Acta Materialia 55 (2007) 19071917

www.actamat-journals.com

A real-space non-local phase-eld model of ferroelectric

domain patterns in complex geometries
Kaushik Dayal *, Kaushik Bhattacharya
Division of Engineering, California Institute of Technology, Pasadena, CA, USA

Received 26 August 2006; received in revised form 10 October 2006; accepted 23 October 2006
Available online 10 January 2007

Abstract

Ferroelectric perovskites are used in various transducer, memory and optical applications due to their attractive properties. In these
applications, the ferroelectric materials often have complex geometries and electrode arrangements, and are subjected to domain switch-
ing. Therefore, it is important to understand the domain pattern that forms in these geometries. However, models of domain evolution
often make assumptions like periodicity and complete shielding which make them unrealistic for these applications. In this paper, we
develop a phase-eld approach that can study domain patterns and domain evolution in complex geometries with no a priori assumption
on geometry, electrode arrangement, and dielectric properties. The key idea is a boundary element method to resolve the electrostatic
elds. We illustrate the method by examining the closure domains that form at a free surface and domain switching under cyclic electric
eld in a device with interdigitated electrodes.
2006 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: Ferroelectrics; Phase-eld models; Boundary element method (BEM); Domain patterns; Free surfaces

1. Introduction applied by various researchers to understand domain pat-

terns and their evolution in barium titanate [810], lead
Ferroelectric materials are widely used in transducers in titanate [11] and the mechanics of domain switching
sensing and actuation [1,2]. More recently, they have found [12,9,10]. While these have provided important insight,
applications in high-speed memories [3], and have been they have made substantial assumptions on the geometry
proposed as elements of microwave circuits [4], as photonic to simplify the treatment of electrostatics. In particular,
switches at small length scales [5] and as substrates for self- these works either assume periodicity so that they can use
assembly [6]. In many of these newer applications, the pro- Fourier techniques, or assume that the ferroelectric mate-
posed designs exploit domain patterns as device elements rial is completely shielded by electrodes. Unfortunately nei-
[7]. Further, fracture and failure processes in ferroelectrics ther is realistic in real devices with complex geometry,
are strongly coupled to domain patterns. All of these moti- arrangement of electrodes and free surfaces.
vate the need for understanding domain patterns in realis- The electrostatic elds generated by ferroelectric bodies
tic and complex geometries. are in general nonlocal, and elds can exist not only within
Phase eld models based on the time-dependent Devon- the material but over all space. Specically, consider a fer-
shireLandauGinzburg (TDGL) framework have been roelectric body with a domain pattern such that the polar-
ization at some point on the free surface has a component
*
normal to the surface. This causes surface charges, and in
Corresponding author. Present address: Aerospace Engineering and turn an electric eld outside the body. The energetic cost
Mechanics, University of Minnesota, Minneapolis, MN, USA. Tel.: +1
612 625 6053.
of such elds gives rise to closure domains and eects that
E-mail addresses: [email protected] (K. Dayal), bhatta@caltech. inuence the domain pattern and properties of the ferro-
edu (K. Bhattacharya). electric material. Therefore, a realistic understanding of

1359-6454/$30.00 2006 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.actamat.2006.10.049
1908 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917

domain patterns in complex geometries requires an accu- However, charge compensation is a complex environment-
rate calculation of the electrostatic potential both inside and time-dependent process and is seldom complete. Also,
and outside the ferroelectric region. This, however, requires modern devices are often packaged to operate in clean
care since it is simply not computationally feasible to dis- environments. Furthermore, since domain patterns can
cretize and compute the electric eld in all space. change, charges that compensate one pattern can increase
In this paper, we propose a computational method to the barrier for others. Therefore, this requires examination
study domain patterns in complex geometries while accu- in a specic context. We indicate in the nal section how
rately and eciently taking into account the stray electric our method can be adapted when one has charge
elds. The main idea is to use a boundary element method compensation.
for resolving the electrostatic eld in the phase-eld model We begin by presenting the phase-eld equations in Sec-
developed in Ref. [9]. A key feature of this model is that it tion 2, and the solution technique emphasizing the details
formulates the entire problem, including the electrostatics, of the boundary element method in Section 3. We present
in real space. It requires the solution of an electrostatic examples to illustrate the approach in Sections 4 and 5.
problem in all space, but requires the knowledge of the eld We conclude with a discussion in Section 6.
only inside the body to evolve the polarization. Boundary
element techniques are widely used in electromagnetism,
2. The phase-eld formulation
elasticity, acoustics and other areas that require the knowl-
edge of elds only over a nite region, but which are deter-
The potential energy E of a ferroelectric body X with
mined by a problem in all space [1315]. Therefore this
elastic strain eld e and polarization eld p may be written
method is ideally suited for our purpose. The method we
following Ref. [19] as
develop has no a priori restrictions on geometry, electrode Z Z
arrangement or dielectric moduli. 0 2
Ee; p U rp W e; pdX jr/j dV 2:1
We demonstrate our method by studying two problems X 2 R3

chosen to give a glimpse of the competition between var- where a domain wall (Ginzburg) energy density U penalizes
ious contributions to the energy in determining domain gradients in p and models the energy contributions from
patterns. The rst problem concerns closure domains domain walls, a stored (DevonshireLandau) energy den-
when we have equally spaced 180 domain walls emerging sity W penalizes p and e deviating from the spontaneous
normal to the surface of a body. The electrostatic eld polarization and strain, and the nal term represents the
generated by the surface charges gives rise the formation electrostatic energy contained in the electric eld
of small 90 closure domains, but these cost elastic and E r/, where / is obtained by solving Maxwells
domain wall energy. We examine the various scalings, equation:
and the dierence between bulk bodies and thin lms.
The problem of closure domains has been carefully stud- r  p  0 r/ 0 over R3 2:2
ied in ferromagnetism [16,17], but there is no comparable The rst two terms in the energy (2.1) are local to the body
literature in ferroelectricity. While some aspects are simi- being considered, while the electrostatic contribution re-
lar, there are dierences there are no free charges in fer- quires evaluation over all space.
romagnetism, the magnitude of the magnetization (the We obtain the governing equations as a gradient ow of
analog of polarization) is xed and the spontaneous the potential energy [9]:
strains tend to be higher in ferroelectric materials and !
these can lead to very dierent closure domain micro- dpi oU oW
l   /;i over X 2:3a
structures. Our calculations may be viewed as a rst step dt opi;j opi
;j
in this direction.  
The second problem concerns a device with interdigita- oW
0 over X 2:3b
ted electrodes which is of interest for mechanical actua- oeij ;j
tion, for optical applications to produce periodically pi;i  0 /;ii 0 over R3 2:3c
poled lithium niobate (PPLN) ferroelectric crystals for
use in optical second-harmonic generation [18], and has The boundary conditions for the polarization evolution
been proposed for tunable optical second-harmonic gener- (2.3a) are rp  ^n 0, where ^n is the outward normal to
ation devices [5]. Once again, we see that the free surfaces the boundary. Conventional elasticity boundary conditions
give rise to complex domain and switching patterns as are used for the elastic equilibrium equation (2.3b). Finally,
well as points of stress and eld concentration that can the boundary conditions associated with Maxwells electro-
lead to failure. static equation (2.3c) are potential (voltage) specied at the
In the presentation of our method, as well as in the electrodes and decaying at innity.
examples, we assume for simplicity that there is no charge We focus on barium titanate in two dimensions, and use
compensation at free surfaces. When exposed over long the same model as in Ref. [9]:
periods, dust and other sources of free charges in the envi- a0
U rp p21;1 p21;2 p22;1 p22;2 2:4
ronment can lead to compensation of the surface charges. 2
 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917 1909

a1 2 a2 a3 To be precise, the electrostatics equation and its bound-

W e; p p p22 p41 p42 p21 p22
2 1 4 2 ary conditions are
a4 6 6 a5 4 4
p1 p2 p1 p2 0 r  r/x r  px : qx over R3 3:1a
6 4
1 /x V 0 x on oX/ 3:1b
e  eT p  C  e  eT p 2:5
2 0 r/x  n^ px  n^ : rx on oXr 3:1c
We also use the constants a0 ; a1 ; a2 ; a3 ; a4 ; a5 ; C and the r/x ! 0 as x!1 3:1d
expression for eT p as in Ref. [9]. In that non-dimensional-
ized system, the spontaneous polarization is 1, the elastic The regions oX/ and oXr are the portions of the boundary
moduli are O102 and the breakdown voltage across the with electrodes and free surfaces, respectively, as shown in
thickness is O(10). Fig. 1. We assume oX/ [ oXr oX.
We begin by replacing the voltage boundary conditions
3. Solving the phase-eld evolution equations over oX/ by unknown surface charges r* so that the equa-
tions become
Our strategy to solve the governing equations is as fol- 0 r  r/x qx over R3 3:2a
lows. We begin with an initial condition for the polariza- 
rx on oXr
tion and evolve the polarization evolution equation (2.3a) 0 r/x  ^n 3:2b
in time. At each time step, we solve the elastic equilibrium r x on oX/
(2.3b) and electrostatic (2.3c) equations to nd the strain r/x ! 0 as x ! 1 3:2c
and electric potential caused by the updated polarization 0
eld. These new elds are then used to update the polariza- Let Gx; x be the fundamental (Greens) solution to this
tion, and the process is repeated. equation that satises Gx; x0 ! 0 as jx  x0 j ! 1. For
We use a square nite-dierence grid to discretize the the electrostatics equation in two dimensions,
polarization, displacement and electric potential inside the Gx; x0 lnjx0  xj=0 . In three dimensions,
body. The polarization update is achieved using a simple Gx; x0 jx0  xj1 =0 . We can then write the solution
explicit time marching scheme, with conventional nite dif- for the potential as
Z Z
ferences to model spatial derivatives. Similarly, the elasticity 0 0 0
/x Gx; x qx dx Gx; x0 rx0 dx0
equations are solved by writing the equations in terms of the X oXr
displacement and then using conventional nite dierences. Z
The electrostatic equation requires care. It is nonlocal in Gx; x0 r x0 dx0 3:3
oX/
the sense that it is dened, and determined by solving Max-
wells equation, on all space. However, we need to know Applying this relation to the boundary region oX/ where
the electric eld only inside the body to update the polari- we know that /x V 0 x:
zation. This leads to the boundary element technique, a Z Z
method of transforming the nonlocal problem dened in V 0 x Gx; x0 qx0 dx0 Gx; x0 rx0 dx0
X oXr
all space to a local problem that provides the solution only Z
over the body of interest, as an ecient means of solving Gx; x0 r x0 dx0 3:4
this equation. The key idea behind this method is the oX/
Dirichlet to Neumann map that allows us to transform
We can now solve this equation for the unknown function
our Dirichlet boundary conditions from being partially
r x0 . We do this on the nite dierence grid by assuming
specied on the boundary (given voltage at the electrodes)
that r x0 is piecewise-constant, i.e. it has a specic
and partially specied at innity (decay condition) to sim-
constant value over the region associated with the
pler Neumann boundary conditions where the surface
charge is specied on the entire boundary, including at
the electrodes.
The idea is as follows. We are given the charge density in
the interior due to the imbalance of the polarization point-
wise and, further, we are given the surface charge density
by the change in polarization at the exposed surfaces. How-
ever, we are not given the charge at regions covered by elec-
trodes; instead, we are given the voltages there. So we
replace the electrodes by unknown surface charge distribu-
tions and then solve for these charges such that the total
potential generated by all charges is exactly equal to that
specied in the original boundary conditions. Once we
know this charge distribution, we can easily obtain the
potential at any point in space. Fig. 1. Boundary conditions for the electrostatics problem.
1910 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917

corresponding grid point. Applying the equation above at polarizations are not completely compatible at the triple
each grid point in oX/ gives us a system of linear equations junctions and lead to elastic elds and elastic energy.
that are to be solved for the rm at each grid point: Roughly, the horizontal domain tries to elongate to open
X X the wedge, and we can view each triple junction as a
V 0l Gxl ; xm qm Dx2 Gxl ; xm rm Dx
disclination.
fm:xm 2Xg fm:xm 2oXr g
X
Gxl ; xm rm Dx 3:5 4.1. Scaling
fm:xm 2oX/ g

This system of linear equations can be solved to nd the A simple scaling argument demonstrates the competition
unknown rm . In two dimensions, the reference potential between two situations. Consider a crystal of unit thickness
is not well dened, and the potentials are dened relative into the plane. In the situation shown in Fig. 2a, the elastic
to each other; this requires an additional equation to be energy is eectively zero but the electrostatic energy per unit
added to the system, coming from balance of charges. length along the free surface is given by:
Now that we know r*, we can use Eq. (3.3) to nd the Z
1 0 2
electrostatic potential at any point in space, in particular, jr/j dV subject to r  0 r/ p 0
h 2 h1;1
on the surface or the interior of the body. However, we
have found that it is ecient to use Eq. (3.3) to nd the 4:1
potential / x on the surface oX and then solve the
where h is the spacing between the domains. From the elec-
boundary value problem
trostatic equation and using the non-dimensionalization
0 r  r/x qx in X 3:6a x hx0 and p p0 P, where p0 is the spontaneous polariza-
/x / x on oX 3:6b tion, we get
1   
0
using conventional and ecient techniques. rx0  rx0 / p0 P 0 4:2
h h
4. Closure domains at a free surface and to balance the terms in the bracket we have / hp0 /0 .
So, the electrostatic energy contribution in this limit can be
If the domain pattern in the body is such that the polar- written
ization at some point on the boundary has a component Z
1 0
normal to the boundary, then it gives rise to charges as jr/j2 dV
h 2 h1;1
we saw in the previous section. These charges, if uncom- !
Z
pensated, give rise to an electrostatic eld which in turn 1 2 2 0 2
p0 h jrx0 /0 j dV 0 C 1 p20 h 4:3
contributes to the potential energy. The material can avoid h 2 11;1
this energetic penalty by forming new patterns near the
boundary, but these may cost domain wall and elastic So we conclude that, in this conguration, the electrostatic
energy. Therefore the patterns near the boundary, closure energy scales as the square of the spontaneous polarization
domains, can be quite complex as the materials balance and linearly with spacing.
these various energies. We examine a very simple situation In contrast, in the situation shown in Fig. 2b, the elec-
in this section that demonstrates the various issues that are trostatic energy is eectively zero but the elastic energy
involved. per unit length along the free surface is
Consider the situation shown in Fig. 2a consisting of Z
1
equally spaced 180 domain walls emerging normal to the e  eT  Ee  eT dV
h h1;0
surface of a semi-innite body. This generates alternate Z
regions of positive (respectively, negative) surface charges 1 2 2
ah T E e0  eT 0  E0 e0  eT 0 dV 0
associated with the up or c+ (respectively, down or c) h 11;0
polarizations. This costs electrostatic energy. One way of C 2 Eh2T 4:4
avoiding this is to form closure domains with 90 domain
walls as shown in Fig. 2b. This has no surface charges. where eT is the transformation strain and E is the elastic
However, the strains associated with the spontaneous modulus, and we use the non-dimensionalization e T e0 ,

Fig. 2. Emerging 180 domains on a surface (a) without and (b) with closure domains. The computational region is marked in gray.
 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917 1911

eT T eT 0 and E EE0 . T is a scalar of the same order of face is xed to values corresponding to the spontaneous
magnitude as the spontaneous strain, and E is a scalar with strain of the domain pattern and the top surface is trac-
the same order of magnitude as the elastic modulus. Thus, tion-free. The vertical faces have zero shear traction and
in this conguration, the elastic energy scales as the square given horizontal displacement.
of the spontaneous strain and linearly with spacing. 2. Mechanically unconstrained: These boundary conditions
Comparing the two situations, materials with large represent a thin lm with the thickness of the computa-
spontaneous polarization prefer the closure domains tional domain on the order of the closure domain size.
shown in Fig. 2b while materials with large spontaneous The entire boundary is traction-free.
strain and elastic modulus prefer to retain the emerging
domain walls shown in Fig. 2a. We start with the alternating polarization distribution
In either case, both contributions increase linearly with described above and evolve the governing equations till
h. However, the domain wall energy scales with the number we reach equilibrium. For both sets of boundary condi-
of domain walls or as h1. So the energy can be written as tions, we see the nucleation of small horizontal domains
C 1 p20 h or C 2 E2T h C 3 h1 C 4 4:5 at the surface due to the large electrostatic elds generated
by the emerging polarization, and these eventually form the
where C3 depends on a0, and C4 is the DevonshireLandau closure domains. The polarization distribution is as shown
energy. Optimizing this expression for the optimal value of in Fig. 4. This gure shows the mechanically constrained
the domain wall spacing gives case, but the other case is similar.
 12 However, the elastic energy and stress that builds up in
C3
h : 4:6 the two cases are quite dierent, as can be seen in Fig. 5.
C 1 p20 or C 2 E2T
The horizontal polarization in the closure domains induce
a surface layer of compressive stress as it attempts to reach
4.2. Closure domains in barium titanate its spontaneous strain. The mechanically constrained case
corresponding to the bulk material (Fig. 5a) is unable to
We examine closure domains further using the numer- relieve this stress, while the mechanically unconstrained sit-
ical method we have developed. We use a rectangular uation corresponding to the thin lm can at least partially
computational domain corresponding to the gray area relieve this stress (Fig. 5b) by bending. The deformed shape
in Fig. 2, and the initial state shown in Fig. 3. We use of the mechanically unconstrained case is shown in Fig. 6,
a grounded electrode on the bottom surface to approxi- and demonstrates the bending.
mate an innite body, in the electrical sense. We also use This dierence in mechanical stresses also has a conse-
periodicity in the horizontal direction. However, we quence on the electric eld that builds up outside the
assume that the top surface has no electrodes. The equal body. From Fig. 7, we see that the potential outside
volume fractions of the c+ and c domains ensures that the bulk body in the constrained case is considerably
we do not have any long range eects. Further, the crys- higher than in the thin lm. Since the bulk body has
tallographic basis is oriented along the rectangular direc- large elastic energy, the polarization tries to optimize
tions. We use two dierent sets of mechanical boundary the elastic energy as much as possible with higher electric
conditions. eld. In contrast, since the thin lm has lower elastic
energy, the polarization has room to optimize the electric
1. Mechanically constrained: These boundary conditions eld at least partially. This dierence may be of rele-
are shown in Fig. 3 and represent the surface of a large vance to current eorts to use such domain patterns
ferroelectric body. The displacement of the bottom sur- for self-assembly [6].

Fig. 3. Boundary conditions and initial polarization eld

1912 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917

Fig. 4. Equilibrium closure domain structure (p2 eld). The arrows are only to guide the reader.

Fig. 5. von Mises stress magnitude in the closure domains that form in the (a) mechanically constrained and (b) mechanically unconstrained situation.

Fig. 6. Bending of the lm caused by incompatibility disclinations of the closure domains (vertical displacements magnied by a factor of 100).

5. Interdigitated electrode device changed in such a manner that the energy barrier for
switching is much lower. In our system, we have positioned
As a second example, we study a device with interdig- them just below the center of each electrode. For a discus-
itated electrodes (IDE) shown in Fig. 8. We approximate sion on why nucleation in phase-eld models may be di-
the problem as being two-dimensional in this cross- cult and why soft spots and other techniques are required
section, and we use a rectangular computational domain to induce switching, see Ref. [20].
that is periodic in the horizontal direction. We ground We begin with a crystal that is entirely poled along the
the entire bottom face, apply a constant DC voltage to a-axis, with the polarization pointing towards the right.
the left electrode that extends over the rst quarter of As we begin cycling, we nd that the crystal forms complex
the top surface and a cyclic AC voltage to the right elec- domain patterns that evolve cyclically after an initial
trode that extends over the third quarter of the top transient. Fig. 9 shows snapshots from an entire cycle, rep-
surface, and leave free surfaces between the electrodes. resented by plots of both components of the polarization
All the faces are traction-free. vector eld at dierent times.
Nucleation of new domains plays an important role in In the rst snapshot at time t, Fig. 9a, the left electrode
this example, and we use the method of soft spots proposed has constant positive DC bias and the right electrode is at
in Ref. [9] to enable nucleation. The soft spots are small zero and climbing upwards. The positive voltage on the left
regions that have the DevonshireLandau parameters electrode creates a c domain structure beneath it and the
 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917 1913

Fig. 7. Electrostatic potential eld caused by closure domains (note the dierent magnitudes).

Fig. 8. Schematic of an IDE device and the idealized two-dimensional calculation.

negative voltage from the just concluded half cycle has left As the voltage on the right electrode drops, the domain
behind a c+ domain under the right electrode. There are pattern does not change much. At half-cycle beyond t,
a-axis closure domains on the exposed regions of the Fig. 9e, the right electrode is again at zero and falling.
surface with no electrode cover and some in the interior The domain patterns do not show much change, though
to satisfy compatibility approximately. As before, this the magnitude of the polarization in the c domain beneath
domain pattern is not exactly compatible at the triple junc- the AC electrode has reduced. Also, the c+ has grown
tions is and thus stressed, as shown in Fig. 10. slightly.
As the voltage increases on the right electrode, Fig. 9b, The voltage on the right electrode subsequently goes
the c+ domain beneath it shrinks and the c domain from negative and the c+ domain grows and reaches the elec-
the bottom face grows upwards and reaches the electrode. trode. At three-quarter-cycle beyond t, Fig. 9f, the right
As the positive AC voltage at the right electrode further electrode is at the lowest point of the AC cycle. The domain
increases, Fig. 9c, the a domain to the right of the AC beneath the right electrode is completely c+. The domain
(right) electrode is unfavorably oriented against the eld geometry looks very similar to that in the rst snapshot
between the electrodes and has nucleated a surface closure at the beginning of the cycle, but the magnitudes of the
a domain with the opposite orientation. The c+ domain has polarization are higher. The evolution from half-cycle
further shrunk and is replaced by the growth of the beyond t to three-quarter-cycle beyond t is approximately
c domain from the bottom face. a reversal of the process described to go from t to quar-
As the voltage reaches the peak value on the right elec- ter-cycle beyond t.
trode, at quarter-cycle beyond t, Fig. 9d, the domain As the voltage decreases beyond the negative peak, the
beneath the right electrode has completely switched to c. domain pattern does not change much and again reaches
The right closure domain is also now completely formed. Fig. 9a as the voltage becomes zero. This pattern repeats
The other domains change shape and size slightly to on further cycling.
accommodate the new domain patterns and approximately In summary, we see signicant switching under the right
roughly divergence-free. electrode as the voltage is cycled. The domain geometry is
1914 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917

Fig. 9. Snapshots of the polarization eld. In each subgure, the upper plot represents the horizontal component of p and the lower plot represents the
vertical component of p. The position on the AC cycle is schematically marked above the right electrode. The arrows are only to guide the reader. The
entire movie is available from the authors. (a) Domain patterns at time t. (b) Domain patterns between t and quarter-cycle beyond t. (c) Domain patterns
between t and quarter-cycle beyond t. (d) Domain patterns at quarter-cycle beyond t. (e) Domain patterns at half-cycle beyond t. (f) Domain patterns at
three-quarter-cycle beyond t.
K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917 1915

Fig. 9 (continued)
1916 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917

Fig. 10. von Mises stress eld at time t.

somewhat independent of the magnitudes of the applied sion of Eq. (3.6) at the beginning of the computation, large
voltages and most of the domain motion occurs in short deformation elasticity will not require much extra compu-
intervals. We always have horizontal polarization under tational eort for the electrostatics component. The extra
the free surfaces. Interestingly, the left closure domain does eort will come from having to evaluate the coecients
not change orientation even when the eld is unfavorable. in Eq. (3.5) for each time instant as the electrodes may also
This is possibly due to that fact that there is not much deform along with the ferroelectric, and hence being unable
domain rearrangement in that region. The complex domain to store the inverse for that system. An alternate approach
patterns lead to internal stresses even though the region is by Li and Aluru [21] and others in the context of MEMS
subjected to traction-free boundary conditions. Finally, we shows that formulating the electrostatics in the reference
see large polarizations at the edges of the electrodes due to conguration through a change of variables is also feasible.
the high electric elds in that region. Hence, we anticipate In this presentation, we have neglected any charge com-
failure in these regions. More sophisticated models of fer- pensation for simplicity. The method can readily be
roelectrics that incorporate fracture, breakdown and the adapted to include them: we simply replace the denition
motion of charged defects are required to understand this of r in Eq. (3.1c) with r : p  ^n rcomp where rcomp is
failure. the compensating charge.
Finally, we note that we have treated ferroelectrics as
6. Discussion insulators while they are in reality wide-bandgap semicon-
ductors [22]. It is known that space charges can play a role
We have formulated a phase eld model with a bound- at electrodes and domain walls, but they may also play a
ary element method to understand the eect of free surfaces role at free surfaces. This can be addressed by the integra-
in ferroelectric crystals. We have used this approach to tion of our boundary element approach with the model
solve two examples, closure domains and interdigitated presented in Ref. [22].
electrodes. These examples show the interesting interplay
between crystallography, electrostatics, mechanics and Acknowledgements
domain walls when one has complex device and electrode
geometry. This work draws from the doctoral dissertation of Kaus-
This technique can be extended to other geometries hik Dayal at the California Institute of Technology. We are
such as cracks and notches. This can be seen by recalling glad to acknowledge the partial nancial support of the US
the solution procedure, where after nding the unknown Army Research Oce through the MURI Grant No.
charge density at the electrodes from Eq. (3.5), the funda- DAAD 19-01-1-0517.
mental solution Eq. (3.3) is used to compute the potential
over the boundary of the body. However, Eq. (3.3) can
References
just as easily be used to compute the potential over an
arbitrary surface that encloses the ferroelectric as well as [1] Xu Y. Ferroelectric materials and their applications. Amster-
free space rather than the boundary of the body. Hence, dam: North-Holland; 1991.
we can choose a bounding surface that allows ecient [2] Uchino K. Piezoelectric Actuators and Ultrasonic Motors. Dor-
solution of Eq. (3.6). Similarly, this approach can also drcht: Kluwer; 1996.
[3] Scott JF. Ferroelectric Memories. Heidelberg: Springer; 2000.
be used to understand piezo-force microscopy experiments
[4] Tagantsev AK, Sherman VO, Astaev KF, Venkatesh J, Setter N. J
by modeling the probe tip as a constant potential spheri- Electroceram 2003;11:5.
cal electrode. [5] Dayal K, Bhattacharya K. Tunable quasi-phase-matching through
This idea can also be used for the ecient solution of ferroelectric domain switching; in preparation.
problems with large deformations, where the bounding sur- [6] Kalinin S, Bonnell D, Alvarez T, Lei X, Hu Z, Ferris J, Zhang Q,
Dunn S. Nano Lett 2002;2:589.
face can be large enough that the deformations of the body
[7] Bhattacharya K, James RD. Science 2005;307(5706):53.
are entirely within it. While large deformation elasticity [8] Hu HL, Chen LQ. J Am Ceram Soc 1998;81(3):492.
poses numerous challenges, the electrostatics will be rela- [9] Zhang W, Bhattacharya K. Acta Mater 2005;53:185.
tively straightforward. By computing and storing the inver- [10] Zhang W, Bhattacharya K. Acta Mater 2005;53:199.
 K. Dayal, K. Bhattacharya / Acta Materialia 55 (2007) 19071917 1917

[11] Li YL, Hu SY, Liu ZK, Chen LQ. Acta Mater 2002;50(2):395. [16] Hubert A, Schafer R. Magnetic Domains. Berlin: Springer; 1998.
[12] Wang J, Shi S-Q, Chen L-Q, Li Y, Zhang T-Y. Acta Mater [17] DeSimone A, Kohn RV, Muller S, Otto F. In: Bertotti G,
2004;52(3):749. Magyergyoz I, editors. Science of Hysteresis. Elsevier; 2005.
[13] Shu YC, Lin MP, Wu KC. Mech Mater 2004;36(10):975. [18] Nakamura K, Kurz J, Parameswaran K, Fejer MM. J Appl Phys
[14] Wrobel LC. The Boundary Element Method: Applications in 2002;91(7):4528.
Thermo-Fluids and Acoustics, vol. 1. New York: Wiley & Sons; [19] Shu YC, Bhattacharya K. Philos Mag B 2001;81(12):2021.
2002. [20] Dayal K, Bhattacharya K. J Mech Phys Solids 2006;54(9):1811.
[15] Kane JH. Boundary element analysis in engineering continuum [21] Li G, Aluru NR. J Microelectromech S 2002;11(3):245.
mechanics. Englewood Clis, NJ: Prentice-Hall; 1994. [22] Xiao Y, Shenoy VB, Bhattacharya K. Phys Rev Lett 2005;95:247603.