Dielectric Anomalies in Ferroelectric Nanostructures

I. Ponomareva,
1,
*
L. Bellaiche,
1
and R. Resta
2,3
1
Department of Physics, University of Arkansas, Fayetteville, Arkansas 72701, USA
2
Dipartimento di Fisica Teorica, Universita` di Trieste, Strada Costiera 11, 34014 Trieste, Italy
3
INFM DEMOCRITOS National Simulation Center, via Beirut 2, 34014 Trieste, Italy
(Received 2 August 2007; published 29 November 2007)
First-principles-based methods are used to determine the external dielectric susceptibility (i.e., the
polarization response to the external electric eld) and the internal susceptibility (i.e., the polarization
response to the average internal eld) in ferroelectric dots, wires, and lms, as a function of the electrical
boundary conditions. While the external susceptibility is obviously positive, we nd that the internal one
is negative over a wide range of boundary conditions for all kinds of nanostructures. A Landau-type
phenomenological model provides a rationale for all of our ndings.
DOI: 10.1103/PhysRevLett.99.227601 PACS numbers: 77.22.Ej, 71.15.m, 77.55.+f, 77.84.s
Ferroelectric nanostructures (FEN) are of increasing
technological interest because of their need in miniaturi-
zation of devices [1]. The low dimensionality is respon-
sible for exotic phenomena, due to depolarizing elds
effects [27]. Recently, the occurrence of negative static
dielectric susceptibilities has been predicted in thin layers
[8,9]. Interestingly, a negative dielectric susceptibility is
essential in the quest for systems exhibiting a negative
index of refraction [10], but it is unclear how a static
dielectric susceptibility can be negative without violating
any stability requirement [11]. One may also wonder if this
negative sign is only typical of thin lms or rather if other
FENs can also exhibit such anomaly. Similarly, it is worth-
while to know the inuence of the magnitude of the resid-
ual depolarizing eld and its related different dipole
patterns [4,5,12,13] on this negative sign.
Here, we show that more than one dielectric suscepti-
bility can be dened in FENs, in analogy with the magnetic
case [14]. We dene
(ext)

as the polarization response of

the nite sample of volume V to the external eld E
0
and

(int)

as the response to the internal (screened) eld (E):

(ext)

@(P

)
@E
0;
;
(int)

@(P

)
@(E

)
; (1)
where () denotes a double average: thermal average and
average over the nite sample. refers to Cartesian coor-
dinates and V(P) is the dipole of the nite sample.
We then perform microscopic simulations via a model
Hamiltonian, where the depolarizing eld can be tuned by
means of a screening parameter. We nd that
(ext)

, being
proportional to the dipolar polarizability of a nite sample,
is always positive in any studied FEN. At variance with this
and for almost the whole screening parameter range, the
internal susceptibility is negative in thin lms but also in
rods and dots. Finally, we establish a phenomenological
model based on a Landau-type approach, which nicely
reproduces (and helps understand) all of the main features
emerging from the simulations, as well as leads to another
result: there exists an intrinsic or bulk susceptibility
(i.e., that is independent of boundary conditions and di-
mensionality) that neither coincides with
(ext)

nor with

(int)

.
Here, we focus on ferroelectric nanostructures made of
disordered Pb(Zr
0:4
Ti
0:6
)O
3
(PZT60) solid solutions. We
consider two-, one-, and zero-dimensional stress-free
PZT60 nanostructures, represented by a thin lm (periodic
in two directions, x and y), a wire (periodic along one
direction, z), and a cubic dot, respectively. The x, y, and z
axes are chosen along the [100], [010], and [001] direc-
tions, respectively, and the lateral size of the structures
along nonperiodic directions is 48 A

. The Hamiltonian of
such nanostructures can be written as [4,12,13,15]:
H H
(0)
VP jE
0
(E
(MD)
)|; (2)
where E
(MD)
is the maximum depolarizing eld. [Notice
that H is a self-consistent Hamiltonian since a mean
value of E
(MD)
occurs in the denition of Eq. (2).] H
(0)
gathers all the energetic terms that do not explicitly depend
on E
0
and that correspond to an ideal open-circuit (OC)
boundary condition. More precisely, H
(0)
is an explicit
function of the local dipoles, as well as of the atomic
conguration of the alloy, macroscopic strain, and strain-
related microscopic variables [16]. The expression and
rst-principles-derived parameters of H
(0)
are those given
in Refs. [16] for PZT bulk, except for the dipole-dipole
interactions for which we use the analytical expressions of
Refs. [12,17] for nanostructures under ideal OCconditions.
The parameter allows to consider the effect of any
electrical boundary condition [12,13]. The limiting value
0 corresponds to a free sample in vacuo, that is to OC
boundary condition. The other limiting value 1, called
short circuit (SC) boundary condition, corresponds to
the sample embedded in a perfect metallic medium that
fully screens the polarization-induced charges at the sur-
faces or interfaces. The parameter thus controls the
magnitude of the residual depolarizing eld. As a result,
PRL 99, 227601 (2007)
P HYS I CAL RE VI E W L E T T E RS
week ending
30 NOVEMBER 2007
0031-9007=07=99(22)=227601(4) 227601-1 2007 The American Physical Society
the component of (E) can be written as:
(E

) E
0;
+ (1 )(E
(MD)

): (3)
(E
(MD)
) is calculated at an atomistic level via the approach
of Ref. [12] and depends on the dimensionality, size, and
dipole pattern of the investigated nanostructure. The ther-
mal averages of strain, net polarization (P), and residual
depolarizing eld (1 )(E
(MD)
) are obtained by running
Monte Carlo simulations [18] for a given temperature,
external electric eld E
0
, and screening parameter .
We calculate the susceptibilities by applying an electric
eld E
0
to all the systems and calculating the change in
polarization in response to this eld. This is achieved by
writing the change in the component of (P) as:
(P

)
(ext)

E
0;
(P

)
(int)

(E

)
(int)

jE
0;
+ (1 )(E
(MD)

)|:
(4)
These equations, and the initial conditions of our consid-
ered nanostructures, imply that the derivatives appearing in
Eq. (1) are taken at E
0;
0 and (E

) 0 in general. In
our calculations, we successively apply different electric-
eld values E
0;
, then calculate the resulting average po-
larization (P

) and maximum depolarizing eld (E

(MD)
),
and nally extract
(ext)

and
(int)

from Eq. (4). E

0;
is
increased from 0 to 100 kV=cm in a series of 5 equivalent
steps. After each step, the system is allowed to relax during
40 000 Monte Carlo steps20 000 of which are used to
calculate the polarization and the depolarizing eld.
Figures 1(a)1(c) show the dependence of
(ext)

, ob-
tained at T 300 K, on the screening parameter for
lm, wire, and dot, respectively. Figure 2 show similar data
but for
(int)

. Before commenting on these results, it is

useful to realize that, for small , the dipoles adopt a
pattern that annihilates the depolarizing eld. For instance,
wires and lms develop a homogeneous polarization along
a periodic direction whereas a ferroelectric dot develops a
vortex of dipoles [4,5,7,12,13] (see left insets of Fig. 1). As
the screening coefcient increases and approaches its
upper 1 limit, all of the nanostructures undergo a
transition at a critical value
crit
(indicated by a vertical
line in Figs. 1 and 2), beyond which all exhibit a homoge-
neous polarization along a nonperiodic direction
[4,5,7,12,13]see right insets of Fig. 1.
Let us now comment some striking results displayed in
Figs. 1 and 2. We restrict our attention to the
(ext)

and
(int)

for which the component corresponds to a direction

along which the depolarizing eld can exist that is along
z in the lm, along x or y in the wire, and x, y, or z in the
dot. One can rst realize that the
(int)

coefcients neither
diverge nor peak at
crit
for any system, but rather exhibit a
discontinuity with a change of sign for a value very close
to 1, and to be denoted by
disc
. As a result, these
(int)

coefcients are negative for most of the range. These

unusual features contrast with
(ext)

that diverges or peaks

at
crit
, and that is always positive for any electrical
boundary condition. It can also be seen that
(ext)

strongly
depends on for a given nanostructure. It is also sensitive
to the dimensionality of the material at a xed . On the
other hand, the
(int)

associated with the depolarizing eld

are much less sensitive to the materials shape and electri-
cal boundary conditions when is lower than
crit
.
Interestingly, features depicted in Figs. 1 and 2 may be
experimentally checked since measuring
(ext)

simply
0
50
100
150
200

e
x
t

xx
ext

yy
ext

zz
ext
0
50
100
150
200

e
x
t
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0
50
100
150
200

e
x
t
(a)
(b)
(c)
FILM
WIRE
DOT
z
z
y
x
y
x
FIG. 1 (color online). External dielectric susceptibility
(ext)

of
a PZT60 lm [part (a)], wire [part (b)], and dot [part (c)] vs the
screening parameter . Left and right insets show the projections
of the dipole patterns in the structures under OC-like and SC-like
boundary conditions, respectively. The vertical lines characterize
the transition from SC-like conditions to OC-like conditions.
0 0.2 0.4 0.6 0.8

-6
-4
-2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

-20
-10
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

-20
-10
0
-40
0
40
80
120
160

i
n
t

zz
int
-80
-40
0
40
80
120
160

i
n
t

xx
int

yy
int
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

-80
-40
0
40
80
120
160

i
n
t
(a)
(b)
(c)
FILM
WIRE
DOT
FIG. 2 (color online). Internal dielectric susceptibility
(int)

of
a PZT60 lm [part (a)], wire [part (b)], and dot [part (c)] vs the
screening parameter . The vertical lines characterize the tran-
sition from SC-like conditions to OC-like conditions. The insets
emphasize that the
(int)

coefcients are slightly negative for

0, and become more negative as increases. Note that

(int)
xx

(ext)
xx
and
(int)
yy

(ext)
yy
in the lm, while
(int)
zz

(ext)
zz
in the wiresince no depolarizing eld exists along these
directions. These latter coefcients are already shown in Fig. 1.
PRL 99, 227601 (2007)
P HYS I CAL RE VI E W L E T T E RS
week ending
30 NOVEMBER 2007
227601-2
involves the observation of the change in polarization
under an applied eld, while experimentally determining

(int)

can be accomplished via the new development of

Ref. [19] that yields the total internal eld.
In order to provide some rationale about the aforemen-
tioned points, we address quite generally a homogeneous
sample of ellipsoidal shape, in which case we have [20]
(E
(MD)

) 4n

(P

); (5)
where n

are the depolarizing coefcients [21]. Special

cases are the 0D-like case of the sphere (n
x
n
y
n
z

1=3), the 1D-like case of a cylinder along z (n
x
n
y

1=2, n
z
0), and the 2D-like case of a slab normal to z
(n
x
n
y
0, n
z
1). In terms of such coefcients,
Eq. (3) becomes
(E

) E
0;
4n

(1 )(P

): (6)
Combining Eqs. (4) and (6) then yields:

(ext)

(int)

1 4n

(1 )
(ext)

: (7)
Next, we are going to establish a phenomenological
model in which the free energy F (per unit volume) of
the nanostructures can be approximated by a Landau-type
expression [21] as follows:

F

_
a

(P

)
2
2
+
b

(P

)
4
4
(P

)E
0;
+ 2n

(P

)
2
(1 )
_
(8)
with a

negative (since we are interested in temperatures

below the Curie point) and b

positive. We neglect the

coupling between the different Cartesian components of
the polarization because PZT60 (at T 300 K) has a
polarization pointing along a (001) direction in the bulk
and in its stress-free nanostructure forms (when they adopt
a homogeneous dipole pattern, see insets of Fig. 1) [22].
Minimizing F with respect to the order parameter (P

), we
obtain the following equilibrium condition:
ja

+ 4n

(1 )|(P

) +b

(P

)
3
E
0;
0: (9)
In the absence of an external electric eld, we thus have:

(P
/

) 0 for a

+ 4n

(1 ) 0;
(P
/

)
...............................................

+ 4n

(1 )
b

_
for a

+ 4n

(1 ) 0; (10)
where the notation (P
/

) denotes the component of the

polarization under no external electric eld.
For a direction along which a depolarizing eld can
exist, the transition between a state with zero and nonzero
(P
/

) thus occurs for

crit
such as:

crit
1 +
a

4n

: (11)
Moreover, one can obtain from Eq. (9):

@(P

)
@E
0;

1
a

+ 4n

(1 ) + 3b

(P

)
2
: (12)
From Eq. (10) and (12), it follows that:

(ext)

@(P

)
@E
0;

E
0;
0

1
a

+4n

(1)
for
crit
;

(ext)

@(P

)
@E
0;

E
0;
0

1
2ja

+4n

(1)|
for
crit
; (13)
where the }
E
0;
0
symbol emphasizes that these results are
valid near E
0;
0.
Equations (11) and (13) explain why (i)
(ext)

is positive
for any , (ii) diverges at the transition point
crit
, and
(iii) is dependent on both the material shape (via the n

depolarization factor) and the electrical boundary condi-

tion (via the coefcient) as consistent with Figs. 1.
Furthermore, one gets from Eqs. (7) and (13) that:

(int)

1
a

for
crit
;

(int)

1
j2a

+ 12n

(1 )|
for
crit
:
(14)
Equations (11) and (14) imply that
(int)

diverges at

disc
(1 + 2
crit
)=3, which is always larger than
crit
therefore explaining why
(ext)

and
(int)

diverge/peak at
different values. Interestingly, Eqs. (11) and (14) also
imply that
(int)

is (i) negative for <

disc
, (ii) continuous
across
crit
, and (iii) is only independent of n

and for
<
crit
, as consistent with our rst-principles-based
simulations. Moreover, it is obvious from Eq. (8) that
(P

)
@F
@E
0;
: (15)
As a result, we have:

(ext)

@(P

)
@E
0;

E
0;
0

@
2
F
@E
2
0;

E
0;
0
;

(int)

@(P

)
@(E

E
0;
0

@
2
F
@E
0;
@(E

E
0;
0
;
(16)
where the second derivatives of F are taken at E
0;
0 and
(E

) 0 in general.
The stability condition (near the equilibrium case for
which E
0;
0 and (E

) 0) requires [11,23]:
PRL 99, 227601 (2007)
P HYS I CAL RE VI E W L E T T E RS
week ending
30 NOVEMBER 2007
227601-3

@F
@E
0;

E
0;
0
0;
@F
@(E

E
0;
0
0;

@
2
F
@E
2
0;

E
0;
0
>0;
@
2
F
@(E

)
2

E
0;
0
>0;
@
2
F
@E
2
0;

E
0;
0
@
2
F
@(E

)
2

E
0;
0

_
@
2
F
@E
0;
@(E

E
0;
0
_
2
>0:
(17)
Comparing Eqs. (16) and (17) implies that
(ext)

must be
positive whereas
(int)

can adopt any sign without violating

the stability requirements.
Next, we dene the intrinsic susceptibility as

@(P

)
@(E

(E

)0
; (18)
which differs from
(int)

because the former involves the

derivative of the polarization taken at (E

) 0 (and
E
0;
0, in general) while the latter is related to the
derivative of the polarization at E
0;
0 (and (E

) 0
in general).

thus corresponds to a situation in which

the unperturbed nanostructure is in zero average internal
eld. This happens if the nanostructure is under ideal SC
conditions (which is a situation that is most likely never
achieved in a real nanostructure [2] ), or if the applied
external eld exactly overcomes the residual depolarizing
eld when moving away from ideal SC conditions. This
latter condition was not satised in our simulations de-
picted in Figs. 1 and 2 (which prevents us from predicting

from our numerical approach).

To obtain an analytical expression for

, let us rewrite
Eq. (9) as:
a

(P

) +b

(P

)
3
(E

) 0: (19)
In the absence of a total internal eld, the nonzero solution
for the component of the polarization is thus given by:
(P
//

)
...........

_
; (20)
where we use the notation (P
//

) to denote the component

of the polarization under zero total internal eld. As a
result, we get from Eqs. (18)(20):

@(P

)
@(E

(E

)0

1
a

+b

(P
//

)
2

1
2a

: (21)
In other words,

(which is positive since a

<0) is the
susceptibility of the FEN that is independent of both the
shape and electrical boundary conditions. It is equal to the
susceptibility of the bulk, assuming that a

does not differ

between the bulk and nanostructures.
In summary, rst-principles-based simulations con-
rmed that the external susceptibility is always positive
and showed that it peaks at the critical value of the depolar-
izing eld associated with a change in dipole arrangement.
For the internal susceptibility, our simulations show that it
is negative for almost any electrical boundary condition
(except close to the ideal SC case). Such simulations also
indicated that the external and internal susceptibilities both
depend on electrical boundary condition and dimensional-
ity. A simple phenomenological model was developed to
(i) understand such features, (ii) explain why they do not
contradict stability requirement, and (iii) determine the
susceptibility that is intrinsic to the material.
This work is supported by NSF Grants No. DMR-
0701588, No. DMR-0404335, and No. DMR-0080054
(C-SPIN), by ONR Grants No. N00014-04-1-0413,
No. N00014-07-1-1095, and No. N00014-01-1-0365
(CPD), and by DOE Grant No. DE-FG02-05ER46188.
*iponoma@uark.edu
[1] J. F. Scott and C. A. P. de Araujo, Science 246, 1400
(1989).
[2] J. Junquera and P. Ghosez, Nature (London) 422, 506
(2003).
[3] D. Fong et al., Science 304, 1650 (2004).
[4] I. Kornev et al., Phys. Rev. Lett. 93, 196104 (2004).
[5] I. Ponomareva and L. Bellaiche, Phys. Rev. B 74, 064102
(2006).
[6] H. Fu and L. Bellaiche, Phys. Rev. Lett. 91, 257601
(2003).
[7] I. Naumov et al., Nature (London) 432, 737 (2004).
[8] A. Bratkovsky and A. Levanyuk, Phys. Rev. B 63, 132103
(2001).
[9] A. Bratkovsky and A. Levanyuk, Appl. Phys. Lett. 89,
253108 (2006).
[10] A. Pimenov et al., Phys. Rev. Lett. 95, 247009 (2005).
[11] O. Dolgov et al., Rev. Mod. Phys. 53, 81 (1981).
[12] I. Ponomareva et al., Phys. Rev. B 72, 140102 (2005).
[13] I. Ponomareva et al., Phys. Rev. B 72, 214118 (2005).
[14] R. Waldron, Br. J. Appl. Phys. 14, 700 (1963).
[15] B.-K. Lai et al., Phys. Rev. Lett. 96, 137602 (2006).
[16] L. Bellaiche et al., Phys. Rev. Lett. 84, 5427 (2000).
[17] I. Naumov and H. Fu, arXiv:cond-mat/0505497.
[18] N. Metropolis et al., J. Chem. Phys. 21, 1087 (1953).
[19] D. J. Kim et al., Phys. Rev. Lett. 95, 237602 (2005).
[20] Reference [12] conrmed that Eq. (5) is valid for ultrathin
nanostructures, but the depolarization coefcients can
slightly deviate from their continuum-derived values.
[21] L. Landau and E. Lifshitz, Electrodynamics of Continuous
Media (Pergamon, New York, 1984), Chap. 2.
[22] We do not consider any gradient term of polarization in
Eq. (8) because we deal with properties (such as (P

) and
(E
(MD)

)) that are averaged over the sample.

[23] B. Strukov and A. Levanyuk, Ferroelectric Phenomena in
Crystals: Physical Foundations (Springer, New York,
1998).
PRL 99, 227601 (2007)
P HYS I CAL RE VI E W L E T T E RS
week ending
30 NOVEMBER 2007
227601-4