Acta Materialia 54 (2006) 21092120

www.actamat-journals.com

Dynamic strain loading of cubic to tetragonal martensites

Rajeev Ahluwalia, Turab Lookman *, Avadh Saxena
Theoretical Division and Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
Received 24 June 2005; received in revised form 23 December 2005; accepted 29 December 2005
Available online 13 March 2006

Abstract
We present three-dimensional simulations of the microstructure and mechanical response of shape memory alloys undergoing cubic to
tetragonal transitions, using FePd as an example. The simulations are based on a nonlinear elastic free-energy in terms of the appropriate
strain elds. The dynamics is simulated by force balance equations for the displacement elds with a damping term derived from a dissipation function. Stressstrain properties in the pseudoelastic as well as the shape memory regime are investigated using strain loading.
We also study the eects of defect-induced heterogeneous nucleation and motion of twin boundaries during deformation. Thus, we probe
the inuence of the microstructure on the mechanical response and investigate how the stressstrain behavior changes as a function of
strain rate.
 2006 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Shape memory alloys; Martensitic phase transformation; Microstructure; Modelling; Nucleation

1. Introduction
Shape memory alloys exhibit interesting mechanical
properties due to a diusionless structural phase transformation from a high temperature austenite phase (e.g.,
cubic) to a low temperature martensite phase that is often
tetragonal, orthorhombic or monoclinic in structure. This
transition is usually rst order and is accompanied by a
spontaneous strain. The martensitic transition is also
responsible for the shape memory eect, which refers to
the recovery by heating of an apparently permanent deformation undergone below a critical temperature [1]. This
property makes shape memory materials suitable for a
large number of technological applications [2]. Another
important property of shape memory materials is the socalled pseudoelastic behavior which arises due to a reversible stress/strain-induced martensitic transformation at a
temperature that is higher than the austenite nish temperature of the material. In pseudoelastic deformation, a high
temperature cubic austenite phase typically transforms to
*

Corresponding author. Tel.: +1 505 665 0419.

E-mail address: txl@lanl.gov (T. Lookman).

the martensite under deformation so that there is a plateau

in the stressstrain curves. On removing the deformation,
the material transforms back to the cubic austenite and
the deformation is recovered upon unloading. In fact, some
shape memory materials can recover strains of up to 10%
under tension, making these materials suitable for actuator
applications [2].
The martensitic transformation results in the formation
of a complex microstructure consisting of twin boundaries
between the crystallographic variants of the transformation. This microstructure inuences the eective mechanical properties of these materials. For example, if a stress
is applied to a material in the martensitic phase, there is
motion of the twin boundaries as the favored variants grow
at the expense of the unfavored variants [1]. Even during
pseudoelastic deformation, the strain-induced transformation involves nucleation and growth of the martensitic variants [2,3]. The motion of the domain walls can inuence the
strain rate dependence of the mechanical response and thus
it is important to incorporate this aspect in a theoretical
framework. In the present paper, we study theoretically
the role of the microstructural evolution on the mechanical
response of shape memory materials.

1359-6454/$30.00  2006 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.actamat.2005.12.040

2110

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

A number of approaches have been used to model shape

memory materials. These include energy minimization
techniques that are subject to certain constraints that
ensure the integrity of the lattice [1] and micro-mechanical
models that require prior knowledge of the habit planes
and their volume fractions [4,5]. Such techniques have been
successful in estimating the recoverable strains for many
martensitic transformations. Over the past few years, continuum models based on the GinzburgLandau approach
have also been used to study the mechanical response of
shape memory alloys [6,7]. The eect of microstructure
on the eective stressstrain response has also been simulated using time-dependent GinzburgLandau approaches
with dynamics that is purely dissipational [810]. However,
these studies do not consider the strain rate dependence of
the stressstrain behavior. In the present paper, we use a
three-dimensional displacement eld based dynamic model
that includes inertial forces to study the eect of defects
and microstructure on the stressstrain properties of shape
memory materials. We also investigate how microstructural evolution inuences the strain rate dependence of
the mechanical response.
The paper is organized as follows. Section 2 describes
the GinzburgLandau free energy for a cubic to tetragonal
transformation and Section 3 analyzes the properties of the
homogeneous part of the free energy. In Section 4, we
introduce a dynamic model for the displacement elds. Section 5 describes the simulation of microstructure during a
temperature-induced cubic to tetragonal transformation
using the dynamic model. In Section 6, we describe our
simulations of stressstrain response in the pseudoelastic
regime and in Section 7 we discuss the mechanical response
in the shape memory regime. Finally, we end the paper in
Section 8 with a summary and discussion of our results.
2. Free energy functional for cubic to tetragonal transition
A free energy functional for cubic to tetragonal transformations was proposed by Barsch and Krumhansl [11]. The
theory is formulated in terms of symmetry adapted combinations of the components of the strain tensor. For the linou
oui
earized strain tensor ij 12 ox
oxji , the tetragonal
j
distortions are described by the deviatoric strains given
by e2 p12 xx  yy and e3 p16 xx yy  2zz . The bulk
strain is e1 p13 xx yy zz and the three shear strains
are denoted by e4 = xy, e5 = xz and e6 = yz. The appropriate free energy functional is then written as
Z
F d~
rftetragonal fgradient fbulk-shear fload ;
1
where ftetragonal represents the free energy cost for a cubic to
tetragonal distortion and is given by
ftetragonal Ae22 e23 Be3 e23  3e22 Ce22 e23 2 .

We note that the above functional form preserves the cubic

symmetry as the strains e2, e3 transform appropriately under the three generators 3[1 1 1], 4[0 0 1] and 2[1 1 0] of the

high symmetry cubic group. The origin of the cubic term

lies in the threefold rotation. If the coecient of the harmonic term changes sign, e2, e3 take on non-zero values,
thus transforming the cubic phase to the tetragonal phase.
The coecient A is the deviatoric modulus expressed in
terms of the elastic constants as A = (C11  C12)/2. This
modulus is temperature dependent and controls the cubic
to tetragonal transition. The quantities B and C are the
appropriate nonlinear elastic constants. As we have discussed earlier, more than one crystallographic variants
are formed as a result of the transition. Thus, there are domain walls between these variants and the energy cost associated with these variants is assumed to be
G
2
2
re2 re3 ;
3
2
where the coecient G can be obtained from phonon dispersion curves or from experimental microstructure. The
transformation can also result in bulk and shear strains
and the free energy cost associated with such deformations
is given by
Abulk
Ashear 2
2
fbulk-shear
e1  E0 e22 e23
e4 e25 e26 .
2
2
4

fgradient

The quantities Abulk and Ashear are the bulk and shear moduli, respectively, that can be expressed in terms of the linear
12
elastic constants. In particular, Abulk C11 2C
and
3
Ashear = 4C44. Notice that the above form ensures that
the transformation described by the deviatoric strains results in non-zero e1 in equilibrium. Thus, the free energy
also incorporates volume changes associated with the
transformation. The volume change depends on the parameter E0 that can be determined from the transformation
strains obtained from the lattice parameter data.
We rst rescale the free energy by introducing strain
variables
i ei =e0 ;

where the condition F = 0 at s = 1 gives e0 = |B|/2C. Since

the transformation is determined by a softening of the
deviatoric elastic constant A, we assume A to be temperature dependent, i.e. A = A0s where A0 = B2/4C is the value
of this elastic constant at the transformation start temperature and s is a dimensionless temperature dened by
s 4AC=B2 T  T c =T 0  T c ;

where T0 is the temperature at which the transformation

starts and Tc is the temperature at which the austenite
phase becomes unstable. The rescaled free energy can be
written as
Fe F =F 0
Z
~rf~ tetragonal f~ gradient f~ bulk-shear f~ load ;
d~
7
where F0 is given by
F 0 d2 B4 =16C 3 d2 A0 e20 .

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

2111

Here d is the length rescaling factor dened by

~
~r.
r d~

With these denitions, the contributions to the free energy

can be expressed in terms of dimensionless variables. The
free energy for tetragonal distortions is expressed as
2
f~ tetragonal s22 23 23 23  322 22 23

10

and the contribution to the strain gradients is now

e
G
~ 2 2 r
~ 3 2 ;
f~ gradient 2 r
11
2d
e G=A0 is the rescaled gradient coecient. Simiwhere G
larly, the contributions to the bulk and shear deformations
take the form
e bulk
e
A
e 0 2 2 2 A shear 2 2 2 ;
1  E
f~ bulk-shear
2
3
4
5
6
2
2
12
e bulk Abulk =A0 , E
e shear Ashear =A0 .
e 0 E0 e2 and A
where A
0

Fig. 1. Stressstrain curves in the pseudoelastic regime calculated from the

homogeneous free energy fhomo in Eq. (13). The thick lines are the result of
stress loading and the thin lines are obtained by strain loading. The black,
red and blue curves correspond to T = 313, 305, 300 K, respectively. (For
interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

3. Analysis of the homogeneous free energy

It is instructive to analyze the homogeneous free energy
(f~ gradient 0) as it provides insight into the nature of the
stressstrain properties. To specify the free energy for the
FePd system, we need the parameters A0, Abulk, Ashear,
E0, e0, T0 and Tc. The transformation strains are calculated
, ac = 3.725 A

from the lattice parameters [12] ao = 3.756 A

at ao
ac ao

and at = 3.795 A using xx yy ao and zz ao .

o o 2o
These allow us to obtain e0 xx pyy zz 0:0152 and
o

  
f0 e1 0:47,
E
where
e01 xx pyy3 zz .
The
values
eO
10
2
A0 = 1.97 10 N/m , Abulk = 19.23 1010 N/m2, Ashear
= 28 1010 N/m2, T0 = 295 K and Tc = 270 K are taken
from temperature-dependent elastic constant data [13].
We then calculate the stressstrain curves for a uniaxial
tensile loading (xx = applied) by minimizing the energy
fhomo with respect to yy and zz, where

fhomo s22 23 B3 23  322 C22 23

e bulk
A
f0 2 2 2
1  E
2
3
2

Fig. 2, we show the stressstrain curves in the shape memory regime. These curves correspond to the situation where
at zero load the material is initially in the austenite state
and the martensite is either a local or global minima. Thus,
upon unloading, these curves exhibit a residual strain.
Notice that the residual strain increases as the temperature
is decreased. We have also calculated the stressstrain
curves by applying an external stress. In this case, an
external stress contribution, rxxxx is added to the energy
and the free energy is minimized to calculate the strains

13

as the shear strains xy, xz, yz are zero. The strains min
yy and
min
represent
the
stress-free
transverse
strains
if
an
external
zz
xx is applied. These stress-free strains are obtained by
numerically solving _ yy  dfdhomo
, _ zz  dfdhomo
so that the
yy
zz
system may be driven to the free energy minima. At convergence, the stress rxx = ofhomo/oxx is calculated at
xx = applied, which is varied from 0 to 0.03. Fig. 1 shows
the calculated stressstrain curves in the pseudoelastic
regime for three dierent temperatures. It is clear that the
stressstrain curves for T = 305 and 300 K exhibit regions
of negative slope corresponding to mechanically unstable
regions. It is also observed that the stress required for the
transition increases as the temperature increases. In

Fig. 2. Stressstrain curves in the shape memory regime calculated from

the homogeneous free energy fhomo in Eq. (13). The black, red and blue
curves correspond to T = 297, 295, 290 K, respectively. (For interpretation of the references to color in this gure legend, the reader is referred to
the web version of this article.)

2112

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

for dierent values of the applied stress. Note that the

stressstrain curves obtained by stress-loading and strain
loading are identical in the stable regions, as seen in
Figs. 1 and 2.
The analysis in this section has shown that the stress
strain curves obtained by minimizing the homogeneous free
energy predict mechanically unstable regions. Clearly, the
behavior of the system in these unstable regions cannot
be described by this simple approach and a dynamic framework that can describe the inhomogeneous microstructure
is essential. In Section 4, we describe this framework which
is used to simulate microstructural evolution in the unstable regions in Sections 57.
4. Dynamics
Dynamic equations for the elastic degrees of freedom
can be written in terms of the displacement elds by appropriate force balance equations [1416]. The advantage of
working with the displacement eld approach is that the
elastic compatibility constraints that inuence the microstructure are automatically satised [17,18]. Explicitly,
$

$
q~
u r  r r  r0 .

14

Here, the elastic stresses are obtained from the free-energy

in Eq. (1) as
X dF dek
15
rij
dek dij
k
$

and r0 represents the dissipative part of the stress tensor

that is determined from a dissipation functional R by [16]
X dR d_ek
.
16
r0ij
d_ek d_ij
k
The dissipation functional R is considered to be of the form
 0

Z
A
A0
A0
R d~
r 1 e_ 21 2 _e22 e_ 23 3 _e24 e_ 25 e_ 26 ;
17
2
2
2
where A01 ; A02 ; A03 are the damping coecients. For simplicity, we choose A01 A02 A03 g and the equations of
motion for the rescaled displacement variables ~ui
e0 dui i 1; 2; 3 become
o2 ~ux o~
rxx o~
rxy o~
rxz

2
~
o~x
o~y
o~z
ot
o2 ~uy o~
rxy o~
ryy o~
ryz

2
o~x
o~y
o~z
o~t

18

o2 ~uz o~
rxz o~
ryz o~
rzz

o~x
o~y
o~z
o~t2
where ~x; ~y ; ~z are rescaled space variables given by x d~x,
y d~
y
and z d~z. The rescaled time variable ~t is written as
q
t ~t qd2 =A0 . The total (elastic + dissipative) rescaled
stress elds can be determined from the elastic free energy
and the dissipation functional using

G1
G2 G3
~xx p p p ;
r
3
6
2
G1 G2 G3
~yy p  p p ;
r
3
6
2
G1 2G3
~zz p  p ;
r
3
6
~yz G4 ;
r
~xz G5 ;
r
~xy G6 ;
r

19

where
e bulk 1  E
e 0 2 2 cm o1 ;
G1 A
2
3
o~t
2
2
G2 2s2  122 3 42 2 3
e
e bulk 1  E
e 0 2 A
e 0 2 2  G r~2 2 cm o2 ;
 2E
2
3
o~t
d2
2
2
2
2
G3 2s3  63  2 43 2 3
e
e bulk 1  E
e 0 3 A
e 0 2 2  G r~2 3 cm o3 ;
 2E
2
3
o~t
d2
o
e shear 4 cm 4 ;
G4 A
o~t
e shear 5 cm o5 ;
G5 A
o~t
o
e shear 6 cm 6
G6 A
o~t
20
q
and the rescaled damping constant cm g=A0 A0 =qd2 .
The above set of equations of motion can be used to simulate the microstructural evolution in shape memory alloys.
In the following section, we simulate the microstructure
associated with the cubic to tetragonal transformation as
a function of temperature using these equations of motion.
5. Temperature-induced transformation
We apply the present theory to obtain the microstructures for temperature-induced cubic to tetragonal transformation in FePd shape memory alloys which have attracted
attention due to interesting magnetoelastic properties [19].
To completely specify the FePd system, we need experimental values for e0, E0, A0, Abulk and Ashear used in
Section 3. To specify the spatial length scales, we use the
value G = 3.15 108 N from microstructural data
obtained by transmission electron microscopy for FePd
using 30 per cent Pd [20]. The value is chosen such that
the length scales of the simulated microstucture match
experiment. With this choice, the smallest length scale
being simulated becomes d  4 nm based on a simulation
box size of 64 64 64. Phonon dispersion data provide
a value [13,21] of G = 3.38 1010 N, corresponding to
d  0.3 nm. The system size would then be too small and
therefore we choose G from microstructural data. To solve
the full dynamic equations we need an estimate of the

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

damping constant g that can be obtained from measurements of ultrasonic attenuation as a function of frequency
[22,23], x. The attenuation varies as gx2/vs, where vs is the
sound velocity and thus this phonon viscosity model provides an estimate for g. In the absence of ultrasonic measurements for FePd, we use the estimate g  0.015 N s/m2
for V3S as it also undergoes a very similar cubic to tetragonal transformation [22].qAs
the rescaled time ~t is related to

2
the real time t by t ~t qd =A0 , we use the estimate for
density q  104 kg m3 to obtain t=~t  2:8  1012 s.
The form of the damping used in the simulation is a generalization of the result for a body with nite velocity
undergoing internal friction to continuous elastic bodies
[24]. The damping coecient corresponds to a domain wall
mobility of 104 m3/N s at the transformation temperature
[22,23]. The damping is 103  104 larger than theoretically
calculated room temperature values for several metals, so
that the attenuation measured is not directly due to merely
phonon viscosity. Dislocation drag, thermoelastic eect,
phonon scattering, electronic damping would all make
some contribution. Ultrasonic measurements [23] indicate
rapid but continuous increase in attenuation below the
structural transformation. However, there are limitations
in the interpretation of attenuation data. As the damping
term helps to drive the system to the free energy minima,
we do not expect the damping to aect the transformation
strains. However, anharmonicities in the dissipational
function may inuence the dynamics and transients.
Eqs. 1820 are discretized using nite dierences
D~x 1; D~t 0:02 and periodic boundary conditions.
Starting from small amplitude random initial conditions
(corresponding to austenite) for the displacement elds, a
quasi-static cooling process is simulated by varying the

2113

temperature T from T = 270 to 230 K in steps of

DT = 5 K. Note that after each change the system is
allowed to relax for t  0.67 ns. Fig. 3 shows the development of the microstructure at two dierent temperatures
during cooling. The columns show the spatial distribution
of the strains xx, yy, zz. The yellow/red regions in the rst
column correspond to tetragonal variants stretched along
the x direction characterized by xx 6 0. Similarly, the red
regions in the second column depict the variant distorted
along the y direction (yy 6 0), and in the third column
these red shaded regions represent the tetragonal variant
distorted along z (zz 6 0). The blue regions in the rst, second and third columns represent areas where xx = 0,
yy = 0 and zz = 0, respectively. A cubic austenite phase
exists at T = 270 K where all the strains are close to zero.
The austenite phase becomes unstable below this temperature and the three tetragonal variants are formed. We can
see the emergence of the three variants in Fig. 3 for the
snapshots corresponding to T = 260 K. However, the interfaces are not yet well dened and there are still traces of the
remnant austenite phase as the microstructure is still evolving at this stage. By T = 230 K, the microstructure equilibrates and the interfaces become well dened. Little
austenite is left by this time. Notice that on average the
interfaces adopt specic orientations that depend on the
variants that they separate. This is seen in the second
row of Fig. 3. For example, the interface between the variant stretched along y direction and that stretched along the
z direction makes an angle of 45 in the yz plane. In the
xy and xz planes, this interface is parallel to one of the
edges of the simulation box. These orientations are consistent with the predictions of Sapriel [25] that are based on
strain matching. In our model, the interfaces spontane-

Fig. 3. Simulated microstructures in the martensite phase at T = 260 and 230 K. The columns show the spatial distribution of strains xx, yy and zz. The
variant distorted along x is represented by red/yellow regions in the rst column (xx 6 0), the variant distorted along y is represented by red/yellow
regions in the second column (yy 6 0) and the variant distorted along z is represented by red/yellow regions in the third column (zz 6 0). (For
interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

2114

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

ously assume these specic orientations due to the tendency

of the system to maintain elastic compatibility or produce
coherent interfaces.
Earlier studies [17,18] of the cubic to tetragonal transition based on this free energy considered an ideal martensitic transformation where the volume change associated
with the transformation was zero. In the present work,
the volume change of 0.7% leads to interfaces which are
irregular, although the overall pattern is similar to the case
of ideal martensitic transformations.
6. Strain-induced transformation and pseudoelasticity
A martensitic transformation can also be induced by an
external deformation which results in large recoverable
strains. This behavior is referred to as pseudoelasticity. In
this section, we study strain-induced transformations using
the displacement eld approach. A loading process is simulated by dening the strain


1 oui ouj
ij

.
21
applied
ij
2 oxj oxi
This corresponds to a homogeneous strain applied everywhere in the system. In the present simulation, we apply
a uniaxial tensile strain and so applied
applied
applied

yy
zz
xy
applied
applied

0.
The
uniaxial
strain
along
x
is
given
by
yz
xz
c_ t
applied
xx

22

where c_ represents the strain rate.

Before we describe our results, we remark that because
we do not use stress free boundary conditions, all load-

ingunloading simulations reported in this paper correspond to a system that is clamped transverse to the
loading direction. A stressstrain curve can be calculated
for this situation by computing the stress analytically from
the free energy in Eq. (13) using rxx = dF/dxx, subject to
the constraint that yy = zz = 0. This analytical solution
is plotted in Fig. 4. The analytical solution shows that
for strain values in the range 0.006 to 0.017, the austenite is mechanically unstable, as is clear from the negative
slope of the stressstrain curve in that region. The point
C corresponds to where the austenite and martensite have
the same energy and so between C and D, the austenite is
metastable. As the behavior in the metastable and unstable
regions is inuenced by the dynamics associated with
nucleation and growth, a framework that describes the
dynamics is essential.
The strain loading simulations are performed using the
denitions in Eqs. (19) and (20) by solving Eq. (18) as
described in Section 5. We simulate a system in the cubic
austenite phase at T = 313 K. The maximum applied strain
for this case is applied
0:025 and the loading is completed
xx
in t  2.8 ns corresponding to a strain rate of
c_ 8:92  106 =s. Thereafter, the applied strain is decreased
at the same strain rate to simulate the unloading process.
The stressstrain curves are computed by plotting the average stress hddFxx i vs. the average strain xx. Fig. 5 depicts the
microstructural evolution during the loading process. Here,
the red regions in the columns also represent strains with
xx 6 0, yy 6 0 and zz 6 0. At t = 0, the system exists in
the austenite state given by xx = yy = zz = 0. On loading,
the homogeneous austenite phase persists until at

Fig. 4. The homogeneous stressstrain curve calculated by plotting ofohomo

vs. xx, subject to the constraint that yy = zz = 0. The plots A, B, C and D show
xx
the corresponding potential energy proles.

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

2115

Fig. 5. Microstructural evolution during uniaxial loading along the x direction at T = 313 K for a strain rate c_ 8:92  106 =s. The snapshots A, B and C
correspond to average stresses and strains indicated on the stressstrain curve. Notice that a multi-domain state is formed beyond the mechanical stability
limit corresponding to a stress level equal to the stress at point D in Fig. 4. (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

xx  0.006 (point D in Fig. 4), the austenite becomes

unstable and a non-equilibrium multi-domain state is
formed. This is clear by comparing the microstructures at
points A and B on the loading part of the stressstrain
curve in Fig. 5. Notice that in the multi-domain state all
three tetragonal distortions occur, although the variant
stretched along the x direction appears to be the favored
variant. It is instructive to compare the orientation of

domain walls for this multi-domain state to the low temperature microstructures without an external load that
are depicted in Fig. 3. For example, the angle between
the variant stretched along x and the variant stretched
(not as much) along y, as seen in the top xy plane, is dierent from 45, unlike the low temperature microstructure
shown in Fig. 3. This may be due to the two variants being
unequally distorted as the variant distorted along x is the

2116

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

favored variant. It is also interesting to note that the stress

required to maintain this multi-domain state is lower than
the stress at the instability at which the variants are
expected to start forming. The jump in the stress at the
onset of the transformation is typically associated with this
stress drop. The favored variant in the multi-domain state
grows at the expense of the other variants, as can be seen
from the microstructure corresponding to the point C on
the stressstrain curve. Thereafter, the system stays in a single variant state of the favored variant and deforms according to linear elasticity. During the unloading, the material
remains in the single variant state until that state becomes
unstable. A non-equilibrium state is formed again in which
the system starts transforming to the austenite and there is
a jump in the stressstrain curve at the onset of the reverse
transition. At the end of the unloading, the material is
cubic and all strains become zero, resulting in complete
strain recovery.
We should remark that the stressstrain curve of the
type obtained in the present simulations has indeed been
seen in experiments on NiTi and CuZnAl shape memory
alloys [26,27]. Our simulations show that if there are no
nucleating defects present in the system, the stress required
to cause the transition corresponds to the intrinsic limit of
stability of the system. However, as the austenite becomes
unstable, a jump in the stressstrain curve is observed. Similar behavior is seen at the onset of the reverse transformation upon unloading. The present simulations show that
the jumps in stress at the onset of the transitions are a generic feature of strain loading as it places the system in the
mechanically unstable regions that result in multi-variant
states. Note that no jumps were observed in our earlier
stress loading calculations as the unstable regions could
not be accessed using that method [8,9].
As the deformation involves the formation and motion
of domain walls, the rate of loading may inuence the
stressstrain response. To investigate this, we have performed dynamic strain loading simulations described
above for three dierent strain rates corresponding to
c_ 8:92  106 =s, c_ 4:46  106 =s and c_ 2:23  106 =s.
Fig. 6 shows the simulated stressstrain curves for the three
strain rates. We nd that the stressstrain curves for all
three strain rates match the analytical solution in the stable
regions. However, a strain rate dependence is observed in
the unstable region of the stressstrain curve. As observed
in Fig. 5, the loading process in this regime is governed by
variant formation and domain wall motion. As this process
is not instantaneous, a competition between the time scales
of the domain dynamics and the loading rate is responsible
for this strain rate dependence. For example, if the strain
rate is fast enough that domain formation is avoided, the
stressstrain curve is identical to the analytical solution in
Fig. 4. For slower strain rates, twin boundaries appear
and the stress starts to deviate from the analytical solution.
In fact, as seen in Fig. 6, the slower is the strain rate, the
smaller is the strain at which the stressstrain behavior
begins to deviate from the analytical solution.

Fig. 6. Simulated stressstrain curves at T = 313 K for three dierent

strain rates. The blue curve corresponds to c_ 8:92  106 =s, the red curve
corresponds to c_ 4:46  106 =s, and the green curve corresponds to
c_ 2:23  106 =s. For comparison, the homogeneous stressstrain curve of
Fig. 4 is also shown. We see that the deviation from the homogeneous
response occurs at lower strain values for slower strain rates. (For
interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

The next question we want to address is the eect of

defect-induced nucleation on the stressstrain curves. It is
expected that the presence of defects will nucleate the transformation even before the system is in the unstable region,
thereby reducing the stress required to cause the transition.
To study defect-induced transformation, we repeated our
simulations of the loading process with an initial
quenched seed of the transformed phase that is embedded in the initial austenite matrix. The total uniaxial strain
eld in this direction is given by
xx

oux
applied
seed
xx
xx
ox

23

is the strain due to the defect.

where seed
xx
seed
jx  x0 j 6 L0 ; jy  y 0 j 6 L0 ; jz  z0 j 6 L0
xx 0
0 jx  x0 j > L0 ; jy  y 0 j > L0 ; jz  z0 j > L0 .
24
where the midpoint of the simulation box of size
64 64 64 is given by x0 = y0 = z0 = 64/2 = 32. This represents an inclusion of a cube of the martensite phase of
length L0 in the austenite matrix. A loading process for this
system with strain rate c_ 2:23  106 =s (L0  24 nm) is
shown in Fig. 7. This gure shows the stressstrain curve
and the spatial distribution of xx at points A, B, C, D
and E on the stressstrain curve. In the snapshots A, B,
C, D and E, the red/yellow regions correspond to regions
distorted along x (xx 6 0). To clearly show the evolution
of the microstructure around the defect region, we display
two mutually intersecting perpendicular planes that pass
through the defect. There is a point (C in Fig. 4) on the
stressstrain curve where austenite and martensite have

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

2117

Fig. 7. Microstructural evolution during uniaxial loading (_c 2:23  106 =s) along the x direction for the case with a defect embedded at the center of the
cubical simulation box. The snapshots A, B, C, D and E show the distribution of the strain xx (along the loading direction) corresponding to the average
stresses and strains indicated on the stressstrain curve. To clearly show the defect region, two mutually intersecting planes through the defect are shown.
Unlike the defect free case, a multi-domain state is created before the limit of mechanical instability due to defect-induced nucleation in the metastable
region. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

the same energy. For strains higher than this critical strain,
the nucleation of martensite can take place as martensite
has lower energy than the austenite. This nucleation
process can be observed by comparing the microstructure
at points B and C. We can see the growth of the martensite
domains (regions shaded red) from a seed defect that is
embedded in the austenite matrix (regions shaded blue).
On further loading, the favored variant grows, as can be
observed in the snapshot corresponding to D. Eventually,
a single variant state of the variant stretched along the x
direction is established as can be observed in the snapshot

E. Notice that the strains in the defect regions are higher

(4%) than the bulk value (3%). Even during this loading
process, the unfavored variants are also formed. Fig. 8
shows the full microstructures for xx, yy and zz corresponding to the point C of Fig. 7. Notice that the magnitudes of the distortions along the y and z directions are
smaller than those along the favored x direction.
To test the strain rate dependence of the stressstrain
response in the presence of the defect, Fig. 9 compares
the behavior for strain rates c_ 8:92  106 =s, c_ 4:46 
106 =s and c_ 2:23  106 =s with the analytical solution

Fig. 8. Distribution of xx, yy and zz on the surfaces for snapshot C of Fig. 7. It is seen that although all three variants appear to form, tetragonal
distortions along the x direction are larger than those along y and z directions. (For interpretation of the references to color in this gure legend, the reader
is referred to the web version of this article.)

2118

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

for the defect free case. It is clear that the slower the rate of
loading, the smaller is the strain at which the stress deviates
from the analytical curve. This is due to the fact that nucleation is not instantaneous and so the time dependence of
the nucleation will inuence the stressstrain response.
For example, if the loading rate is fast enough that there
is little time for the martensite to nucleate, the apparent
stress required to cause the transition will be higher and
the jump at the onset of the transition will be larger. Thus,
the jump in stress becomes less pronounced as the strain
rate becomes smaller. These results show how nucleation
and growth of variants crucially inuence the mechanical
response of shape memory alloys in the pseudoelastic
regime.
To further illustrate the role of defects on the mechanical response, we have performed loading simulations with
three dierent sizes of defects. The stressstrain curves
for defect sizes L0 = 16, 24, 40 nm with the defect free case
at the xed strain rate c_ 2:23  106 =s are shown in
Fig. 10. As expected, nucleation occurs at smaller strains
for larger defect sizes. In fact, for the defect free case, the
martensite is formed only after the applied strain is higher
than the intrinsic limit of mechanical stability. Thus, the
presence of defects inuences the magnitude of the jump
in stress at the onset of the transformation.
We note that the strain rates used in the simulations
(106/s) are relatively higher than those typically observed
in practice. The reason for the high strain rates lies in the
time scales that we are simulating, which are relatively
short (1012 s). The real time is related to the scaled time
via the smallest length (d  0.110 nm). This limits the size

Fig. 10. Simulated stressstrain curves for dierent sizes of the embedded
defect at T = 313 K. The cyan, blue and red curves correspond to defect
sizes L0 = 16, 24, 40 nm, respectively. The green curve is for the defect free
case. The size of the defect determines the magnitude of the jump as the
larger the defect, the easier is nucleation. (For interpretation of the
references to color in this gure legend, the reader is referred to the web
version of this article.)

of the system and the times that can be simulated. A

coarse-grained model which averages over several twin
boundaries could describe slower strain rates. The physics
at the coarse-grained scale would then be governed by
the movement of austenitemartensite interfaces rather
than twin boundaries. Our simulations therefore model fast
processes at short length scales.
7. Deformation in the martensite phase and the shape
memory eect

Fig. 9. Simulated stressstrain curves for the system with a defect at

T = 313 K for three dierent strain rates. The blue curve corresponds to
c_ 8:92  106 =s, the red curve corresponds to c_ 4:46  106 =s and the
green curve corresponds to c_ 2:23  106 =s. The black curve represents
the homogeneous curve in Fig. 4. We observe that the magnitude of the
jump at the onset of transformation is higher for a higher strain rate. (For
interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

Here we investigate the mechanical response at low temperatures when the material has already transformed to
martensite in the absence of external deformation. Our initial condition is the microstructure in the second row of
Fig. 3, corresponding to T = 230 K. The loading conditions are identical to those in the previous section, although
we load to higher values of the maximum strain to span the
regimes of interest. Fig. 11 shows the simulated loading
and unloading process. The stressstrain curve is calculated
in the same manner as the previous section but the average
stress at applied
0 is subtracted for all points so that the
xx
stress is zero for the rst point. The strain rate used in
the present simulations is c_ 12:5  106 =s. The conguration at the rst point on the stressstrain curve is the as
cooled martensite state of the second row in Fig. 3. As
the strain is applied, there is a short linear elastic regime
during which the microstructure does not change signicantly. This is clear by comparing the microstructure for
the point A in Fig. 11 to that of the second row in
Fig. 3. Beyond the linear elastic region, the twin boundaries
start to move, as can be seen in the microstructure

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

2119

Fig. 11. Microstructural evolution during uniaxial loading (_c 12:5  106 =s) along x in the shape memory regime for T = 230 K, starting from a twinned
state at T = 230 K depicted in Fig. 3. The snapshots A, B and C correspond to the stresses and strains indicated on the stressstrain curve. Note that after
a transient linear elastic region, the favored variant starts to grow at the expense of the unfavored variants. (For interpretation of the references to color in
this gure legend, the reader is referred to the web version of this article.)

corresponding to point B. Notice that the variant stretched

along the x direction grows at the expense of the other variants. This variant continues to grow and we can observe in
the snapshot corresponding to C that most of the material
has transformed into this variant. Eventually, all variants
switch in favor of this variant and we obtain a single
domain state. The unloading process is continued until
the stress is zero. We observe that the material remains in

a single domain state with a residual strain even when the

stress is zero. This residual strain can be recovered on heating giving rise to the shape memory eect.
We performed simulations of the loadingunloading
(starting from the same initial state shown in the second
row of Fig. 3), for three dierent stain rates given by
c_ 12:5  106 =s, c_ 6:25  106 =s and c_ 3:125  106 =s.
Fig. 12 shows the strain-rate dependence of the simulated

2120

R. Ahluwalia et al. / Acta Materialia 54 (2006) 21092120

Fig. 12. Simulated stressstrain curves for three dierent strain rates at
T = 230 K. The black, red and blue curves correspond to
c_ 12:5  106 =s, c_ 6:25  106 =s and c_ 3:125  106 =s, respectively.
(For interpretation of the references to color in this gure legend, the
reader is referred to the web version of this article.)

stressstrain curves. During the loading part, we can see

that the detwinning stress appears to be lower for lower
strain rate. This is due to the fact that for slower strain rates,
the twin boundaries have time to move by their own inertia
and the stress required to move the interfaces is lower.
8. Summary and conclusions
We have carried out a detailed study of the microstructure of cubic to tetragonal martensites and the inuence
of microstructural evolution on mechanical properties.
The simulations are based on a nonlinear elastic free energy
in terms of the appropriate strain components that drive the
cubic to tetragonal transition. The dynamics is simulated by
solving the force balance equations for the displacement
elds with a dissipational function providing the necessary
damping. The advantage of this method is that the elastic
compatibility relations are naturally satised.
We have applied the framework to the case of FePd
shape memory alloys by using estimates for the free energy
parameters and damping constants that are obtained from
existing experimental data. We have simulated the microstructure development as the material is cooled from the
high temperature cubic austenite phase. The model has
been applied to study strain-induced cubic to tetragonal
transformation and the associated pseudoelastic behavior.
Finally, the loadingunloading process in the shape memory regime, starting from a low temperature martensite
state has been studied. Although the stressstrain curves
simulated are in reasonable agreement with experiments,
there are limitations on our simulations, such as spatial
and time scales accessed as well as loading and boundary
conditions, that make a more quantitative comparison difcult. Nevertheless, our simulations capture important fea-

tures of the deformation process that are not described by

existing theoretical models. For example, the experimentally observed jumps in stress at the onset of the transformation during loading are explained by the current
model. The simulations show that these are due to domain
formation that reduces the total stress in the system. Thus,
the stress required to initiate the transformation is higher
than the stress required to maintain the multi-domain state.
The simulations also show that due to time-dependent phenomena such as nucleation and domain wall motion, a
strain rate dependence of the mechanical properties is
observed in these materials.
Our present simulations describe processes at relatively
small length (250 nm) and time (25 ns) scales. As larger sizes and longer times will not appreciably increase
the scales, a coarse-graining is required to allow access to
engineering length scales. However, the fact that we are
able to capture the salient aspects of experiments at macroscopic length scales, indicates that the essential physics
(that of reversible elastic processes) is also valid at the
smaller scales we have simulated. Renement of our
scheme to a more coarse grained model in terms of volume
fractions of the transformed phase is a possible approach
to describing larger length scales.
References
[1] Bhattacharya K. Microstructure of martensite. Why it forms and how
it gives rise to the shape-memory eect? Oxford: Oxford University
Press; 2003.
[2] Otsuka K, Wayman CM, editors. Shape memory materials. Cambridge (UK): Cambridge University Press; 1998.
[3] Vaidyanathan R, Bourke MAM, Dunand DC. Acta Mater
1999;47:3353.
[4] Huang M, Brinson LC. Int J Plast 2000;16:1371.
[5] Levitas VI, Idesman AV, Preston DL. Phys Rev Lett 2004;93:105701.
[6] Falk F. Acta Metall 1980;28:1773.
[7] Levitas VI, Preston DL. Phys Rev B 2002;66:134206.
[8] Ahluwalia R, Lookman T, Saxena A. Phys Rev Lett 2003;91:055501.
[9] Ahluwalia R, Lookman T, Saxena A, Albers RC. Acta Mater
2004;52:209.
[10] Artemev A, Wang Y, Khachaturyan AG. Acta Mater 2000;48:2503.
[11] Barsch GR, Krumhansl JA. Phys Rev Lett 1984;53:1069.
[12] Seto H, Noda Y, Yamada Y. J Phys Soc Jpn 1998;57:3668.
[13] Sato M, Grier BH, Shapiro SM, Miyajima H. J Phys F 1982;12:2117.
[14] Lookman T, Shenoy SR, Rasmussen KO, Saxena A, Bishop AR.
Phys Rev B 2003;67:024114.
[15] Ahluwalia R, Ananthakrishna G. Phys Rev Lett 2001;86:4076.
[16] Bales GS, Gooding RJ. Phys Rev Lett 1991;67:3412.
[17] Jacobs AE, Curnoe SH, Desai RC. Phys Rev B 2003;68:224104.
[18] Ahluwalia R, Lookman T, Saxena A, Shenoy SR. Phase Trans
2004;77:457.
[19] Cui J, Shield TW, James RD. Acta Mater 2004;52:35.
[20] Oshima R, Sugiyama M, Fujita FE. Metall Trans A 1988;19A:803.
[21] Saxena A, Barsch GR. Physica D 1993;66:195.
[22] Barsch GR. J de Physique IV (Colloque) 1995;5:119.
[23] Testardi LR, Bateman TB. Phys Rev 1967;154:402.
[24] Landau LD, Lifshitz EM. Theory of elasticity, 3rd ed., vol.
7. Oxford: Butterworth-Heinemann; 1998.
[25] Sapriel J. Phys Rev B 1975;12:5128.
[26] Iadicola MA, Shaw JA. Int J Plast 2004;20:577.
[27] Otsuka K, Wayman CM, Nakai K, Sakamoto H, Shimizu K. Acta
Metall 1976;24:207.