen, and H. Krakauer, J. Geophys. Res.
93, 8002 (1988); 94, 1977 (1989).
R. M. Martin, Phys. Rev. B 9, 1998 (1974).
R. M. Martin and K. Kunc, Phys. Rev. B 24, 2081 (1981); M.
Pasternak, A. Baldereschi, A. Catellani, and R. Resta, Phys.
Rev. Lett. 64, 1777 (1990).
~~P. Giannozzi, S. de Gironcoli, and R. Resta, Proceedings of
the Phonons 89 Conference, Heidelberg iWorld Scientific,
Singapore, 1990), in press.
L. L. Boyer, M. J. Mehl, J. W. Flocken, and J. R. Hardy, Jpn.
J. Appl. Phys. 24, Suppl. 24-2, 204 (1985).