Amr2013 065 06 060802

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/260633648
Advances in Developing Electromechanically Coupled Computational

Methods for Piezoelectrics/Ferroelectrics at Multiscale
Article in Applied Mechanics Reviews · October 2013

DOI: 10.1115/1.4025633
CITATIONS READS
11 2,999
6 authors, including:
Faxin Li Bin Liu

Peking University Tsinghua University
123 PUBLICATIONS 1,181 CITATIONS 136 PUBLICATIONS 3,535 CITATIONS
SEE PROFILE SEE PROFILE
Yihui Zhang Jiawang Hong

Tsinghua University Beijing Institute of Technology
180 PUBLICATIONS 7,741 CITATIONS 97 PUBLICATIONS 2,069 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Flexoelectricity View project
3D structure View project
All content following this page was uploaded by Faxin Li on 02 December 2014.
The user has requested enhancement of the downloaded file.

Advances in Developing
Electromechanically Coupled
Daining Fang1
State Key Laboratory for
Computational Methods
Turbulence and Complex Systems,
College of Engineering,
Peking University,
for Piezoelectrics/Ferroelectrics
Beijing 100871, China;
Department of Engineering Mechanics,
at Multiscale
Tsinghua University,
Beijing 100084, China Piezoelectrics and ferroelectrics have been widely used in modern industries because of
e-mail: [email protected] their peculiar electromechanical coupling properties, quick response, and compact size.
In this work, we give a comprehensive review of our works and others’ works in the past
Faxin Li decade on the multiscale computational mechanics methods for electromechanical cou-
State Key Laboratory for pling behavior of piezoelectrics and ferroelectrics. The methods are classified into three
Turbulence and Complex Systems, types based on their applicable scale (i.e., macroscopic methods, mesoscopic methods,
College of Engineering, and atomic-level methods). In macroscopic methods, we first introduce the basic linear fi-
Peking University, nite element method and employ it to analyze the crack problems in piezoelectrics. Then,
Beijing 100871, China the nonlinear finite element methods are presented for electromechanically coupled de-
e-mail: [email protected] formation and the domain switching processes were simulated. Based on our developed
nonlinear electromechanically coupled finite element method, the domain switching
instability problem was specially discussed and a constrained domain-switching model
Bin Liu was proposed to overcome it. To specially address the crack problem in piezoelectrics,
we further proposed a meshless electromechanical coupling method for piezoelectrics. In
Yihui Zhang mesoscopic methods, the phase field methods (PFM) were firstly presented and the simu-
lation results on the defects effect and size effect of deformation in ferroelectrics were
Jiawang Hong given. Then, to solve the computational complexity problem of PFM in polycrystals, we
proposed an optimization-based computational method taking the interactions between
Department of Engineering Mechanics, grains in an Eshelby inclusion manner. The domain texture evolution process can be cal-
Tsinghua University, culated, and the Taylor’s rule of plasticity has been reproduced well by this optimization-
Beijing 100084, China based model. Alternatively, the domain switching in polycrystalline ferroelectrics can be
simulated by a proposed Monte Carlo method, which treated domain switching as a sto-
chastic process. In atomic-level methods, we firstly introduce the first-principles method
Xianghua Guo to calculate polarization and studied the topological polarization and strain gradient
State Key Laboratory Explosion and effect in ferroelectrics. Then, we present a modified electromechanically coupled molecu-
Safety Science, lar dynamic (MD) method for ferroelectrics based on the shell model and investigated
Beijing Institute of Technology, the size effect of electromechanical deformation in ferroelectric thin films and nanowires.
Beijing 100081, China Finally, we introduced our recently proposed novel atomic finite element method
(AFEM), which has higher computational efficiency than the MD. The deformation as
well as domain evolution processes in ferroelectrics calculated by AFEM were also pre-
sented. The development of electromechanically coupled computational mechanics meth-
ods at multiscale is greatly beneficial, not only to the deformation and fracture of
piezoelectrics/ferroelectrics, but also to structural design and reliability analysis of smart
devices in engineering. [DOI: 10.1115/1.4025633]
1 Introduction electric field. The ferroelectricity was first discovered in 1920 in

Rochelle salt by Valasek. The prefix ferro, meaning iron, was bor-
The piezoelectricity, first discovered by the Curie brothers in
rowed from ferromagnetism, as they both exhibit hysteresis loops,
1880 in quartz, is the linear response of materials between the me-
although most ferroelectric materials do not contain iron. When a
chanical (electric) input and the electric (mechanical) output. The
ferroelectric material is subjected to an applied stress (or electric
piezoelectric effect in all crystals is reversible. The electric
field), its polarization (or strain) will change linearly with the
response to the mechanical stimulus is called direct piezoelectric
applied load; thus, all ferroelectric materials are piezoelectric.
effect, and its inverse is called converse piezoelectric effect. Pie-
Ferroelectric materials usually have very strong piezoelectric
zoelectric materials have had wide applications in the past 100
effect compared to the nonferroelectric piezoelectrics. So far, the
years as oscillators, transducers, sensors, actuators, ultrasonic
most widely used piezoelectric materials are lead titanate zircon-
motors, etc. [1].
ate (PZT) ceramics, which are also ferroelectric. Besides the pie-
Ferroelectrics are such crystals that, under a certain temperature
zoelectric applications, ferroelectric materials can also be used as
(Curie temperature), spontaneous polarization can appear in the
nonvolatile memories because of their characteristics of polariza-
crystal and the polarization direction can be oriented by an applied
tion switching.
The spontaneous polarization directions of ferroelectrics are
1
Corresponding author.
crystal symmetry–dependent. The number of equivalent polar
Manuscript received November 19, 2012; final manuscript received September axes is six for tetragonal ferroelectrics, eight for rhombohedral,
23, 2013; published online October 28, 2013. Assoc. Editor: Bart Prorok. twelve for orthorhombic, etc. [2,3]. The polarization directions
Applied Mechanics Reviews Copyright V

C 2013 by ASME NOVEMBER 2013, Vol. 65 / 060802-1
Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

inside a material may be different in different locations. The small ferroelastic domain switching may also generate large internal
zones inside the materials with all polarization oriented along the stress, which is apt to degrade or even break the material. In ferro-
same direction are call domains, and different domains are repre- electric applications, such as nonvolatile memories, domain
sented by their polarization directions. The interface between two switching is expected to accomplish completely at the designated
different domains is called domain wall, which is a thin layer with voltage to ensure the signal-writing process stable and repeatable.
the thickness usually of only several nanometers. The formation Therefore, in both the piezoelectric and ferroelectric applications,
of domain walls in ferroelectric materials is to reduce the depola- domain switching should be accurately controlled, before which
rization energy (by forming 180 deg domain walls) or the elastic the domain-switching mechanism must be clearly understood
energy (by forming non-180 deg domain walls) of the whole ma- under electric loading and/or stress loading.
terial system. Obviously, the types of domain walls are also de- In order to depict the mechanism of complicated domain
pendent of the crystal symmetry. In ferroelectric single crystals, switching under coupled electromechanical loading, quite a lot of
the domain and domain wall directions are limited by the crystal theoretical work has been done in the past decades on the nonlin-
orientations, as illustrated in Fig. 1(a). In polycrystalline ferro- ear constitutive models of ferroelectrics [5–17]. These models can
electric ceramics, because of the random orientations of numerous be classified into two types, the phenomenological models [5–11]
grains and the grain boundary effect, the domain patterns are very and the micromechanical models [12–17]. The former usually
complicated, as shown in Fig. 1(b). takes the remnant polarization and/or remnant strains as internal
In ferroelectrics, the spontaneous polarization is always accom- variables, and their evolution laws are treated in a similar way as
panied by a spontaneous strain, which is usually transversally iso- that of the plastic strain in elastoplastic materials. Such models
tropic around the polar axis. For ferroelectrics with multiple polar can well describe the material behavior if using suitable fitting pa-
axes, the spontaneous strain also has several equivalent orienta- rameters while the detailed mechanism of domain switching can-
tions and can be reoriented by an applied stress. Such properties not be included. The rather limited experimental data make the
are called ferroelasticity, which can also exist independent of fer- phenomenological models mostly concentrate on the uniaxial
roelectricity. For example, the martensitic phases in shape mem- loading cases, and the three-dimensional model is just at the hy-
ory alloys are actually ferroelastic. Most ferroelectric materials, pothesis stage [11]. In the micromechanical models, the material
such as lead titanate zirconate (PZT) ceramics, BaTiO3 crystals, is usually composed of numerous grains/domains; the evolution
and other lead magnesium niobate titanate ceramics, are both fer- law of each domain is based on a domain-switching criterion that
roelectric and ferroelastic. can easily take into account both the electric and mechanical con-
The polarization directions of ferroelectric domains can be tributions. These models are very easy to understand, while the
switched by a large electric field or by a high stress if the material real domain-switching process is very difficult to reproduce,
is also ferroelastic, which is called domain switching. The 180 because of the large computational complexity. The fracture stud-
deg domain switching will cause polarization variations in materi- ies of ferroelectric materials were also conducted [18–24], but
als but will not change the strain state, because the spontaneous among which most are based on the linear piezoelectric constitu-
strains are the same for domains with opposite directions. Non- tive model to get the analytical field solutions near the crack tip.
180 deg domain switching will both polarization change and strain The small-scale yielding model in elastoplastic materials has been
change in the materials, and their types are obviously crystal sym- extended to ferroelectric materials to study the effect of domain
metry–dependent. Therefore, from the energetic point of view, an switching on the fracture behavior [25–28], while in these models,
electric field can cause both 180 deg and non-180 deg domain usually the compatible conditions between domains are neglected
switching, while a stress can only induce non-180 deg domain and the materials are treated as isotropic, thus not applicable for
switching. ferroelectric single crystals or poled ferroelectric ceramics. Gao
Domain switching is an irreversible and energy dissipation pro- et al. [29] investigated the effect of electric yielding based on the
cess. It usually has a considerably large energy barrier, below assumption that there is a strip saturation zone in front of the crack
which domain switching is not possible, very similar to the stress tip. This model provides a relative good estimation of electrome-
induced detwining process in martensitic shape memory alloys chanical coupling on fracture toughness of ferroelectric ceramics.
[4]. Under a low electric field or a small stress, no domain switch- Computational study is an alternative approach for the nonlin-
ing occurs and ferroelectric materials will exhibit linear deforma- ear electromechanical coupled behavior of piezoelectric/ferroelec-
tion and piezoelectric response, and their constitutive laws can be tric materials. It can deal with arbitrary materials geometries,
well characterized by linear piezoelectric equations. When do- which are not possible for the theoretical models. It can also take
main switching occurs, ferroelectric materials show intensively into account the interactions between domains or grains in a
nonlinear response in both the polarization and strain. In piezo- straightforward manner, which is difficult for the theoretical mod-
electric applications, large nonlinearities are prohibited, because it els. Furthermore, with the decreasing dimensions of devices in
makes the devices or system difficult to control. Furthermore, the modern industries, the piezoelectric and ferroelectric behavior of
Fig. 1 Schematic depictions of typical domain structures observed in ferroelectric (a) single
crystal and (b) polycrystal ceramic
060802-2 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

materials at micro- or even nanoscale are more and more con- multiple domains is minimized using an optimization method, tak-
cerned. As the dimension goes down, experimental works are ing the volume fractions of domains as the optimization variables.
rather difficult to conduct, and computational investigation is the The computational complexity of this model is fairly reduced,
most promising approach at present. making the model suitable for the 3D case using numerous grains.
In the past decades, multiscale computational studies on the Like that, the domain patterns can be calculated step by step in the
electromechanical coupled behavior of ferroelectric materials PFM and the domain texture evolution process can be calculated
have been intensively conducted by researchers in the fields of at each loading step in this model. Meanwhile, the model can
mechanics, materials, physics, and other related disciplines. With- reproduce the Taylor’s rule of plasticity very well [52]. Tang et al.
out loss of generality, the existing computational methods for fer- [51] treated domain-switching in ferroelectricity as a stochastic
roelectrics can be classified to be three types based on their process and proposed a Monte Carlo method to simulate the
applicable dimensions (i.e., macroscopic methods [30–34], meso- domain-switching process in polycrystalline ferroelectrics. The
scopic methods [35–54], and atomic-level methods [55–70]). potential energy of the whole material system is evaluated for all
The macroscopic computational methods for ferroelectrics are the possible states, and the variations of potential energy are taken
mainly based on the finite element method (FEM), both linear and as the driving force for domain switching. Their model can well
nonlinear. The linear electromechanical coupled FEM for piezo- simulate the domain-switching behavior of tetragonal, rhombohe-
electrics have been developed for about 40 years, since the pio- dral, and morphotropic ferroelectric ceramics. However, because
neer works by Allik and Hughes [71]. In the 1990s, the linear of the large computational complexity, the grain number for cal-
piezoelectric FEM (LPFEM) was intensively studied to address culation is only limited to about 100, which is too small for tex-
the linear fracture problem in piezoelectrics [30,32,72]. The FEM ture analysis.
calculation results were systematically compared with the analyti- The atomic-level computational methods for ferroelectrics may
cal results with different electric boundary conditions for cracks, include the first-principles method [66–70], the molecular dynam-
including the impermeable crack, permeable crack, and semi- ics method (MD) [55–65], and the recently proposed atomic-scale
permeable crack. So far, the linear piezoelectric element has been finite element method (AFEM) [88]. In this type of method, the
available in several commercial FEM softwares, like ANSYS, ABA- material consists of numerous atoms and the complicated interac-
QUS, etc. As ferroelectric materials will show intensive nonlinear- tions between atoms and even electrons (in first-principles calcu-
ities and hysteresis behavior under high fields and/or bipolar lation) are specially addressed; thus, the effects of surface,
fields, the LPFEM cannot address such common problems in fer- defects, and interface on the electromechanical behavior of low-
roelectrics. The nonlinear FEM for ferroelectrics (NLFEM), tak- dimensional ferroelectrics can be effectively taken into account.
ing into account the nonlinearities or the domain switching These methods can well compensate for the conventional theoreti-
behaviors, was then proposed during the mid-1990s to mid-2000s, cal models, such as addressing the scale effect, the complicated
which can further be classified to be two types: one is based on a boundary conditions, etc. On the other hand, new physics, espe-
nonlinear phenomenological constitutive model and solves the cially in the field of micro/nano ferroelectric materials, may be
field variables using a nonlinear iterative process [30,31,73], very discovered using these calculation methods, which could act as
similar to the nonlinear FEM for conventional elastoplastic mate- the basis of building new models in this field. However, because
rials, and the other is based on a domain-switching energy crite- of the rather large computational complexity of this type of meth-
rion, and the domain-switching–induced polarization jump and ods, the affordable computational dimension is usually limited to
strain jump are treated as the eigenproblems in an inclusion man- several tens of nanometers and direct experimental verification at
ner [33,34,74,75]. The nonlinear FEM has been used to analyze this scale is very difficult.
the electrode-tip stress concentration and domain-switching prob- This paper is intended to present a comprehensive review of the
lem in multilayer piezoelectric actuators and obtained some useful multiscale computational mechanics methods for piezoelectrics/
results. However, currently there are still many problems to be ferroelectrics and provide a clear picture of the recent advances in
addressed in the NLFEM, such as the hysteresis problem, the this field. In Sec. 2, the macroscopic computational methods are
domain-switching–induced numerical instability problem, the presented from the method formulation to the practical applica-
rate-dependent problem, etc. Thus, the NLFEM are still under de- tion, including the linear and nonlinear FEM for piezo/ferroelec-
velopment and cannot be commercially available at present. Other trics and the linear meshless method for piezoelectrics. The
macroscopic computation, methods for ferroelectric materials mesoscopic computational methods are introduced in Sec. 3,
may include the linear piezoelectric meshless methods [76–78], including the phase field model (PFM) and the micromechanical
which have the advantage in addressing the crack problem or models. The models are described in detail, and the evolution of
other singular problems. domain structures in single crystals and domain textures in poly-
Currently, the mesoscopic computational methods for ferroelec- crystals are both presented. In Sec. 4, we firstly present the theo-
trics mainly include the phase field methods (PFM) retical basis of the atomic-scale computational methods for
[35–50,79–82] and the micromechanical computational methods ferroelectrics and then give some examples of these methods in
[51–54]. The PFM for ferroelectrics is based on the Landau–De- the application of low-dimensional ferroelectrics. Finally, the cur-
vonshire theory of phase transition and searches the energy min- rently existing problems are summarized and a perspective of the
ima state of the whole material system using the time-dependent computational mechanics methods for ferroelectrics is discussed.
Ginzburg–Landau (TDGL) evolution equation [35]. No prior
domain-switching criterion is needed for the switching process,
and the domain structures under any prescribed electromechanical 2 Macroscopic Electromechanically Coupled
loading can be obtained simultaneously by minimization of the Computational Methods
free energy. The PFM is very powerful in simulating the domain Currently, the macroscopic computational methods for piezo-
structure evolution process and related phenomena in normal fer- electrics/ferroelectrics mainly include the linear and nonlinear fi-
roelectrics [36] and relaxors [83]. However, due to the computa- nite element methods [32,34] and the linear meshless method
tional complexity, PFM is usually used for single crystals in the [78], which will be introduced, respectively, in this section.
2D case, and few studies have been addressed for the polycrystal-
line [84–86] or the 3D case [87]. To address the complicated
domain-switching process in polycrystalline ferroelectric 2.1 Linear Electromechanically Coupled Finite Element
ceramics, Li et al. [52,53] proposed an optimization-based compu- Method. Finite element method (FEM) has been proven to be a
tational model that takes the interactions between grains in an powerful tool for analyzing complicated systems. For piezoelec-
elastic Eshelby inclusion manner. Under any prescribed electro- tric devices, such as piezoelectric multilayer actuators with more
mechanical loading, the free energy of a whole grain consisting of complex geometries, the electromechanical-coupled FEM is
Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-3

ð
required to simulate the stress/electric field concentration near a
ðrij;i þ Tij;i þ fi Þduj þ ðDi;i pe Þd/ dV
crack or the electrode tip and further the failure or degradation V
behavior. Since the original work of Allik and Hughes [71], linear ð
finite element method for piezoelectric materials has been inten- ¼ ðrij dui;j Di d/;i þ Tij;i duj ÞdV
V
sively studied. Wang and Shen [89] obtained the variational ð ð ð
theory for thermal, electric, magnetic, and elastic materials. Qi þ ðfi dui þ pe d/ÞdV þ T j duj dS þ pe /dS (2.1.7)
et al. [32] built a full coupled linear piezoelectric finite element V Sr SD
method and analyzed an elliptical cavity in infinite transversely
isotropic media under fat-field electrical or mechanical load. The Considering the constitutive equation, Eqs. (2.1.3a) and (2.1.3b),
LPFEM results coincide with the analysis solution [32], indicating and the linear geometry equation, Eq. (2.1.4a), the first term of the
that LPFEM is applicable to a variety of piezoelectric samples right hand of Eq. (2.1.7) can be written as
and structures with geometrical singularity. ð
In this subsection, starting from the basic equilibrium equation,
ðrij dui;j Di d/;i þ Tij;i duj ÞdV
the electromechanical potential function is obtained and then a V
FEM formulation is presented. Some problems encountered in the ð
numerical formulation have been discussed. As a numerical exam- ¼ ðrij dei;j þ Di dEi þ Tij;i duj ÞdV
V
ple, an elliptical cavity in infinite transversely isotropic media ð
under fat-field electrical or mechanical load is considered. 1 1
¼d Cijmn emn eij þ eijm Em eij þ dij Ei Ej dV
2 2
ðV
2.1.1 Formulation of Linear Finite Element Method Based on
Piezoelectric Constitutive Equations. The governing equations of þ Tij;i duj dV (2.1.8)
V
the electromechanical coupling problem for piezoelectric materi-
als are presented as follows [32]: The last term of Eq. (2.1.8) is a nonlinear term, but in static analy-
Equilibrium equations sis, it can be omitted. Thus, the potential of the system in domain
with its V boundary @V is obtained,
rij;i þ Tij;i þ fi ¼ 0 (2.1.1)
Di;i ¼ p e
(2.1.2) Y ð 1 1

¼ Cijmn eij emn eijm eij Em dij Ei Ej dV
V 2 2
Constitutive equations for linear piezoelectrics ð ð ð
ðfi ui pe /ÞdV T i ui dS pe /dS (2.1.9)
V Sr SD
rij ¼ CEijmn emij Em (2.1.3a)
e The real solution makes potential function reach the minimum
Dj ¼ ejmn emn þ kjm Em (2.1.3b)
value, and the following variational equation exists:
Linear geometry equations Y
d ¼0 (2.1.10)
1
eij ¼ ui;j þ uj;i (2.1.4a)
2 Considering the mechanical equilibrium equation and electric
@u equilibrium equation, the following symmetric relationship can be
Ei ¼ (2.1.4b) found:
@xi
ui /; eij Ei ; rij Di (2.1.11)
Boundary conditions
Then, a displacement-electric potential–type finite element for-
rij nj ¼ T j on Sr (2.1.5a)
mulation can be obtained. Taking the nodal displacement ui and
ni Di ¼ pe on SD (2.1.5b) nodal potential / as basic variables, the displacement and poten-
tial at point Pðn1 ; n2 ; n3 Þ in the element can be interpolated as
ui ¼ ui on Su (2.1.5c)
/ ¼ / on S u
(2.1.5d) X
n
ðjÞ ðuÞ
ui ¼ Nj ui ¼ Nik ak ; k ¼ 1; 2; 3; ::::; 3n (2.1.12a)
j¼1
in which ui is the displacement vector and / is the electric poten-
tial, eij is the linear strain tensor and Ei the electric field vector, rij X
n
ð/Þ
/¼ Nj /ðjÞ ¼ Nj aj (2.1.12b)
is the stress tensor, Di is the electric displacement vector and Tij j¼1
the Maxwell electrostatic stress, and fi and pe are body force and
body electric charge, respectively. Tj is the surface stress and pe is
the surface charge per unit area, ui are components of the pre- where n is the node number in an element, Nj is an interpolation
scribed displacement, and / is the prescribed electric potential. function, aðuÞ is the nodal displacement vector, and að/Þ is the
Based on the virtual work principle, for the real solution of the nodal potential vector.
system in the domain V with the boundary @V, the following var- The element nodal generalized displacement vector can be
iational equation exists: defined as
ð h T i
ae ¼ aðuÞ ð/ÞT
ðrij;i þ Tij;i þ fi Þduj þ ðDi;i pe Þd/ dV e ae (2.1.13)
V
ð ð
ðrij T j Þduj dS ðDi ni pe Þd/dS ¼ 0 (2.1.6) then,
Sr SD

1 ðuÞT ðuÞ ðuÞ 1 T KeðuÞ 0
Applying the generalized Green–Gauss theorem, the first term of a Ke ae ¼ ae ae (2.1.14a)
the left hand in Eq. (2.1.6) can be written as 2 e 2 0 0

1 ðuÞT ðu/Þ ð/Þ 1 T 0 Keðu/Þ Table 1 Comparison of the stress concentration factors at the
a Ke ae ¼ ae ae (2.1.14b) rim of an elliptical cavity in PZT-4 ceramics using three different
2 e 2 0 0
solutions
1 ðuÞT ðu/ÞT ð/Þ 1 T 0 0
a Ke ae ¼ ae T a (2.1.14c)
2 e 2 Keðu/Þ 0 e Electroelastic Electroelastic
Elastic solution solution Error Difference
1 ðuÞT ð/Þ ð/Þ 1 T 0 0 a/b solution (FEM) (analytical) (%) (%)
ae Ke ae ¼ ae a (2.1.14d)
2 2 0 Keð/Þ e
3 1.622 1.724 1.743 1.09 7.5
1 2.870 3.270 3.230 1.24 12.5
So the first term of P in Eq. (2.1.9) can be written as 1/3 6.610 7.518 7.700 2.36 16.5
" # 1/10 19.7 22.42 23.26 3.61 18.0
X1 KeðuÞ Keðu/Þ X1 0
aTe T
ae ¼ aTe Ke ae (2.1.15)
e
2 K ðu/Þ
K ð/Þ
e
2
e e After taking the above dimension-changing procedure, the ratio of
the maximum and minimum decreases to 10192n and the illness
In the above formulation, the generalized displacement vector is of the equation is eliminated.
not in sequence of nodal
0
number. Therefore, in programming the
row and column, Ke must be adjusted to meet the request of the
assembling of global stiffness matrix. 2.1.2 Results and Discussion
For the electromechanical-coupled problems, the elements on
the diagonal of the global stiffness matrix are not always positive, Stress concentration factor of an elliptical cavity in infinite
because the electric field is the negative gradience of the electric piezoelectric media. To show the validity of the electro-
potential, which means that the global stiffness matrix is not posi- mechanical-coupled FEM, the stress concentration factor at the
tively defined. But it still remains a sparse matrix. In ordinary rim of an elliptical cavity in an infinite PZT-4 piezoelectric media
static and dynamic problems, it will not raise any difficulty in nu- is considered first. The PZT-4 ceramic is poled along the z-axis,
merical solutions, and in the eigenvalue problem, it can be over- and thus, the problem is a typical plane strain case in the Y-Z
come by adding a positive value to the diagonal element. So the plane, in which the Z-direction is the transversely isotropic axis.
nonpositively defined matrix will not introduce extra difficulty in The problem considered here is the same as that addressed by
the linear FEM solution. Sosa analytically [90]. The tensile load is applied at infinity along
In an electromechanical FEM analysis, the fact that the elastic the Y-direction, and the two main axes of the ellipse, a and b, are
constants, the dielectric permittivity constants, and the piezoelec- along the Y- and Z-direction, respectively. The comparison
tric constants have different orders will make the stiffness matrix between the numerical solution of FEM and the analytical solu-
extremely ill conditioned and leads to unstable results. For exam- tion [90] is presented in Table 1, where the elastic solution with-
ple, in PZT-4 material, the constants have the following orders, out the electromechanical-coupling effect is also presented.
respectively: It can be seen from Table 1 that the FEM results fit well with
the analytical solution, while, when the ratio of a/b increases, the
error of the FEM solution increases because the conventional
C 1010 Nm2 ; e 100 Cm2 ; d 109 N1 m2 C2 (2.1.16)
mesh is difficult to catch the sharping tip of the cave. Reason ally,
the FEM error will decrease if a more dense mesh is adopted. Ta-
and the three parts of the element stiffness matrix have the orders ble 1 also shows that the stress concentration factor calculated by
using purely elastic consideration (i.e., neglecting the electrome-
KðuÞ 10
e 10 ; Kðu/Þ
e 100 ; Kð/Þ
e 10
9
(2.1.17) chanical coupling) is smaller than that given by an electromechan-
ical method. For the PZT-4 ceramics, the difference between
It can be seen that the ratio of the maximum elements to the mini- these two methods can reach 18% when a/b reduces to 0.1. There-
mum elements can reach 1019 (i.e., the condition number of the fore, it is necessary to take into account the electromechanical
stiffness matrix is very large and the equation is extremely ill con- effect when analyzing piezoelectric media.
ditioned). For an ill-conditioned equation, the conventional Gaus- Electric and mechanical field at the rim of a circular hole. To
sian method cannot give reliable results, because of the numerical further validate our built electromechanical-coupling FEM, we
errors. If taking a high-precision solution method, the computa- also analyze the electric and mechanical field near the rim of a cir-
tional efficiency of the FEM will dramatically decrease. Here, a cular hole in the infinite large PZT-4 media. The mechanical load
dimension-changing method or a rescaling method is proposed to is applied at infinity in the Z-direction (i.e., the poling direction of
overcome the matrix illness. the material). Figure 2(a) shows the curve of the circumferential
The new force unit, N 0 , is defined as stress rh at the rim of the hole versus the angle. In Fig. 2(a),
Sosa’s [90] analytical solution is also presented and a good agree-
N 0 ¼ 10n N (2.1.18) ment between the FEM solution and analytical solution is
achieved. In this case, the mechanical load is applied along the
poling direction of the material, the tensile stress concentration
where n is an arbitrary number and can be assigned as 8, 9, 10,
factor decreases to 2.89 at h ¼ 0 deg, and the compressive stress
etc. Then, the new material constants take the following orders as
concentration factor is 1.46 at h ¼ 90 deg, which are both
smaller than that of the purely elastic solution without electrome-
2
C 1010 Nm2 1010n N0 m chanical coupling. Together, considering results in Table 1, it can
be concluded that the stress concentration factor will increase or
e 100 Cm2 (2.1.19) decrease due to the intrinsic electromechanical coupling, depend-
ing on the direction of the applied stress and the poling direction.
1
d 109 N1 m2 C2 10n9 N0 m2 C2 The curve of the circumferential electric displacement Dh ver-
sus the angle is shown in Fig. 2(b). Consistent with the analytical
solution by Sosa [90], Dh is antisymmetric about the z- and y-axes
and the new stiffness matrix takes the orders as and changes its direction at the point of h ¼ 45 deg and
h ¼ 135 deg. Dh reaches its minimum value at about h ¼ 66 deg
KðuÞ
e 10
10n
; Kðu/Þ
e 100 ; Kð/Þ
e 10
n9
(2.1.20) and a maximum value at about h ¼ 114 deg. The values of Dh at

Fig. 2 Variations of (a) rh and (b) Dh on the rim of the hole subjected to tension at infinity.
Reprinted from Ref. [32] with the kind permission of Elsevier Science Ltd.
Fig. 3 Variations of (a) Er and (b) Eh on the rim of the hole subjected to tension at infinity.
Fig. 4 Variations of (a) Dh and (b) Eh on the rim of the hole subjected to electric displacement
loading at infinity. Reprinted from Ref. [32] with the kind permission of Elsevier Science Ltd.
h ¼ 0 deg and h ¼ 180 deg are the local maxima and minima, is relatively simple, and Eh takes its minima and maxima at
respectively. h ¼ 0 deg and h ¼ 180 deg.
Figure 3 shows the curves of the tension-induced electric field The case of applied electric displacement at infinity along the
versus the angle h. The radial component Er has a maxima at Y-direction is also investigated. The curve of the electric displace-
h ¼ 45 deg and h ¼ 135 deg and a minima at h ¼ 90 deg, and ment versus angle h is shown in Fig. 4(a). Dh reaches its maxima
these three points are just the points at which Dh changes its direc- at h ¼ 0 deg and its minima at h ¼ 180 deg with Dh max ¼ 2:19D0
tion. The curve of the circumferential component Eh versus angle and Dh ¼ 0 at h ¼ 90 deg. Figure 4(b) gives the distribution of Eh

versus angle. It can be seen that the distribution is the same as Dh the piezoelectric effects. Based on a similar domain-switching cri-
but has a larger difference with the applied load at infinity. terion by Hwang et al. [13,33], Chen and Lynch [74] developed a
From the above numerical results using the electromechanical fully coupled, two-dimensional (2D) nonlinear finite element
coupling FEM, it can be seen that the applied mechanical and model to analyze the cracks in ferroelectric ceramics. Steinkopff
electric loads can result in various concentrations of the stress, the [96] introduced the original domain-switching criterion of Hwang
electric field as well as the electric displacement, which are all et al. [12] into the piezoelectric elements of ANSYS; interactions
sensitive to the directions of the applied electric field, applied me- between grains can then be considered effectively. Also, based on
chanical loads, and the poling direction of the material. The pretty the switching criterion of Hwang et al. [12], Zeng et al. [97] for-
good agreement between the FEM results and the analytical solu- mulated a thermal-electromechanical-coupled nonlinear FE model
tions by Sosa [90] indicates that the developed FEM formulation and studied the effect of domain switching on the force displace-
is applicable to a variety of piezoelectric samples and structures ment of a PZT-4 cantilever beam. Li and Fang [34] give an
with geometric singularity. improved computational domain-switching criterion based on the
Gibbs free energy and formulate a fully coupled, three-
2.2 Nonlinear Electromechanically Coupled Finite Ele- dimensional (3D), nonlinear FE model to study the effects of
ment Methods. The real piezoelectric materials, mostly being intergranular interactions during domain switching.
piezoelectric ceramics, will show intensive nonlinearities under a The nonlinear FE models based on domain-switching criteria
large electric field or a high stress. Also, the material may show usually have simple formulations and can be used in three-
hysteresis loops upon loading and unloading or subjected to a dimensional cases without any additional efforts. But, limited to
large bipolar electric field. The LPFEM cannot treat the very com- the algorithm of such models, usually only one domain can be
mon nonlinear and hysteresis behavior in piezoelectric materials; allowed to switch in each analyzing step [13,33,34], thus leading
thus, nonlinear FEM methods are required to capture these to a rather lower numerical efficiency. Another problem for these
characteristics. FE models is that the energy barrier for 90 deg (or non-180 deg)
Currently, the nonlinear FEM for ferroelectrics can mainly be domain switching is difficult to prescribe, although the energy
classified into two types: one is based on a phenomenological ma- barrier for 180 deg switching has been accepted by most scholars
terial model [30,31] similar to the nonlinear FEM for elastoplastic to be 2P0 Ec (P0 is the magnitude of the spontaneous polarization
materials and the other is based on a domain-switching criterion and Ec the magnitude of coercive field).
[33,34], in which all the nonlinearities are caused by domain Recently, Kamlah et al. [98] conducted finite element simula-
switching. tions of polycrystalline ferroelectrics using a multidomain single
crystal–switching model.
2.2.1 Nonlinear Finite Element Method Based on Nonlinear Formulation of fully coupled nonlinear finite element method
Constitutive Equations. Hom and Shankar [31] and Gong and Suo based on a domain-switching criterion. The nonlinear FEM for-
[30], independent of each other, formulated a nonlinear finite ele- mulations for ferroelectrics are briefly presented as follows [34]:
ment method for analyzing the electrostrictive behavior of multi- Constitutive equations. For ferroelectric materials with remnant
layer actuators based on relaxor ferroelectric ceramics. A fully polarization and remnant strain, the constitutive equations can be
coupled nonlinear finite element model for phase transition mate- written as
rials was first proposed by Ghandi and Hagood [91]. They formu-
lated a three-dimensional (3D) eight-node element with nodal rij ¼ CEijmn ðemn ermn Þ emij Em (2.2.1)
displacement and electric potential degrees of freedom using iso-
e
parametric shape functions. The phase/polarization state of mate- Dj ¼ ejmn ðemn ermn Þ þ kjm Em þ Prj (2.2.2)
rials is represented by internal variables in each element, which
are updated at each simulation step based on a phenomenological where rij and emn are components of the stress tensor and the
model. Later, in Ref. [92], to improve the numerical efficiency, strain tensor, respectively; Em and Dj are components of the elec-
they formulated a four-node hybrid element based on the Hu–Wa- tric field vector and the electric displacement vector, respectively;
shizu functional, taking the electric displacement as nodal varia- e
and CEijmn , ejmn , and kjm are components of the elastic stiffness ten-
bles as well. Recently, Kamlah and Bohle [73] presented a sor, piezoelectric tensor, and the dielectric permittivity tensor,
nonlinear finite element model for ferroelectric materials based on respectively. ermn and Prj are components of the remnant strain ten-
a bilinear phenomenological model. Their FE analysis is carried sor and the remnant polarization vector, respectively.
out in a two-step scheme: first, the purely dielectric boundary Equilibrium equations. Consider a piezoelectric body that
value problem is solved for the history of the electric potential. occupies a volume, V, with a closed surface S. Inside the volume
Second, prescribing this electric potential, the electromechanical V, the following field relations can be given. If body force and
stress analysis for the mechanical boundary conditions yields the body charge can be neglected, the equilibrium equations may be
electromechanical field. written as
All the above-mentioned nonlinear FE models are based on
phenomenological constitutive models, explicit or implicit. This rij;j ¼ 0 (2.2.3a)
type of models, despite their formal complexity because of involv-
ing internal variables, usually have high numerical efficiencies, as Di;i ¼ 0 (2.2.3b)
long as the nonlinear iterative process is convergent. However,
because there are few experimental results for ferroelectric mate- The linear geometry equations and boundary conditions are the
rials under multiaxial loading [10,93–95], these FE models are same as listed in Eqs. (2.1.4) and (2.1.5) and will not be reiterated
usually limited to one-dimensional cases and the generalized 3D here.
models are usually too troublesome to manipulate [73]. Weak forms for incremental FE formulations. The general dis-
placement vector to be solved in the finite element is described as
2.2.2 Nonlinear Finite Element Method Based on a Domain-
Switching Criterion. Since the birth of energy-based, domain- u ¼ ½ ux uy uz / T (2.2.4)
switching criterion for ferroelectric materials by Hwang et al.
[12], several nonlinear FE models based on similar domain- The corresponding general stress matrix, general strain matrix,
switching criteria have been raised [13,33,34,74,96,97]. Using an and general prescribed traction matrix are written as
improved domain-switching criterion based on potential energy,
Hwang et al. [13,33] formulated two nonlinear FE models for fer- T
R ¼ rx ; ry ; rz ; rxy ; rxz ; ryz ; Dx ; Dy ; Dz (2.2.5)
roelectric and ferroelastic materials, respectively, by neglecting

T
C ¼ ex ; ey ; ez ; 2exy ; 2exz ; 2eyz ; Ex ; Ey ; Ez (2.2.6) Table 2 Material constants of a PZT-51 ceramic used in the
nonlinear finite element simulation
T ¼ ½tx ; ty ; tz ; pe T (2.2.7)
Physical properties Value
The general constitutive law can be written in an incremental for- Piezoelectric coefficient d33 1520
mulation as Piezoelectric coefficient d31 570
Dielectric permittivity k33 11,300e0
dR ¼ C dC þ DCðC Cr Þ þ dRr (2.2.8) e0 ¼ 8:85 1012 F=m
Elastic modulus SE33 (m2 =N) 3:03 1011
where C is the matrix of the general stiffness tensor, which Elastic modulus SE31 (m2 =N) 2:9 1011
includes elastic, piezoelectric, and dielectric coefficients. Saturated remnant polarization Psat (C=m2 ) 0.1938
Inside the element, the general displacement can be approxi- Coercive electric field Ec (MV/m) 0.676
mately obtained in terms of the interpolations of the nodal
displacement,
Du ¼ N Dae (2.2.9)
where N is the interpolation matrix. The general strain of the ele-

ment can be described as
DC ¼ LDu ¼ LNDae ¼ BDae (2.2.10)
where L is the linear differential factor matrix and B is the matrix

obtained from the differentiation of N.
The final weak forms for incremental FE formulations are
K Dae ¼ DP (2.2.11)
where
X
K¼ GT Ke G (2.2.12)
e
ð
Ke ¼ BT CBdve (2.2.13) Fig. 5 Stable electric field versus electric displacement hyster-
ve esis loops (the thin line is the simulation result and the bold
ð X ð line the experimental result). Reprinted from Ref. [34] with the
DP ¼ NT DTds þ GT BT CDCr dve kind permission of Elsevier Science Ltd.
St þSD e ve
X ð X ð
þ GT BT DCðC Cr Þdve GT BT DRr dve
e ve e ve
(2.2.14)
where G is the matrix that assembles the element stiffness matrix

to the global stiffness matrix and also assembles the element nodal
traction matrix to the global nodal traction matrix.
It must be noted that, for ferroelectric materials during domain
switching, the second item on the right side of Eq. (2.2.14) is
caused by the change of the remnant strain, the third item is
caused by the change of the general stiffness tensor C (i.e., by the
change of materials’ properties), and the last item is caused by the
change of remnant polarization.
Simulation results and discussions. To simulate the nonlinear
hysteresis behavior of ferroelectrics under different loadings, a
soft PZT-51 ceramic is used here for simulation [34]. The material
constants of the PZT-51 ceramic are obtained from Fang and Li
Fig. 6 Stable longitudinal strain versus electric field curves
[99] and shown in Table 2.
(the thin line is the calculation result and the bold line the
The electric hysteresis loop and butterfly loop Firstly, an experimental result). Reprinted from Ref. [34] with the kind
unpoled PZT-51 ceramic is used to reproduce the electric hystere- permission of Elsevier Science Ltd.
sis loops and butterfly loops under purely electric loading. The
unpoled specimen is modeled as an aggregation of numerous
random-oriented grains, each of which is a single domain. The domain switching, as shown in Figs. 7–9. Note that there are little
simulations are carried out in a block with 10 10 10 elements. differences between the simulated D–E hysteresis loops and but-
To fit the experimental data best, the parameter b ¼ 0:14 is terfly curves obtained using the 4 4 5 block and that using the
adopted. Figures 5 and 6 show the simulation results (thin lines) 10 10 10 block.
and the experimental results (bold lines). As it is difficult to dis- To show the details of domain switching under bipolar electric
play the three-dimensional plots for the calculated output of the field loading, the electric hysteresis loop and butterfly loop with
block with 10 10 10 elements, simulations in a block with some key loading points marked are plotted in Fig. 7. The marked
4 4 5 elements are also carried out to display the process of points, which indicate the macroscopic state of the material,

Fig. 7 The simulated D–E hysteresis loop (a) and butterfly loop (b) for an unpoled specimen with marked loading points.
correspond to each image related to the local domain states in other states (see Figs. 8 and 9). With the above notion, it is known
Figs. 8 and 9. Figure 8 shows the whole process of the domains that switching between 1 and 2, 3 and 4, and 5 and 6 is 180 deg
switching from the initial unpoled state subjected to the bipolar switching and other types of switching are 90 deg switching.
electric field. Figure 9 demonstrates the particular process of do- It can be seen from Figs. 8 and 9 that there is no direct 180 deg
main switching when the applied electric field is close to the coer- domain switching, but two successive 90 deg switchings occur
cive electric field. In Fig. 9, three states of the domains with when the applied electric field reverses, which is caused by the
continuous loading steps are displayed in detail (i.e., (a), (d), and fact that the energy barrier of the 90 deg switching used here is
(g) correspond to the point F in Fig. 7; (b), (e), and (h) correspond about 1/4 of the barrier of 180 deg switching [34]. In Fig. 9, it can
to point G; and (c), (f), and (i) correspond to point G’. be found that the domains near the free boundaries can switch
As denoted in Ref. [34], in Figs. 8 and 9, the initial state of more easily than those inside the material and those near the
each domain is marked 1, and its 180 deg reorientation state is strongly constrained boundaries. In Figs. 9(a) and 9(b), when the
marked 2. Four possible 90 deg reorientation states are marked 3, macroscopic switching begins, several domains at or near the top
4, 5, and 6, respectively. The state 3 can be reached from the state face boundary switch, while the domains inside the material
4 by a 180 deg switching and vice versa, and the same for state 5 remain unchanged (see Figs. 9(d) and 9(e)). The domains near the
and state 6. Each state is specified by a single color different from bottom face boundary do not switch at all, even after two loading
steps (see Figs. 9(g)–9(i)). All these may indicate that the 90 deg
Fig. 8 The whole process of domain switching under uniaxial Fig. 9 Domain-switching process when the applied electric
electric loading. Reprinted from Ref. [34] with the kind permis- field is close to the coercive field. Reprinted from Ref. [34] with
sion of Elsevier Science Ltd. the kind permission of Elsevier Science Ltd.

stress field in a practical ceramic, while it is rather difficult to
address in the simulation.
2.2.3 Domain-Switching–Induced Instabilities in Nonlinear

Finite Element Methods. It should be noted that, in both of the
above-mentioned nonlinear finite element methods for ferroelec-
trics, there exist “domain-switching instabilities,” which are called
“domain-switching–induced iterative instability” in nonlinear FE
models based on phenomenological material models and
“domain-switching instability” in nonlinear FE models based on
domain-switching criteria and will be addressed respectively in
this subsection.
Domain-switching–induced iterative instability. For those FE
models based on phenomenological material models [30,31], the
material responses to the applied field, such as the D–E hysteresis
loops, the stress-strain curves, can be obtained directly from the
material models, explicitly or implicitly. Domain switching will
Fig. 10 Longitudinal strain versus uniaxial stress curves (the not be involved in the nonlinear FE formulations; thus, the second
thin line is the calculation result and the bold line the experi- and the last items on the right side of Eq. (2.2.14) will vanish. The
mental result). Reprinted from Ref. [34] with the kind permis- FE formulations of such models have similar forms with the non-
sion of Elsevier Science Ltd. linear FE formulations for conventional elastoplastic materials,
only with additional dielectric items and the coupled piezoelectric
switching will result in a fairly strong constraint from the neigh- items.
bor domains, which makes the switching more difficult or some- Because the behavior of ferroelectric materials is always load-
times impossible [34]. ing history–dependent, an incremental nonlinear iterative process
Strain versus stress curve and electric displacement versus is required in the FE analysis. Among the nonlinear iterative
stress curve. A fully poled PZT-51 ceramic is used to obtain the methods, the Newton–Raphson method is usually fairly conven-
strain versus stress curve and the electric displacement versus ient, with a satisfying accuracy and convergent rate [100]. Fur-
stress curve. The ceramic is subjected to an increasing compres- thermore, for the conventional elastoplastic materials, the
sive stress parallel to its poling direction, and then the applied Newton–Raphson iteration is always stable [101].
stress is gradually unloaded. The simulation results are present in As indicated in Ref. [75], for ferroelectric materials characteris-
Figs. 10 and 11. tics of domain switching, the slope of D–E curves near the coer-
It can be seen from Figs. 10 and 11 that the simulations do not cive field is usually much larger than that of the linear dielectric
agree with the experimental results well. The calculated strain period. The Newton–Raphson iterative process is usually unstable
increases abruptly at about 25 MPa, which is determined to the co- in solutions of nonlinear constitutive curves containing such an
ercive stress. While in the experimental results, the strain abrupt change. To illustrate such numerical instability, the poling
increases smoothly with the increasing stress and the coercive curve of a practical PZT ceramic is present in Fig. 12 [75]. For a
stress is not distinct. The nonconsistency between the numerical single-degree-of-freedom problem that takes the electric potential
results and the experimental data is reasonable. The polarization as nodal variables with prescribed electric displacement D0 , it can
state of a practically poled ceramic is different from the initial be seen from Fig. 12(a) that the Newton–Raphson iterative pro-
poled state prescribed in the simulations. In the simulated poled cess to solve the exact solution Ex is unstable (i.e., Ei ! Eiþ1
state, it is assumed that all the domains will orient as close as pos- ! divergence, where Ei and Eiþ1 are the electric field computed
sible to the poling field, while in experiments, a poled polycrystal- at the ith and i þ 1th iteration [75].
line ceramics can never reach the theoretically saturated poling To solve this problem, Landis [102] proposed that, by taking
state. Moreover, typically there will be a nonuniform residual the vector electric displacement potential instead of the scalar
electric potential as nodal variables, a straightforward iterative
procedure can be stable for ferroelectric materials with a simpli-
fied bilinear D–E response. However, as indicated by Li and
Rajapakse [75], if the saturation period of the E–D curve (BC in
Fig. 12(b)) is also taken into account, for problems that take the
electric displacement potential as nodal variables with prescribed
electric field, the Newton–Raphson iterative process cannot be
stable, even if electric displacement computed at the ith iteration
(i.e., Di1 in Fig. 1(b) is very close to the exact solution Dx ). That
is, Di1 ! Diþ11 ! divergence and Di ! Diþ1 ! divergence [75].
It should be noted that the above iterative instabilities are
merely caused by the steep slopes of AB curve in Fig. 12(a) and
BC curve in Fig. 13(b), although the real cause is domain switch-
ing. This type of iterative instability, as shown in Fig. 12, cannot
be avoided whether taking the electric potential or the electric dis-
placement potential as nodal variables. Although the special tech-
niques to stabilize the Newton–Raphson iterative procedure, such
as the line search techniques of Matthies and Strang [103] and the
arc length control procedures developed by Crisfield [101],
Wempner [104], and Riks [105], may solve this iterative instabil-
ity, they usually lead to a lower rate of convergence. An alterna-
Fig. 11 Simulated depolarization versus uniaxial stress curve. tive choice to avoid this iterative instability is replacing the D–E
Reprinted from Ref. [34] with the kind permission of Elsevier curve of ferroelectric materials with a bilinear approximation
Science Ltd. [73]. In this case, the iterative procedure will be much more

Fig. 12 Illustrations of Newton–Raphson iterative instability in nonlinear FE mod-
els: (a) taking the electric potential as nodal variables and (b) taking the electric
displacement potential as nodal variables. Reprinted from Ref. [75] with the kind
permission of Elsevier Science Ltd.
simplified and stable at the sacrifice of numerical accuracy to As to the case of 90 deg domain switching illustrated in Fig.
some extent. 13(b), besides the large depolarization field Ed by the spontaneous
polarization change, due to the spontaneous strain change during
Domain-switching instability in FE models based on domain- domain switching, a large mechanical field (i.e., a tensile stress r1
switching criteria. As indicated by Li and Rajapakse [75], for the and a compressive r2 ) will be generated by the adjacent domains.
nonlinear FE models based on micromechanical domain- r1 and r2 are both on the order of Dc=s (1 GPa, where Dc is the
switching criteria [13,33,34,74,96,97], the equivalent nodal load magnitude of the spontaneous strain change and s is the elastic
induced by the change of spontaneous strain and spontaneous compliance constant) and also at least one order higher than the
polarization during domain switching (i.e., the second and the last corresponding coercive stress, making the switched domain
item on the right side of Eq. (2.2.14)) can be rather large com- switch back. Therefore, in an unmodified, nonlinear FE formula-
pared to a moderate electromechanical loading. For the case of tion, such as Eq. (2.2.14), domain switching will always be
180 deg domain switching without change of spontaneous strain, unstable.
as shown in Fig. 13(a), a large depolarization field Ed will be gen- Actually, the simulated domain-switching instability has been
erated [75]. If the charge screening effect is not included (this is noticed by several groups [13,33,34,74,106], while a profound ex-
the usual case in conventional simulations based on FEM), the planation is still lacking. To weaken the rather high electric/me-
depolarization field Ed will be on the order of DP=k (1050 kV/ chanical fields induced by ferroelectric/ferroelastic switching,
mm), where DP is the magnitude of the change of spontaneous Hwang et al. [13] employed a much smaller Yong’s modulus
polarization during switching and k is the dielectric constant of (about 20% of the true value) and a rather larger dielectric con-
the materials. Such a large (usually higher than ten times of the stant (about 14 times of the true value) in their simulations. Chen
coercive field) depolarization field will make the domain switch and Lynch [74] introduce a factor kð0 < k < 1Þ in the last item on
back to its original state; thus, domain-switching instability arises. the right side of Eq. (2.2.14). Kessler and Balke [106] discussed
the unstable switching in their simulations (not FE model), and
they suggested that the switching instability can be removed by
introducing ferroelectric “hardening” in the material model, which
is somewhat equivalent to the simplified bilinear D–E curve
assumption by Kamlah and Bohle [73].
Recently, to overcome the problem of domain-switching insta-
bilities, Li and Rajapakse [75] proposed a finite element model for
ferroelectrics based on constrained domain switching. In their
model, a ferroelectric polycrystalline is made up of numerous
random-oriented grains, each of which contains multidomains and
is represented by a finite element. Charge-screening effect in real
ferroelectrics is taken into account in the model; thus, 180 deg do-
main switching will not be constrained, while the internal stress
induced by non-180 deg switching cannot be compensated in a
similar way and is considered naturally by the FE method. The
fraction of non-180 deg switching in each grain is obtained
through a dichotomy approximation. In this way, the domain-
switching instability problem can be effectively avoided.
2.3 Electromechanically Coupled Meshless Method.

Because of the advantages over conventional finite element
method, meshless method has become one of the hot topics in
computational mechanics. First, only the node information,
instead of the linkage between nodes and elements, is used to dis-
cretize the computation field. During the numeric analysis, the
Fig. 13 High depolarization electric field and mechanical field nodes can be added or deleted conveniently; as a result, the pre-
induced by (a) 180 deg switching or ferroelectric switching and processing is simplified significantly. Secondly, with the shape
(b) 90 deg switching or ferroelastic switching. Reprinted from function based on moving least square method, the field function
Ref. [75] with the kind permission of Elsevier Science Ltd. is of smoothness, high precision, and high-order differential

continuity. Especially, some nodes can be added in the interested namely wðx xI Þ > 0. uI is the nodal value at xI . When J reaches
domain where field variables change significantly, regardless of its minimum, the derivative of J with respect to aðxÞ equals zero.
the shapes of elements. Moreover, the choice of base function, So
weight function, and its influence radius in meshless methods are
relatively flexible, and the boundary conditions can be imple- AðxÞaðxÞ ¼ BðxÞu (2.3.3)
mented in many ways. Therefore, the meshless method is of high
openness in theoretical analysis and application; thus, it can be where
operated and extended easily.
X
n
The first meshless method, i.e., smoothed particle hydrodynam- AðxÞ ¼ wðx xI ÞpT ðxÞpðxÞ (2.3.4)
ics method (SPH), was proposed [107] and was successfully I¼1
applied to simulate astrophysical phenomena, nonboundary gal- BðxÞ ¼ ½wðx x1 Þpðx1 Þ; wðx x2 Þpðx2 Þ; ;
axy explosion. Then, the SPH was developed into the kernel func-
tion method with better precision and robustness, and thus, a wide wðx xn Þpðxn Þ (2.3.5)
application of SPH was found [108]. Based on the moving least u ¼ ½u1 ; u2 ; ; un (2.3.6)
square method that was first introduced into mechanical disci-
plines [109], a meshless method with various integral schemes,
The matrix AðxÞ is often called the moment matrix. Solving Eq.
which is called element-free Galerkin method (EFGM), was
(5), we can get
developed [110–112]. Since it alleviates the burdensome remesh-
ing that is unavoidable in finite element method, the EFGM is
good at simulating dynamic crack propagation. The background aðxÞ ¼ A1 ðxÞBðxÞu (2.3.7)
grids in EFGM are used to implement gauss integration; thus, to
some extent, it is not a true meshless method. The finite point Substituting Eq. (8) into Eq. (1) yields the following equation:
method (FPM) [113], a true meshless method without background
grids, is an efficient computation tool for fluid dynamics prob- X
n X
n
uh ðxÞ ¼ pT ðxÞA1 ðxÞBI ðxÞuI ¼ /I ðxÞuI (2.3.8)
lems. In FPM, the approximate function is constructed by the least I¼1 I¼1
square method, and computation field is discretized by collocation
scheme. On the basis of aforementioned meshless methods, some where
improvements are achieved in many numerical schemes, such as
local boundary integral equation method, least-square collocation /I ðxÞ ¼ pT ðxÞA1 ðxÞBI ðxÞ (2.3.9)
meshless method [114], meshless method based on B-spline
wavelet function [115], and so on. With direct shape functions, an is called a shape function, and its spatial derivatives can be
improved meshless method that is of higher efficiency and preci- expressed as
sion than conventional meshless methods was developed for com-
putational electromagnetics [116]. Due to its advantages over /I;i ðxÞ ¼ pT;i ðxÞA1 ðxÞBI ðxÞ þ pT ðxÞA1
;i ðxÞBI ðxÞ
classical finite element method, meshless method is also found in
wide applications in solid mechanics, especially for multifield þ pT ðxÞA1 ðxÞBI ; iðxÞ (2.3.10)
coupled analysis. The simulation of interface cracking in piezo-
electric layers, the singular stress, and electric field were carried where A1 1 1
;i ðxÞ ¼ A ðxÞA;i ðxÞA ðxÞ.
out by using meshless method [76–78]. Moreover, the meshless The constitutive equations for piezoelectric materials are
method can also be applied to study the fracture and cracks in
magnetoelectroelastic solids [117] and the fracture of electrother- R ¼ C : e eT E (2.3.11)
moelastic plates subjected to thermal shock [118].
In view of the advantages of meshless method, such as flexibil- D¼e:eþdE (2.3.12)
ity, variability, and inherent superiority, we developed an
electromechanical-coupled meshless Galerkin method to simulate where R, e, E, and D are stress, strain, electric field intensity, and
the stress and electric field of hole, layer actuator, and cracks sub- electric displacement tensors, respectively. C, d, and e are the
jected to electromechanical loadings. elastic constant tensor, the dielectric constants, and the piezoelec-
tric constants, respectively. The strain is related to the displace-
2.3.1 Formulation of Electromechanically Coupled Meshless ment by linear geometry equations,
Method. In meshless method, only a scattered set of nodal points
is required in the domain of interest, and the structure mesh is not 1
e ¼ ðru þ urÞ (2.3.13a)
used. A function uðxÞ over the domain can be approximated by 2
the parameter values at discrete nodes, so the approximation E ¼ r/ (2.3.13b)
uh ðxÞ of uðxÞ at any point x can be written as
When it reaches equilibrium, the general stresses meet governing
uh ðxÞ ¼ pðxÞaðxÞ (2.3.1) equations
where pðxÞ is a basis function and aðxÞ are coefficients. Generally rrþf ¼0 (2.3.14a)
speaking, linear basis will bring forth satisfactory results, and
high-order polynomial makes few contributions to accuracy. In r D ¼ pe (2.3.14b)
the case of cracks, however, a few interpolation nodes can capture
the field singularity by a basis function with singular items. in the domain and conditions
The moving least square method is employed to determine the
coefficients vector, aðxÞ. The weighted residue of field variable in r n ¼ T on Sr (2.3.15a)
whole domain is e D
D n ¼ p on S (2.3.15b)
X
n
J¼ wðx xI Þ½pðxÞaðxÞ uI 2
(2.3.2) u ¼ u on Su (2.3.15c)
I¼1
/ ¼ / on S /
(2.3.15d)
where wðx xI Þ is the weight function of node I at point x and n
is the number of nodes that support a domain including point x, on the boundary.

2 3
For simplicity, a 2D domain filled with barium titanate and lead uI;x 0 0
zirconate titanate (PZT) ceramics is investigated as plain strain 6 7 2 3
6 0 uI;y 0 7 uI 0 0
sample. The constitutive equations for transversely isotropic PZT 6 7
material can be simplified as 6 7 6 7
6 0 7
BI ¼ 6 uI;y uI;x 7; uI ¼ 6
4 0 uI 0 7
5
6 7
6 7
r0 ¼ G e0 (2.3.16) 6 0 0 uI;x 7 0 0 uI
4 5
0 0 uI;y
where
2 3 2 3
8 9 8 9 Q1 0 0 cosa 0 sin a 0 0
>
> r11 >
> >
> e11 >> 6 7 6 7
>
> >
> >
> >
> Q ¼ 4 0 Q2 0 5 n ¼ 4 0 sin a cosa 0 0 5
>
> >
> >
> >
>
>
> > > >
> r33 >
> >
>
>
>
> e33
>
>
> 0 0 Q3 0 0 0 cosa sin a
>
> >
> >
> >
>
>
< >
= >
< >
=
r0 ¼ r13 ; e0 ¼ 2e13 if the displacement ui of iði ¼ 1; 2; 3Þ direction is specified,
>
> >
> >
> >
> Qi ¼ 1, else Qi ¼ 0. a is the direction angle of the normal of the
>
> >
> >
> >
>
>
> >
> >
> >
> interface.
>
> D 1 >
> >
> E 1 >
>
>
> >
> >
> >
>
>
> >
> >
> >
>
>
: >
; >
: >
;
D3 E2 2.3.2 Results and Discussion. An infinite PZT-5H ceramic
(2.3.17) plate with a circular hole was studied to verify the electrome-
2 3 chanically coupled meshless algorithm. As shown in Fig. 14, both
C11 C13 0 0 e31
6 7 a vertical stress r0 ¼ 7:5 MPa and nominal electric field intensity
6 7 E ¼ 0:18 MV/m are applied to the ceramic plate. The hoop stress
6 C13 C33 0 0 e33 7
6 7 and dielectric displacement around the hole are shown in Fig. 15,
6 7 which indicates a good agreement between the EFG results, and
6 7
G¼6
6 0 0 C44 e15 0 77 analytical ones [119] are achieved. To demonstrate the application
6 7 of the electromechanically meshless model, as shown in Fig. 16,
6 7
6 0 0 e15 d11 0 7 an interface crack locates in the piezoelectric film and substrate is
6 7
4 5 investigated. The thickness ratio of piezoelectric film to the sub-
e31 e33 0 0 d33 strate is h1 : h2 ¼ 1 : 10. The up and down sides of the specimen
are made from PZT-5H ceramic and PZT-4 ceramic, respectively,
and the material parameters are listed in Table 3.
are general stress vector, stiff matrix, and strain vector, respec-
A normal tensile stress r0 ¼ 0:72 MPa is applied to up and
tively. The total system potential in a domain V with its boundary
down sides of the specimen, so the various ratios of electric-to-
can be written as
mechanical loading are obtained by different applied external
ð ð ð electric intensities. The distributions of stress ryy , as shown in
1
P¼ ru0T G ru0 dV u0T f 0 dV u0T T0 dS Fig. 17, indicates an extensive stress concentration near the crack
2 V V Sr tip. Because of the material discontinuity and thickness discrep-
ð
1 ancy, the stress ryy is not symmetrical, and with the electric field
þ a ðu0 u0 ÞT ðu0 u0 ÞdS
2 Su increasing, the asymmetry is more evident. Different from vertical
ð interface crack, the distribution of electric field is also asymmet-
1
þ b ðuþ u0 ÞT ðuþ u0 ÞdS (2.3.18) ric, as shown in Fig. 18. The electric field reaches its maximum
2 Si
value in the film near to crack tip, and the value of electric field in
where f 0 ¼ ½f pe T , T0 ¼ ½T
pe T , and u0 ¼ ½u/T are general body
force, boundary force, and displacement, respectively. a and b are
penalty coefficients.
With the variational principle, we can get the element-free
Galerkin formulation,
Ku0 ¼ f 0 (2.3.19)
where K and f 0 are made up of 2 2 submatrix kIJ and 2 1 sub-

matrix fI , respectively.
ð ð
kIJ ¼ BTI G BJ dV þ a UI Q UJ dS
V Su
ð
T
þb ðUþ þ
I UI Þ Q ðUJ UJ ÞdS
Si
ð
þ ðUþ
I UI Þ n G BJ dS
Si
ð
T
þ ðn G B þ
I Þ ðUJ UJ ÞdS (2.3.20a)
Si
ð ð ð
f 0I ¼ UI f 0 dV þ UI T0 dS þ a u0 Q UI dS (2.3.20b)
V Sr Su
where Fig. 14 An infinite PZT-5 H plate with circular hole

Fig. 15 Comparison of EFG results with analytical results: the variation of hoop (a) stress rh
and (b) dielectric displacement Dh with angle h. Reprinted from Ref. [78] with the kind permis-
sion of Elsevier Science Ltd.
Fig. 17 The distribution of stress ryy near the crack tip [76]
Fig. 16 Schematic diagram of interface crack
Table 3 Material parameters of PZT-4 and PZT-5H ceramics
PZT-5H PZT-4
2
C11 12:6 1010 Nm 13:9 1010 Nm2
C12 5:5 1010 Nm2 7:78 1010 Nm2
C13 5:3 1010 Nm2 7:43 1010 Nm2
C33 11:7 1010 Nm2 11:3 1010 Nm2
C44 3:53 1010 Nm2 25:6 1010 Nm2
e31 6.5 Cm2 6.98 Cm2
e33 23.3 Cm2 13.84 Cm2
e15 17.0 Cm2 13.44 Cm2
d11 151 1010 CV1m1 60:0 1010 CV1m1
d33 130 1010 CV1m1 54:7 1010 CV1m1
the film is larger than that in the substrate; namely, in Fig. 19, the
electric intensity illustrates more concentration near to the crack
than that at the crack tip.
The distribution of hoop stress rh , when various electric
mechanical loadings are applied to the specimen, is illustrated in Fig. 18 The distributions of electric field Ey around the crack
pffiffiffiffiffiffiffiffiffi
ffi tip [76]
Fig. 20, where the coefficient k ¼ 2r=a and a and r are the
crack length and radial distance of the point from the crack tip,
respectively. For low values of the electrical-to-mechanical load and its magnitude is getting larger. The asymmetry of rh along
ratio, the hoop stress is tensile on the whole and does not reach its the crack surface becomes more and more evident. Because of the
maximum value at h ¼ 0 . With the increase of electromechanical large thickness difference between the thin film and substrate, the
load ratio, the hoop stress rh has the tendency to be press stress, stresses in the film are greater than that of in the substrate.

evolution of domain structure have profound influence on the
macroscopic behavior of ferroelectrics, especially their switching
characteristics. From an energetic point of view, domain forma-
tion/evolution is a result of energy minimization, including bulk
free energy, elastic energy, electrostatic energy, and external
potential induced by electromechanical loads. Because of the non-
linear feature of domain evolution, numerical approaches are
employed, and the phase field method (PFM) is the most success-
ful method in modeling configurations and evolutions of ferro-
electric domains [35].
The phase field method describes the microstructure using a set
of field variables that are continuous across interfacial regions
[35]. The time-dependent spatial evolutions are governed by the
Cahn–Hilliard nonlinear diffusion equation for the conserved field
variables and the Allen–Cahn relaxation equation for the noncon-
served ones. Since PFM is based on a diffuse-interface descrip-
tion, there is no need of tracking the interface positions, which is
impractical for complicated three-dimensional microstructures.
With the maturity of numerical methods in solving the evolution
equations, the phase-field approach has emerged as one of the
most powerful methods for modeling many types of microstruc-
ture evolution processes [35]. In simulations of ferroelectric do-
main structure, the early phase field model was developed by
Chen’s group [79,81]. They introduced phase field method in pre-
Fig. 19 The electric field vector distribution near the crack dicting the coherent domain evolution in constrained thin films.
tip [76] Since then, phase field method has attracted extensive interests in
various areas of ferroelectrics [36,37,40,41] and was adopted or
developed to study several important issues related to fundamental
The piezoelectric fracture under coupled electromechanically
physics and mechanics of ferroelectrics, as introduced in the
coupled loading is simulated by using the meshless model. As a
following.
recently developed numerical analysis method, the meshless
method finds a wide application in fields concerning large defor-
mations (i.e., high-velocity impact). Great efforts should be made Phase Diagrams. Li et al. [82] and Li and Chen [120] proposed
to improve the algorithm, owing to the burdensome matrix an eighth-order polynomial of Landau–Devonshire expansion for
inversion. describing the bulk free energy of BaTiO3 single crystals, which
was employed to construct a temperature-strain phase diagram.
This study also indicated that the eighth-order polynomial was ca-
3 Mesoscopic Electromechanically Coupled pable of providing better domain stability in the full range of
Computational Methods Based on Microstructures experimentally accessible compressive and tensile strains, as com-
and Micromechanics pared to the widely adopted sixth-order polynomial. By incorpo-
rating a new set of structural order parameters, Li et al. [121]
The mesoscopic computational methods could address the elec- further extended the phase field approach in constructing the
tromechanical coupling behavior of ferroelectrics at the grain temperature-strain phase diagram for SrTiO3 thin films. Liu et al.
level or the domain level. These methods may include the phase [122] and Sheng et al. [123] analyzed a more general case, where
field method (PFM), which is mostly for single crystals, and the the biaxial misfit strains are nonequal along two principal direc-
micromechanics-based methods, usually for polycrystals. tions, and they provided the misfit strain-misfit strain diagram of
epitaxial BaTiO3 thin films. To describe the pressure dependence
3.1 Electromechanically Coupled Phase Field Method. It of ferroelectric transition temperatures, Wang et al. [124] adopted
is well known that the ground state of a ferroelectric material is a modified eighth-order Landau potential to take into account the
not uniformly polarized and ferroelectrics tend to lower their elec- quantum mechanical effects at low temperature and constructed
trostatic energy through the formation of domains. The macro- the temperature-hydrostatic pressure phase diagram for BaTiO3
scopic properties of ferroelectrics reflect the collective behavior single crystal.
of domains at micro- and nanoscale, and the configuration and
Piezoelectric/Ferroelectric Fracture. Wang and Zhang [39,87]
introduced phase field approaches to elucidate the interactions of
a permeable static crack with the crack-tip domain structures for
both cracks perpendicular and parallel to the original polarization
direction and explored the polarization switching–induced tough-
ening in ferroelectrics based on the simulations. The analyses
were further extended to the case with an impermeable notch in a
three-dimensional ferroelectric single domain [125]. Song et al.
[42] and Li and Landis [50] employed the phase field methods in
simulating the crack-tip domain structures under an electrical
loading, and the latter implemented an energy-consistent bound-
ary condition in their simulations. Xu et al. [126] further studied
the influences of the applied electric field and the different crack-
face boundary conditions on the fracture behavior. Different from
the above-mentioned studies, where a sharp crack model was
adopted, Miehe et al. [127] developed a framework of diffusive
Fig. 20 The variation of hoop stress around the crack tip when fracture in piezoelectric solids, which enabled the simulation of
different electromechanical loadings are applied [76] diffusive electromechanical crack propagation in piezoelectric

solids. Abdollahi and Arias [47–49] initially incorporated the dif- fbulk ðPi Þ ¼ a1 P21 þ P22 þ P23 þ a11 P41 þ P42 þ P43
fusive crack model in ferroelectrics and evaluated the effects of 2 2
þ a12 P1 P2 þ P22 P23 þ P21 P23 þ a111 P61 þ P62 þ P63
different crack-face conditions on the crack propagation in a fer- h i
roelectric single crystal. þ a112 P21 P42 þ P43 þ P22 P41 þ P43 þ P23 P41 þ P42
Domain Pattern and Evolutions in Ferroelectric Nanostructure. þ a123 P21 P22 P23 (3.1.2)
Wang and Zhang [38] investigated the size effects on domain con-
figuration and polarization switching in two-dimensional epitaxial where a1, a11, a12, a111, a112, and a123 are the expansion coeffi-
ferroelectric islands and in thin films by introducing an extrapola- cients, which determine the thermodynamic behavior of the bulk
tion length in the boundary conditions of the phase field model. paraelectric and ferroelectric phases, such as the Curie tempera-
Hong et al. [128] employed a similar method to explore the inter- ture, the spontaneous polarization, and the stability of the parent
face and surface effects on ferroelectric nanothin films. By intro- paraelectric phase. For example, a1 ¼ (T – T0)/2j0C0 5 1/2j is
ducing an unconventional phase field model and a new set of related to the dielectric constant, j, where T0 is the Curie tempera-
order parameters to describe the surface effect, Zhang et al. [46] ture, C0 is the Curie constant, and j0 is the dielectric constant of
presented a set of size-dependent domain configurations in ferro- vacuum. For BaTiO3 films, the six-order free energy Eq. (3.1.2) is
electric ultrathin films and provided their evolutions under an al- only applicable under relatively small compressive strains
ternative electrical loading. Besides ultrathin films, Wang et al. (0.4%), as there is no solution for equilibrium polarization under
[129,130] also introduced a phase field method in studying the larger compressive strains in a certain temperature range
interactions of surface effects and domain configurations in nano- [139,140]. To overcome this problem, Li et al. [82] proposed an
wires and nanodots. Ferroelectric superlattices are another type of eight-order polynomial of polarization as the bulk free energy
nanostructure where phase field methods have found meaningful function, which is applicable in predicting the ferroelectric phase
applications [131]. transitions of BaTiO3 thin films under large compressive biaxial
strains.
Interactions of Oxygen Vacancy With Ferroelectric Domain The gradient energy density fgrad penalizes the spatial variation
Structures and Evolutions. Xiao et al. [132] and Xiao and Bhatta- in the order parameter P and can be understood as the energetic
charya [133] developed phase field models to study the interac- cost of forming domain walls separating different variants [79],
tions of oxygen vacancies with domain structures in BaTiO3,
revealing the existence of depletion layers at the metal- 1
fgrad Pi;j ¼ G11 P21;1 þ P22;2 þ P23;3
ferroelectric interfaces and vacancy accumulation in 90 deg do- 2
main wall. Hong et al. [134] further extended this model to study þ G12 P1;1 P2;2 þ P2;2 P3;3 þ P3;3 P1;1
the distribution of charge density and oxygen vacancy density in 1 h
90 deg and 180 deg domain structures. In these works, the influ- þ G44 P21;2 þ P22;1
ence of oxygen vacancy was modeled as point charge and the 2 i
memory effect and large recoverable electrostrain, as observed in þ P22;3 þ P23;2 þ P21;3 þ P23;1
experiments [135,136], cannot be well captured. To overcome this h
1
difficulty, Zhang et al. [45] proposed to model the oxygen va- þ G044 P21;2 P22;1
cancy as dipoles in the phase field approach, and their simulations 2 i
reproduced well the reversible domain switching and large recov- þ P22;3 P23;2 þ P21;3 P23;1 (3.1.3)
erable electric-field-induced strain observed in experiments with
both engineered domain structure and single domain structure.
Besides these works, Su and Landis [43] employed phase field where G11, G12, G44, and G044 are the gradient energy coefficients
simulations in investigating the polarization distributions near and the commas in the subscripts denote spatial differentiation.
straight 180 deg and 90 deg domain walls and determining the The elastic energy density felas can be expressed as
electromechanical pinning strength of an array of line charges.
In this section, we will first introduce the general framework of felas ¼ cijkl eij ekl =2 ¼ cijkl eij e0ij ekl e0kl =2 (3.1.4)
phase field model for ferroelectrics and then describe their appli-
cations in two specific physical problems.
where cijkl is the stiffness tensor. eij, eij , and e0ij are the elastic
strain, the total strain, and the transformation strain, respectively.
3.1.1 Phase Field Modeling of Ferroelectric Domain For a cub-to-tetragonal ferroelectric transformation, the transfor-
Structure mation strain e0 can be determined from spontaneous polarization
Model formulation. In the Landau–Devonshire phenomenologi- P as follows:
cal framework, the polarization vector is usually used as the order
parameter to calculate thermodynamic energies of the ferro- e011 ¼ Q11 P21 þ Q12 P22 þ P23 ; e022 ¼ Q11 P22 þ Q12 P21 þ P23 ;
electrics. The total free energy of a ferroelectric system can be
expressed as [35,80,137] e033 ¼ Q11 P23 þ Q12 P21 þ P22 ;
e013 ¼ Q44 P1 P3 ; e023 ¼ Q44 P2 P3 ; e012 ¼ Q44 P1 P2 (3.1.5)
ð

G¼ fbulk þ fgrad þ felas þ felec r0 e E0 P dV (3.1.1)
V where Q11, Q12, and Q44 are the electrostrictive coefficients. The
strain field e is determined by solving the following mechanical
where fbulk, fgrad, felas, and felec denote the bulk energy density, the equilibrium equation:
gradient energy density, the elastic strain energy density, and the
electrostatic energy density, respectively, and the last two terms in cijkl ekl;j e0kl;j ¼ 0 (3.1.6)
Eq. (3.1.1) represent the potential energies contributed by the
external stress r0 and the applied electric field E0. In the equation,
e and P are the transformation strain and spontaneous polarization subject to proper boundary conditions.
of the ferroelectric. The electrostatic energy felec is due to the electric field gener-
The bulk energy density fbulk is usually expanded as a six-order ated by the polarization distribution in the ferroelectrics, which
polynomial of polarization P [138], can be expressed as

felec ¼ Ei;p Pi =2 (3.1.7) separately. On the time scale, a semi-implicit finite difference
scheme [35,79] is commonly adopted to solve Eq. (3.1.9). On the
where the electrostatic field Ep induced by the spontaneous polar- spatial scale, several algorithms, such as the fast Fourier transform
ization can be further expressed as a negative gradient of the elec- [35,36], the finite element method [38,43], the finite difference
trostatic potential / (i.e. Ei,p ¼ /,i, which can be determined by method [46], and the boundary element method [44], can be uti-
solving the Maxwell’s equation [38,80], lized, depending on the physical problem considered. Within these
methods towards spatial discretization, the fast Fourier transform
jj0 /;11 þ /;22 þ /;33 ¼ P1;1 þ P2;2 þ P3;3 (3.1.8) is especially effective for solving problems subject to periodical
boundary conditions, since the partial difference equations in Eqs.
subject to proper boundary conditions. (3.1.6) and (3.1.8) can be transformed into simple algebraic equa-
After introducing the various energy terms, we turn to the equations in the Fourier space. To solve the problem with finite bound-
tions governing the evolutions of polarization that lead to the min- ary conditions, the other three methods can be adopted. There are
imization of total free energy. In the phase field model, the also some other cases in which periodical boundary conditions
temporal variations of polarization field and thus the domain should be used in some directions while finite boundary condi-
structure evolution is described by the classical time-dependent tions should be used in other directions, such as the ferroelectric
Ginzburg–Landau (TDGL) equation, film and wire. In this case, the fast Fourier transform and finite
difference method (or finite element method) should be combined
@Pðr; tÞ=@t ¼ LdG=dPðr; tÞ (3.1.9) to solve the governing equations.
In order to simplify the numerical analyses, normalizations of
field variables and material parameters are usually adopted, as
where r is the spatial vector, t the generalized time, and L the ki-
described by Hu and Chen [145]. The simulation parameters for
netic coefficient. dG/dP(r,t) is the thermodynamic driving force
typical ferroelectric materials can be found in Rabe et al. [146]. In
for the spatial and temporal evolution of P(r,t).
subsections 3.1.2 and 3.1.3, we will introduce the applications of
To this end, we have introduced the basic equations of phase
phase field model in studying specific physical problems.
field model for ferroelectrics. Given an initial polarization distri-
bution, the stable domain configuration can be determined by
repeating solving Eqs. (3.1.6), (3.1.8), and (3.1.9). 3.1.2 Extension of Phase Field Model for the Oxygen Va-
Besides the conventional phase field model, which adopts the cancy Effect on Ferroelectric Behaviors. This subsection demon-
polarization component as the order parameter, an unconventional strates a two-dimensional ferroelectric model [45], which couples
phase field model was also proposed, which is based on the the classical Ginzburg–Landau theory and the oxygen vacancy
description of multirank engineering domain configuration [141]. diffusion, to demonstrate the influences of oxygen vacancies on
Since the bulk free energy density is expanded as a polynomial of domain evolutions under electric field in a BaTiO3 single crystal.
polarization in the conventional phase field model, as shown in The oxygen vacancies are modeled as dipoles that possess the
Eq. (3.1.2), the energy well of the ferroelectric is implicit and the symmetry-conforming property [135] and thus can effectively
expansion coefficients have to be fine-tuned to yield correct mate- capture the influences of vacancies on domain evolutions. The
rial symmetry. To overcome these difficulties, the characteristic simulations can reproduce well the reversible domain switching
functions of ferroelectric variants can be used as order parameters. and large recoverable electric-field-induced strain observed in
For a ferroelectric with N variants, the local spontaneous polariza- experiments with both engineered domain structure [135,136] and
tion and transformation strain can be expressed in terms of charac- single domain structure [147].
teristic function ci of variant (i), Dipolar field induced by oxygen vacancy. Oxygen vacancies
and acceptor impurities commonly coexist in ferroelectric lattices,
X
N X
N
P¼ ci PðiÞ ; e0 ¼ ci eðiÞ (3.1.10) especially after doping [135]. Since the oxygen vacancy is in one
i¼1 i¼1
of the faces of the perovskite cell while the impurity cation is in
the center of the cell, they form additional defect dipoles besides
where P(i) and e(i) are spontaneous polarization and the spontaneous ones. Thus, it makes more sense to model the
P transformation influences of oxygen vacancy as defect dipole instead of monop-
strain of variants (i). Since it is necessary that Ni¼1 ci ¼ 1, only
N 1 characteristic functions are independent. To incorporate this ole. Such defect dipole results in additional dipolar electric field
constraint, we introduce N 1 li that are independent of each in the ferroelectric in addition to the usual depolarization field,
other and let through which oxygen vacancies interact with ferroelectric
domains and domain walls. In order to describe such interactions,
c1 ¼ l1 we can adopt a continuum model to analyze the oxygen vacan-
cy–induced local defect polarization Pdefect and the resulted defect
ci ¼ ð1 l1 Þ ð1 li1 Þli ; 1 < i < N (3.1.11) dipolar field Edefect and their contribution to the free energy of the
cN ¼ ð1 l1 Þ ð1 lN2 Þ ð1 lN1 Þ ferroelectric system.
By examining the defect dipole in a two-dimensional lattice, a
so that the constraint is automatically satisfied. The construction general expression between the local defect polarization and the
is motivated by the multirank lamination that is proven to be corresponding defect dipolar field can be derived as [45]
energy minimizing for ferroelectrics [142] and is shown to result
Ei;defect ¼ C1 Pi;defect =ðpj0 Þ (3.1.12)
in correct domain structure in bulk ferroelectrics [141]. The
approach is particularly convenient to couple different physical
processes, as shown for ferromagnetic shape memory alloys and where C1 is a constant approximately equal to 1.02. Then, a
multiferroic bismuth ferrite [143,144]. The expressions of total probability-based homogenization method can be adopted to
free energy are similar to Eq. (3.1.1) and thus will not be iterated. determine the macroscopic defect polarization. Assuming that the
The detailed description of the unconventional phase field model oxygen vacancy density in ferroelectric is Nd, the probability of
can be found in Refs. [46] and [141]. finding oxygen vacancy in a unit cell can be obtained as Nd V0 ,
where V0 is the unit cell volume. For a tetragonal BaTiO3 lattice,
Numerical implementation. Due to the nonlinearity and com- there are six possible oxygen vacancy sites, indexed as (1)–(6). If
plexity of governing equations, we usually cannot obtain analyti- the polarization is two-dimensional, the possible oxygen vacancy
cal solutions and thus need to resort to numerical methods [35]. sites are reduced to (1)–(4). Denoting qi as the probability of find-
The discretization of time and spatial scales are often performed ing oxygen vacancy at site (i) and using P(i),o to represent the local

defect polarization induced by the oxygen vacancy at site (i), the The evolutions of the spontaneous polarization and oxygen
relation between the macroscopic defect polarization and the oxy- vacancy distribution are controlled by different kinetics, as
gen vacancy density can be derived as described below. The temporal and spatial variations of spontane-
ous polarizations are governed by the classical time-dependent
X
4 Ginzburg–Landau equation (i.e., Eq. (3.1.9)). On the other hand,
Pdefect ¼ Nd V0 qi PðiÞ;o (3.1.13) the density of oxygen vacancy is governed by the diffusion equa-
i¼1 tion [132–134],
Note that the value of P(i),o can be calculated by first-principles

@Nd =@t ¼ r ½bNd rð@Wd =@Nd þ e#/Þ (3.1.17)
method, while the probability qi, which in general varies with
position, can be determined by minimizing the total free energy of
the ferroelectric system with oxygen vacancies. where b is the mobility, Wd is the contribution to the free energy
due to defects, which is assumed to be the usual free energy of
Phase field model of ferroelectrics with oxygen vacancies. For mixing at small concentrations, and # represents the donor
a ferroelectric system with oxygen vacancies, we need to consider valency. In the diffusion dynamics, while the impurity atoms can
not only the electrostatic interactions due to spontaneous polariza- hardly diffuse, since their atomic weights are much larger than
tion, but also the electrostatic interactions due to defect polariza- that of the oxygen atom, the localization of electrons can occur in
tions. This is where the model deviates from the classical theory, new adjacent Ti atoms after the diffusion of oxygen vacancies. As
as introduced in Sec. 3.1.1. In particular, the electrostatic energy such, it is expected that the distribution of defect dipole can vary
in ferroelectricity with oxygen vacancy can be expressed as in accordance with the diffusion of oxygen vacancy. Since the
variation of oxygen vacancies are governed by the diffusion pro-
felec ¼ Ei;p þ Ei;defect Pi þ Pi;defect =2 (3.1.14) cess, it is much slower than the polarization evolution and the
distribution of oxygen vacancies requires substantially longer
Under typical oxygen vacancy density [133], in the order of time to reach the equilibrium.
1 1024 m3, the defect polarization is three orders smaller than
the spontaneous polarization as determined from Eq. (3.1.13), and Results and discussions. The phase field model introduced
thus the electrostatic energy density can be simplified as above was employed to study two important issues towards the
interactions of oxygen vacancies with domain evolutions in

felec ¼ Ei;p þ Ei;defect Pi =2 (3.1.15) BaTiO3 single crystal [45]: (i) the distribution of defect density in
different domain structures and (ii) the influence of oxygen va-
cancy on the electromechanical responses of different domain
Thus, the influence of oxygen vacancy is described by the vacancy
structures under an electric field. In the first question, the ferro-
dipolar field in this framework. In an equilibrium state with mini-
electric is assumed to be sufficiently aged, such that the time is
mum free energy, both the defect polarization Pdefect and the cor-
long enough for the oxygen vacancies to diffuse to the equilibrium
responding defect dipolar field Edefect should be along the same
state. In the second question, the time scale is small relative to the
direction as the spontaneous polarization, given sufficient time for
oxygen vacancy diffusion process, so that only domain switching
relaxation. This implies that the defect dipoles should conform
takes place while oxygen vacancy distribution remains unchanged
with the direction of spontaneous polarization as postulated by
away from the equilibrium state.
Ren [135] and also confirmed by experimental observations [148]
The interactions of oxygen vacancies with three different do-
and first-principles calculations [149]. As such, the probabilities qi
main structures were studied, including a rank-2 domain structure
can be determined from the spontaneous polarization and the cor-
with both 90 deg and 180 deg domains, a rank-1 domain structure
responding defect polarization is derived as
with 180 deg domains, and a single domain state. Here, only the
simulation results of the rank-2 domain structure are demon-
P
Pdefect ¼ Nd V0 Pdefect0 (3.1.16) strated. This domain is shown in Fig. 21(a), where four ferroelec-
jP j tric variants separated by 90 deg and 180 deg domain walls
coexist in the ferroelectric. Such domain structure was indeed
where jPj denotes the magnitude of the spontaneous polarization observed in experiments in BaTiO3 single crystal [150], as shown
and Pdefect0 refers to the magnitude of the local defect in Fig. 21(b). Using such domain configuration as the initial con-
polarization. dition for the polarization distribution and assuming the oxygen
Fig. 21 The engineering domain structures with both 90 deg and 180 deg domain
walls: (a) calculated by the phase field simulation and (b) observed by experiments
in BaTiO3 single crystal. The arrow represents the polarization direction. Reprinted
from Ref. [45] with the kind permission of American Physical Society.

Fig. 22 The distributions of electrostatic potential (a) and oxygen vacancy density (b) in the
rank-2 domain structure. The initial oxygen vacancy density is uniformly distributed with a mag-
nitude of Nd 5 1 3 1024 m–3. Reprinted from Ref. [45] with the kind permission of American Physi-
cal Society.
vacancies are uniformly distributed initially with random noises, the redistribution of oxygen vacancies (Fig. 22(b)). It can be found
the resulting distribution of the electrostatic potential and oxygen that the oxygen vacancies are attracted to different sides of the 90
vacancy density are shown in Fig. 22. The electric potential deg domain wall, but no such accumulation occurs near the 180
changes abruptly at the 90 deg domain walls, but not at the 180 deg domain wall. This can be used to explain the domain wall pin-
deg domain walls. Thus, a large depolarization field exists at the ning and the resulted polarization fatigue.
90 deg domain wall, with the maximum magnitude as high as 5.1 To examine the influences of the oxygen vacancies on the
MV/m. The inhomogeneous electrostatic field furthermore drives electromechanical responses of the BaTiO3 single crystal, an
Fig. 23 The electric hysteresis loops (a) and butterfly loops (b) of rank-2 domain structures with different oxygen vacancy den-
sities, and the domain evolutions for Nd 5 1.2 3 1024 m–3 under a positive alternating field (c). Domains indexed by (I) and (II) in
(c) correspond to the two states (I) and (II) in (a) and (b). Reprinted from Ref. [45] with the kind permission of American Physical
Society.

alternating electric field is applied to pole and depole the @W a ðlÞ
Fgrad
l ¼ 2A1 r2 l; Fanis
l ¼ ;
material periodically. The hysteresis and butterfly loops are @l
shown in Fig. 23 for three different oxygen vacancy
@e0 ðlÞ
densities, which clearly demonstrate that the electromechanical Felas
l ¼ g1 ð f Þ C e g1 ð f Þe0 ðlÞ
responses are sensitive to oxygen vacancy density Nd. When Nd @l
is relatively small (i.e., Nd ¼ 0.5 1024 m3), the electromechan- @PðlÞ @PðlÞ
F0l ¼ f E0 ; Fdl ¼ f Ed ;
ical responses are similar to that of the traditional ferroelectric @l @l
material, since the defect dipolar field is not sufficient to switch
@2f
the domain to the original configuration. When Nd is relatively Fgrad
f ¼ 2A2 2 ; Fdielec f ¼ 2A3 f ;
large (i.e., Nd ¼ 0.9 1024 m3 or 1.2 1024 m3), a double hys- @x3
ð
teresis loop can be found, which indicates that the domain 1 @g1 ð f Þ 0
returns to the original configuration after the removal of electric Felas
f ¼ 2 e ðlÞ C e g1 ð f Þe0 ðlÞ dx1 dx2 ;
l0 x1 x2 @f
field. Such domain evolution process is indeed observed as ð ð
1 1
shown in Fig. 23(c). Since a large amount of 90 deg domain F0f ¼ 2 E0 PðlÞdx1 dx2 ; Fdf ¼ 2 Ed PðlÞdx1 dx2
switching occurs during the poling and depoling processes, large l0 x1 x2 l0 x1 x2
recoverable strain induced by the electric field can be found (3.1.20)
in Fig. 23(b). Such double hysteresis loop and large recoverable
strain were indeed observed in aged BaTiO3 single crystal with l0 being the in-plane dimension of computational super
[135,136]. cell. Notice that the order parameter l(x1,x2) satisfies periodic
boundary conditions in the x1 x2 plane, while f(x3) satisfies the
3.1.3 Extension of Phase Field Model for the Finite Size
traditional Landau boundary condition [151] (i.e., @f =@x3 jx3 ¼6h2
Effect on Ferroelectric Ultrathin Films. This subsection demon-
¼ f =d), with d being the extrapolation length and h being the
strates a computational model [46] to investigate the configuration
film thickness. The boundaries are also assumed to be electrically
and evolution of domain structure in ultrathin ferroelectric films,
short-circuited and mechanically clamped in-plane.
with particular emphasis on the interplay between the configura-
The phase field model introduced above was applied to study
tion and evolution of domain structure and the size effect.
ultrathin ferroelectric films with rhombohedral structure [46]. Fig-
The key idea to capture the size and surface effects in ultrathin
ure 24(a) shows the domain configuration obtained from the simu-
ferroelectric film is to introduce a new order parameter, f(x3),
lation for film with thickness 2.4 nm < h < 4.8 nm, where it is
which reflects the atomic relaxation along film thickness, so that
observed that the domain exhibits a stripe pattern composed of
local spontaneous polarization P** and transformation strain e**
two variants forming 180 deg domain walls, which agrees with
can be expressed in thickness-dependent form as P** ¼ f(x3)P and
that calculated from MD simulations [152], as shown in Fig.
e** ¼ f2(x3)e0, where P and e0 are nominal spontaneous polariza-
24(b). On the other hand, the domain structure obtained from sim-
tion and transformation strain of bulk materials. To simplify the
ulation of thicker films with h > 20 nm, when neglecting the out-
analyses, we can assume that li is independent of x3, implying
of-plane depolarization effect, is nearly identical to that obtained
that the domain walls are perpendicular to film thickness, which is
by Shu et al. [141] for bulk ferroelectrics, as shown in Figs. 24(c)
reasonable for ultrathin film. The total energy of ferroelectric film
and 24(d). The zigzag domain pattern appearing in this case has
can then be expressed in terms of the order parameter l(x1,x2) and
also been observed in PZT in the vicinity of a morphotropic phase
f(x3) as
boundary (MPB) [153].
ð Then, the phase field method is employed to analyze the size

Gðl; f Þ ¼ W grad ðrl; rf Þ þ W anis ðlÞ þ W dielec ð f Þ effect in ultrathin ferroelectric films. The variations of average
out-of-plane polarization along the thickness direction for films
with different thicknesses are shown in Fig. 25. The distributions
þ W elas ðl; f Þ R0 e E0 P dx
ð are almost similar for all films, with larger polarization inside and
j smaller one outside, which is consistent with the prediction by tra-
þ jr/j2 dx (3.1.18) ditional univariate Landau-Ginzburg-Devonshire (LGD) theory
2 R3
[151]. Figure 26 further demonstrates the variation of the domain
In the first integral of the equation, W grad ¼ A1 jrlj2 þA2 jrf j2 is pattern and the average out-of-plane polarization with respect to
the gradient energy that penalizes the film thickness. With decreasing film thickness, the domain
P the changes in the order param-
configuration undergoes three transitions: from a zigzag pattern
eters l and f . W anis ¼ K N1 2 2
i¼1 li ð1 li Þ is the anisotropy
energy of double-well type, ensuring that the characteristic func- with eight variants coexistence to a zigzag pattern with four var-
tions take either 0 or 1. The third term, Wdielec(f) ¼ A3f2, represents iants coexistence at h ¼ 7.6 nm, then to a vortex pattern with four
the internal energy induced by linear dielectric effect, correspond- variants coexistence at h ¼ 6.8 nm, and finally to a stripe pattern
ing to the quadratic term of polarization in the Landau polyno- with two variants coexistence at h ¼ 4.4 nm. When the film thick-
mial. The fourth term, Welas(l), is the stored elastic energy. The ness falls further below 2.8 nm, the domain pattern vanishes and
fifth and sixth terms are the potential energies caused by the the polarization becomes zero, indicating the suppression of fer-
applied electric field and stress field. The second integral in Eq. roelectricity for films thinner than 2.8 nm. This critical thickness
(3.1.18) denotes the depolarization energy, which is equivalent to of 2.8 nm is similar to that predicted by ab initio calculations [67]
Eq. (3.1.7). for BaTiO3 films between two metallic SrRuO3 electrodes. It can
The evolutions of the order parameters, l(x1,x2) and f(x3), are also be found from Fig. 26 that the number of coexisting variants
governed by the gradient flow of the total energy density G(l,f) in decreases with the thickness.
Eq. (3), i.e.
3.2. Optimization-Based Computational Model for Poly-
d
@l=@t ¼ L1 Fgrad
l þ Fanis elas 0
l þ Fl þ Fl þ Fl ; crystalline Ferroelectrics. The above-presented phase field
(3.1.19) method (PFM) seems more promising compared to others, as no
@f =@t ¼ L2 Fgrad þ Fdielec þ Felas þ F0f þ Fdf imposition of any priori domain-switching criterion is required
f f f
and domain switching is a natural process during minimization of
the free energy of the whole material system. PFM has achieved
where L1 and L2 are the mobility coefficients and the various driv- great success in modeling domain evolutions in ferroelectric sin-
ing forces are given by gle crystals [35,39,42], while it has hardly been used for modeling

Fig. 24 Two extreme cases simulated by the phase field method: (a) The domain pattern of a
ultrathin film calculated by the phase field method with 2.4 nm < h < 4.8 nm [46]; (b) the domain
pattern of the ultrathin film calculated by molecular dynamics with h 5 4.4 nm [152]; (c) the do-
main pattern of a thin film calculated by the phase field method with h > 20 nm without consider-
ing the out-of-plane depolarization [46]; (d) the domain pattern of the thin film calculated by the
2D phase field simulation without concerning the size effect [141]. The symbol
() denotes
the direction of polarization flowing out of (into) the x1x2 surface. Reprinted from Ref. [46] with
the kind permission of The American Institute of Physics.
polycrystalline ferroelectrics [84,85], probably because of the dif-

ficulties encountered in addressing the complicated interactions,
especially the elastic interactions between grains. Moreover, the
rather large computational complexity encountered in handling
three-dimensional cases may be another challenge for the use of
PFM. To solve these problems in conventional PFM, we proposed
an optimization-based computational model for polycrystalline
ferroelectrics [52–54].
3.2.1 Method Description and Optimization Methodology

Material model. In this model, a polycrystalline ferroelectric
ceramic is made up of numerous randomly oriented grains, each
of which contains N types of domains (N ¼ 6 for the tetragonal,
N ¼ 8 for the rhombohedral, and N ¼ 14 for the morphotropic
ceramics, where tetragonal and rhombohedral phases coexist).
The volume fraction of each type ofPdomain is denoted by
fi ði ¼ 1; 2:::NÞ, and obviously we have Ni¼1 fi ¼ 1. For the mor-
photropic ceramics, field-induced phase transformations between
tetragonal and rhombohedral phases are permissible; thus, both
the in-phase and interphase domain switching can occur during
Fig. 25 The size-dependent distribution of the out-of-plane electric and/or mechanical loading. The charge screening effect
polarization at different thicknesses. Reprinted from Ref. [46] [3,154] in real ceramics is included in this model; thus, the depo-
with the kind permission of The American Institute of Physics. larization electric field induced by polarization switching

1
UðE; rÞ ¼ E Pr r : er þ kE E
2
1 1 1 v
þ r:r ½trðrÞ2
2 2l 2 2lð1 þ vÞ
1 7 5v
þ 2ld ðer eÞ:ðer eÞ
2 15ð1 vÞ
þ f180 W180 þ fnon180 Wnon180 (3.2.1)
where E and r are the applied electric field vector and applied
stress tensor, respectively. k is the isotropic dielectric constant and
landv are the isotropic shear modulus and Poisson’s ratio. e is the
mean strain tensor of the whole material system or the matrix. ld
is the equivalent shear modulus during domain switching and
should be calibrated experimentally using the stress-strain curve.
As the material will become “softer” when domain switching
occurs, ld < l always hold true.
Pr and er are the remnant polarization and remnant strain of a
Fig. 26 The size-dependent out-of-plane polarization and grain, which can be expressed as linear functions of the volume
domain pattern. The domain exhibits a zigzag pattern with fractions of domains as follows:
eight variants coexisting when h > 7.6 nm, a zigzag pattern with
four variants coexisting when 6.8 nm < h < 8.0 nm, a vortex X
N X
N
pattern with four variants coexisting when 4.4 nm < h < 7.2 nm, Pr ¼ fi Pi ; er ¼ fi ei (3.2.2)
and a stripe pattern with two variants coexisting when
i¼1 i¼1
2.4 nm < h < 4.8 nm; when the thickness is smaller than 2.8 nm,
the domain disappears, indicating that the ferroelectricity is
suppressed. Reprinted from Ref. [46] with the kind permission where Pi and ei are the spontaneous polarization and spontaneous
of The American Institute of Physics. strain tensor of the ith domain.
The six types of tetragonal domains and eight types of rhombo-
hedral domains in quasicubic coordinates are illustrated in
Fig. 28.
vanishes. The mechanical interactions between grains, however, For the tetragonal ferroelectric crystals, we have
cannot be treated similarly and are considered in a self-
consistent Eshelby inclusion manner [155]. That is, each grain is Pð1Þ ¼ Pð2Þ ¼ P0 ð0; 0; 1ÞT ; Pð3Þ ¼ Pð4Þ ¼ P0 ð1; 0; 0ÞT ;
regarded as an inclusion surrounded by an infinite large matrix
with the materials properties the same as the whole material, as Pð5Þ ¼ Pð6Þ ¼ P0 ð0; 1; 0ÞT (3.2.3)
shown in Fig. 27. As the spherical inclusion case is the simplest 2 3
1
and no evidence indicates that other shapes of inclusion are more S0 6 7
accurate, we use the spherical inclusion case in this model. Fur- eð1Þ ¼ eð2Þ ¼ 4 1 5;
3
thermore, to grasp the major responses of domain switching and 2
to make the computational complexity affordable, we also 2 3 2 3
assume that a ferroelectric ceramic is dielectrically and elasti- 2 1
S0 6 7 S0 6 7
cally isotropic and shows linear dielectric and elastic behavior eð3Þ ¼ eð4Þ ¼ 4 1 5; eð5Þ ¼ eð6Þ ¼ 4 2 5
3 3
unless domain switching occurs (i.e., all the nonlinear items are 1 1
caused by domain switching).
(3.2.4)
Free energy of a single grain. In the present model, as For rhombohedral crystals, we have
the interactions between grains have been considered in an
Eshelby inclusion manner, we use the free energy of each grain P0 P0
Pð1Þ ¼ Pð2Þ ¼ pffiffiffi ð1; 1; 1ÞT ; Pð3Þ ¼ Pð4Þ ¼ pffiffiffi ð1; 1; 1ÞT ;
as the optimization objective and sweep the optimization process 3 3
over all grains to get the domain structures and properties of the P0 P0
whole material system. The free energy of a specific grain is P ¼ P ¼ pffiffiffi ð1; 1; 1Þ ; P ¼ P ¼ pffiffiffi ð1; 1; 1ÞT
ð5Þ ð6Þ T ð7Þ ð8Þ
3 3
expressed as
(3.2.5)
Fig. 27 2D Illustration of material model for morphotropic fer- Fig. 28 (a) Six types of tetragonal domains and (b) eight types
roelectric ceramics of rhombohedral domains in ferroelectric crystals

2 3
0 1 1 Obviously, the free energy is a nonlinear function of volume
ð1Þ ð2Þ S0 6 7 fractions of domains; thus, problem I turns to be a constrained
e ¼e ¼ 4 0 1 5;
3 nonlinear optimization problem. To solve such an optimization
sym 0 problem, the sequential quadratic programming (SQP) method
2 3
0 1 1 [159] is a very effective algorithm with high accuracy and quick
S0 6 7 convergence. As the expression of the free energy varies from
eð3Þ ¼ eð4Þ ¼ 4 0 1 5
3 grain to grain and also depends on the applied loading, at each
sym 0 loading step, the optimization process is applied over all grains,
2 3 (3.2.6)
0 1 1 and the volume fractions of domains in each grain are obtained to
S0 6 7 get the overall polarization and strain.
eð5Þ ¼ eð6Þ ¼ 4 0 1 5;
3
sym 0 Domain texture evolution. During domain switching in single
2 3 crystals, the topological structure of domains may be the most
0 1 1 concerned, as the material properties are usually very sensitive to
S0 6 7
eð7Þ ¼ eð8Þ ¼ 4 0 1 5 it, while in polycrystals, the detailed topological domain structure
3
sym 0 of a specific grain makes little sense, as the material consists of
numerous grains usually oriented in all directions. Li et al.
[52–54] introduced the concept of “topology insensitive structure”
where P0 is the spontaneous polarization and S0 is the single crys-
in ferroelectric polycrystals (i.e., the domain structure inside a
tal deformation in ferroelectric crystals [2]. For the tetragonal,
grain can be determined only by the volume fraction of each type
S0 ¼ Slattice ¼ c=a 1 (where c and a are tetragonal lattice con-
domain and is independent of the detailed topological structures).
stants), while for the rhombohedral, we have [16,17,156]
In their work, the domain structure evolution is represented by the
ð8=9ÞS0 ¼ Slattice ¼ d½111 =d½111 1 (3.2.7) domain texture (pole figure of the polar axis) evolution calculated
by this optimization-based model.
The pole figure [160] of the polar axis (i.e., (001) for tetragonal
Similarly, the spontaneous polarization vectors and strain tensors and (111) for rhombohedral) are calculated at each loading step.
for other domains can be obtained by permutation and will not be For the morphotropic ferroelectric ceramics, the cumulative pole
presented here. figures of [001] axis in tetragonal phase and [111] axis in rhombo-
Returning to Eq. (3.2.1), W180 and Wnon180 are the energy bar- hedral phase are obtained to get the domain texture evolution pro-
rier (per volume) for 180 deg and non-180 deg domain switching, cess. Under electric loading or uniaxial compression (tension), the
respectively. Obviously, all the interphase domain switchings pole figure of the polar axis is axisymmetric around the loading
between tetragonal and rhombohedral phases are non-180 deg axis and thus can be described by an angular distribution function
switchings. For simplicity, it is assumed that all the non-180 deg gðhÞð0 h p=2Þ [161] and gðhÞ 1 for the random uniform
domain switchings have the same energy barrier. Following previ- distribution, and the normalization condition requires
ous models [157,158], here we adopt W180 ¼ 2P0 EC and
Wnon180 ¼ P0 EC , where EC is the nominal coercive field. f180 and ð p=2
fnon180 are the volume fractions of 180 deg and non-180 deg gðhÞ sin hdh ¼ 1 (3.2.9)
domain switching, respectively, and are calculated as follows: 0
As indicated by Li et al. [52,53], gðhÞ completely depends on non-

1 X
N=2

fnon180 ¼ 0
f2i1 þ f2i ðf2i1 0
þ f2i1 Þ (3.2.8a) 180 deg domain switching and is independent of 180 deg switch-
2 i¼1 ing. In the present work, gðhÞ is calculated using 10,000 grains,
and the interval of (0, 90 deg) is divided into 18 subintervals, giv-
1X N
fi f 0 fnon180 ing rise to equal interval of 5 deg. The accumulated volume frac-
f180 ¼ i (3.2.8b)
2 i¼1 tions of domains at all these subintervals give the values of gðhÞ at
discrete values of h, which are midpoints of the 18 subintervals.
where fi0 is the volume fraction of the ith domain calculated at the 3.2.2 Simulation Results and Discussions. Firstly, both tetrag-
last loading step. onal and rhombohedral single-phase PZT ferroelectric ceramics
In the expression of the free energy (i.e., Eq. (3.2.1)), the first under electric/mechanical poling are simulated using the proposed
two items on the right side are the potential energy; the third to optimization model. The materials constants used in the simula-
fifth items are the linear dielectric energy and strain energy, tions are listed in Table 4, referring to Hoffmann et al. [162], but
respectively; the sixth item is the inclusion strain energy or misfit with slight modifications because of the isotropic dielectric and
strain energy; and the last two items are the dissipation energies elastic assumptions. For a better comparison, the elastic proper-
during domain switching. From Eqs. (3.2.2)–(3.2.8), it can also be ties, dielectric properties, the spontaneous polarization, and the
seen that, under any prescribed electromechanical loading, the nominal coercive field of the two types of ceramics are taken to
only unknown variables to be solved in Eq. (3.2.1) are the volume be the same, as they did not differ much in reality.
fractions of each type domain (i.e., fi ði ¼ 1; 2:::NÞ).
Optimization methodology. In this model, the volume fraction

of each type domain under any specific loading is calculated by Table 4 Material constants of tetragonal and rhombohedral
minimizing the total free energy of a specific grain in polycrystal- PZT used in the model
line ferroelectrics. The mathematical optimization problem for
Material constants Rhombohedral Tetragonal
this model can be written as
Problem I Shear modulus l(GPa) 30
Poisson’s ratio v 0.3
Min Uðf1 ; f2 ; ; fN Þ Dielectric constant kðF=mÞ 1:0 108
s:t: 0 fi 1 ði ¼ 1; 2; ; NÞ Spontaneous polarization P0 (C/m2) 0.52
Lattice deformation Slattice 0.65% 2.77%
X
N
Single crystal deformation S0 0.73% 2.77%
fi ¼ 1 Coercive field EC (kV/mm) 1.0
i¼1

Fig. 29 Simulated D-E hysteresis loops (a) and butterfly curves (b) of tetragonal PZT ceramics during the first three cycles of
electric loading. The overlapped curves after 1.5 cycles loading indicates the good convergence of the proposed model.
Convergence and computational complexity. The simulation Ferroelastic domain switching under mechanical loading. We
results on tetragonal and rhombohedral PZT ceramics have shown also simulated the ferroelastic domain-switching process in
that both the P–E hysteresis loops and strain curves will stabilize unpoled tetragonal and rhombohedral PZT ceramics under uniax-
only after 1.5 cycles of electric loading (see Fig. 29 for the case of ial tension/compression loading using the proposed model [52].
tetragonal PZT ceramics) [163]. The hysteresis loops and strain The material constants used here are the same as listed in Table 4,
curves for subsequent cyclic loading completely overlap the for- except that the energy barrier for ferroelastic domain switching is
mer curves, which indicate good convergence of this computa-
tional model.
It should be mentioned that, by focusing on the volume frac-
tions of domains instead of the detailed domain patterns in grains,
this optimization-based model has a much smaller computational
complexity compared to other methods, such as the phase field
model (PFM). Therefore, it is feasible to study the 3D cases with
a very large number of grains using this model. For the case of a
polycrystalline ceramic with 10,000 grains and 61 loading steps,
the computation time of this model on a Pentium 3.0 GHz perso-
nal computer is only 1.5 h for tetragonal PZT and about 2.5 h for
rhombohedral PZT ceramics.
Ferroelectric switching under electric loading. Figure 30
shows the simulated polarization and longitudinal strain curves
under cyclic electric loading up to 5EC for both tetragonal and
rhombohedral PZT ceramics. Both the remanent polarization and
strain of the rhombohedral PZT ceramics are obviously larger
than that of the tetragonal PZT ceramics. The rather small strain
in the tetragonal ceramics indicates that little 90 deg domain
switching occurs, although the 180 deg switching is almost
complete, which can be estimated from the polarization curve in
Fig. 30(a). In comparison, the large remanent strain of rhombohe-
dral PZT ceramics shows that a considerable amount of non-180
deg switching has occurred during electric poling. From Fig. 30, it
can also be seen that, in both ceramics, domain switching is a
gradual process with the applied electric loading.
Figure 31 shows the domain textures evolution process in both
tetragonal and rhombohedral PZT ceramics corresponding to the
electric loading points indicated in Fig. 30(a). It can be seen that,
during electric loading, the domain textures of tetragonal PZT
ceramics change little from the initial unpoled state, which is con-
sistent with the small remanent strain shown in Fig. 29(b). As
expected, the domain textures of rhombohedral PZT ceramics
change a lot from the unpoled state during electric poling
(R1!R2), as shown in Fig. 31(b). It also shows that, upon remov-
ing the electric field, partial domain switching reverses, which
causes domain texture variations (R2!R3). Nevertheless, in Fig. 30 Simulated (a) polarization and (b) strain curves of tet-
rhombohedral PZT ceramics, after electrical poling, the domain ragonal and rhombohedral PZT ceramics. Reprinted from Ref.
textures approach the theoretically saturated state [161] and can [53] with the kind permission of The American Institute of
never return to the initial unpoled state in subsequent loading. Physics.

is typically no more than 50 MPa [94]; thus, the results under large
tension are only presented to show the tendency and make a com-
parison with the theoretical studies [161]. As had been shown by
Li and Soh [52], for tetragonal PZT ceramics, the switching strain
under tension and compression is symmetric and very small. The
maximum switching strain in both cases is about 0.1%, and the re-
manent strain is only about 0.04%. The switching strain is far from
saturation, even under a high stress (5rC ), which indicates that fer-
roelastic domain switching in tetragonal PZT ceramics is very dif-
ficult to accomplish. In comparison, considerable amount of
switching strains can be realized in rhombohedral PZT ceramics
via ferroelastic domain switching [52]. Moreover, in rhombohedral
PZT ceramics, the switching strains under uniaxial tension and
compression are asymmetric, which accords well previous theoret-
ical studies [161]. The calculated different domain-switching
behavior of tetragonal and rhombohedral PZT ceramics coincides
with what we get from our analytical model [16,17] and is also
consistent with experimental observations [164].
Figure 33 shows the calculated domain texture evolution pro-
cess in both tetragonal and rhombohedral PZT ceramics under
uniaxial tension/compression. As expected, in tetragonal PZT
ceramics, the domain texture evolution under tension and com-
pression is symmetric and does not deviate much from the initial
unpoled state, which is consistent with the symmetric and small
switching strains in Fig. 32(a). In comparison, in rhombohedral
PZT ceramics, the domain texture changes a lot from the initial
unpoled state during mechanical loading, reaching the theoreti-
cally saturated state [161] under maximum tension. While under
Fig. 31 (a) (001) pole figure of tetragonal PZT and (b) (111) pole
figures of rhombohedral PZT ceramics during cyclic electric
loading. Reprinted from Ref. [53] with the kind permission of
The American Institute of Physics.
expressed as Wnon180 ¼ S0 rC , where rC is the coercive stress and

is taken to be 70 MPa for both tetragonal and rhombohedral PZT
ceramics in this paper.
Figure 32 shows the calculated switching strain curves of tetrag-
onal and rhombohedral PZT ceramics under uniaxial tension/com-
pression. The simulation results under a very high tensile stress
may not make any sense, as the tension strength of PZT ceramics
Fig. 32 Switching strain curves of tetragonal and rhombohe- Fig. 33 (a) (001) pole figures of tetragonal PZT and (b) (111)
dral PZT ceramics under uniaxial tension/compression. pole figures of rhombohedral PZT ceramics during uniaxial ten-
Reprinted from Ref. [52] with the kind permission of Elsevier sion/compression. Reprinted from Ref. [52] with the kind per-
Science Ltd. mission of Elsevier Science Ltd.

maximum compression, the domain texture is not so saturated. It nal PZT and 48% for rhombohedral PZT, which may indicate that
can also be seen from Fig. 32(b) that, in rhombohedral PZT the ferroelastic switching–induced internal stress in morphotropic
ceramics upon removing stress, reverse domain switching PZT ceramics is not very large.
(R2!R3) occurs and it is more obvious during tension than that The domain textures of morphotropic PZT ceramics under
during compression (R4!R5). applied stress, as shown in Fig. 34(b), show that, under maximum
tension, most domains switch to a state with their elongation axis
Reproducing the Taylor’s rule of plasticity. We further investi- oriented as close as possible to the tensile loading direction. In the
gate the domain evolution under mechanical loading in PZT narrow range of h < 10 deg, gðhÞ reaches a value higher than six,
ceramics near the MPB where tetragonal and rhombohedral which is close to the mathematical saturated value of seven. After
phases coexist. In the simulation, it is assumed that the morpho- removing the uniaxial tension, some domains return to their initial
tropic PZT ceramics contains equal fractions of tetragonal and states, making curve C slightly lower than curve B within a small
rhombohedral phases with the material constants listed in Table 4. range of h (say h < 30 deg). In comparison, under maximum
Figure 34 shows the simulated switching strain-stress curve and uniaxial compression, most domains switch to a state with their
the cumulative pole figures of (001) and (111) axes in morpho- elongation axis oriented as perpendicular as possible to the com-
tropic PZT ceramics under uniaxial tension and compression. It pression direction, referring to as state D in Fig. 34(b), where gðhÞ
can be seen from Fig. 34(a) that, similar with the rhombohedral vanishes from h ¼ 0 deg until h 55 deg and after that increases
PZT case shown in Fig. 32, the switching strain of morphotropic gradually with h. However, the maximum value of gðhÞ at 90 deg
PZT under maximum tension (about 0.62%) is considerably larger is only about three, considerably smaller than the mathematical
than that under maximum compression (0.4%). However, differ- saturated value of seven, which may be an indication that it is
ent from the case in rhombohedral ceramics, the remnant switch- more difficult to reach the saturated domain orientation state
ing strain under tension (about 0.5%) is also larger than that under under uniaxial compression than that under uniaxial tension. In
compression (about 0.35%). The calculated remnant strain under Fig. 34(b), the almost overlapped curves at states D and E may
compression (0.35%) is very close to the few experimental data indicate that only a little reverse ferroelastic switching occurs
obtained very recently (i.e., 0.3% [165] and 0.33% [166]) on upon removing the compressive stress.
unpoled morphotropic PZT ceramics under compression. The It can be seen from Figs. 32 and 34(a) that, after removing the
reverse ferroelastic switching that occurs in morphotropic PZT mechanical stresses, the remnant switching strain of the morpho-
ceramics upon removing the applied stress is not as severe as that tropic PZT ceramics (0.5% under tension and 0.35% under com-
in the rhombohedral and tetragonal PZT ceramics. Over 80% of pression) is much larger than that of the tetragonal (0.04%) and
maximum switching strain is retained in morphotropic PZT rhombohedral PZT ceramics (0.12%). This tendency can be more
ceramics, in comparison with that of only about 40% for tetrago- clearly seen in Fig. 35, in which the switching strain versus stress
curves for three types of PZT ceramics are replotted for compari-
son. As indicated by Li and Soh [52], the remnant strain in
polycrystalline ferroelastics is crystal symmetry–dependent.
According to the Taylor’s rule of plasticity [167], a crystal must
have at least five slip systems for its polycrystalline to be ductile.
In terms of the deformation modes, a tetragonal ferroelastic crys-
tal has three slip systems (among which two are independent) and
a rhombohedral ferroelastic crystal has four (three of which are in-
dependent). Thus, in tetragonal or rhombohedral polycrystalline
ferroelastics, the change of spontaneous strain in a particular grain
during ferroelastic domain switching cannot be accommodated by
neighboring grains and, therefore, will generate large internal
stresses inside the grain (i.e., such ferroelastic switching will be
constrained by the surrounding grains and may not occur) [52].
The tetragonal and rhombohedral PZT ceramics are thus brittle
materials from this point of view, while in PZT ceramics near the
MPB, where tetragonal and rhombohedral phases coexist, there
Fig. 34 (a) Switching strain-stress curves and (b) the cumula- Fig. 35 Comparison of switching strain versus stress curves
tive pole figures of (001) and (111) axes in morphotropic PZT of tetragonal, rhombohedral, and MPB PZT ceramics (the nomi-
ceramics under uniaxial tension and compression. Reprinted nal coercive stress rC 5 70 MPa). Reprinted from Ref. [52] with
from Ref. [52] with the kind permission of Elsevier Science Ltd. the kind permission of Elsevier Science Ltd.

are totally six independent deformation modes, which causes the
material to be ductile. During ferroelastic switching in morpho-
tropic PZT ceramics, most of the switching strains in a particular
grain can be accommodated by neighboring grains. Thus, the fer-
roelastic domain switching does not generate very large internal
stress in the grain and can be almost completely accomplished in
morphotropic PZT ceramics [164].
Based on the Taylor’s rule of plasticity, the remnant switching
strain for tetragonal and rhombohedral PZT ceramics should be
zero. Actually, the nonzero simulated remnant strain in the tetrag-
onal and rhombohedral PZT ceramics (refer to Fig. 35) is due to
the large coercive stress (70 MPa) for ferroelastic domain switch-
ing. If a very small coercive stress (say 5 MPa) is employed for
simulation, the remnant strain for both tetragonal and rhombohe-
dral ceramics will reduce to nearly zero, leading to the almost
hysteresis-free strain-stress curves, as shown in Fig. 36. In com-
parison, the remnant strain of the morphotropic PZT ceramics is
still fairly large, even if a small coercive stress is used, and this is
because the Taylor bound of the remanent strain in morphotropic
PZT ceramics is not zero but a fairly large value. For the morpho-
tropic PZT ceramics with the values of single-crystal deformation
listed in Table 1, the Taylor bound of remnant switching strain is
0.32% under uniaxial tension and 0.22% under uniaxial compres-
sion, which, as expected, are smaller than the remnant strain in
real cases (0.5% under uniaxial tension and 0.35% under uniaxial
compression, as shown in Fig. 35). From both Figs. 35 and 36, it
can be seen that the proposed model can reproduce the Taylor’s
rule of plasticity very well.
Domain switching under coupled electromechanical loading.
As PZT ceramic is a brittle material and its tensile strength (about
40 MPa) is usually much smaller than its compression strength
(say above 400 MPa), in some applications, such as actuators, to
prevent it from tensile fracture, a precompression is usually
applied; thus, the ceramic is mostly under uniaxial electromechan-
ical loading. Therefore, we further studied both the nonlinear per-
formance and domain texture evolution in morphotropic PZT
ceramics under uniaxial electromechanical loading using our pro-
posed optimization-based computational model [54].
Here, the morphotropic PZT ceramic still has equal volume
fractions of tetragonal and rhombohedral phases with the material
constants listed in Table 4, except that the single crystal deforma-
tion S0 is taken as 2.0% for the tetragonal and 1.0% for the rhom-
bohedral, referring to the XRD results [168]. Furthermore, to
compare our simulation results with existing experiments
[99,169], the precompression is applied up to 400 MPa and the
maximum applied electric field is 2 kV/mm.
Fig. 37 (a) D-E hysteresis loops, (b) butterfly curves, and (c)
reversed butterfly curves of a morphotropic PZT ceramic under
electric loading with different levels of precompression.
Reprinted from Ref. [54] with the kind permission of The Ameri-
can Institute of Physics.
Figure 37 shows the simulated stabilized D–E hysteresis loops,

butterfly curves, and reversed butterfly curves of a morphotropic
PZT ceramic under electric loading with different levels of pre-
compression (i.e., 0, 25, 50, 75, 100, 200, and 400 MPa). Consist-
ent with the experiments [169], with the increasing prestress, the
D–E hysteresis loops become more flat, indicating that the switch-
able polarization becomes smaller. The coercive field becomes
Fig. 36 Simulated switching strain versus stress curves of tet- less distinct, and obviously, the polarization cannot saturate under
ragonal, rhombohedral, and MPB PZT ceramics using a small the maximum applied field of 2 kV/mm. When the prestress is
nominal coercive stress of rC 5 5 MPa. Reprinted from Ref. [52] above 200 MPa, the remnant polarization becomes very small
with the kind permission of Elsevier Science Ltd. (<0.04 C/m2) and the hysteresis loops become less obvious.

In Fig. 37(b), with the increase of the prestress, the simulated
butterfly curves show somewhat different tendencies from the
polarization curves. The switchable strain does not decrease but
increases a little bit when the prestress is below 50 MPa. Such a
phenomenon had also been observed experimentally by Zhou
et al. [169] but was not observed in Fang and Li’s [99] experi-
ments. When the prestress is above 50 MPa, the switchable strain
decreases steadily with the increasing compression and almost
vanishes above 200 MPa. The reversed butterfly curves under uni-
axial electromechanical loading in Fig. 37(c) show the same tend-
ency as the butterfly curves for different levels of prestress, also
consistent with the experimental observations [99,169].
To study the domain texture evolution process of morphotropic
PZT under electromechanical loading, at each loading step, the
cumulative pole figures of [001] and [111] axes were calculated
by using the volume fractions of domains obtained by the SQP
algorithm. For each level of prestress, the domain textures at three
typical loading points were presented (i.e., before electric loading,
under maximum electric field, and after removing the electric
field), as shown in Figs. 38(a)–38(c).
It can be seen from Fig. 38(a) that, when only the precompres-
sion is applied, domains will reorient with their elongation axes
more close to the plane perpendicular to the compressive direc-
tion. The mathematically saturated domain texture under uniaxial
compression, which can act as the upper bound of domain texture
in this case, is also calculated and plotted using the dashed line in
Fig. 38(a). It can be seen that a uniaxial compression of 25 MPa is
large enough to drive the non-180 deg domain switching and
change the domain texture significantly. The domain texture under
400 MPa is almost identical to the saturated texture under com-
pression. In comparison, the domain texture under 5 EC [54] is
still considerably deviated from the corresponding saturated state.
This may indicate that, to switch the domain texture in a morpho-
tropic PZT ceramic, a uniaxial compression of 400 MPa is more
effective than an electric field of 5 EC (or 5 kV/mm). Note that
such a tendency has been observed by us recently [170].
Figure 38(b) shows the domain texture of morphotropic PZT
ceramics at the maximum field of 2 kV/mm with different levels
of precompression. The electric field tends to orient the domains
“stand up” along the field direction, while the uniaxial compres-
sion tends to push the domains “lie down” on the plane perpendic-
ular to the field direction. The competition between the electric
field and the uniaxial compression determines whether the domain
will “stand up” or “lie down”. Therefore, when the precompres-
sion is small (say 25 MPa), the electric field of 2 kV/mm is domi-
nant and most domains will “stand up” close to the field direction.
As the compression reaches 50 MPa, most domains stay at the in-
termediate states between “standing up” and “lying down”. Under
a rather high compression, almost all the domains have been con-
fined within those planes and ferroelastic domain switching can
hardly be driven by the limited amplitude of electric loading, lead-
ing to the almost same domain textures in Figs. 38(a) and 38(b)
with precompressions above 200 MPa.
Figure 38(c) shows the domain texture of the morphotropic
PZT ceramics after removing the maximum electric field of 2 kV/
mm with different levels of precompression. When the compres-
sion is high enough (above 200 MPa), almost no ferroelastic do-
main switching occurs; thus, no back switching occurs upon Fig. 38 The cumulative pole figures of [1] and [111] axes of a
removing the electric field, while under a lower compression morphotropic PZT ceramic with different levels of precompres-
(100 MPa or lower), the prestress is not large enough to confine sion: (a) before electric loading; (b) under maximum electric
the domain; thus, non-180 deg domain switching occurs under field; (c) after removing the electric field. Reprinted from Ref.
applied electric field, and reverse switching occurs upon removing [54] with the kind permission of The American Institute of
the field. Physics.
It should be noted that, in this optimization-based computational
model, there are no fitting parameters and all the used parameters
are materials constants. All the simulations results can fit well with [162], respectively, while the simulated maximum strain is only
the experiments’ results [162,164], but there still exist deviations 0.03% and 0.21% for the tetragonal and rhombohedral [53], which
between the simulations and experimental results. For example, are considerably smaller than the experimental results. We think
the experimentally observed electric field–induced strain can reach this is due to the fact that the misfit elastic energy between grains
0.2% and 0.35% for tetragonal and rhombohedral PZT ceramics in Eq. (3.2.1) had been considerably overestimated, and thus, the

ð pÞ ð pÞ
volume fractions of switchable domains had been underestimated where ci 0 and ci 1 are volume fractions of variant p in grain i
under a prescribed loading. In reality, when domain switching before and after the switching. The energy dissipation for 180 deg
occurs, the materials become “softer” and a smaller elastic con- and non-180 deg switching is distinguished, i.e.,
stant l should be used in the misfit energy item in Eq. (3.2.1),
therefore multiplying l by a fitting parameter less than unit can fit D ¼ ci D90 jDci j90 þD180 jDci j180 (3.3.5)
the experimental results better. While we just want to propose a
more general computational model without any fitting parameters where
that can capture the major characteristics of domain switching in
polycrystalline ferroelectrics, a model with fitting parameters will
1 X ð2p1Þ1
K=2
ð2pÞ ð2p1Þ0 ð2pÞ
inevitably become less predictive. jDci j90 ¼ ci þ ci 1 ci þ ci 0 (3.3.6)
2 p¼1
3.3 Monte Carlo Method for Polycrystalline Ferroelec-
trics. Alternatively, domain switching in ferroelectrics can be is the accumulated volume fraction change contributed by non-
regarded as a stochastic process. Tang et al. [51] proposed a 180 deg switching and
Monte Carlo method to simulate the domain-switching process in
polycrystalline ferroelectrics. In their work, similar to Li and jDci j180 ¼ jDci j jDci j90 (3.3.7)
Rajapakse [16,17], a ferroelectric polycrystal consists of numer-
ous grains, each containing several types of domains depending is the accumulated volume fraction change contributed by non-
on the crystal structure. The potential energy of the whole mate- 180 deg switching. It is assumed that
rial system is evaluated for all the possible states, and the varia-
tions of potential energy are taken as the driving force for domain D90 ¼ PS Ec ; D180 ¼ 2PS Ec (3.3.8)
switching. At each step of loading, the probability of each type of
domain switching is evaluated to get the volume fractions of where PS is the spontaneous polarization of the ferroelectric crys-
domains. The polarization, strain, and the domain texture can then tal and Ec is the corresponding coercive field of the single crystal.
be obtained with the applied loading. It is assumed that domain switching can only occur if it leads to
3.3.1 Domain Switching Based on Monte Carlo Method. reduction in potential energy, and such reduction should be suffi-
Here, we consider domain switching from variant p to variant q in ciently large to overcome the associated energy dissipation. As a
grain i, such that the volume fraction of variant p in grain i result, it is necessary that
ð pÞ ð pÞ
changes from ci to ci Dpq i , while that of variant q changes
ðq Þ ðqÞ pq
from ci to ci þ Di . As a result, we have the change of trans- DFin þ D 0 (3.3.9)
field in grain i given as [51]
for the switching to occur. In fact, this inequality has often been
ðqÞ ð pÞ used as a switching criterion for ferroelectrics. Here, we propose a
D YS i ¼ Dpq
i Yi Y i (3.3.1)
probabilistic switching criterion instead, such that an allowable
domain switching occurs according to probability Pds , given by
where Y ¼ ½eDT is the generalized strain matrix, which includes 8
the strain tensor and the electric displacement vector. hii is used >
< 1; DFin þ D 0
to denote volume averaging in grain i. The corresponding trans-
Pds ¼ DFin þ D (3.3.10)
field change in the ceramics is given as >
: exp ; DFin þ D 0
ckT

ðqÞ ð pÞ
D YS ¼ ci Dpq i Yi Yi (3.3.2)
In the equation, k is the Boltzmann constant, T is the absolute
temperature, and c is a constant connecting microscopic kT with
These changes in the transfield in turn lead to change in potential macroscopic energy variation.
energy, and after some manipulation, it can be shown that the It can be seen from Eq. (3.3.10) that the switching probability
energy change per unit volume due to the specific domain switch- depends on temperature and the extent of energy variation,
ing is given by increases with higher temperature, and decreases with higher energy
barrier. Such a probabilistic switching criterion makes it possible

S S S 1 ci S for the ferroelectric to overcome energy barriers and approach
DFin ¼ ci D Y i Q Y i Y þ D Y i global minimum in the energy profile of ferroelectric ceramics and
2
S can be easily implemented using Monte Carlo simulations.
c i X0 D Y i (3.3.3)
3.3.2 Results and Discussion. The micromechanics-based
where X0 ¼ ½r0 E0 T is the generalized stress matrix, which probabilistic switching model has been implemented into Monte
includes the stress tensor and the electric field vector. Q is the Carlo simulation using MATLAB, where N ¼ 100 different grain ori-
generalized stiffness coefficient matrix, which includes the elastic entations are randomly generated first to model a polycrystalline
stiffness coefficients, the piezoelectric coefficients, and the recip- ferroelectric ceramics. For ceramics near the MPB where tetragonal
rocal of the dielectric constants. Note that, in the equations, all the and rhombohedral phases coexist, it is more likely to switch to
averaged quantities are evaluated before the domain switching. rhombohedral variants, since there are more rhombohedral variants
Obviously, there exists energy dissipation associated with do- than tetragonal ones, resulting in a bias favoring rhombohedral
main switching. Here, it is assumed that the energy dissipation D phase over tetragonal ones. To avoid such a bias, it is first deter-
(or energy barrier for domain switching) per unit volume is pro- mined randomly whether the upcoming switching is in-phase or
portional to the accumulative change in volume fractions of var- interphase. The corresponding switching probability is then eval-
iants. For a particular grain i, the accumulative change in volume uated according to Eq. (3.3.10), and a random number 0 < R < 1 0
fractions can be evaluated as is also generated. The switching will be accepted if Pds > R and
rejected otherwise. This completes one simulation step. We define
K L ¼ 102 NK simulation steps as one Monte Carlo step, with
1X ð pÞ1 ð pÞ
jDci j ¼ ci ci 0 (3.3.4) N ¼ 100 and K taking 6, 8, and 14 for tetragonal, rhombohedral,
2 p¼1
and MPB PZT, respectively. Thus, on average, each variant in each

Fig. 39 The simulated hysteresis (left) and butterfly (right) loops of PZT ceramics. Reprinted
from Ref. [51] with the kind permission of Elsevier Science Ltd.
grain is visited 102 times during one Monte Carlo step. The simula- r < 1 describes the reorientation of polar axis of domains toward
tion continues until a stable domain configuration emerges, when poling direction during electric poling, and r > 1 describes the
all the switchings are rejected, and thus, no switching occurs during reorientation of polar axis away from poling direction during me-
one Monte Carlo step. It should be noted that a physical switching chanical depoling. Since ODF is usually measured from X-ray or
step is accomplished through many computational switching steps neutron diffraction, only non-180 deg domain switching is
in Monte Carlo simulation. To speed up the computation, a large reflected in the change of March coefficient, due to the inherent
initial ckT ¼ 103 J=m3 is chosen at beginning and then is reduced centrosymmetry of the diffraction technique [51].
to 95% of the previous value after every Monte Carlo step. This is The variation of March coefficient versus electric field is shown
similar to a physical annealing process. Also note that the simula- in Fig. 40, where in all PZT phases before poling, the March coeffi-
tion results of this model are not sensitive to the initial temperature. cient is close to unit. After applying the electric field, very little
change in March coefficient is observed in single-phase tetragonal
Electric Poling. We consider the electric poling process of
PZT, which is consistent with the simulated domain textures evolu-
three different PZT ceramics, including the tetragonal, the rhom-
tion using the optimization-based model. In comparison, the March
bohedral, and the morphotropic, in which all possible variants
coefficient for tetragonal phase at MPB decreases to about 0.2 after
exist in grains with equal volume fractions. A bipolar poling elec-
poling, suggesting extensive 90 deg switching that reorients polar-
tric field of 3 MV/m is applied to the ceramic with an increment
izations toward the poling direction, which contributes to large
of 0.1 MV/m. The obtained D–E hysteresis loops and butterfly
strain observed in PZT near the MPB during poling. In single-
loops are shown in Fig. 39, in which EC ¼ 1MV=m is used for all
phase rhombohedral PZT, the March coefficient decreases to about
the PZT ceramics. It can be seen that the difference in polarizabil-
0.63 after poling, suggesting modest extent of 71 deg and 109 deg
ity across MPB is evident. Tetragonal PZT has the largest coerciv-
domain switchings that are responsible for the observed modest
ity and is least polarizable, with remanent polarization calculated
strain. Also, the rhombohedral phase at MPB shows more extensive
as 0.21 C/m2. This is only 44% of the single crystalline spontane-
71 deg and 109 deg switchings, as seen from larger change in its
ous polarization and is close to 0.17 C/m2 measured in experiment
March coefficient. The differences in macroscopic behavior of PZT
[171]. The calculated switching strain is only 0.037%, much
ceramics across MPB during poling thus merely reflect the differ-
smaller than single crystalline value of 2.0% and comparable to
ences in their microscopic domain switching. To further validate
the measured value of 0.047% in tetragonal BaTiO3 ceramics,
this Monte-Carlo model, simulations were also done on a rhombo-
which have similar single crystalline transformation strain [172].
hedral, soft PLZT ceramics under bipolar electric loading, and the
Rhombohedral PZT shows improved polarizability, with remanent
results can fit well with the experiments, as shown in Fig. 41 [51].
polarization calculated as 0.31 C/m2, about 64.6% of single
crystalline spontaneous polarization, and agrees well with that of Mechanical loading. The behavior of mechanical depoling of
0.31 C/m2 measured in experiment [171]. Also consistent with the poled PZT ceramics is also investigated by using the Monte Carlo
experiments, the calculated switching strain in rhombohedral model. The starting point is a ceramic poled by an electric field of
ceramics is 0.14%, much larger than tetragonal ceramic. As 3 MV/m, and then the poling field is removed. The stable domain
expected, PZT at MPB has much larger remanent polarization configuration is obtained by Monte Carlo simulation as discussed
(0.36 C/m2) and switching strain (0.375%), both in excellent agree- previously. Then, a compressive stress up to 300 MPa is gradually
ment with experimental values (0.36 C/m2 and 0.37%) [171,172]. applied to the poled ceramic along the poling direction, and then
To understand the difference in polarizability of PZT ceramics it is gradually removed. During both loading and unloading, the
across MPB, we examine the texture evolution of PZT during increment of stress is 10 MPa. The simulated strain-stress curves
electric poling. Here, the orientation distribution function (ODF) and stress-depolarization curves for tetragonal, rhombohedral, and
PðhÞ of domains is employed, from which the March coefficient r MPB PZT ceramics are shown in Fig. 42. It can be seen that PZT
of tetragonal and rhombohedral phases can be fitted according to at MPB shows very large reduction in remanent polarization and
the following equation [164]: is almost completely depoled under the maximum compressive
stress of 300 MPa. This indicates that extensive non-180 deg do-
PðhÞ ¼ ðr 2 cos2 h þ r 1 sin2 hÞ3=2 (3.3.11) main switching occurs, which leads to large nonlinear strain as
high as 0.62% that is not recovered after stress unloading. In com-
As indicated by Tang et al. [51], the March coefficient r is particu- parison, tetragonal PZT only exhibits a polarization reduction of
larly convenient to describe the texture evolution in ferroelectric 0.08 C/m2 and rhombohedral PZT has a modest polarization
ceramics due to domain switching. For example, r ¼ 1 describes reduction of 0.24 C/m2, suggesting modest extent of 71 deg and
random orientation distribution of domains in an unpoled ceramic, 109 deg switchings that lead to modest nonlinear strain.

Fig. 40 Evolution of March coefficient of PZT ceramics in tetragonal (left) and rhombohedral (right) phases dur-
ing electric poling. Reprinted from Ref. [51] with the kind permission of Elsevier Science Ltd.
Fig. 41 Comparison between simulation and experiment to hysteresis (left) and butterfly (right) loops of soft
PLZT during electric poling. Reprinted from Ref. [51] with the kind permission of Elsevier Science Ltd.
The change of March coefficient during mechanical depoling is model, the Monte Carlo model also overestimates the coercive
also calculated and shown in Fig. 43 [51]. In all three PZT, the stress, and thus, a more accurate model addressing the softening
March coefficient increases beyond unit, suggesting that polariza- during domain switching is required in the future.
tions are reoriented away from the poling direction by the com-
pressive stress. As expected, the increase is largest for PZT at
MPB and smallest in single-phase tetragonal PZT, which agrees 4 Atomic-Level Electromechanically Coupled
well with the neutron diffraction data [164]. As indicated by Tang Computational Methods
et al. [51], if we set the spontaneous polarization in all the variants
to be zero, very little changes in March coefficient and strain are 4.1 First-Principles Method of Ferroelectrics. There has
observed in single-phase tetragonal PZT. These results suggest been rapid development in the atomistic modeling of materials in
that the strain compatibility plays a dominant role in domain the last decades, driven by the fast increase of the computational
switching in ferroelectric ceramics and polarization compatibility power and important progresses in the development of efficient
is less influential [51]. algorithms. Nowadays, it is possible to describe very accurately
To further compare the simulation with the experimental results, the properties of materials by using methods directly based on
we also conducted simulations on a soft PLZT ceramic under me- solving the fundamental equations of quantum mechanics. Even if
chanical loading and compared the simulated depolarization curve it may require some approximations for the study of complex sys-
and stress-strain curve with the experiments, as shown in Fig. 44 tems, these methods are free of empirically adjustable parameters.
[51]. It can be seen that, overall, the simulated results can fit well Therefore, they are referred to as “first-principles“or “ab initio”
with the experimental data, although they differ in some details methods. In recent years, first-principles computations have
near the transition regions. Similar to the optimization-based become ubiquitous in materials science, and it is hard to find a

Fig. 42 Depolarization (left) and stress–strain (right) curves of PZT ceramics during mechanical depoling.
Fig. 43 Evolution of March coefficient of PZT ceramics in tetragonal (left) and rhombohedral (right) phases dur-
ing mechanical depoling. Reprinted from Ref. [51] with the kind permission of Elsevier Science Ltd.
material of scientific interest that has not been studied increases but piezoelectric constant decreases as thickness reduces
computationally. [180]. And for BaTiO3 nanowires, the critical size is 1.2 nm,
Ferroelectric materials have been intensively studied from first- determined from first-principles [69], which agrees with the data
principles since the 1990s, when the modern theory of polariza- extrapolated from experiment [181]. Recent first-principles calcu-
tion was developed [173,174]. The accuracy of density functional lations show that some novel polarization patterns appear in ferro-
theory for characterizing the ground-state structures of ferroelec- electric nanodots and nanowires, which have never been found in
trics was first determined for BaTiO3 and PbTiO3 [66] and later ferroelectric bulk before [68,70,177,182], and all these patterns
extended to other ferroelectric materials [175]. First-principles are found to correspond to the topological point defects with dif-
methods have been remarkably successful in modeling many of ferent winding numbers [177]. Bousquet et al. [179] found an un-
the physical properties of ferroelectric perovskite materials, usual improper ferroelectricity in short-period PbTiO3/SrTiO3
including the ground state structure, the spontaneous polarization superlattices due to the octahedra rotation in the interfaces, which
and the related dynamical effective charges, the dielectric and pie- stimulates the research for the interface engineering in superlatti-
zoelectric response, etc. ces and hybrid improper ferroelectricity in multiferroics [183].
As computational power and algorithmic efficiency have In this section, we first briefly review the modern theory of
improved, it has become possible to study increasingly complex polarization and then review its applications in topological polar-
systems, such as surfaces [176], ultrathin films [67], nanowires ization in nanostructures, flexoelectricity, and size-dependence
[69,177], and superlattices [178,179]. As the size reduces, the fer- properties of films.
roelectric nanostructures show some novel properties, which the
bulk materials do not have. The first-principles calculations show 4.1.1 Polarization Calculation From First-Principles. The
that the ferroelectricity disappears below 2.4 nm for BaTiO3 films macroscopic polarization is a fundamental concept in the theory
at short circuit boundary conditions [67], and its elastic stiffness of electrostatics, particularly in describing the response of systems

Fig. 44 Comparison between simulation and experiment of depolarization (left) and stress-strain (right) curves of soft PLZT
during mechanical depoling. Reprinted from Ref. [51] with the kind permission of Elsevier Science Ltd.
to applied electric fields. The presence of a spontaneous (and changes in polarization are rigorously defined and can be calcu-
switchable) macroscopic polarization is the defining property of a lated quantum mechanically using electronic structure methods.
ferroelectric material, and the macroscopic polarization is thus For example, in Fig. 45 the polarization (dipole moment per
central to the whole physics of ferroelectrics. Despite its primary unit length for one-dimensional case) is 1=2 and 1=2 (in unit jej)
role in electrostatics theories and its overwhelming importance, if choosing different unit cells for centrosymmetry (nonpolar)
the macroscopic polarization has long evaded microscopic under- case (top panel). For the distortion (polar) case (bottom panel, cor-
standing, not only at the first-principles level, but even at the level responding to the ferroelectric case) with cation displacement of
of microscopic models. A typical definition of polarization is that D, the polarizations are 1=2 þ D=a and 1=2 þ D=a for different
the macroscopic polarization of a solid is the electric dipole unit cells. It can be seen that the change in polarization between
moment per unit volume. However, such a quantity is neither polar and nonpolar cases is the same (D=2), which is independent
measurable nor model-independent: the dipole of a periodic of the choice of unit cells. Actually, the changes in polarization
charge distribution is in fact ill-defined [184]. For example, it can also correspond to experimentally measurable observables. For
be seen from the top panel of Fig. 45, without any calculation, example, the spontaneous polarization is measured by the half of
that two equal unit cells show completely opposite polarizations the difference between up- and down-polarized states.
based on this definition. The cation and anion are considered as point charges in the pre-
This difficulty is solved thanks to the so-called modern theory vious simple ionic model. However, in a real solid, there is more
of polarization, which was developed in the 1990s [173,174,185]. physics to take care of, such as the electronic contribution to the
It was realized that, even at the theoretical level, polarization dif- polarization. In real materials, the charge (nuclear and electronic)
ferences are conceptually more fundamental than the “absolute” can be decomposed into localized contributions whose dipoles
polarization. This change led to the development of a new theoret- determine polarization. The polarization can be obtained by the
ical understanding, involving formal quantities, such as Berry sum over localized charges multiplied by their positions. This is
phases and Wannier functions, that have come to be known as the straightforward for the ions, which can be still treated as point
“modern theory of polarization”. According to this theory, the charges. For the electrons, this procedure also works through
Fig. 45 Calculation of the polarization for different unit cells in one-dimensional

chain of alternating anions and cations. In the top panel, anions and cations are
spaced a distance a/2 apart (a is the lattice constant). In the bottom panel, the cati-
ons are displaced by a distance D. The dashed rectangles indicate unit cells that
are used for calculation of the polarization (see text).

Wannier representation. The Wannier functions are localized @Pa
Zj;ab ¼X (4.1.7)
functions wn ðrÞ, labeled by band n and unit cell R, @uj;b
ð ð
X ikR X
wn ðr RÞ ¼ dke w nk ðrÞ ¼ dkeikðrRÞ unk ðrÞ Born effective charge is a tensor. For ferroelectric materials, it is
ð2pÞ3 ð2pÞ3 quite different from the formal charge. For example, for cubic
(4.1.1) BaTiO3, the formal charges are Ba þ2, Ti þ4, and O –2, but the
Born effective charges are Ba þ 2.75, Ti þ 7.16, O? –2.11, and
where wnk ðrÞ ¼ eikr unk ðrÞ are the Bloch functions, in terms of Ok –5.69 [189].
the cell-periodic part unk ðrÞ, X is the unit cell volume, and the in- Equation (4.1.7) suggests that the Born effective charges can be
tegral is over the Brillouin zone. used to calculate the local polarization in each unit cell,
Since the Wannier functions are localized, we can work out the
1X
average position of the electrons in the Wannier function. This av- Pa ¼ Z dj;b (4.1.8)
erage position of the Wannier function is called the Wannier cen- X j;b j;ab
ter, rn . The Wannier center associated with band n is defined as
ð where dj;b is the displacement of atom j in the unit cell along the
rn ¼ wn ðrÞrwn ðrÞdr (4.1.2) b direction relative to the reference structure (high-symmetry
structure).
It is useful to write this expression in terms of the periodic cell It is useful to use Born effective charge to calculate local polar-
functions using the momentum representation of the position op- ization in single unit cells in which the polarization is different in
erator r ¼ ið@=@kÞ, each unit cell. In this case, it is difficult to use Berry phase method
to calculate local polarization in each unit cell. This method is
ð
X @unk used in Sec. 4.1.3 for flexoelectric response calculation.
rn ¼ i dke ikR
u (4.1.3)
ð2pÞ3
nk @k The capability of computing polarization is now available in
almost all commonly used software packages for bulk electronic-
With this concept of Wannier center, the electronic contribution structure calculations. These methods allow for the computation
to polarization Pel is of numerous quantities of interest, including spontaneous polar-
ization, Born effective charges, linear piezoelectric coefficients,
1Xocc etc. The modern theory of polarization provides a robust and
Pel ¼ ðqn rn ÞWFs (4.1.4) powerful foundation for modern computational studies of the
X n polarization-related properties of ferroelectric materials.
where the sum over is for all the electronic charges centered at the 4.1.2 Topological Polarization in Ferroelectric Nanowires.
Wannier centers of each occupied Wannier function. As the size of ferroelectric nanostructures continues to decrease, it
Therefore, the polarization is was initially assumed that the polarization would vanish because
ferroelectricity is a collective phenomenon. However, seminal
! works by Fu and Bellaiche [182] and Naumov et al. [68] showed
1 X ions
Xocc
WFs that the polarization may first form a ferroelectric vortex before
P ¼ Pion þ Pel ¼ ðqi ri Þ þ ðqn rn Þ
X i n disappearing. This has attracted much theoretical [69,70,190,191]
occ ð and experimental work [181,192] to establish critical diameters
1X ions 2ie X ikR
@unk

¼ ðqi ri Þ dke unk for ferroelectricity and understand the characteristics of ferroelec-
X i ð2pÞ3 n @k tric vortices. First-principles method plays an important role in
(4.1.5) determining critical size of ferroelectrics and exploring novel
properties in ultrathin films and nanowires. Recent study shows
that some novel topological polarization patterns appear in thin
Using the Wannier function picture, the polarization can be nanowires which do not exist in the bulk [177].
written as the sum over the charges times their positions: the con- The two-dimensional (2D) topology of the polarization field in
tribution from the positively charged ion cores and the contribu- ferroelectric nanostructures can be mapped onto the topology of
tion from the negatively charged valence electrons and the ðPx ; Py Þ for wires if the field remains constant along z, the axial
“position” of each valence electron is its Wannier center. direction of the wire. Topological defects in such a 2D field [193]
The polarization of Eq. (4.1.5) is referred to as the “formal are reviewed here. The winding number is defined as n ¼ /=2p,
polarization” to distinguish it from the “effective polarization,” / being the total angle the 2D field vector rotates when going
which is defined as the difference between two states, around a closed circuit, and n must be an integer for a continuous
field. Some novel polarization textures with n ¼ 0, 1, 1, and 3
Peff ¼ Pð1Þ Pð0Þ (4.1.6) in ultrathin BaTiO3 nanowires are found from first-principles
calculations.
where Pð1Þ and Pð0Þ are the formal polarization for state 1 and state The Ti displacement from the center of its coordination octahe-
0. Usually, the state 1 is the low-symmetry state and state 0 is the dron was taken to define PðrÞ from the relaxed structures as a
high-symmetry state (reference state). Effective polarization is measure of the local dipole. For Ti atoms on surfaces or edges, the
well-defined, and it corresponds to the experimentally measurable off-center displacement was determined by the remaining coordi-
polarization. nating O atoms. Four kinds of nanowires with different surface
Actually, the integrals in Eq. (4.1.5) are the Berry phase devel- terminations have been considered, as shown in Fig. 46. Figure 47
oped by the wavefunction unk ðrÞ as it evolves along the path k. (left column) shows the polarization patterns of the wire types of
As a result, the formalism for calculating polarization using this Fig. 46, with qualitative sketches of the corresponding topological
method is often called the Berry phase theory of polarization. defects for a 2D vector field (right column). It shows that, even
Interested readers are referred to the original papers [173–175] for the small size wires, the field textures appear very clearly. All
and reviews [186–188] for details. textures preserve inversion symmetry at the center of the wire if it

Next, we introduce the Born effective charge Zj;ab , which is is not previously broken by the surface termination.
defined as the change in polarization along a direction (Pa ) due to The inward and outward radial patterns in the first two cases in
the displacement of atom j along b direction (uj;b ), Fig. 47 correspond to the same kind of topological defect of

Fig. 46 The largest balls are for Ba ions, the intermediate ones
represent Ti ions, and the smallest ones are for O. (a) is TiO2 ter-
minated; (b) is BaO terminated, except the edges, where the
corner Ba atoms have been removed; (c) is stoichiometric, with
two BaO surfaces and two TiO2 ones. (d) is BaO terminated.
Reprinted from Ref. [177] with the kind permission of American
Physical Society.
winding number 1. They are thus homotopical with each other but
also with the vortex structures shown in Naumov et al. [68] and
Pilania et al. [70]. Figures 47(c) and 47(d) correspond to different
winding numbers as yet not observed or proposed in this field,
namely 1 and 3. In Fig. 47(c), the defect is displaced from the
center. Figure 47(d) shows the n ¼ 3 case of the BaO-terminated
wire. Figure 47(e) is for the same BaO termination as 47(d) but
smaller thickness. Although a different discretization, the winding
around the outer circuit clearly shows the same n ¼ 3: a central
defect with n ¼ þ1 plus four other defects of n ¼ 1, as illustrated
in Fig. 47(j).
These textures are induced by surface and edge effects. The
phenomenology described can be understood as originated by the
following three tendencies: (i) Ti-terminated surfaces and edges
induce inwards Ti off-centering patterns, (ii) Ba surface termina-
tion induces outward Ti off-centering patterns, and (iii) Ba edge
termination induces an inward tendency. The competition of (ii)
and (iii) in the Ba-terminated wires gives rise to the rich n ¼ 3
pattern. By removing the Ba edges (Fig. 47(b)), the competition
disappears and the fully outward pattern is established. Similarly,
when removing the Ba edge in the stoichiometric case, the polar- Fig. 47 Left column (a)–(d): Ti off-center displacements for the
four respective nanowires in Fig. 46. (e) is similar to (d) but for a
ization field becomes topologically homogeneous (n ¼ 0), with the thinner wire. The right column sketches the corresponding 2D
field lines crossing the wire diagonally pointing left and down field lines around topological point defects of winding numbers
(not shown here). Special chemical tendencies at edges and surfa- n 5 1, 1, 21, and 23, respectively, (j) showing the n 5 23
ces have been previously reported [181]. decomposition into a central n 5 11 and four n 5 21 defects.
The respective largest arrows correspond to (a) 21 pm, (b) 35
pm, (c) 23 pm, (d) 6.8 pm, and (e) 6.6 pm off-centering displace-
ment. Reprinted from Ref. [177] with the kind permission of
4.1.3 Strain Gradient Effects in Ferroelectrics. Inhomogene- American Physical Society.
ous strains can induce dielectric polarization in all insulators due
to flexoelectricity (FxE), which linearly couples polarization to caused a recent explosion of experimental and theoretical interest
strain gradient [194]. FxE is quite different from piezoelectricity, in FxE [195–207].
which arises only in noncentrosymmetric materials. Although it is FxE refers to the linear response of electric polarization to an
a universal property of all dielectrics, FxE is nevertheless a rela- applied strain gradient,
tively small effect compared with piezoelectricity and it is nor-
mally negligible on conventional length scales. However, it may @ejk
Pi ¼ lijkl jkl ¼ lijkl (4.1.9)
become very strong at the nanoscale, where huge strain gradients @xl
can significantly affect the functional properties of dielectric
nanostructures. The possibility of large FxE effects in ferroelec- where Pi is the FxE-induced polarization, jkl ¼ @ejk =@xl is the
trics at the nanoscale with application to functional devices has strain gradient, and lijkl are the flexoelectric coefficients (FECs).

Fig. 48 Supercell and strain profile used in the calculations. For BaTiO3, the unit
cell contains only one formula unit (a), while for SrTiO3, due to its antiferrodistor-
tive transition at low temperature, the in-plane size of the unit cell is doubled (b).
(c) and (d) supercell in xz plan, (e) cosine gradient strain in the supercell, (f) atom
displacement in the supercell. Reprinted from Ref. [200] with the kind permission
of Institute of Physics (IOP) Publishing.
In materials with simple cubic symmetries, such as cubic perov- the fixed ones but also that the difference between the two values
skites, there are only three independent tensor coefficients: longi- decreases with increasing the height of supercell, suggesting that
tudinal l1111 and transverse l1122 and l1212 . such a difference was the artifact of an insufficiently big supercell.
In order to keep the periodic boundary conditions required by The strain gradients of the fixed and relaxed atoms thus converge,
the first-principles calculations, a special supercell is chosen in and an extrapolation of the FECs allows estimating l3333
which the strain gradient is periodic. Specifically, the strain profile 0.37 6 0.03 nC/m for BaTiO3 and l3333 1.38 6 0.65 nC/m
along the z direction is chosen in the cosine form, for SrTiO3.
z The FECs obtained here are all in the order of nC/m, which is
eðzÞ ¼ emax cos 2p (4.1.10) consistent with the theoretical estimation for a simple dielectric
h [194] and also with recent measurements in SrTiO3 single crystals
and ceramics [198]. These values are also on the same order of
where h is the height of the supercell and emax is the maximum magnitude as the effective Hamiltonian calculations of Maran-
strain in the supercell. The supercell is shown in Fig. 48. The posi- ganti and Sharma [208]. Our calculated FEC of BaTiO3 agrees in
tion of h/4 of the supercell is chosen to calculate FECs where the magnitude with the calculations of Maranganti and Sharma [208]
strain gradient is maximum, and the piezoelectric polarization in and are comparable also to those measured for SrTiO3. However,
this unit cell is zero, because the average strain is zero by con- the longitudinal FEC of BaTiO3 has not yet been directly
struction. Thus, choosing the unit cell at h/4 not only provides the
maximum FxE but also eliminates the piezoelectric contributions.
Given a strain profile (Eq. (4.1.10)), the displacement of each
atom in the supercell is given by
ðz
emax h z
dðzÞ ¼ eðnÞdn ¼ sin 2p (4.1.11)
0 2p h
After applying these displacements, the system is relaxed, except

for A-site atoms (Ba or Sr), which are fixed in order to impose the
strain gradient in the supercell. A-site atoms are fixed because
they are the ones that participate least in the polarization in
BaTiO3 and SrTiO3. All the other atoms are relaxed to their equi-
librium positions. The local polarization in each unit cell is calcu-
lated by using Born effective charges.
After the relaxation, the gradient of the relaxed atoms is found
to be smaller than that of the fixed A-site atoms (Ba or Sr) for
small supercells. This has a considerable impact on the calculated
value of the FEC: since it is calculated dividing the polarization
over the gradient, a smaller gradient results in a bigger apparent Fig. 49 Longitudinal FEC l3333 of rhombohedral BaTiO3 for dif-
ferent supercell sizes (N 5 6, 10, 14) under various strain gra-
FEC. The discrepancy between the gradient values of the fixed
dients (emax 5 0.5%, 1.0%, 1.5%, 2.0%). In each panel, the line
and relaxed atoms is largely an artifact caused by the small size with square symbols (up) is for the applied strain, the line with
of the supercell, and it disappears as supercell size increases. dot symbols (down) is for the “relaxed strain” (see text). The two
Figure 49 shows the FEC for rhombohedral BaTiO3 with different values converge as the supercell size is increased. Reprinted
numbers of unit cells in the supercell. It can be seen that the calcu- from Ref. [200] with the kind permission of Institute of Physics
lated FEC is larger when using the relaxed atoms than when using (IOP) Publishing.

measured, so direct comparison with experiment is impossible for
this material. Nevertheless, the transverse FEC of BaTiO3 has
been measured and is reported to be of the order of several lC/m
[209], which is several orders of magnitude bigger than the theo-
retical estimate for the longitudinal one. Further discussions for
this difference can be found in Ref. [200].
4.1.4 Size Effect of Ferroelectric Thin Films. When the size
of ferroelectric material reduces, the anomaly of many functional
properties of ferroelectrics comes up, such as the collapsed magni-
tude of dielectric constant [210], the increased remanent polariza-
tion [211], and appearing novel polarization patterns [68,177].
Recently, it was found that the piezoelectric and elastic properties
are also size-dependent in ferroelectric BaTiO3 thin films from
first-principles calculations [180].
The elastic constants are focused on for the mechanical proper-
ties of ferroelectric thin films. The theoretical elastic constants are
calculated from the energy variation by imposing small strains to
the equilibrium unit cell. The elastic energy under strain is given
by
VX 6 X 6
DE ¼ Cij ei ej (4.1.12) Fig. 50 Size effect of elastic stiffness C11 for BaTiO3 films.
2 i¼1 j¼1 Reprinted from Ref. [180] with the kind permission of Taylor &
Francis.
where V is the volume of the initial unit cell undistorted, DE is the
energy increment between the unit cell energy with and without
the strain e ¼ ðe1 ; e2 ; e3 ; e4 ; e5 ; e6 Þ, and Cij is the elastic stiffness
tensor.
The stoichiometric BaTiO3 film is chosen in the study, with one
face terminated with BaO and the other face terminated with
TiO2. The film thickness is specified by an integer n that corre-
sponds to the number of unit cells along the z direction (out of
plane). Periodic replicas of the films are separated by 16 Å of vac-
uum. For the tetragonal BaTiO3, Cij can be conveniently obtained
by applying different strains ei on an initial unit cell and calculat-
ing the energy difference. The density functional theory calcula-
tion shows that the in-plane lattice constants are size-dependent,
which are 0.3945 nm (n ¼ 2), 0.3968 nm (n ¼ 3), 0.3981 nm
(n ¼ 4), and 0.3989 nm (n ¼ 5) for different films. The in-plane lat-
tice constants increase with increments of film thickness, and it is
very close to that of bulk (0.3999 nm) when film has thickness of
five unit cells.
BaTiO3 is brittle material, and its tensile stiffness is different
from compressive stiffness. Therefore, the tensile and compres-
sive stiffnesses (C11) are fitted separately through Eq. (4.1.12) for Fig. 51 The piezoelectric properties of BaTiO3 bulk and thin
films of n ¼ 2–5, which are shown in Fig. 50. It shows that the ten- films, e33 for bulk and e11 for films of n 5 2 and 4. Reprinted
sile stiffness is smaller than compressive stiffness for a film, but from Ref. [180] with the kind permission of Taylor & Francis.
they are both size-dependent. The elastic stiffness C11 of BaTiO3
film reduces as film thickness increases. It should be noted that environment and thus are effective in characterizing the surface
the compressive stiffness is very close to that of BaTiO3 bulk effect and the behaviors of nanoscale ferroelectrics.
when the thickness reaches n ¼ 5 (i.e., 2.0 nm). Molecular dynamics (MD) method is a widely used atomic-
Piezoelectric stress coefficients are calculated for BaTiO3 bulk level simulation method that has been playing an important role in
and films of n ¼ 2 and 4, as shown in Fig. 51. For the bulk, e33 is solid-state and material science. Since the motion of the particles
17.9 C/m2, which agrees with the previous result of 18.6 C/m2 in are governed by the Newtonian equations, this method is easy to
Ref. [212]. The piezoelectric properties of BaTiO3 films also implement and more efficient than the first-principles method.
show size-dependence: e11 decreases with reduction of film Given the expressions characterizing interatomic interactions, the
thickness. variations of particle positions can be easily determined. A key
point of this method is to find an appropriate potential model to
4.2 Molecular Dynamics Method of Ferroelectrics. With reflect the macroscopic material properties. For a ferroelectric,
rapid progress of nanoscale manufacturing technology, low- one defining characteristic is that the material should have a spon-
dimensional ferroelectrics, such as ultrathin films, superlattices, taneous polarization that can be switched under an electromechan-
nanowires, nanoparticles, and nanorings, have attracted increasing ical load [3]. From the microscopic point of view, ferroelectricity
attention [67,68,181,213–216], due to their intriguing properties stems from the delicate balance between the long-range Coulomb
and prospective applications. In these nanostructures, the role of and short-range repulsive interactions [66]. Such complicated
surface atoms becomes outstanding, and the induced surface behavior is not easy to characterize using the traditional intera-
effect, either in the physical or chemical aspects, usually cannot tomic potential model. To overcome this difficulty, the shell
be fully understood using continuum modeling approach. The model [55–57,59,217,218] was introduced in ferroelectric perov-
atomic-level simulation methods are capable of reflecting the skites, which decomposes each atom into a core and a shell, mod-
reconstruction of surface atoms as well as their different atomic eling an atomic nucleus and electron shell, respectively. This

model has been proven to be able to depict the spontaneous polar- anisotropic spring potential is usually adopted to model the core-
ization of ferroelectric perovskites and also their temperature- shell interaction of the anion O2– [59]. Take BaTiO3 as an exam-
induced phase transition behavior [57,60,219]. Although the shell ple, in the direction parallel
with the O–Ti bond, the potential is
nonlinear
model MD is not as accurate as the first-principles method, its given by Uspring rij ¼ K2 rij2 =2 þ K4 rij4 =24, while in the other
computational efficiency is enhanced considerably and the finite two directions perpendicular to the O–Ti bond, the potential is
temperature behavior is easier to cope with based on the shell written as
model MD. As such, the shell model MD has been widely used in
linear
studying the unique property of nanoscale ferroelectrics Uspring rij ¼ K2 rij2 =2 (4.2.3)
[58,61–63,152,219,220] and the phase transition behavior
[221,222]. where the superscript means a linear spring interaction. The long-
For ferroelectric ultrathin films, Tinte and Stachiotti [220] ini- range force represents the electrostatic interaction among all cores
tially introduced the shell model MD in studying their finite tem- and shells, except between the core and shell of the same atom.
perature behaviors and verified the applicability of shell model This interaction is characterized by the Coulomb potential, i.e.,
potential by comparing the calculated polarizations with first-
principles results. Sepliarsky et al. [219] further explored the 1 qi qj
effects of surface reconstruction in PbTiO3 ultrathin films, which Ulongrange rij ¼ (4.2.4)
4pj0 rij
well-reproduced the antiferrodistortive surface reconstruction
observed in experiment. Using similar interatomic potentials,
Sang et al. [223] and Zhang et al. [64] carried out a systematic where qi and qj are the charges of particle i and j, respectively.
study on size-dependent ferroelectric properties (e.g., polarization j0 ¼ 8.85 1012 C2N1m2 is permittivity of the free space.
distribution, Curie temperature, and hysteresis loop) in BaTiO3 Different from the long-range interactions, the short-range interac-
ultrathin films. For ferroelectric nanowires, Zhang et al. [61,62] tion only exists between shells of different atoms. Formally,
investigated the size and strain effects of polarization distributions short-range interactions represent the van der Waals attraction and
and hysteresis loops, revealing the existences of two critical diam- the Pauli repulsion due to the overlap of the electron densities of
eters and an unconventional stepwise hysteresis loop. This study two ions [59]. But in practice, the short-range interactions usually
also illustrated the possible equivalence of size and strain effects represent all the atomic interactions, excluding the electrostatic
on the ferroelectric performances of BaTiO3 nanowire. An exten- interactions. Therefore, the function form of short-range interac-
sion of this study was on the electromechanically coupled behav- tion varies for different material, including the Buckingham, Ryd-
iors of BaTiO3 nanowires [63], which demonstrated three-stage berg, Born–Mayer, and Morse potentials. For the K(Nb, Ta)O3
stress-strain and stress-polarization relations and size- and (Ba, Sr)TiO3 systems, the short-range interaction is usually
dependences of piezoelectric coefficient and elastic modulus. described by the Buckingham potential, i.e.,
Besides the geometries of ultrathin film and nanowires, Sepliarsky
rij C
and co-workers [58,65,221] also introduced shell model MD in UBuck rij ¼ A exp 6 (4.2.5)
studying the long-ranged interactions, phase transitions, and q rij
dynamic behaviors of KNbO3/KTaO3 superlattices and size-
dependent polarization distributions of PbTiO3 nanoparticles. where A, B, and C are empirical coefficients. This interaction only
We will introduce the shell model and the interatomic exists between the shell of O2– and the other ion-shells (i.e., O(shel-
potentials that have been developed for ferroelectric perovskites l)–A(shell), O(shell)–Bshell, and O(shell)–O(shell)). For PbTiO3, the
in Sec. 4.2.1. Some computational skills for calculation of the short-range interaction between O(shell) and O(shell) is still
long-range force will be presented in Sec. 4.2.2. In Sec. 4.2.3, we described by the Buckingham potential, while all the other short-
will review some representative atomic simulations using shell range interactions, including O(shell)–A(shell), O(shell)–Bshell, and
model MD, with the emphasis on the study of ferroelectric A(shell)–B(shell), are described by the Rydberg potential, which
nanostructures. takes the following form:

4.2.1 Shell Model and Interatomic Potentials. The shell rij
UBuck rij ¼ A þ Crij exp (4.2.6)
model is a widely used method in atomistic simulations of oxides, q
which phenomenologically describes the deformation of the elec-
tronic structure of an ion due to the interactions with other atoms
In the shell model potentials, the parameters required for perov-
[57,59]. In this model, each atom is described by two charged par-
skite ABO3 can be summarized as follows: X, Y, K2, and K4 for
ticles: a massive ion-core (with a charge X) and a massless ion-
cores and shells of all atoms and A, C, and q for all the short-
shell (with a charge Y) that are linked by a spring. Electronic
range potentials. Note that a charge neutrality condition should be
polarization effects are captured by the dipolar moment produced
satisfied, requiring that the sum of core and shell charges in a unit
by the relative core-shell displacements. In the ABO3 perovskite
cell must be zero. Then it can be found that there are totally 21 in-
structure, there are three types of particle interactions, i.e.,
dependent parameters for K(Nb,Ta)O3 and (Ba,Sr)TiO3 systems
and 23 independent parameters for PbTiO3 [59,60]. These param-
Uinner ¼ Uspring þ Ulongrange þ Ushortrange (4.2.1)
eters can be fitted to reproduce the stable crystal structure, the fer-
roelectric soft mode behavior, the elastic properties, and phonon
where Uspring, Ulong-range, and Ushort-range are the spring, long- frequencies, as derived from first-principles calculations. The
range, and short-range potentials, respectively. The spring poten- groups of Sepliarsky, Tinte, Stachiotti, and Migoni have made im-
tial describes the interaction between the core and shell of the portant contributions in generating effective potential parameters
same atom, which is commonly assumed to be isotropic, as given of several ABO3 ferroelectrics [57,217,224], and these parameters
by have been concluded in Ref. [59]. Recently, Shimada et al. [60]
nonlinear
developed a more delicate potential model for PbTiO3 and applied
Uspring rij ¼ K2 rij2 =2 þ K4 rij4 =24 (4.2.2) it in studying domain walls.
Given the shell model parameters, the classical integration
where K2 and K4 denote the second-order and fourth-order spring algorithm can be adopted to calculate the variations of atom posi-
nonlinear tions with times, such as the Verlet method and leap frog method.
stiffness and the superscript of Uspring means a nonlinear spring
interaction. However, to describe the hybridization effects on the The Verlet method estimates the atom position and velocity using
polarizability, which is important for ferroelectric behavior, an the following equation:

f ðtÞ 2 an atomic system with N particles, the system Coulomb potential
r ðt þ DtÞ 2r ðtÞ r ðt DtÞ þ Dt ; vðtÞ
m can be simplified as
r ðt þ DtÞ r ðt DtÞ
(4.2.7) 1 X N
2Dt UCoul ðRc Þ
8pj0 i¼1

where m is the atom mass, v the velocity, f the total force applied X qi qj erfc arij
qi qj erfc arij
at the atom, t the time coordinate, and Dt the time increment. The lim
computational accuracy based on the Verlet method is on the j6¼i
rij rij !Rc rij
order of Dt4 for the atom position and Dt2 for the atom velocity. ð ij c Þ
r <R
The leap frog method adopts a different algorithm, and the inte- N
gration expressions of kinematic equations are written as 1 erfcðaRc Þ a X
þ 1=2 q2i (4.2.9)
4pj0 2Rc p i¼1
r ðt þ DtÞ r ðtÞ þ Dtvðt þ Dt=2Þ; vðt þ Dt=2Þ
f ðtÞ where Rc is the cutoff length and a is a coefficient tuning the
vðt Dt=2Þ þ Dt (4.2.8) attenuation speed. erfc is the complementary errorpfunction (i.e.,
m Ð ar 2 ffiffiffi
erfc(ar) ¼ 1 erf(ar)), and erf ðar Þ ¼ 2 0 er dr= p.
where the velocity at t Dt/2 is obtained from Secondly, the expression of the approximate Coulomb potential
vðt Dt=2Þ ½r ðtÞ r ðt DtÞ=Dt. It can be found that the ve- (Eq. (4.2.9)) contains the error function, which cannot be inte-
locity and position are defined at different times in the leap frog grated explicitly. Too much time will be cost if the numerical inte-
method. Thus, it is not appropriate to calculate the total energy of gration technique is used to cope with the error function in each
the atomic system based on this method, since the obtained kinetic time when calculating the Coulomb force. To overcome this prob-
energy and potential energy are not for the same time. lem, an interpolation method can be adopted to evaluate the error
After calculating the atom positions, the stable atomic configu- function [88]. Detailed numerical calculations show that the com-
ration and the structural parameters (e.g., lattice parameter and the putational speed of the optimized direct summation method is
angle between different bonds) can be determined. An important about an order higher than the Ewald method [227,228]. In sub-
physical quantity of ferroelectrics is the polarization, which is section 4.2.3, we will discuss the applications of the shell model
usually calculated in a single crystallographic unit cell. Taking MD.
BaTiO3 as an example, there are two typical cells in such an
4.2.3 Results and Discussion. The shell model MD introduced
atomic system, as shown in Fig. 52 (i.e., the Ba corner cell and the
above has been used to simulate the finite temperature behavior of
Ti corner cell). For the Ba corner cell, the polarization component
ferroelectric bulk materials [57,60,218], ultrathin films
in thePaP direction for unit cell m is defined as
2 [64,219,220,223], superlattices [58,221], nanowires [61,62], and
Pam ¼ i v¼1 qivm rivam =Vm , where Vm is the cell volume and nanoparticles [65]. Note that the applications of shell model MD
qivm and rivam are the charge and coordinate along a direction,
in bulk ferroelectrics have been detailed and reviewed by Sepliar-
respectively. For a given particle i, v ¼ 1 and v ¼ 2 correspond to
sky et al. [59], including the phase transition sequences, the char-
the ion core and ion shell, respectively. The polarization for the Ti
acteristics of phonon frequencies, etc. Thus, this section will
corner cell can be derived similarly.
mainly review the studies on ferroelectric nanostructures, includ-
ing ultrathin films and nanowires.
4.2.2 Computational Skills of Long-Range Force. Since the Ferroelectric ultra-thin films. It is known that the emergence of
Coulomb potential is a long-range interaction attenuated by r1, free surfaces induces distinct electromechanical behaviors for fer-
the cutoff length should be large enough, so as to calculate accu- roelectric ultrathin films. As the surface proportion changes tre-
rately the Coulomb force, which costs a large amount of computa- mendously when the film thickness decreases into nanoscale, the
tional time. This section introduces two skills that can enhance ferroelectric behaviors become size-dependent. Tinte and Sta-
obviously the computational efficiency of Coulomb force. chiotti [220] initially introduced the shell model MD in studying
Firstly, the direct summation method proposed by Wolf et al. the finite temperature behaviors of BaTiO3 ultrathin films. The
[225] is introduced to compute the Coulomb force. This method potential parameters obtained for bulk BaTiO3 were carefully tes-
reduces cutoff length to a relative small value while maintains the tified for ultrathin films with both the BaO and TiO2 terminations.
accuracy of calculation. Specifically, this method evaluates the The calculated atomic positions and ferroelectric distortions at
Coulomb force by the classic expression (Eq. (4.2.4)) for the parti- different layers were compared with the first-principles results
cle with distance smaller than the cutoff length, while it uses a [229], which demonstrated that a very good description of surface
neutralization projection method to calculate approximately the properties is obtained for the TiO2 surface based on the shell
Coulomb force for the particle with distance larger than the cutoff model, while the description is not so accurate for the BaO sur-
length. A rigorous proof for convergence of the Wolf method was face. Nevertheless, the relaxation patterns obtained with the shell
presented by Angoshtari and Yavari [226] such that, for an arbi- model for both films were in full qualitative agreement with the
trary lattice of unit cells, the lattice sum obtained via Wolf method first-principles ones.
converges to the one obtained via Ewald method [227,228]. For After verifying the applicability of shell model potential in
ultrathin BaTiO3 films, the polarization distributions and their size
dependences were investigated [152,220,223]. The TiO2-termi-
nated films with different thicknesses were studied, and the size-
dependent polarization distributions of the out-of-plane compo-
nent are demonstrated in Fig. 53. The temperature in this set of
simulations was kept at 10 K in order to estimate the ferroelectric
behavior near 0 K. The results show that, for a thicker film (e.g.,
4.4 nm thick), the polarization near the surface is smaller than that
of inner cells because of the surface atomic relaxation. A 180 deg
strip domain pattern forms in this case, as shown in Figs. 53(b)
and 54(c). However, when the film thickness is equal to or thinner
than 2.8 nm, the polarization in a z-direction chain varies monot-
Fig. 52 Two different representative unit cells of BaTiO3 onically, as shown in Fig. 53(a) and there is no domain formation

Fig. 53 (a) The cell-by-cell out-of-plane polarization profile of a z-direction chain
in film with different thicknesses. (b) Schematic side views of the out-of-plane
polarization pattern for film in which h 3.2 nm. (c) Schematic side views of the
out-of-plane polarization pattern for film in which h £ 2.8 nm. Reprinted from Ref.
[152] with the kind permission of Elsevier Science Ltd.
in the film (Fig. 53(c)). The critical thickness is therefore 2.8 nm, dependent on film thickness. Figure 56(a) shows that the average
which is very close to the results (2.5 nm) derived from experi- polarization decreases as the temperature increases, because the
ments [230] and the first-principles calculations (2.4 nm) [67]. atoms become more active at higher temperature. Figure 56(b)
The electric-field-driven domain evolution and the size- shows an almost linear relation between the Curie temperatures
dependent hysteresis loops were also obtained for BaTiO3 ultra- and the reciprocal of film thicknesses, as indicated by MD simula-
thin film. The domain evolutions in the poling and depoling proc- tions, Landau–Ginzburg theory, and experiments.
esses are illustrated in Figs. 54(c)–54(k) for the 4.4-nm-thick film, The in-plane strain, induced by the mismatch between the sub-
and the variations of polarization are plotted in Figs. 54(a) and strate and the film or defects formed during film deposition, is an
54(b). In the spontaneous state, an 180 deg strip domain, consist- important factor that obviously influences the ferroelectric proper-
ing of periodically alternating positive and negative domains with ties of a thin film. In the MD simulations, the strain loading is usu-
2 to 3 unit cells in width, is clearly shown in the top view of the ally applied by controlling the x-y plane boundary displacements
film (Fig. 54(c)). Such strip domain structure is in reasonable of the films [64,152,220]. Meanwhile, the computational system is
agreement with the experimental observations on ultrathin allowed to expand or contract freely in the z-direction to keep the
BaTiO3 and PbTiO3 films [214,231]. In the electrical poling pro- out-of-plane stress zero.
cess, as shown in Fig. 54(a), three stages are distinguished. Stage The out-of-plane polarization profiles were analyzed for
I: When the electric field is less than 0.4 108 V/m, it is too small BaTiO3 film with thickness h ¼ 4.4 nm under different in-plane
to make any negative domain switch from –z to þz direction, as strains. The results show that in-plane compression strains
shown in Figs. 54(c) and 54(d). As a result, the average polariza- strengthen the polarization of thin films along the z-direction,
tion of the film increases very slowly in Fig. 54(a). Stage II: while in-plane tensile strains decrease it. More interestingly, there
When the electric field reaches the value between exists a “critical strain,” above and below which the polarization
0.41.6 108 V/m, domain switching occurs (Figs. 54(e)–54(h)), distributions present two different stable patterns, similar to Figs.
leading to a sharp increase of the system polarization, as shown in 53(b) and 2(c). The dependence of the hysteresis loop on the in-
stage II of Fig. 54(a). At the end of this stage, the strip domains plane strain was also investigated. For films under the in-plane
have merged into a single domain, as shown in Fig. 54(i). Stage compression strain, the restriction along x- and y-direction makes
III: As the electric field further increases above 1.6 108 V/m, the domain switching difficult, so the area of the corresponding
there is no domain switching anymore and the system polarization hysteresis loop is larger as compared with the free-standing ones.
increases slowly with the electric field again. When the external In other words, a larger in-plane compression strain induces a
electric field decreases to zero, the entire polarization is still finite greater remanent polarization and coercive field. By contrast, the
positive, which is just the remanent polarization Pr. The polariza- tensile strain has an opposite effect. The external strain also influ-
tion falls down quickly when the electric field decreases further. ences the Curie temperature as predicted by the MD simulations,
Once the electric field reaches the coercive field Ec (about such that a tensile in-plane strain decreases the Curie temperature
0.29 108 V/m), the entire polarization becomes zero again, as while a compressive one increases it.
shown in Figs. 54(a) and 54(k).
The size-dependent hysteresis loops of the ultrathin BaTiO3 Ferroelectric nanowires. Different from the ultrathin film, the
film are shown in Fig. 55(a). For the 3.6-nm-thick film, the hyster- nanowire has free surfaces in two principle directions and thus
esis loop is obviously smaller than that of 4.4-nm-thick film, may have more evident surface effect than the ultrathin film. In
which is also reflected by the value of corresponding remanent order to explore the size dependence of ferroelectric behaviors,
polarization and the coercive field. Moreover, for the film thinner the ferroelectric nanowires with different diameters (i.e., consist-
than the critical thickness, such as 2.8 nm, there is no hysteresis ing of n n 7 unit cells (2 n 10)) were numerically studied.
loop and the polarization has nearly linear relation with the exter- Periodical boundary conditions are applied in the axial direction,
nal electric field. Interestingly, a quasilinear relation of remanent and a surface-free boundary condition is applied in the other two
polarization Pr and coercive field Ec with the reciprocal film thick- directions. For a nanowire with a small diameter, the polarization
ness was revealed, as shown in Figs. 55(b) and 55(c). An extrapo- distributions calculated by shell model MD simulations were care-
lation of remanent polarization at 1/h ¼ 0 gives 17.5 lC/cm2 that fully tested by comparison with the ab initio results, which dem-
is in reasonable agreement with the experiment measurements of onstrated relatively good accordance both qualitatively and
remnant polarization for rhombohedral BaTiO3 bulk material. quantitatively [61].
Besides the spontaneous polarization and hysteresis loop, The size dependence of polarization distribution was studied by
another important quantity (i.e., the Curie temperature) is also the MD method, and the relation between the average polarization

Fig. 54 The evolution of domains in the 4.4-nm-thick film under the external electric field. (a)
and (b): The average polarization of the film as a function of the external electric field. (c) (k):
The sequent top views of the out-of-plane polarization patterns under different electric fields
marked in (a) and (b). The value in each grid represents the average z-direction polarization
over a chain along z-direction (unit: lC/cm2). Reprinted from Ref. [152] with the kind permission
of Elsevier Science Ltd.
components and the nanowire diameter is shown in Fig. 57. The lengths, the stable polarization in the x-y plane emerges, while the
polarization distribution significantly depends on the wire diame- axial component is still near zero (i.e., Pz 0). The x-y plane
ter, and three scenarios can be distinguished. Scenario I: When polarization exhibits a quasiaxisymmetric radial distribution. Sce-
the diameter is only one cell length, the nanowire is too thin to nario III: When the diameter is larger than two cell lengths, the
support any ferroelectric distortion along any directions. There- magnitude of polarization along the x-, y-, and z-directions are on
fore, the spontaneous polarization is almost zero everywhere (i.e., the same order and not negligible. The z-component of polariza-
Px ; Py ; Pz 0). Scenario II: When the diameter is two cell tion Pz over the cross section is along the same direction,

Fig. 55 The hysteresis loops (a) of BaTiO3 films with thicknesses of 4.4 nm, 3.6 nm, and 2.8 nm
and the variations of remnant polarization (b) and coercive field (c) with the reciprocal of film
thickness. Reprinted from Ref. [152] with the kind permission of Elsevier Science Ltd.
Fig. 56 (a) Out-of-plane average polarization as a function of temperature for stress-free

BaTiO3 nanofilms with different thicknesses, and (b) the Curie temperature as a function of the
reciprocal of film thickness. Reprinted from Ref. [64] with the kind permission of American Sci-
entific Publishers.
indicating that a monodomain is formed, which is consistent with between dcr1 and dcr2. When the nanowire is thicker than dcr2, the
the experimental observation [192]. The axial polarization distri- polarization has both radial and axial components and the axial
bution is similar to that of scenario II. polarization is larger at the surface region and smaller at the inner
According to the three scenarios of different polarization distri- region of the nanowire. Such nonuniform polarization distribution
butions, two critical diameters of the BaTiO3 nanowire, can be understood by a lattice mismatch mechanism [232].
dcr1 ¼ 0.8 nm and dcr2 ¼ 1.2 nm, can be identified. No polarization As shown in Fig. 58, both the size and the shape of the hystere-
exists when the diameter is less than dcr1. The radial polarization sis loop vary with the diameter of the nanowires. When the diame-
appears while no axial polarization exists when the diameter is ter d 0:8nm, there is no loop, since nearly no axial remnant

electric nanowire is similar to that of ferroelectric ultrathin film,
the MD results on this facet are not iterated.
The ferroelectric nanowires, similar to nanofilms, are easily
influenced by the extrinsic strain, which may be generated in the
fabricating process or application environment. In order to obtain
a clear picture of the strain effect on the ferroelectric behaviors of
nanowires, axial prestrains were applied to BaTiO3 nanowires
with different sizes until atomic relaxation, and then the polariza-
tion and hysteresis loop were calculated [62].
The MD simulations show that a tensile strain increases the
axial polarization, while a compressive one reduces it. The nano-
wire with d ¼ 0.8 nm has no spontaneous axial polarization at its
freestanding state, but a tensile strain can recover its ferroelectric-
ity. When the magnitude of the compressive strain reaches a criti-
cal value, the axial polarization vanishes. Similar to the definition
of the critical diameter, a critical strain can be also introduced for
the BaTiO3 nanowire to characterize the vanishing of the ferroe-
lectricity. The critical strain increases gradually with the diameter
and asymptotically tends to a stable value. By analyzing the polar-
Fig. 57 The size-dependent polarization components of ization variation under a strain/stress loading, the piezoelectric
BaTiO3 nanowire. Reprinted from Ref. [61] with the kind permis-
coefficient can be calculated. The MD results indicate that the lin-
sion of Institute of Physics (IOP) Publishing.
ear piezoelectric effect is usually valid for a ferroelectric nano-
wire, even under a relatively large strain (e.g., ez ¼ 2.5%), and the
piezoelectric coefficient changes slowly with the diameter.
Due to the strain effect, the nanowires under different prestrains
may have different domain-switching behaviors, and thus the hys-
teresis loop may exhibit different characteristics. The MD simula-
tions of 2.0-nm-thick nanowire [62] demonstrated that both the
size and the shape of the hysteresis loop vary with the axial pre-
strain. When the nanowire is freestanding, the hysteresis loop
exhibits an unusual stepwise shape. This stepwise shape becomes
more obvious for the nanowire under tensile prestrain ez ¼ 0.5%,
while it becomes almost vanishing for the nanowire under com-
pressive prestrain ez ¼ 0.5%. The atomic scale polarization
switching process can well explain these differences induced by
axial strains.
The size effect is found to be able to tailor the size and shape of
the hysteresis loop for ferroelectric nanowires [61], while the
strain effect is found to have a similar function [62]. Therefore,
the equivalent roles of the size effect and strain effect were
explored for ferroelectric nanowire [62]. A comparison was made
Fig. 58 The size-dependent ferroelectric hysteresis loops of
between hysteresis loops of two different thick nanowires, one
BaTiO3 nanowires. Reprinted from Ref. [61] with the kind per-
mission of Institute of Physics (IOP) Publishing. freestanding and another under an axial strain, which demon-
strated that the two loops are very close to each other. Especially,
the corresponding axial spontaneous polarizations are approxi-
polarization exists in this case. When d > 0:8nm, the size of hys- mately the same. Such equivalence might have some implications
teresis loop and the remnant polarization increase with the diame- and applications. For example, in a theoretical aspect, it can help
ter. In order to understand the various hysteresis loops, the us understand the mechanisms of finite size effect and strain
evolution of domain structure was analyzed at different electrical effect, which can be utilized to enhance the electromechanical
loading levels. In particular, the domain evolutions of 2.0-nm- properties of nanoscale ferroelectric devices via controlling the
thick nanowire are demonstrated here as an example. From applied strain. In a computational aspect, we can predict the ferro-
Fig. 59, three-stage domain switching behavior and a core-shell electric behavior of a thick nanowire by simulating a thin nano-
structure with a negative polarization core and a positive polariza- wire under proper pretension strain, which can save
tion shell can be observed. When the electric field is in the range computational cost and may pave the way for investigating ferro-
of 0.05 108 V/m to 0.1 108 V/m, domain switching only electric properties of large-dimension nanostructure by means of
occurs in the center unit cell of the nanowire, as displayed in ab initio method or other atomic computational methods.
Fig. 59(c). As the negative electric field magnitude increases to
above 0.1 108 V/m, the subsequent domain switching occurs in
the eight near-center unit cells of the nanowire, leading to an 4.3 Atomic-Scale Finite Element Method of Ferroelectrics.
obvious polarization drop (c!d in Fig. 59). When increasing the Section 4.2 introduced the shell model MD method of ferroelec-
negative electric field magnitude to 2.3 108 V/m, the 16 unit tric perovskites, which can save much computation cost compared
cells at the surface region finally switch, which gives rise to to the first-principles method. However, there are still two con-
another sudden polarization drop (e!f in Fig. 59). Therefore, the straints that restrict the computational efficiency of shell model
hysteresis loop for d ¼ 2:0nm nanowire exhibits a two-step shape. MD method. Firstly, the shell mass of each atom is very small in
It can then be expected that, when the diameter of the nanowire is the shell model, and thus the time step should be small enough to
larger, more steps may appear in the hysteresis loop. In the avoid the divergence of simulations. Secondly, with the increase
extreme case (i.e., bulk material), the hysteresis loop may be of simulation system, the central processing unit (CPU) time of
regarded as the composition of infinite steps and becomes very the MD simulation usually increases much quicker than linear
smooth since the step height is too small to be distinguished. Since estimation. Thus, the CPU time will increase rapidly when the
the influence of finite size effect on Curie temperature of ferro- dimension of simulation system increases to several nanometers.

Fig. 59 Hysteresis loop of nanowire with diameter d 5 2.0 nm and the domain structures at dif-
ferent electrical loading levels. Reprinted from Ref. [61] with the kind permission of Institute of
Physics (IOP) Publishing.
These two limitations can be overcome if the atomic-scale finite Etot is composed of two parts: the inner potential Uinner character-
element method (AFEM) [233,234] is adopted to determine the izing the atomic interactions between different particles and the
atomic positions at the equilibrium state, instead of MD method. external potential Uextenner induced by the applied physical field,
AFEM is a molecular statics method that solves the energy mini- such as electric field or stress field. The total energy Etot can be
mization problem in the finite element framework. Thus, the parti- given by
cle mass is not needed in the numerical calculation, and more
importantly, the CPU time of this method increases almost line- X
N
arly with the atom number [233,234]. Etot ðxÞ ¼ Uinner ðxÞ þ Uexterner ðxÞ ¼ Uinner ðxÞ i xi ;
F
The combination of AFEM and shell model leads to an effec- i¼1
tive atomic-level computational method of ferroelectric perov- (4.3.1)
skites, as we will introduce in this section. In Sec. 4.3.2, we first
present the formulation of electromechanical AFEM and demon-
strate its applicability in simulations of ferroelectrics and its where x is an integrated coordinate vector of all particles and xi is
the coordinate vector of particle i. F is the generalized force
advantage in computational efficiency. This developed electrome-
chanical AFEM is further utilized to analyze the nanoscale do- induced by an external conserved field, such as the electrome-
main configurations of BaTiO3 and the electric field–driven chanical field. The stabilization of the atomic system requires
evolutions in Sec. 4.3.3. minimization of the total energy, which yields
4.3.1 Formulation of Atomic Finite Element Method. We con- @Etot

¼0 (4.3.2)
sider an atomic system consisting of N particles. The total energy @x

A Taylor expansion of the energy gradient at point xð0Þ gives both used in the two methods for calculating the stable atomic
configuration of BaTiO3 nanowire with different diameters. In the
@Etot @Uinner @ 2 Uinner i ¼ 0 simulations, the same initial atomic coordinates are input and the
¼ þ u F (4.3.3)
@xi @x x¼xð0Þ @x@x x¼xð0Þ same precisions of nonequilibrium force are prescribed as the con-
vergence condition. After running the two programs in the same
where the displacement u ¼ x xð0Þ . The governing equation for computer with a single CPU, the time costs were recorded, as
determination of the atomic configuration can be derived from Eq. shown in Fig. 60. For different nanowire systems, the computa-
(4.3.3), i.e., tional speeds of the developed AFEM are always higher than that
of the MD method by almost ten times. With the computational
Ku¼P (4.3.4) scale becoming larger, the computational efficiency of the present
AFEM becomes even better than that of the MD method. It should
be also noticed that, although the present AFEM possesses a supe-
where
rior computational efficiency to that of shell model MD, it is only
applicable for simulating material properties at 0 K. The develop-
@ 2 Etot @ 2 Uinner @Etot
K¼ ¼ ; P¼ ment of a novel AFEM with consideration of atomic oscillation
@x@x x¼xð0Þ @x@x x¼xð0Þ @x x¼xð0Þ that may study atomic behaviors at finite temperature is ongoing.

¼F @Uinner (4.3.5)
@x x¼xð0Þ 4.3.3 Results and Discussion. From the above discussions, it
can be found that the shell model AFEM provides an effective
and reliable atomic-level simulation method of perovskite ferro-
Above, K is the stiffness matrix of the atomic system, depending
electric material, by which a larger atomic system can be simu-
on the potential between different atoms, as well as the coordi-
lated compared with the MD method. In this section, we introduce
nates of each atom. P is the nonequilibrium force, reflecting the
the application of this method in simulations of nanoscale domain
gap between the present configuration and the equilibrium one.
structure in BaTiO3, which is rather difficult to achieve using the
Compared with the MD method, the AFEM takes into account not
MD method.
only the first order derivative of the energy function with respect
To reduce the computational scale and the complexity in analy-
to the coordinate (i.e., the nonequilibrium force), but also its sec-
sis, only the two-dimensional polarization configurations in ferro-
ond order derivative (i.e., the stiffness matrix), and they can be
electrics are considered. As such, the z-directional dimension of
easily obtained analytically. Thus, the AFEM may reach the stable
the BaTiO3 atomic system is fixed as six lattice parameters. In the
state of the system quicker than the MD.
x-y plane where the domain structure forms, the dimension ranges
In order to characterize the spontaneous polarization of the fer-
from 6 to 20 lattice parameters. The atom number of the largest
roelectric material, the shell model [55–57] that describes phe-
BaTiO3 system is 12,000 in the simulations, which is difficult to
nomenologically the deformation of the electronic structure of
deal with by shell model MD using a single CPU. In order to sim-
each ion should be adopted. The detailed descriptions of shell
ulate the behavior of bulk material and avoid the finite boundary
model interatomic potentials have been provided in Sec. 4.2.1 and
effect, periodical boundary conditions are applied in the three
thus are not iterated here. After giving the initial atomic coordi-
axial directions.
nates and coefficients of potential, the stable atomic configuration
By changing the initial atomic coordinates, different polariza-
can be then determined by again solving Eq. (4.3.4) with updated
tion structures can be stabilized after relaxation. The domain pat-
stiffness and force matrices. The applicability of this method was
terns of these stable polarization structures are shown in Fig. 61.
carefully testified for both BaTiO3 bulk material and nanowire
The simulations demonstrate that the single domain, 90 deg do-
[88] by comparing the calculated lattice parameter and polariza-
main, and vortex domain can be stabilized even for the 6 6 6
tion with MD and first-principles results.
atomic system, and the thickness of 90 deg domain or vortex do-
main ranges from 1.2 nm to 4.0 nm. The thickness of the 90 deg
4.3.2 Comparison of Computational Efficiency With Molecu- domain wall is determined as about one lattice parameter
lar Dynamics Method. In order to compare the computation effi- (0.4 nm) from Fig. 61, which is comparable to the experimental
ciency of electromechanical AFEM and MD methods, the results (1.0 nm–2.5 nm) [235] and the theoretical prediction [236],
calculation skills of long-range force as described in Sec. 4.2.2 are both for BaTiO3 single crystal. Note that, in more recent
Fig. 60 A comparison of computational efficiency between the developed AFEM

and MD method. The inset table demonstrates the ratio of time cost in the MD to
that in the AFEM. Reprinted from Ref. [88] with the kind permission of Elsevier Sci-
ence Ltd.

Fig. 61 Three types of stable nanoscale domain structure: (a) single domain; (b) 90 domain;
(c) vortex domain. The three figures above illustrate the polarization distributions with each
arrow representing the polarization of a unit cell, while the ones below demonstrate the sche-
matic domain patterns. Notice that only 1/4 area of each figure are calculated by the present
AFEM, and based on the periodical boundary conditions, four identical patterns are packed to-
gether to obtain a better image of the polarization topology. Reprinted from Ref. [88] with the
kind permission of Elsevier Science Ltd.
theoretical and experimental studies [237–239], the presented do- the polarization of variant i to the electric field (E) direction anti-
main wall thickness is nearly an order larger than 0.4 nm. In order clockwise (180 deg hi 180 deg). The domain evolution
to obtain a larger domain wall thickness from AFEM simulations, results are demonstrated in Fig. 62 for different electrical loading
a more reasonable set of potential parameters is required. In the directions. When h1 ¼ 0 deg or 45 deg, the polarization always
vortex domain structure, the clockwise and anticlockwise vortexes remains unchanged, no matter how large the electric field is,
are distributed alternatively along the 645 deg directions with because domain switching would increase the angle between the
respect to the x-axis, similar to the experimental results of Tsou polarization and electric field and thus enhances the system
et al. [240], who observed vortex domain pattern in tetragonal energy. When h1 ¼ 90 deg, as the electric field approaches the
BaTiO3. Interestingly, besides the vortex polarization structure, coercive field, a new domain parallel with E first nucleates auto-
the streamline polarization structure also exists in the vortex do- matically in the original domain, and then the domain wall moves
main configuration. Since the polarizations along opposite direc- laterally until the domain completely switches to the direction of
tions counteract each other in the center of the streamline electric field. When h1 ¼ 135 deg, the domain evolution process is
structure, this domain configuration also satisfies the continuity similar to that of h1 ¼ 90 deg. Interestingly, when the domain is
condition of normal polarization at the domain wall, thus avoiding applied to an opposite electric field, the domain undergoes a con-
the emergence of large depolarization field. In the other ab initio secutive two-step, 90 deg domain-switching process. The original
calculations, single vortex structure is revealed in the PZT nano- domain first switches to variant 4 via nucleation and movement of
wire [68] and PbTiO3 nanowire [241]. Hong et al. [177] initially horizontal 90 deg domain wall (or switches to variant 2 with
reported the existence of streamline polarization pattern in nucleation and movement of vertical 90 deg domain wall), and
BaTiO3 nanowire by first-principles method. Different from then the whole domain switches simultaneously to variant 3.
these polarization structures, the domain structure as shown in Notice that a similar two-step domain-switching process has been
Fig. 61(c) has a coexistence of both vortex and streamline polar- observed in the in situ experiment on tetragonal BaTiO3 single
ization configurations. crystal using polarized light microscopy [242]. From the simula-
By inputting the stable atomic coordinates of different domain tions, it can be found that, when |h| > 45 deg, the domain switch-
configurations as the initial state and applying electric fields along ing can be triggered to reduce the system energy as long as the
different directions, the electric field–driven domain evolutions external electric field surpasses the coercive field.
can be simulated. As an example, we introduce the various evolu- For the 90 deg and vortex domains, the electric field–driven
tion paths of single domain structure. The atomic systems ranging evolution diagrams are more complicated than that of single do-
from 6 6 6 to 20 20 6 are considered, and the simulations main and can be found in Zhang et al. [88]. Besides the qualitative
show that the variation of domain size almost has no influence on evolution patterns, the variations of overall polarizations with the
the domain evolution sequences. Thus, only the simulation results electric field can be also obtained. The P–E curves during the
for the atomic system of 12 12 6 are exemplified to demon- electrical poling and depoling processes were calculated for vor-
strate the domain evolution diagrams. tex domain as the initial pattern. The calculations indicate that, for
Define the unit cell with two positive polarization components two different loading directions (horizontal and diagonal),
(Px > 0 and Py > 0) as variant 1, the cell with Px > 0 and Py < 0 as although the microscopic domain evolution processes are differ-
variant 2, the cell with Px < 0 and Py < 0 as variant 3, and the cell ent, the variation trends of macroscopic polarization responses are
with Px < 0 and Py > 0 as variant 4. Denote hi the angle rotating similar. During the poling process, when the electric field is small

Fig. 62 Evolution of single domain structure under electric fields with different
loading directions. Reprinted from Ref. [88] with the kind permission of Elsevier
Science Ltd.
enough so that the domain switching cannot be activated, the [81,120–123], where the misfit strain due to lattice mismatch
polarization increases slowly and almost linearly with the electric between film and substrate is related to the lattice constants of
field, due to the dielectric effect. As the electric field exceeds the both materials. The elastic energy is generally expressed as a
coercive field, the domain wall starts moving and the polarization function of the stiffness tensor for anisotropic materials (depend-
increases quickly until the domain is poled into the saturated state. ing on the phases). Since some components of the stiffness tensor
During the depoling process, when the magnitude of negative are not available or not easy to measure via experiments, first-
electric field is smaller than the coercive field, the domain config- principles calculations could provide an effective alternative route
uration remains unchanged and the polarization decreases rela- to determine the entire stiffness tensor. Besides, the double-well
tively slowly. As the electric field reaches the coercive value, the free energy function (in terms of polarization or electric displace-
domain undergoes a drastic change. ment), as adopted in phase field method, can also be fitted from
first-principles calculations with the recently developed fixed
5 Connections Among Electromechanically Coupled polarization method [243] and fixed electric displacement method
[244–246]. The output from first-principles method can also be
Computational Methods at Different Scales used as input for MD method to obtain the proper potential. As we
The various computational methods introduced above are only know, good potential is essential for the accurate MD simulation,
applicable within their own scales, because of the different and for different materials, the potential has different forms or dif-
assumptions, formulations, and the computational complexity. ferent coefficients. After assuming the potential form for specific
However, in some special cases, simple interfaces or connections materials, its coefficients can be fitted by comparing the properties
between the computational methods at different scales are possi- (free energy, spontaneous distortion, etc.) at different phases
ble (e.g., by passing on the material properties calculated at a and strained states from first-principles calculations and MD
lower-scale method to serve as input parameters for a higher-scale simulations.
method). In the following, we take some representative examples The mobility coefficient of domain wall movement required in
to illustrate such connections among the computational methods phase field method can be possibly obtained by MD simulations.
of three different scales. McCash et al. [247] used classical molecular dynamics with the
The lattice constant, elastic modulus, polarization, piezoelectric force field derived from a first-principles–based effective Hamil-
coefficient, flexoelectric coefficient, and some other fundamental tonian to investigate the intrinsic dynamics of ultrafast polariza-
material properties can be obtained from first-principles calculation reversal in ferroelectric nanowires. They traced the time
tion and passed to the mesoscale methods. For example, the phase evolution of polarization under applied electric field and found
field method was widely adopted to explore the domain patterns that, in defect-free nanowires, the polarization reversal can occur
of ferroelectric ultrathin films grown on a heterogeneous substrate within picoseconds. Similar analyses can be extended to

ferroelectric ultrathin films to estimate the speed of domain wall addressed. In the mesoscopic methods, the conventional phase
movement, which is related to the kinetic coefficient of the classi- field models (PFM) and the optimization-based computational
cal time-dependent Ginzburg–Landau equation for phase field mechanics methods were both presented and the domain pattern
method. Note that, in previous reports, this kinetic coefficient was evolution in single crystals and domain texture evolution in poly-
usually assumed as a small parameter just for ensuring the conver- crystals were obtained. In the atomic-level methods, we employed
gence of numerical calculations, and thus, the evolution of domain the first-principles method to investigate the topological polariza-
pattern is only related to a generalized time (not the real time). tion and flexoelectricity response and adopted the molecular dy-
The material properties in the constitutive equations of piezo- namics method to address the size effect and surface effect in
electrics and/or ferroelectrics needed for the macroscopic method ferroelectric thin films and nanowires. To enhance the computa-
can be evaluated from mesoscale simulations. It is well known tional efficiency of MD, we proposed a novel atomic finite ele-
that the macroscopic piezoelectric coefficients are highly depend- ment method, which has been proven to be much faster than the
ent on the configurations and layouts of microstructures (e.g., MD when the number of atoms is large. Obviously, the develop-
phases, domain patterns, etc.) for both single crystal and polycrys- ment of electromechanically coupled computational mechanics
talline ferroelectrics. The mesoscale methods, such as phase field methods at multiscale is greatly beneficial, not only to scientific
method and Monte Carlo method, are well suited to describe the research of material properties and electromechanical behavior,
effects of microstructures on the macroscopic material properties such as deformation and fracture for piezoelectric/ferroelectric
that can be then passed to the macroscopic computational meth- materials, but also to the structural design and reliability analysis
ods. To determine the material properties using the mesoscale of smart devices in engineering.
method, we usually need to focus on a representative volume It should be pointed out that all the computational methods pre-
element by analyzing the layout of periodically distributed sented in this review are only valid in their own scale, because of
microstructures. the distinct fundamental assumptions in each method. So far, a
It should also be noted that practically each type of computa- truly cross-scale computational method for ferroelectrics is still
tional method is only valid in its own scale scope, although some lacking, which may hinder the understanding of ferroelectricity
methods are actually scale-independent. In the atomic-level meth- and piezoelectricity at multiscale. The interfaces or transformers
ods, the first-principles method [66–70] can be regarded as the between the computational methods at different scales are really
most accurate one, as it uses only the numbers of the atomic sys- imperative at present. Actually, the idea of such interfaces has
tem and some initial guesses at their coordinates as input without been used recently [63] that employed the lattice constants calcu-
artificial approximation (so it is also called ab initio method). lated by the first-principles method as the input parameters in the
With the rapid development of computer hardwares, the first- MD method, while generally the interface between the macro-
principles method can deal with the systems with hundreds of scopic methods and the microscopic (or atomic-level) methods is
atoms (10 nm in one dimension), making it possible to study the very difficult to build, as the time scale of the former is usually
nanostructures (films, superlattices, nanowires, nanotubes, and tens of orders larger than the latter. Furthermore, at the atomic
nanodots), the effects of defects and the ferroelectric domain scale, the definition of the basic physics quantities, such as tem-
walls, etc. However, it is still difficult to use first-principles perature, stress, strain, etc., is quite different from that at macro-
method to study some dynamic properties (e.g., domain wall scopic scale, which is another difficult problem to be solved in the
movement) and temperature effect. The MD method [55–65] future.
could cope with a larger atomic system than first-principles
method, because of the simplification in describing the intera-
tomic interactions. Its affordable size is millions of atoms on a
powerful supercomputer, which makes it possible for some large Acknowledgment
systems (100 nm), like piezoelectric nanobeams. However, the D.F. acknowledges the support from the Natural Science Foun-
simplification of interatomic interaction and lack of description of dation of China under Grant No. 11090331. F.L. also thanks the
electronic contribution hinder its application in electronic proper- support from the Natural Science Foundation of China under
ties and high-accuracy calculations. Our recently proposed AFEM Grant No. 11002002.
method [88] can effectively enhance the computational size of the
MD method, which in some sense may be treated as a transscale
method from the atomic level to mesoscopic level. Actually, the
References
[1] Xu, Y., 1991, Ferroelectric Materials and Their Applications, North-Holland,
mesoscopic methods [35–54], including the PFM method and the Amsterdam.
micromechanical method, and the macroscopic methods [30–34], [2] Jaffe, B., Cook, W. R., and Jaffe, H., 1971, Piezoelectric Ceramics, Academic,
including the FEM method and the meshless method, are all London.
scale-independent in their original forms, although the PFM [3] Lines, M. E., and Glass, A. M., 1977, Principles and Applications of Ferro-
electrics and Related Materials, Clarendon, Oxford.
method can address the size effect by taking into account the sur- [4] Otsuka, K., and Ren, B., 2005, “Physical Metallurgy of Ti-Ni-Based Shape
face issues [35–50,79–82]. Theoretically, the PFM method and Memory Alloys,” Prog. Mater. Sci., 50, pp. 511–678.
the micromechanical methods are also applicable at the macro- [5] Chen, P. J., and Peerey, P. S., 1979, “One-Dimensional Dynamic Electrome-
scopic level, but the rather large computational complexity makes chanical Constitutive Relations of Ferroelectric Materials,” Acta Mech., 31,
pp. 231–241.
their achievable material size typically below 100 lm or smaller. [6] Chen, P. J., 1980, “Three Dimensional Dynamic Electromechanical Constitu-
Similarly, the macroscopic methods (FEM method, meshless tive Relations for Ferroelectric Materials,” Int. J. Solids Struct., 16, pp.
method, etc.) should be applicable in the scale range as long as 1059–1067.
the continuum mechanics is valid, which is typically larger than [7] Bassiouny, E., Ghaleb, A. F., and Maugin, G. A., 1988, “Thermodynamical
Formulation for Coupled Electromechanical Hysteresis Effects-I. Basic Equa-
100 nm. tions,” Int. J. Eng. Sci., 26(12), pp. 1279–1295.
[8] Cocks, C. F., and McMeeking, R. M., 1999, “A Phenomenological Constitu-
tive Law for the Behaviour of Ferroelectric Ceramics,” Ferroelectrics, 228, pp.
6 Conclusions 219–228.
[9] Kamlah, M., and Tsakmakis, C., 1999, “Phenomenological Modeling of the
In summary, we present a comprehensive review of the multi- Nonlinear Electromechanical Coupling in Ferroelectrics,” Int. J. Solids Struct.,
scale computational methods for ferroelectrics/piezoelectrics 36, pp. 669–695.
developed at our group and other groups in the past decade. In the [10] Huber, J. E., and Fleck, N. A., 2001, “Multi-axial Electrical Switching of a
macroscopic methods, we introduce the linear/nonlinear finite ele- Ferroelectric: Theory Versus Experiment,” J. Mech. Phys. Solids, 49, pp.
785–811.
ment methods and the meshless method for linear piezoelectrics. [11] Landis, C. M., 2002, “Fully Coupled, Multi-axial, Symmetric Constitutive
The rescaling technique in the linear FEM and the domain- Laws for Polycrystalline Ferroelectric Ceramics,” J. Mech. Phys. Solids, 50,
switching instability problem in nonlinear FEM were specially pp. 127–152.

[12] Hwang, S. C., Lynch, C. S., and McMeeking, R. M., 1995, “Ferroelectric/Fer- [44] Dayal, K., and Bhattacharya, K., 2007, “A Real-Space Non-local Phase-Field
roelastic Interactions and a Polarization Switching Model,” Acta Metall. Model of Ferroelectric Domain Patterns in Complex Geometries,” Acta
Mater., 43(5), pp. 2073–2084. Mater., 55, pp. 1907–1917.
[13] Hwang, S. C., Huber, J. E., McMeeking, R. M., and Fleck, N. A., 1998, “The [45] Zhang, Y. H., Li, J. Y., and Fang, D. N., 2010, “Oxygen-Vacancy-Induced
Simulation of Switching in Polycrystalline Ferroelectric Ceramics,” J. Appl. Memory Effect and Large Recoverable Strain in a Barium Titanate Single
Phys., 83(3), pp. 1530–1540. Crystal,” Phys. Rev. B, 82, p. 064103.
[14] Chen, X., Fang, D. N., and Hwang, K. C., 1997, “Micromechanics [46] Zhang, Y. H., Li, J. Y., and Fang, D. N., 2010, “Size Dependent Domain Con-
Simulation of Ferroelectric Polarization Switching,” Acta Mater., 45(8), pp. figuration and Electric Field Driven Evolution in Ultrathin Ferroelectric Films:
3181–3189. A Phase Field Investigation,” J. Appl. Phys., 107, p. 034107.
[15] Lu, W., Fang, D. N., Li, C. Q., and Hwang, K. C., 1999, “Nonlinear Electric- [47] Abdollahi, A., and Arias, I., 2011, “Phase-Field Modeling of the Coupled
Mechanical Behavior and Micromechanics Modeling of Ferroelectric Domain Microstructure and Fracture Evolution in Ferroelectric Single Crystals,” Acta
Evolution,” Acta Mater., 47, pp. 2913–2926. Mater., 59, pp. 4733–4746.
[16] Li, F. X., and Rajapakse, R. K. N. D., 2007, “A Constrained Domain Switch- [48] Abdollahi, A., and Arias, I., 2012, “Numerical Simulation of Intergranular and
ing Model for Polycrystalline Ferroelectric Ceramics: Part I—Model Formula- Transgranular Crack Propagation in Ferroelectric Polycrystals,” Int. J. Fract.,
tion and Application to Tetragonal Materials,” Acta Mater., 55, pp. 174, pp. 3–15.
6472–6480. [49] Abdollahi, A., and Arias, I., 2012, “Phase-Field Modeling of Crack Propaga-
[17] Li, F. X., and Rajapakse, R. K. N. D., 2007, “A Constrained Domain Switch- tion in Piezoelectric and Ferroelectric Materials With Different Electrome-
ing Model for Polycrystalline Ferroelectric Ceramics: Part II—Combined chanical Crack Conditions,” J. Mech. Phys. Solids, 60, pp. 2100–2126.
Switching and Application to Rhombohedral Materials,” Acta Mater., 55, pp. [50] Li, W., and Landis, C. M., 2011, “Nucleation and Growth of Domains Near
6481–6488. Crack Tips in Single Crystal Ferroelectrics,” Eng. Fract. Mech., 78, pp.
[18] Park, S. B., and Sun, C. T., 1995, “Fracture Criteria for Piezoelectric Ceram- 1505–1513.
ics,” J. Am. Ceram. Soc., 78, pp. 1475–1480. [51] Tang, W., Fang, D. N., and Li, J. Y., 2009, “Two-Scale Micromechanics-
[19] Zhang, T. Y., and Tong, P., 1996, “Fracture Mechanics for a Mode III Crack Based Probabilistic Modeling of Domain Switching in Ferroelectric Ceram-
in a Piezoelectric Material,” Int. J. Solids Struct., 33, pp. 343–359. ics,” J. Mech. Phys. Solids, 57, pp. 1683–1701.
[20] Zhang, T. Y., and Qian, C. F., 1998, “Linear Electro-elastic Analysis of a Cav- [52] Li, F. X., and Soh, A. K., 2010, “An Optimization-Based Computational
ity or a Crack in a Piezoelectric Material,” Int. J. Solids Struct., 35, pp. Model for Domain Evolution in Polycrystalline Ferroelastics,” Acta Mater.,
2121–2149. 58, pp. 2207–2215.
[21] Zhang, T. Y., Zhao, M. H., and Tong, P., 2002, “Fracture of Piezoelectric [53] Li, F. X., Zhou, X. L., and Soh, A. K., 2010, “An Optimization-Based “Phase
Ceramics,” Adv. Appl. Mech., 38, pp. 147–289. Field” Model for Polycrystalline Ferroelectrics,” Appl. Phys. Lett., 96, p.
[22] Chen, Y. H., and Lu, T. R., 2003, “Cracks and Fracture in Piezoelectrics,” 152905.
Adv. Appl. Mech., 39, pp. 121–215. [54] Zhou, X. L., and Li, F. X., 2011, “Simulations of Domain Evolution in Mor-
[23] Zhang, T. Y., and Gao, C. F., 2004, “Fracture Behavior of Piezoelectric Mate- photropic Ferroelectric Ceramics Under Electromechanical Loading Using an
rials,” Theor. Appl. Fract. Mech., 41, pp. 339–379. Optimization-Based Model,” J. Appl. Phys., 109, p. 084107.
[24] Zhang, T. Y., Zhao, M. H., and Gao, C. F., 2005, “The Strip Dielectric Break- [55] Migoni, R., Bilz, H., and Bauerle, D., 1976, “Origin of Raman-Scattering and
down Model,” Int. J. Fract., 132, pp. 311–327. Ferroelectricity in Oxidic Perovskites,” Phys. Rev. Lett., 37, pp. 1155–1158.
[25] Yang, W., and Zhu, T., 1998, “Switch-Toughening of Ferroelectrics Subjected [56] Khatib, D., Migoni, R., Kugel, G. E., and Godefroy, L., 1989, “Lattice-Dy-
to Electric Fields,” J. Mech. Phys. Solids, 46(2), pp. 291–311. namics of BaTiO3 in the Cubic Phase,” J. Phys.: Condens. Matter, 1, pp.
[26] Zhu, T., and Yang, W., 1997, “Toughness Variation of Ferroelectrics by Polar- 9811–9822.
ization Switch Under Non-uniform Electric Field,” Acta Mater., 45(11), pp. [57] Tinte, S., Stachiotti, M. G., Sepliarsky, M., Migoni, R. L., and Rodriguez, C.
4695–4702. O., 1999, “Atomistic Modelling of BaTiO3 Based on First-Principles Calcu-
[27] Mao, G. Z., and Fang, D. N., 2004, “Fatigue Crack Growth Induced by Do- lations,” J. Phys.: Condens. Matter, 11, pp. 9679–9690.
main Switching Under Electromechanical Load in Ferroelectrics,” Theor. [58] Sepliarsky, M., Phillpot, S. R., Wolf, D., Stachiotti, M. G., and Migoni, R. L.,
Appl. Fract. Mech., 41, pp. 115–123. 2001, “Long-Ranged Ferroelectric Interactions in Perovskite Superlattices,”
[28] Fang, D. N., Zhang, Y. H., and Mao, G. Z., 2011, “A COD Fracture Model of Phys. Rev. B, 64, p. 060101.
Ferroelectric Ceramics With Applications in Electric Field Induced Fatigue [59] Sepliarsky, M., Asthagiri, A., Phillpot, S. R., Stachiotti, M. G., and Migoni, R.
Crack Growth,” Int. J. Fract., 167, pp. 211–220. L., 2005, “Atomic-Level Simulation of Ferroelectricity in Oxide Materials,”
[29] Gao, H., Zhang, T. Y., and Tong, P., 1997, “Local and Global Energy Release Curr. Opin. Solid State Mater. Sci., 9, pp. 107–113.
Rate for an Electrically Yield Crack in a Piezoelectric Ceramic,” J. Mech. [60] Shimada, T., Wakahara, K., Umeno, Y., and Kitamura, T., 2008, “Shell Model
Phys. Solids, 45, pp. 491–510. Potential for PbTiO3 and Its Applicability to Surfaces and Domain Walls,” J.
[30] Gong, X., and Suo, Z., 1996, “Reliability of Ceramic Multiplayer Actuators: A Phys.: Condens. Matter, 20, p. 325225.
Nonlinear Finite Element Simulation,” J. Mech. Phys. Solids, 44(5), pp. [61] Zhang, Y. H., Hong, J. W., Liu, B., and Fang, D. N., 2009, “Molecular Dy-
751–769. namics Investigations on the Size-Dependent Ferroelectric Behavior of
[31] Hom, C. L., and Shankar, N., 1996, “A Finite Element Method for Electro- BaTiO3 Nanowires,” Nanotechnology, 20, p. 405703.
strictive Ceramics,” Int. J. Solids Struct., 33(12), pp. 1757–1779. [62] Zhang, Y. H., Hong, J. W., Liu, B., and Fang, D. N., 2010, “Strain Effect on
[32] Qi, H., Fang, D. N., and Yao, Z. H., 1997, “FEM Analysis of Electro- Ferroelectric Behaviors of BaTiO3 Nanowires: A Molecular Dynamics
mechanical Coupling Effect of Piezoelectric Materials,” Comp. Mater. Sci., 8, Study,” Nanotechnology, 21, p. 015701.
pp. 283–290. [63] Zhang, Y. H., Liu, B., and Fang, D. N., 2011, “Stress-Induced Phase Transi-
[33] Hwang, S. C., and McMeeking, R. M., 1999, “A Finite Element Model of Fer- tion and Deformation Behavior of BaTiO3 Nanowires,” J. Appl. Phys., 110, p.
roelastic Polycrystals,” Int. J. Solids Struct., 36(10), pp. 1541–1556. 054109.
[34] Li, F. X., and Fang, D. N., 2004, “Simulations of Domain Switching in Ferro- [64] Zhang, Y. H., Sang, Y. L., Liu, B., and Fang, D. N., 2011, “Critical Thickness
electrics by a Three-Dimensional Finite Element Model,” Mech. Mater., 36, and the Size-Dependent Curie Temperature of BaTiO3 Nanofilms,” J. Com-
pp. 959–973. put. Theor. Nanosci., 8, pp. 867–872.
[35] Chen, L. Q., 2002, “Phase-Field Models for Microstructure Evolution,” Annu. [65] Stachiotti, M. G., and Sepliarsky, M., 2011, “Toroidal Ferroelectricity in
Rev. Mater. Res., 32, pp. 113–140. PbTiO3 Nanoparticles,” Phys. Rev. Lett., 106, p. 137601.
[36] Wang, J., Shi, S. Q., Chen, L. Q., Li, Y. L., and Zhang, T. Y., 2004, “Phase [66] Cohen, R. E., 1992, “Origin of Ferroelectricity in Perovskite Oxides,” Nature,
Field Simulations of Ferroelectric/Ferroelastic Polarization Switching,” Acta 358, pp. 136–138.
Mater., 52, pp. 749–764. [67] Junquera, J., and Ghosez, P., 2003, “Critical Thickness for Ferroelectricity in
[37] Wang, J., Li, Y. L., Chen, L. Q., and Zhang, T. Y., 2005, “The Effect of Me- Perovskite Ultrathin Films,” Nature, 422, pp. 506–509.
chanical Strains on the Ferroelectric and Dielectric Properties of a Model Sin- [68] Naumov, I. I., Bellaiche, L., and Fu, H., 2004, “Unusual Phase Transitions in
gle Crystal – Phase Field Simulation,” Acta Mater., 53, pp. 2495–2507. Ferroelectric Nanodisks and Nanorods,” Nature, 432, pp. 737–740.
[38] Wang, J., and Zhang, T. Y., 2006, “Size Effects in Epitaxial Ferroelectric [69] Geneste, G., Bousquet, E., Junquera, J., and Ghosez, P., 2006, “Finite-Size
Islands and Thin Films,” Phys. Rev. B, 73, p. 144107. Effects in BaTiO3 Nanowires,” Appl. Phys. Lett., 88, p. 112906.
[39] Wang, J., and Zhang, T.-Y., 2007, “Phase Field Simulations of Polarization [70] Pilania, G., Alpay, S. P., and Ramprasad, R., 2009, “Ab Initio Study of Ferroe-
Switching-Induced Toughening in Ferroelectric Ceramics,” Acta Mater., 55, lectricity in BaTiO3 Nanowires,” Phys. Rev. B, 80, p. 014113.
pp. 2465–2477. [71] Allik, H., and Hughes, T. J. R., 1970, “Finite Element Method for Piezoelec-
[40] Zhang, W., and Bhattacharya, K., 2005, “A Computational Model of Ferro- tric Vibration,” Int. J. Numer. Meth. Eng., 2, pp. 151–157.
electric Domains. Part I: Model Formulation and Domain Switching,” Acta [72] Hom, C. L., and Shankar, N., 1995, “A Numerical Analysis of Relaxor Ferro-
Mater., 53, pp. 185–198. electric Multilayered Actuators and 2-2 Composite Arrays,” Smart Mater.
[41] Zhang, W., and Bhattacharya, K., 2005, “A Computational Model of Ferro- Struct., 4, pp. 266–273.
electric Domains. Part II: Grain Boundaries and Defect Pinning,” Acta Mater., [73] Kamlah, M., and Bohle, U., 2001, “Finite Element Analysis of Piezoceramic
53, pp. 199–209. Components Taking Into Account Ferroelectric Hysteresis Behavior,” Int. J.
[42] Song, Y. C., Soh, A. K., and Ni, Y., 2007, “Phase Field Simulation of Crack Solids Struct., 38, pp. 605–633.
Tip Domain Switching in Ferroelectrics,” J. Phys. D: Appl. Phys., 40, pp. [74] Chen, W., and Lynch, C. S., 1999, “Finite Element Analysis of Cracks in Fer-
1175–1182. roelectric Ceramic Materials,” Eng. Fract. Mech., 64, pp. 539–562.
[43] Su, Y., and Landis, C. M., 2007, “Continuum Thermodynamics of Ferroelec- [75] Li, F. X., and Rajapakse, R. K. N. D., 2008, “Nonlinear Finite Element Model-
tric Domain Evolution: Theory, Finite Element Implementation, and Applica- ing of Polycrystalline Ferroelectrics Based on Constrained Domain Switch-
tion to Domain Wall Pinning,” J. Mech. Phys. Solids, 55, pp. 280–305. ing,” Comp. Mater. Sci., 44(2), pp. 322–329.

[76] Guo, X. H., 2004, “Study on Mechanical Behavior of Thin Films”, Ph.D. the- [110] Belytschko, T., Gu, L., and Lu, Y. Y., 1994, “Fracture and Crack Growth by Ele-
sis, Tsinghua University, Beijing, China. ment Free Galerkin Methods,” Model. Simul. Mater. Sci. Eng., 2(3A), pp. 519–534.
[77] Guo, X. H., and Fang, D. N., 2004, “Simulation of Interface Cracking in Pie- [111] Belytschko, T., Lu, Y. Y., and Gu, L., 1994, “Element-Free Galerkin Meth-
zoelectric Layers,” Int. J. Nonlin. Sci. Numer. Simul., 5(3), pp. 235–242. ods,” Int. J. Numer. Meth. Eng., 37, pp. 229–256.
[78] Guo, X. H., Fang, D. N., Soh, A. K., Kim, H. C., and Lee, J. J., 2006, [112] Belytschko, T., Organ, D., and Gerlach, C., 2000, “Element-Free Galerkin
“Analysis of Piezoelectric Ceramic Multilayer Actuators Based on an Electro- Methods for Dynamic Fracture in Concrete,” Comp. Meth. Appl. Mech. Eng.,
mechanical Coupled Meshless Method,” Acta Mech. Sinica, 22, pp. 34–39. 187, pp. 385–399.
[79] Li, Y. L., Hu, S. Y., Liu, Z. K., and Chen, L. Q., 2001, “Phase-Field Model of [113] Onate, E., Idelsohn, S., Zienkiewicaz, O. C., and Taylor, R. L., 1996, “A Finite
Domain Structures in Ferroelectric Thin Films,” Appl. Phys. Lett., 78, pp. Point Method in Computational Mechanics: Applications to Convective Trans-
3878–3880. port and Fluid Flow,” Int. J. Numer. Meth. Eng., 39, pp. 3839–3866.
[80] Li, Y. L., Hu, S. Y., Liu, Z. K., and Chen, L. Q., 2002, “Effect of Electrical [114] Zhang, X., Liu, X. H., Song, K. Z., and Lu, M. W., 2011, “Least-Square Collo-
Boundary Conditions on Ferroelectric Domain Structures in Thin Films,” cation Meshless Method,” Int. J. Numer. Meth. Eng., 51(9), pp. 1089–1100.
Appl. Phys. Lett., 81, pp. 427–429. [115] Xu, T., Zou, P., Xu, T. S., and Jiye, C. M., 2010, “Study on Weight Function
[81] Li, Y. L., Hu, S. Y., Liu, Z. K., and Chen, L. Q., 2002, “Effect of Substrate of Meshless Method Based on B-Spline Wavelet Function,” The 3rd Int. Joint
Constraint on the Stability and Evolution of Ferroelectric Domain Structures Conference on Computational Science and Optimization, pp. 36–40.
in Thin Films,” Acta Mater., 50, pp. 395–411. [116] Razmjoo, H., Movahhedi, M., and Hakimi, A., 2010, “Improved Meshless
[82] Li, Y. L., Cross, L. E., and Chen, L. Q., 2005, “A Phenomenological Thermo- Method Using Direct Shape Function for Computational Electromagnetics,”
dynamic Potential for BaTiO3 Single Crystals,” J. Appl. Phys., 98, p. 064101. Proceedings of the Asia-Pacific Microwave Conference, Yokohama, Japan,
[83] Zhao, X. F., Soh, A. K., and Li, L., 2010, “Influence of Dipole Defects on pp. 2157–2160.
Polarization Switching in the Vicinity of a Crack in Relaxor Ferroelectrics,” [117] Sladek, J., Sladek, V., Solek, P., and Pan, E., 2008, “Fracture Analysis of
Philos. Mag. Lett., 90, pp. 251–260. Cracks in Magneto-electro-elastic Solids by the MLPG,” Comput. Mech., 42,
[84] Wang, Y. U., Jin, Y. M., and Khachaturyan, A. G., 2002, “Phase Field Microe- pp. 697–714.
lasticity Theory and Modeling of Elastically and Structurally Inhomogeneous [118] Feng, W. J., Han, X., and Li, Y. S., 2009, “Fracture Analysis for Two-
Solid,” J. Appl. Phys., 92, pp. 1351–1360. Dimensional Plane Problems of Nonhomogeneous Magneto-Electro-Thermo-
[85] Choudhury, S., Li, Y. L., Krill, C. E., and Chen, L. Q., 2005, “Phase-Field Elastic Plates Subjected to Thermal Shock by Using the Messless Local Petrov
Simulation of Polarization Switching and Domain Evolution in Ferroelectric Galerkin Method,” Computer Model. Eng. Sci., 48(1), pp. 1–26.
Polycrystals,” Acta Mater., 53, pp. 5313–5321. [119] Sosa, H. A., and Pak, Y. E., 1990, “Three Dimensional Eigenfunction Analysis
[86] Schrade, D., Mueller, R., Xu, B. X., and Gross, D., 2007, “Domain Evolution of a Crack in a Piezoelectric Material,” Int. J. Solids Struct., 26, pp. 1–15.
in Ferroelectric Materials: A Continuum Phase Field Model and Finite Ele- [120] Li, Y. L., and Chen, L. Q., 2006, “Temperature-Strain Phase Diagram for
ment Implementation,” Comput. Mech. Appl. Mech. Eng., 196, pp. BaTiO3 Thin Films,” Appl. Phys. Lett., 88, p. 072905.
4365–4374. [121] Li, Y. L., Choudhury, S., Haeni, J. H., Biegalski, M. D., Vasudevarao, A., Sharan,
[87] Wang, J., and Zhang, T.-Y., 2008, “Phase Field Simulations of a Permeable A., Ma, H. Z., Levy, J., Gopalan, V., Trolier-McKinstry, S., Schlom, D. G., Jia,
Crack Parallel to the Original Polarization Direction in a Ferroelectric Mono- Q. X., and Chen, L. Q., 2006, “Phase Transitions and Domain Structures in
domain,” Eng. Fract. Mech., 75, pp. 4886–4897. Strained Pseudocubic (100) SrTiO3 Thin Films,” Phys. Rev. B, 73, p. 184112.
[88] Zhang, Y. H., Xu, R., Liu, B., and Fang, D. N., 2012, “An Electromechanical [122] Liu, P.-L., Wang, J., Zhang, T.-Y., Li, Y., Chen, L.-Q., Ma, X.-Q., Chu, W.-
Atomic-Scale Finite Element Method for Simulating Evolutions of Ferroelec- Y., and Qiao, L.-J., 2008, “Effects of Unequally Biaxial Misfit Strains on
tric Nanodomains,” J. Mech. Phys. Solids, 60, pp. 1383–1399. Polarization Phase Diagrams in Embedded Ferroelectric Thin Layers: Phase
[89] Wang, X., and Shen, Y., 1995, “Some Basic Theory for Thermal Magnetic Field Simulations,” Appl. Phys. Lett., 93, p. 132908.
Electric Elastic Media,” Chinese J. Appl. Mech., I2(2), pp. 28–39 (in [123] Sheng, G., Zhang, J. X., Li, Y. L., Choudhury, S., Jia, Q. X., Liu, Z. K., and
Chinese). Chen, L. Q., 2008, “Misfit Strain-Misfit Strain Diagram of Epitaxial BaTiO(3)
[90] Sosa, H., 1991, “Plane Problem in Piezoelectric Media With Defects,” Int. J. Thin Films: Thermodynamic Calculations and Phase-Field Simulations,”
Solids Struct., 28(4), pp. 491–505. Appl. Phys. Lett., 93, p. 232904.
[91] Ghandi, K., and Hagood, N. W., 1996, “Nonlinear Finite Element Modeling of [124] Wang, J. J., Wu, P. P., Ma, X. Q., and Chen, L. Q., 2010, “Temperature-Pres-
Phase Transitions in Electro-mechanically Coupled Material,” Proc. SPIE, sure Phase Diagram and Ferroelectric Properties of BaTiO(3) Single Crystal
2715, pp. 121–140. Based on a Modified Landau Potential,” J. Appl. Phys., 108, p. 114105.
[92] Ghandi, K., and Hagood, N. W., 1997, “A Hybrid Finite-Element Model for [125] Wang, J., and Kamlah, M., 2009, “Three-Dimensional Finite Element Model-
Phase Transition in Nonlinear Electro-mechanically Coupled Material,” Proc. ing of Polarization Switching in a Ferroelectric Single Domain With an Imper-
SPIE, 3039, pp. 97–112. meable Notch,” Smart Mater. Struct. 18, p. 104008.
[93] Chen, W., and Lynch, C. S., 2001, “Multiaxial Constitutive Behavior of Fer- [126] Xu, B.-X., Schrade, D., Gross, D., and Mueller, R., 2010, “Phase Field Simula-
roelastic Materials,” ASME J. Eng. Mater. Technol., 123, pp. 169–175. tion of Domain Structures in Cracked Ferroelectrics,” Int. J. Fract., 165, pp.
[94] Fett, T., and Munz, D., 2003, “Deformation of PZT Under Tension, Compres- 163–173.
sion, Bending, and Torsion Loading,” Adv. Eng. Mater., 5, pp. 718–722. [127] Miehe, C., Welschinger, F., and Hofacker, M., 2010, “A Phase Field Model of
[95] Li, F. X., and Fang, D. N., 2005, “Effects of Lateral Stress on the Electrome- Electromechanical Fracture,” J. Mech. Phys. Solids, 58, pp. 1716–1740.
chanical Response of Ferroelectric Ceramics: Experiments Versus Model,” J. [128] Hong, L., Soh, A. K., Song, Y. C., and Lim, L. C., 2008, “Interface and Sur-
Intell. Mater. Syst. Struct., 16(7–8), pp. 583–588. face Effects on Ferroelectric Nano-thin Films,” Acta Mater., 56, pp.
[96] Steinkopff, T., 1999, “Finite-Element Modeling of Ferroic Domain Switching 2966–2974.
in Piezoelectric Ceramics,” J. Eur. Ceram. Soc., 19, pp. 1247–1249. [129] Wang, J., Kamlah, M., Zhang, T. Y., Li, Y., and Chen, L. Q., 2008, “Size-De-
[97] Zeng, W., Manzari, M. T., Lee, J. D., and Shen, Y. L., 2003, “Fully Coupled pendent Polarization Distribution in Ferroelectric Nanostructures: Phase Field
Non-linear Analysis of Piezoelectric Solids Involving Domain Switching,” Int. Simulations,” Appl. Phys. Lett., 92, p. 162905.
J. Numer. Meth. Eng., 56, pp. 13–34. [130] Wang, J., Kamlah, M., and Zhang, T.-Y., 2009, “Phase Field Simulations of
[98] Kamlah, M., Liskowsky, A. C., McMeeking, R. M., and Balke, H., 2005, “Finite Ferroelectric Nanoparticles With Different Long-Range-Electrostatic and -
Element Simulations of a Polycrystalline Ferroelectric Based on a Multidomain Elastic Interactions,” J. Appl. Phys., 105, p. 014104.
Single Crystal Switching Model,” Int. J. Solids Struct., 42, pp. 2949–2964. [131] Li, Y. L., Hu, S. Y., Tenne, D., Soukiassian, A., Schlom, D. G., Chen, L. Q.,
[99] Fang, D. N., and Li, C. Q., 1999, “Nonlinear Electric-Mechanical Behavior of Xi, X. X., Choi, K. J., Eom, C. B., Saxena, A., Lookman, T., and Jia, Q. X.,
a Soft PZT-51 Ferroelectric Ceramic,” J. Mater. Sci., 34, pp. 4001–4010. 2007, “Interfacial Coherency and Ferroelectricity of BaTiO(3)/SrTiO(3)
[100] Axelsson, O., 1994, Iterative Solution Methods, Cambridge University, Cam- Superlattice Films,” Appl. Phys. Lett., 91, p. 252904.
bridge, UK. [132] Xiao, Y., Shenoy, V. B., and Bhattacharya, K., 2005, “Depletion Layers and
[101] Crisfield, M. A., 1983, “An Arc-Length Method Including Line Searches and Domain Walls in Semiconducting Ferroelectric Thin Films,” Phys. Rev. Lett.,
Accelerations,” Int. J. Numer. Meth. Eng., 19, pp. 1269–1289. 95, p. 247603.
[102] Landis, C. M., 2002, “A New Finite-Element Formulation for Electromechani- [133] Xiao, Y., and Bhattacharya, K., 2008, “A Continuum Theory of Deformable,
cal Boundary Value Problems,” Int. J. Numer. Meth. Eng., 55, pp. 613–628. Semiconducting Ferroelectrics,” Arch. Ration. Mech. Anal., 189, pp. 59–95.
[103] Matthies, H., and Strang, G., 1979, “The Solution of Non-linear Finite Ele- [134] Hong, L., Soh, A. K., Du, Q. G., and Li, J. Y., 2008, “Interaction of O Vacan-
ment Equations,” Int. J. Numer. Meth. Eng., 14, pp. 1613–1626. cies and Domain Structures in Single Crystal BaTiO3: Two-Dimensional Fer-
[104] Wempner, G. A., 1971, “Discrete Approximations Related to Nonlinear Theo- roelectric Model,” Phys. Rev. B, 77, p. 094104.
ries of Solids,” Int. J. Solids Struct., 7, pp. 1581–1599. [135] Ren, X. B., 2004, “Large Electric-Field-Induced Strain in Ferroelectric Crys-
[105] Riks, E., 1972, “The Application of Newtons Method to the Problem of Elastic tals by Point-Defect-Mediated Reversible Domain Switching,” Nature Mater.,
Instability,” J. Appl. Mech., 39, pp. 1060–1066. 3, pp. 91–94.
[106] Kessler, H., and Balke, H., 2001, “On the Local and Average Energy Release [136] Zhang, L. X., and Ren, X., 2005, “In Situ Observation of Reversible Domain
in Polarization Switching Phenomena,” J. Mech. Phys. Solids, 49, pp. Switching in Aged Mn-Doped BaTiO3 Single Crystals,” Phys. Rev. B, 71, p.
953–978. 174108.
[107] Lucy, L. B., 1977, “A Numerical Approach to the Fission Hypothesis,” J. [137] Shu, Y. C., and Bhattacharya, K., 2001, “Domain Patterns and Macroscopic
Astron., 8(12), pp. 1013–1024. Behaviour of Ferroelectric Materials,” Philos. Mag. B, 81, pp. 2021–2054.
[108] Monaghan, J. J., 1992, “Smoothed Particle Hydrodynamics,” Ann. Rev. [138] Devonshire, A. F., 1954, “Theory of Ferroelectrics,” Adv. Phys., 3, pp.
Astron. Astrophys, 30, pp. 543–574. 85–130.
[109] Nayroles, B., Touzot, G., and Villion, P., 1992, “Generalizing the Finite Ele- [139] Pertsev, N. A., Zembilgotov, A. G., and Tagantsev, A. K., 1998, “Effect of
ment Method: Diffuse Approximation and Diffuse Element,” Comput. Mech., Mechanical Boundary Conditions on Phase Diagrams of Epitaxial Ferroelec-
10, pp. 307–318. tric Thin Films,” Phys. Rev. Lett., 80, pp. 1988–1991.

[140] Koukhar, V. G., Pertsev, N. A., and Waser, R., 2001, “Thermodynamic Theory [172] Berlincourt, D. A., and Krueger, H. H. A., 1959, “Domain Processes in Lead
of Epitaxial Ferroelectric Thin Films With Dense Domain Structures,” Phys. Titanate Zirconate and Barium Titanate Ceramics,” J. Appl. Phys., 30, pp.
Rev. B, 64, p. 214103. 1804–1810.
[141] Shu, Y. C., Yen, J. H., Chen, H. Z., Li, J. Y., and Li, L. J., 2008, “Constrained [173] King-Smith, R. D., and Vanderbilt, D., 1993, “Theory of Polarization of Crys-
Modeling of Domain Patterns in Rhombohedral Ferroelectrics,” Appl. Phys. talline Solids,” Phys. Rev. B, 47, pp. 1651–1654.
Lett., 92, p. 052909. [174] Resta, R., 1993, “Macroscopic Electric Polarization as a Geometric Quantum
[142] Li, J. Y., and Liu, D., 2004, “On Ferroelectric Crystals With Engineered Do- Phase,” EPL, 22, pp. 133–138.
main Configurations,” J. Mech. Phys. Solids, 52, pp. 1719–1742. [175] King-Smith, R. D., and Vanderbilt, D., 1994, “First-Principles Investigation
[143] Ma, Y. F., and Li, J. Y., 2007, “Magnetization Rotation and Rearrangement of of Ferroelectricity in Perovskite Compounds,” Phys. Rev. B, 49, pp.
Martensite Variants in Ferromagnetic Shape Memory Alloys,” Appl. Phys. 5828–5844.
Lett., 90, p. 172504. [176] Meyer, B., and Vanderbilt, D., 2001, “Ab Initio Study of BaTiO3 and PbTiO3
[144] Li, L. J., Yang, Y., Shu, Y. C., and Li, J. Y., 2010, “Continuum Theory and Surfaces in External Electric Fields,” Phys. Rev. B, 63, p. 205426.
Phase-Field Simulation of Magnetoelectric Effects in Multiferroic Bismuth [177] Hong, J. W., Catalan, G., Fang, D. N., Artacho, E., and Scott, J. F., 2010,
Ferrite,” J. Mech. Phys. Solids, 58, pp. 1613–1627. “Topology of the Polarization Field in Ferroelectric Nanowires From First
[145] Hu, H. L., and Chen, L. Q., 1997, “Computer Simulation of 90 deg Ferroelec- Principles,” Phys. Rev. B, 81, p. 172101.
tric Domain Formation in Two-Dimensions,” Mater. Sci. Eng. A, 238, pp. [178] Neaton, J. B., and Rabe, K. M., 2003, “Theory of Polarization Enhancement in
182–191. Epitaxial BaTiO3/SrTiO3 Superlattices,” Appl. Phys. Lett., 82, pp. 1586–1588.
[146] Rabe, K. M., Ahn, C. H., and Triscone, J. M., 2007, Physics of Ferroelectrics: [179] Bousquet, E., Dawber, M., Stucki, N., Lichtensteiger, C., Hermet, P., Gariglio,
A Modern Perspective, Springer, Heidelberg. S., Triscone, J.-M., and Ghosez, P., 2008, “Improper Ferroelectricity in Perov-
[147] Zhang, L. X., and Ren, X. B., 2006, “Aging Behavior in Single-Domain Mn- skite Oxide Artificial Superlattices,” Nature, 452, pp. 732–736.
Doped BaTiO3 Crystals: Implication for a Unified Microscopic Explanation [180] Hao, F., Hong, J., and Fang, D., 2011, “Size Effect of Elastic and Electrome-
of Ferroelectric Aging,” Phys. Rev. B, 73, p. 094121. chanical Properties of BaTiO3 Films From First-Principles Method,” Integr.
[148] Zhang, L. X., Erdem, E., Ren, X. B., and Eichel, R. A., 2008, “Reorientation Ferroelectrics, 124, pp. 79–86.
of (Mn-Ti(’)-V-O(Center Dot Center Dot))(x) Defect Dipoles in Acceptor- [181] Spanier, J. E., Kolpak, A. M., Urban, J. J., Grinberg, I., Lian, O. Y., Yun, W.
Modified BaTiO3 Single Crystals: An Electron Paramagnetic Resonance S., Rappe, A. M., and Park, H., 2006, “Ferroelectric Phase Transition in Indi-
Study,” Appl. Phys. Lett., 93, p. 202901. vidual Single-Crystalline BaTiO3 Nanowires,” Nano Lett., 6, pp. 735–739.
[149] Erhart, P., Eichel, R. A., Traskelin, P., and Albe, K., 2007, “Association of [182] Fu, H., and Bellaiche, L., 2003, “Ferroelectricity in Barium Titanate Quantum
Oxygen Vacancies With Impurity Metal Ions in Lead Titanate,” Phys. Rev. B, Dots and Wires,” Phys. Rev. Lett., 91, p. 257601.
76, p. 174116. [183] Benedek, N. A., and Fennie, C. J., 2011, “Hybrid Improper Ferroelectricity: A
[150] Hooton, J. A., and Merz, W. J., 1955, “Etch Patterns and Ferroelectric Mechanism for Controllable Polarization-Magnetization Coupling,” Phys.
Domains in Batio3 Single Crystals,” Phys. Rev., 98, pp. 409–413. Rev. Lett., 106, p. 107204.
[151] Zhong, W. L., Wang, Y. G., Zhang, P. L., and Qu, B. D., 1994, [184] Martin, R. M., 1974, “Comment on Calculations of Electric Polarization in
“Phenomenological Study of the Size Effect on Phase-Transitions in Ferro- Crystals,” Phys. Rev. B, 9, pp. 1998–1999.
electric Particles,” Phys. Rev. B, 50, pp. 698–703. [185] Vanderbilt, D., and King-Smith, R. D., 1993, “Electric Polarization as a Bulk
[152] Sang, Y. L., Liu, B., and Fang, D. N., 2008, “The Size and Strain Effects on Quantity and Its Relation to Surface Charge,” Phys. Rev. B, 48, pp.
the Electric-Field-Induced Domain Evolution and Hysteresis Loop in Ferro- 4442–4455.
electric BaTiO3 Nanofilms,” Comp. Mater. Sci., 44, pp. 404–410. [186] Resta, R., 1994, “Macroscopic Polarization in Crystalline Dielectrics: The
[153] Asada, T., and Koyama, Y., 2007, “Ferroelectric Domain Structures Around Geometric Phase Approach,” Rev. Mod. Phys., 66, pp. 899–915.
the Morphotropic Phase Boundary of the Piezoelectric Material PbZr1- [187] Resta, R., and Vanderbilt, D., 2007, “Theory of Polarization: A Modern
xTixO3,” Phys. Rev. B, 75, p. 214111. Approach,” Physics of Ferroelectrics, Topics in Applied Physics, Springer,
[154] Lupascu, D. C., 2004, Fatigue in Ferroelectric Ceramics and Related Issues, Berlin/Heidelberg, pp. 31–68.
Springer-Verlag, Berlin. [188] Spaldin, N. A., “A Beginner’s Guide to the Modern Theory of Polarization,” J.
[155] Eshelby, J. D., 1957, “The Determination of the Elastic Field of an Ellipsoidal Solid State Chem. (in press).
Inclusion and Related Problems,” Proc. R. Soc. A, 241, pp. 376–396. [189] Zhong, W., King-Smith, R. D., and Vanderbilt, D., 1994, “Giant LO-TO
[156] Uchida, N., and Ikeda, T., 1967, “Electrostriction in Perovskite-Type Ferro- Splittings in Perovskite Ferroelectrics,” Phys. Rev. Lett., 72, pp. 3618–3621.
electric Ceramics,” Jpn. J. Appl. Phys., 6, pp. 1079–1088. [190] Hong, J., and Fang, D., 2008, “Size-Dependent Ferroelectric Behaviors of
[157] Huber, J. E., Fleck, N. A., and McMeeking, R. M., 1999, “A Crystal Plasticity BaTiO3 Nanowires,” Appl. Phys. Lett., 92, p. 012906.
Model for Ferroelectrics,” Ferroelectrics, 228, pp. 39–52. [191] Hong, J., and Fang, D., 2008, “Systematic Study of the Ferroelectric Properties
[158] Li, F. X., Fang, D. N., and Soh, A. K., 2004, “An Analytical Axisymmetric of Pb(Zr0.5Ti0.5)O3 Nanowires,” J. Appl. Phys., 104, p. 064118.
Model for the Poling-History Dependent Behavior of Ferroelectric Ceramics,” [192] Wang, Z. Y., Hu, J., and Yu, M. F., 2006, “One-Dimensional Ferroelectric
Smart Mater. Struct., 13, pp. 668–675. Monodomain Formation in Single Crystalline BaTiO3 Nanowire,” Appl. Phys.
[159] Han, S. P., 1977, “A Global Convergent Method for Nonlinear Programming,” Lett., 89, p. 263119.
J. Optim. Theory Appl., 22, pp. 297–309. [193] Mermin, N. D., 1979, “The Topological Theory of Defects in Ordered Media,”
[160] Bunge, H. J., 1982, Texture Analysis in Materials Science, Butterworth, Rev. Mod. Phys., 51, pp. 591–648.
Berlin. [194] Kogan, S. M., 1964, “Piezoelectric Effect During Inhomogeneous Deforma-
[161] Li, F. X., and Rajapakse, R. K. N. D., 2007, “Analytically Saturated Domain tion and Acoustic Scattering of Carriers in Crystals,” Sov. Phys. Solid State, 5,
Orientation Textures and Electromechanical Properties of Ferroelectric pp. 2069–2070.
Ceramics Due to Electric/Mechanical Loading,” J. Appl. Phys., 101, p. [195] Tagantsev, A. K., 1986, “Piezoelectricity and Flexoelectricity in Crystalline
054110. Dielectrics,” Phys. Rev. B, 34, pp. 5883–5889.
[162] Hoffmann, M. J., Hammer, M., Endriss, A., and Lupascu, D. C., 2001, [196] Cross, L. E., 2006, “Flexoelectric Effects: Charge Separation in Insulating Sol-
“Correlation Between Microstructure, Strain Behavior, and Acoustic Emission ids Subjected to Elastic Strain Gradients,” J. Mater. Sci., 41, pp. 53–63.
of Soft PZT Ceramics,” Acta Mater., 49, pp. 1301–1310. [197] Catalan, G., Noheda, B., McAneney, J., Sinnamon, L., and Gregg, J.,
[163] Li, F. X., and Zhou, X. L., 2011, “Simulations of Gradual Domain-Switching 2005, “Strain Gradients in Epitaxial Ferroelectrics,” Phys. Rev. B, 72, p.
in Polycrystalline Ferroelectrics Using an Optimization-Based, Multidomain- 020102.
Grain Model,” Comput. Struct., 89, pp. 1142–1147. [198] Zubko, P., Catalan, G., Buckley, A., Welche, P. R. L., and Scott, J. F., 2007,
€ undag, E., and Bhattacharya, K., 2005, “Domain
[164] Li, J. Y., Rogan, R. C., Ust€ “Strain-Gradient-Induced Polarization in SrTiO3 Single Crystals,” Phys. Rev.
Switching in Polycrystalline Ferroelectric Ceramics,” Nature Mater., 4, pp. Lett., 99, p. 167601.
776–781. [199] Ma, W., 2008, “A Study of Flexoelectric Coupling Associated Internal Elec-
[165] Webber, K. G., Aulbach, E., Key, T., Marsilius, M., Granzow, T., and R€ odel, tric Field and Stress in Thin Film Ferroelectrics,” Physica Status Solidi B,
J., 2009, “Temperature-Dependent Ferroelastic Switching of Soft Lead Zir- 245, pp. 761–768.
conate Titanate,” Acta Mater., 57, pp. 4614–4623. [200] Hong, J., Catalan, G., Scott, J. F., and Artacho, E., 2010, “The Flexoelectricity
[166] Li, Y. W., Zhou, X. L., and Li, F. X., 2010, “Temperature Dependent Mechan- of Barium and Strontium Titanates From First Principles,” J. Phys.: Condens.
ical Depolarization of Ferroelectric Ceramics,” J. Phys. D: Appl. Phys., 43, p. Matter, 22, p. 112201.
175501. [201] Resta, R., 2010, “Towards a Bulk Theory of Flexoelectricity,” Phys. Rev.
[167] Taylor, G. I., 1938, “Plastic Strain in Metals,” J. Inst. Met., 62, pp. 307–324. Lett., 105, p. 127601.
[168] Noheda, B., Cox, D. E., Shirane, G., Guo, R., Jones, B., and Cross, L. E., [202] Hong, J., and Vanderbilt, D., 2011, “First-Principles Theory of Frozen-Ion
2001, “Stability of the Monoclinic Phase in the Ferroelectric Perovskite Flexoelectricity,” Phys. Rev. B, 84, p. 180101.
PbZr1-xTixO3,” Phys. Rev. B, 63, p. 014103. [203] Lee, D., Yoon, A., Jang, S. Y., Yoon, J.-G., Chung, J.-S., Kim, M., Scott, J. F.,
[169] Zhou, D. Y., Kamlah, M., and Munz, D., 2005, “Effects of Uniaxial Prestress and Noh, T. W., 2011, “Giant Flexoelectric Effect in Ferroelectric Epitaxial
on the Ferroelectric Hysteretic Response of Soft PZT,” J. Eur. Ceram. Soc., Thin Films,” Phys. Rev. Lett., 107, p. 057602.
25, pp. 425–432. [204] Catalan, G., Lubk, A., Vlooswijk, A. H. G., Snoeck, E., Magen, C., Janssens,
[170] Li, Y. W., and Li, F. X., 2010, “Large Anisotropy of Fracture Toughness in A., Rispens, G., Rijnders, G., Blank, D. H. A., and Noheda, B., 2011,
Mechanically Poled/Depoled Ferroelectric Ceramics,” Script. Mater., 62, pp. “Flexoelectric Rotation of Polarization in Ferroelectric Thin Films,” Nature
313–316. Mater., 10, pp. 963–967.
[171] Berlincourt, D. A., Cmolik, C., and Jaffe, H., 1960, “Piezoelectric Properties [205] Lu, H., Bark, C.-W., Esque De Los Ojos, D., Alcala, J., Eom, C. B., Catalan,
of Polycrystalline Lead Titanate Zirconate Compositions,” Proc. Inst. Radio G., and Gruverman, A., 2012, “Mechanical Writing of Ferroelectric Polar-
Eng., 48, pp. 220–229. ization,” Science, 336, pp. 59–61.

[206] Zhou, H., Hong, J., Zhang, Y., Li, F., Pei, Y., and Fang, D., 2012, [226] Angoshtari, A., and Yavari, A., 2011, “Convergence Analysis of the Wolf
“Flexoelectricity Induced Increase of Critical Thickness in Epitaxial Ferro- Method for Coulombic Interactions,” Phys. Lett. A, 375, pp. 1281–1285.
electric Thin Films,” Physica B: Condens. Matter, 407, pp. 3377–3381. [227] Ewald, P. P., 1921, “Die Berechnung Optischer und Elektrostatischer
[207] Zhou, H., Hong, J., Zhang, Y., Li, F., Pei, Y., and Fang, D., 2012, “External Gitterpotentiale,” Ann. Phys., 369, pp. 253–287.
Uniform Electric Field Removing the Flexoelectric Effect in Epitaxial Ferro- [228] Deleeuw, S. W., Perram, J. W., and Smith, E. R., 1980, “Simulation of Elec-
electric Thin Films,” EPL, 99, p. 47003. trostatic Systems in Periodic Boundary-Conditions 1. Lattice Sums and
[208] Maranganti, R., and Sharma, P., 2009, “Atomistic Determination of Flexoelec- Dielectric-Constants,” Proc. R. Soc. A, 373, pp. 27–56.
tric Properties of Crystalline Dielectrics,” Phys. Rev. B, 80, p. 054109. [229] Padilla, J., and Vanderbilt, D., 1997, “Ab Initio Study of BaTiO3 Surfaces,”
[209] Ma, W., and Cross, L. E., 2006, “Flexoelectricity of Barium Titanate,” Appl. Phys. Rev. B, 56, pp. 1625–1631.
Phys. Lett., 88, p. 232902. [230] Strukov, B. A., Davitadze, S. T., Taraskin, S. A., Goltzman, B. M., Shulman,
[210] Parker, C. B., Maria, J.-P., and Kingon, A. I., 2002, “Temperature and Thick- S. G., and Lemanov, V. V., 2003, “Thermodynamical Properties of the Thin
ness Dependent Permittivity of (Ba,Sr)TiO3 Thin Films,” Appl. Phys. Lett., Polycrystalline BaTiO3 Films on Substrates,” Ferroelectrics, 286, pp.
81, pp. 340–342. 967–972.
[211] Sinnamon, L. J., Bowman, R. M., and Gregg, J. M., 2002, “Thickness-Induced [231] Drezner, Y., and Berger, S., 2005, “Thermodynamic Stability of BaTiO3
Stabilization of Ferroelectricity in SrRuO3/Ba0.5Sr0.5TiO3/Au Thin Film Nano-domains,” Mater. Lett., 59, pp. 1598–1602.
Capacitors,” Appl. Phys. Lett., 81, pp. 889–891. [232] Zhang, Y. H., Hong, J. W., Liu, B., and Fang, D. N., 2010, “A Surface-Layer
[212] Ramirez, F., Heyliger, P. R., and Pan, E., 2006, “Discrete Layer Solution to Model of Ferroelectric Nanowire,” J. Appl. Phys., 108, p. 124109.
Free Vibrations of Functionally Graded Magneto-Electro-Elastic Plates,” [233] Liu, B., Huang, Y., Jiang, H., Qu, S., and Hwang, K. C., 2004, “The Atomic-
Mech. Adv. Mater. Struct., 13, pp. 249–266. Scale Finite Element Method,” Comput. Mech. Appl. Mech. Eng., 193, pp.
[213] Ahn, C. H., Rabe, K. M., and Triscone, J.-M., 2004, “Ferroelectricity at the 1849–1864.
Nanoscale: Local Polarization in Oxide Thin Films and Heterostructures,” Sci- [234] Liu, B., Jiang, H., Huang, Y., Qu, S., Yu, M. F., and Hwang, K. C., 2005,
ence, 303, pp. 488–491. “Atomic-Scale Finite Element Method in Multiscale Computation With Appli-
[214] Fong, D. D., Stephenson, G. B., Streiffer, S. K., Eastman, J. A., Auciello, O., cations to Carbon Nanotubes,” Phys. Rev. B, 72, p. 035435.
Fuoss, P. H., and Thompson, C., 2004, “Ferroelectricity in Ultrathin Perov- [235] Zhang, X., Hashimoto, T., and Joy, D. C., 1992, “Electron Holographic Study
skite Films,” Science, 304, pp. 1650–1653. of Ferroelectric Domain Walls,” Appl. Phys. Lett., 60, pp. 784–786.
[215] Lee, H. N., Christen, H. M., Chisholm, M. F., Rouleau, C. M., and Lowndes, [236] Merz, W. J., 1954, “Domain Formation and Domain Wall Motions in Ferro-
D. H., 2005, “Strong Polarization Enhancement in Asymmetric Three- electric BaTiO3 Single Crystals,” Phys. Rev., 95, pp. 690–698.
Component Ferroelectric Superlattices,” Nature, 433, pp. 395–399. [237] Floquet, N., and Valot, C., 1999, “Ferroelectric Domain Walls in BaTiO3:
[216] Zhu, X. H., Evans, P. R., Byrne, D., Schilling, A., Douglas, C., Pollard, R. J., Structural Wall Model Interpreting Fingerprints in XRPD Diagrams,” Ferro-
Bowman, R. M., Gregg, J. M., Morrison, F. D., and Scott, J. F., 2006, electrics, 234, pp. 107–122.
“Perovskite Lead Zirconium Titanate Nanorings: Towards Nanoscale Ferro- [238] Floquet, N., Valot, C. M., Mesnier, M. T., Niepce, J. C., Normand, L., Thorel,
electric ‘Solenoids’?” Appl. Phys. Lett., 89, p. 129913. A., and Kilaas, R., 1997, “Ferroelectric Domain Walls in BaTiO3: Finger-
[217] Sepliarsky, M., Stachiotti, M. G., and Migoni, R. L., 1995, “Structural Insta- prints in XRPD Diagrams and Quantitative HRTEM Image Analysis,” J. Phys.
bilities in KTaO3 and KNbO3 Described by the Nonlinear Oxygen Polariz- III, 7, pp. 1105–1128.
ability Model,” Phys. Rev. B, 52, pp. 4044–4049. [239] Hlinka, J., and Marton, P., 2006, “Phenomenological Model of a 90 deg Do-
[218] Sepliarsky, M., Stachiotti, M. G., and Migoni, R. L., 1997, “Ferroelectric Soft main Wall in BaTiO3-Type Ferroelectrics,” Phys. Rev. B, 74, p. 104104.
Mode and Relaxation Behavior in a Molecular-Dynamics Simulation of [240] Tsou, N. T., Potnis, P. R., and Huber, J. E., 2011, “Classification of Laminate
KNbO3 and KTaO3,” Phys. Rev. B, 56, pp. 566–571. Domain Patterns in Ferroelectrics,” Phys. Rev. B, 83, p. 184120.
[219] Sepliarsky, M., Stachiotti, M. G., and Migoni, R. L., 2005, “Surface Recon- [241] Pilania, G., and Ramprasad, R., 2010, “Complex Polarization Ordering in
struction and Ferroelectricity in PbTiO3 Thin Films,” Phys. Rev. B, 72, p. PbTiO3 Nanowires: A First-Principles Computational Study,” Phys. Rev. B,
014110. 82, p. 155442.
[220] Tinte, S., and Stachiotti, M. G., 2001, “Surface Effects and Ferroelectric Phase [242] Jiang, B., Bai, Y., Chu, W. Y., Su, Y. J., and Qiao, L. J., 2008, “Direct Obser-
Transitions in BaTiO3 Ultrathin Films,” Phys. Rev. B, 64, p. 235403. vation of Two 90 Degrees Steps of 180 Degrees Domain Switching in BaTiO3
[221] Sepliarsky, M., Phillpot, S. R., Stachiotti, M. G., and Migoni, R. L., 2002, Single Crystal Under an Antiparallel Electric Field,” Appl. Phys. Lett., 93, p.
“Ferroelectric Phase Transitions and Dynamical Behavior in KNbO3/KTaO3 Super- 152905.
lattices by Molecular-Dynamics Simulation,” J. Appl. Phys., 91, pp. 3165–3171. [243] Dieguez, O., and Vanderbilt, D., 2006, “First-Principles Calculations for Insu-
[222] Sepliarsky, M., and Tinte, S., 2009, “Dynamical Behavior of the Phase Transi- lators at Constant Polarization”, Phys. Rev. Lett., 96, p. 056401.
tion of Strained BaTiO(3) From Atomistic Simulations,” Physica B, 404, pp. [244] Stengel, M., Spaldin, N. A., and Vanderbilt, D., 2009, “Electric Displacement
2730–2732. as the Fundamental Variable in Electronic-Structure Calculations,” Nature
[223] Sang, Y.-L., Liu, B., and Fang, D.-N., 2008, “Strain and Size Effects on Ferro- Phys., 5, pp. 304–308.
electric Properties of BaTiO3 Nanofilms,” Chin. Phys. Lett., 25, pp. 1113–1116. [245] Hong, J., and Vanderbilt, D., 2011, “Mapping the Energy Surface of PbTiO3
[224] Tinte, S., Stachiotti, M. G., Phillpot, S. R., Sepliarsky, M., Wolf, D., and in Multidimensional Electric-Displacement Space,” Phys. Rev. B, 84, p.
Migoni, R. L., 2004, “Ferroelectric Properties of BaxSr1-xTiO3 Solid Solu- 115107.
tions Obtained by Molecular Dynamics Simulation,” J. Phys.: Condens. Mat- [246] Hong, J., and Vanderbilt, D., 2013, “Electrically Driven Octahedral Rotations
ter, 16, pp. 3495–3506. in SrTiO3 and PbTiO3,” Phys. Rev. B, 87, p. 064104.
[225] Wolf, D., Keblinski, P., Phillpot, S. R., and Eggebrecht, J., 1999, “Exact [247] McCash, K., Srikanth, A., and Ponomareva, I., 2012, “Competing Polarization
Method for the Simulation of Coulombic Systems by Spherically Truncated, Reversal Mechanisms in Ferroelectric Nanowires,” Phys. Rev. B, 86, p.
Pairwise r(-1) Summation,” J. Chem. Phys., 110, pp. 8254–8282. 214108.
DownloadedViewFrom:
publicationhttp://appliedmechanicsreviews.asmedigitalcollection.asme.org/
stats on 01/06/2014 Terms of Use: http://asme.org/terms

Amr2013 065 06 060802

Uploaded by

Copyright:

Available Formats

Amr2013 065 06 060802

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Amr2013 065 06 060802

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Advances in Developing Electromechanically Coupled Computational

Article in Applied Mechanics Reviews · October 2013

Faxin Li Bin Liu

SEE PROFILE SEE PROFILE

Yihui Zhang Jiawang Hong

SEE PROFILE SEE PROFILE

Flexoelectricity View project

3D structure View project

The user has requested enhancement of the downloaded file.

1 Introduction electric field. The ferroelectricity was first discovered in 1920 in

Applied Mechanics Reviews Copyright V

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

060802-2 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-3

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

060802-4 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-5

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

060802-6 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-7

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

where N is the interpolation matrix. The general strain of the ele-

DC ¼ LDu ¼ LNDae ¼ BDae (2.2.10)

where L is the linear differential factor matrix and B is the matrix

where G is the matrix that assembles the element stiffness matrix

060802-8 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-9

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

2.2.3 Domain-Switching–Induced Instabilities in Nonlinear

060802-10 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

2.3 Electromechanically Coupled Meshless Method.

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-11

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

060802-12 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

where K and f 0 are made up of 2  2 submatrix kIJ and 2  1 sub-

where Fig. 14 An infinite PZT-5 H plate with circular hole

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-13

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Table 3 Material parameters of PZT-4 and PZT-5H ceramics

060802-14 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-15

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

060802-16 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-17

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Note that the value of P(i),o can be calculated by first-principles

060802-18 / Vol. 65, NOVEMBER 2013 Transactions of the ASME

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

Applied Mechanics Reviews NOVEMBER 2013, Vol. 65 / 060802-19

Downloaded From: http://appliedmechanicsreviews.asmedigitalcollection.asme.org/ on 01/06/2014 Terms of Use: http://asme.org/terms

where K and f 0 are made up of 2 2 submatrix kIJ and 2 1 sub-