PHYSICAL REVIEW B 68, 134201 共2003兲

Three-dimensional Landau theory for multivariant stress-induced martensitic phase

transformations. III. Alternative potentials, critical nuclei, kink solutions,
and dislocation theory
Valery I. Levitas,1,* Dean L. Preston,2 and Dong-Wook Lee1
1
Texas Tech University, Center for Mechanochemistry and Synthesis of New Materials, Department of Mechanical Engineering, Lubbock,
Texas 79409-1021, USA
2
Applied Physics Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA
共Received 2 April 2003; published 6 October 2003兲
In part III of this paper, alternative Landau potentials for the description of stress-and temperature-induced
martensitic phase transformations under arbitrary three-dimensional loading are obtained. These alternative
potentials include a sixth-degree 共2-4-6兲 polynomial in Cartesian order parameters and a potential in hyper-
spherical order parameters. Each satisfies all conditions for the correct description of experiments. The unique
features of the potentials are pointed out and a detailed comparison of the potentials is made for NiAl alloy.
Analytic solutions of the one-dimensional time-independent Ginzburg-Landau equations for the 2-3-4 and
2-4-6 potentials for a constant-stress tensor and invariant-plane strain are obtained and compared. Solutions
include martensitic and austenitic critical nuclei and diffuse martensite-austenite and martensite-martensite
interfaces. The widths and energies of the nuclei and interfaces are functions of the thermodynamic driving
force, the gradient energy coefficient, and a parameter that characterizes the stability of austenite. The splitting
of a martensite-martensite interface into two austenite-martensite interfaces is interpreted as a potentially new
mechanism—namely, barrierless austenite nucleation—which might be observed experimentally at the inter-
face between two invariant-plane-strain variants. The widths, energies, and gradient energy coefficients of the
martensite-martensite and austenite-martensite interfaces are estimated for NiAl. Finally, we outline a version
of phase field theory for dislocations based on our theoretical framework for phase transformations.

DOI: 10.1103/PhysRevB.68.134201 PACS number共s兲: 64.60.⫺i

I. INTRODUCTION modifies the profiles of 2-4-6 critical martensitic nuclei rela-

In parts I 共Ref. 1兲 and II 共Ref. 2兲 we developed a fifth-
degree polynomial 共2-3-4-5兲 Gibbs 共Landau兲 potential for the
description of multivariant stress- and temperature-induced
martensitic phase transformations 共PT’s兲 in three dimensions.
Our approach was a phenomenological one; that is, the 2-3-
4-5 potential was constructed by requiring that it respects the
experimentally observed features of martensitic PT’s in
shape memory alloys and steels, specifically, constant trans-
formation strain, weakly temperature dependent, or constant,
stress hysteresis, and transformation at nonzero tangent elas-
tic moduli. However, the 2-3-4-5 potential is by no means
the only Landau potential that satisfies these basic require-
ments. In this paper, we develop three alternative potentials:
namely, a 2-4-6 polynomial in Cartesian order parameters
共Sec. II兲 and two potentials in hyperspherical order param-
eters 共Sec. III兲. The symmetry requirements for the Gibbs
potential 共see Sec. IV in Ref. 2兲 are satisfied for the poten-
tials derived in this paper. The phase equilibrium and trans-
formation conditions for all potentials, including the 2-3-4-5
potential, are identical. Consequently, the geometric repre-
sentations of the phase equilibrium and transformation con-
ditions introduced in part II Ref. 2 can be used for the alter-
native potentials as well.
The distinguishing feature of the 2-4-6 polynomial is that
its curvature at the martensitic (M) minimum is 4 times
larger than at the austenitic (A) minimum, while both curva-
tures are the same for the 2-3-4-5 polynomial derived in
Refs. 1 and 2. This difference in the curvatures significantly
tive to the corresponding profiles for the 2-3-4-5 potential. In
contrast to the 2-3-4-5 and 2-4-6 potentials, the hyperspheri-
cal potentials have no unphysical local minima and have no
constants that do not appear in the phase equilibrium and
transformation conditions. Variant-variant transformations
occur along the unit hypersphere, whereas the variant-variant
transformation paths are much more complicated for the 2-3-
4-5 and 2-4-6 polynomial potentials.
It is shown for all potentials that the number of order
parameters can be reduced by a factor of 2 if transformation
strains for pairs of martensitic variants decompose into two
components: one that is the same for both variants and one
that is equal in magnitude but of opposite sign for the two
variants. This allows us to reduce the number of order pa-
rameters by a factor of 2. Examples of applications include
cubic-orthorhombic, cubic-monoclinic-I, and cubic-
monoclinic-II PT’s and PT’s of invariant-plane-strain 共IPS兲
variants.
In Sec. IV, we compare the 2-3-4-5, 2-4-6, and polar 2-3-4
potentials for the NiAl cubic-tetragonal phase transformation
for zero stress and for two three-dimensional stress states.
In Sec. V, analytical solutions of the one-dimensional
static Ginzburg-Landau equations for a constant three-
dimensional stress tensor and invariant-plane strain are found
for the 2-3-4 and 2-4-6 potentials and compared. Analytical
solutions of the Ginzburg-Landau equation for the 2-4-6 po-
tential in strain for the stress-free case were found by Falk.5
Jacobs6 generalized some of them to finite strain and to two-
dimensional problems that can be treated as one dimensional

0163-1829/2003/68共13兲/134201共24兲/$20.00 68 134201-1 ©2003 The American Physical Society

VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

and also calculated displacements. However, the physical in-
terpretation of some of the solutions, such as solitons on A
and M and soliton splitting, was unclear. In Ref. 7, numerical
solutions for critical M nuclei of the 2-3-4 potential were
analyzed in detail under prescribed displacements. It was
used in Ref. 7 to model the nonclassical nucleation of an
ellipsoidal region.
In contrast to Refs. 5 and 6, our theory incorporates ho-
mogeneous multiaxial stresses and we provide physical in-
terpretations of the solutions of the Ginzburg-Landau equa-
tions. Despite the significant differences between our theory
and Falk’s,5 the dimensionless forms of the Ginzburg-
Landau equations for our 2-4-6 potential and Falk’s strain-
based potential coincide. This enables us to borrow some of
Falk’s analytical solutions for the stress-free case and ana-
lyze their counterparts in relevant variables and under a
constant-stress tensor, which is significantly different from
Falk’s results. Analytical solutions for the 2-3-4 potential are
found. All solutions depend on three parameters: s 1 , which
characterizes the stability of austenite; s 2 , which is propor-
tional to the thermodynamic driving force for the M→A PT
共both s 1 and s 2 are stress and temperature dependent兲; and
the gradient energy coefficient ␤ . Phase and transformation
diagrams in s 1 -s 2 coordinates are analyzed. It is proven by
numerical solution of the time-dependent Ginzburg-Landau
equation that a stationary M soliton on A and an A soliton on
M are in fact martensitic and austenitic critical nuclei. The
structure and energetics of the M and A critical nuclei, as
well as diffuse A-M and M⫹ -M⫺ interfaces, are studied in
detail. In particular, the widths of the M and A critical nuclei,
and the thicknesses of their interfaces, and the energies of the
nuclei, and their interfaces are determined in terms of the
aforementioned three parameters. A relation between the en-
scribe PT’s with transformation strains that differ only in
sign. This is usually done by means of an even polynomial
potential in the strain.8,6 The only strain appearing in our
Landau potentials is the transformation strain ␧t , but it can-
not serve as an order parameter because it does not change
with the stress, like the strain does—the transformation strain
is fixed for each martensitic variant. To describe a change in
sign of the transformation strain, the function ␸ 关see Eqs. 共1兲
and 共2兲兴 must be odd in the order parameter ␩ , but the ther-
mal part of the free energy f must be an even function of ␩ .
If ␸ ( ␩ ) is a 1-3-5 polynomial and f ( ␩ ) is a 2-4-6 polyno-
mial, then G( ␩ ) is a complete 共contains all powers兲 sixth-
degree polynomial that cannot be studied analytically and
may have additional unwanted extrema. There is, however,
an alternative to including odd powers of ␩ : ␧t sgn( ␩ ) is
substituted for ␧t in a 2-4-6 polynomial Gibbs potential.
In Sec. II A we derive a 2-4-6 polynomial potential in
order parameters ␩ i 苸 关 0,1兴 . In Sec. II B the range of order
parameter values is extended from 关 0,1兴 to 关 ⫺1,1兴 for PT’s
with ␧t that differ only in sign and more generally for PT’s
where the ␧t for pairs of martensitic variants decompose into
two components: one that is the same for both variants and
one that is equal in magnitude but of opposite sign for the
two variants. This extension reduces the number of order
parameters by a factor of 2. A similar result was obtained for
the 2-3-4-5 potential.
A. Positive order parameters
The 2-4-6 polynomial is subject to the same requirements
as the 2-3-4-5 polynomial Gibbs potential derived in parts I
共Refs. 1兲 and II 共Ref. 2兲. Following the same steps as before
we obtain

冉 兺 冊
ergy and the width of the equilibrium A-M interface and
n
stress hysteresis is found. Two types of M⫹ -M⫺ interfaces
are considered: a simple kink connecting M⫹ to M⫺ and an G⫽⫺ ␴: ␭0 ⫹ 共 ␭k ⫺␭0 兲 ␸ 共 a ␭ , ␩ k 兲 :␴/2
k⫽1
exotic split M⫹ -M⫺ interface comprised of juxtaposed M⫹ -A
and A-M⫺ interfaces. This splitting is interpreted as a poten-
tially new mechanism: barrierless A nucleation. Nucleation
occurs in the region of stability of M near the equilibrium
⫺ ␴: 冉冉 兺 ␭30 ⫹
n

k⫽1
冊冊
共 ␭3k ⫺␭30 兲 ␸ 共 a 3␭ , ␩ k 兲 :␴ :␴/3

M-A line. Such nucleation might be found experimentally at

the interface between two IPS variants.
In Sec. VI, the gradient energy coefficients, energies, and
⫺ ␴: ␴: 冉冉 兺 ␭40 ⫹
n

k⫽1
冊冊
共 ␭4k ⫺␭40 兲 ␸ 共 a 4␭ , ␩ k 兲 :␴ :␴/4

冉 冊
widths of the M⫹ -M⫺ and A-M interfaces are estimated for n n
NiAl alloy.
In Sec. VII, a phase field theory of dislocations is devel-
⫺ ␴: 兺 ␧tk ␸ 共 a, ␩ k 兲 ⫺ ␴:
k⫽1
␧␪ 0 ⫹ 兺共 ␧␪ k ⫺ ␧␪ 0 兲 ␸ 共 a ␪ , ␩ k 兲
k⫽1
oped. The known theory3 is based on a formalism similar to n n⫺1 n
phase field theory of martensitic PT’s 共Ref. 4兲 and has a
similar shortcoming; namely, the equilibrium value of the
⫹ 兺
k⫽1
f 共 ␪,␩k兲⫹ 兺 兺
i⫽1 j⫽i⫹1
F i j共 ␩ i , ␩ j 兲; 共1兲
Burgers vector and the plastic strain depend on stress. We
extend our approach developed for PT’s to dislocations to
eliminate this drawback. ␸ 共 a, ␩ k 兲 ⫽a ␩ 2k /2⫹ 共 3⫺a 兲 ␩ 4k ⫹ 共 a⫺4 兲 ␩ 6k /2, 0⭐a⭐6; 共2兲

II. SIXTH-DEGREE POLYNOMIAL GIBBS POTENTIAL f 共 ␪ , ␩ k 兲 ⫽A ␩ 2k /2⫹ 共 3⌬G ␪ ⫺A 兲 ␩ 4k ⫹ 共 A⫺4⌬G ␪ 兲 ␩ 6k /2; 共3兲

In this section we obtain a sixth-degree polynomial Gibbs

potential G that describe PT’s between austenite (A) and F i j 共 ␩ i , ␩ j 兲 ⫽B ␩ 2i ␩ 2j ⫹ 共 Z i j ⫺B 兲 ␩ 4i ␩ 2j ⫹C ␩ 3i ␩ 3j
martensitic variants (Mi , i⫽1,2, . . . ,n) and between mar-
tensitic variants. One of the requirements of G is that it de- ⫹ 共 Z ji ⫺B 兲 ␩ 2i ␩ 4j ; 共4兲

134201-2
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

a 3 Mi →A:
Z i j ⫽ ␴:␧t j ⫺ ␴:共 ␧t j ⫺ ␧ti 兲
2 2

冋 册
␩i兲
⳵ 2 G 共 ␴, ␪ , ˆ
a␪ 3 ⭐0
⫹ ␴: 共 ␧ ⫺ ␧␪ 0 兲 ⫺ 共 ␧␪ j ⫺ ␧␪ i 兲 ⳵ ␩ 2i
2 ␪j 2

⫹ ␴: 冋 a␭
4
3
册
共 ␭ j ⫺␭0 兲 ⫺ 共 ␭ j ⫺␭i 兲 :␴
4
⇒ 共 6⫺a 兲 ␴:␧ti ⫹ 共 6⫺a ␪ 兲 ␴:共 ␧␪ i ⫺ ␧␪ 0 兲

⫹
6⫺a ␭
␴:共 ␭i ⫺␭0 兲 :␴
⫹ ␴: 冉 冋 a 3␭ 3
6
1
册 冊
共 ␭ j ⫺␭30 兲 ⫺ 共 ␭3j ⫺␭3i 兲 :␴ :␴
2
⫹
2
6⫺a 3␭
␴:„共 ␭3i ⫺␭30 兲 :␴…:␴
⫹ ␴: ␴:冉 冋 a 4␭ 4
8
3
册 冊
共 ␭ j ⫺␭40 兲 ⫺ 共 ␭4j ⫺␭4i 兲 :␴ :␴
8
⫹
3
6⫺a 4␭
␴:„␴:共 ␭4i ⫺␭40 兲 :␴…:␴
4
⫹Ā/2⫺A/2. 共5兲
⭐6⌬G ␪ ⫺A; 共6兲
Here ␴ is the stress tensor, ␧ti and ␧␪ i are the transformation
and thermal strains of the ith variant, i⫽0 corresponds to A, Mi →M j :
␧t0 ⫽0, ␭ki is the elastic compliance tensor of order k for
variant i (␭i ⬅␭2i ), and ⌬G ␪ is the difference between the ␩i兲
⳵ 2 G 共 ␴, ␪ , ˆ
⫽⫺6 ␴:共 ␧t j ⫺ ␧ti 兲 ⫺6 ␴:共 ␧␪ j ⫺ ␧␪ i 兲
thermal parts of the Gibbs energies of M and A. The param- ⳵ ␩ 2j
eters A and Ā characterize the thresholds for A↔Mi and
M j ↔Mi transformations, while B and C control the Gibbs ⫺3 ␴:共 ␭ j ⫺␭i 兲 :␴⫺2„␴:共 ␭3j ⫺␭3i 兲 :␴…:␴
energy away from both the A and Mi minima and the
3
minimum-energy paths between the minima; therefore, they ⫺ ␴:„␴:共 ␭4j ⫺␭4i 兲 :␴…:␴⫹2Ā⭐0. 共7兲
do not affect phase equilibrium and transformation condi- 2
tions. The material parameters a, a ␪ , a 2 ␪ , a 3 ␪ , and a 4 ␪
govern the variations of ␧ti , ␧␪ i , and the elastic compliances The transformation strain is equal to ⫺ ⳵ G(0, ␪ , ␩ i )/ ⳵␴ at
between the A and Mi minima. zero thermal strain:
Define ␩ˆ i ⫽(0, . . . ,0,␩ i ⫽1,0, . . . ,0), the vector from the n n⫺1 n
1
origin to Mi . The Gibbs potential was constructed to have ␧t ⫽ 兺 ␧ti ␸ 共 a, ␩ i 兲 ⫺
2 兺 兺 ␩ 2i ␩ 2j 关 3 共 ␩ 2i ␧ti ⫹ ␩ 2j ␧t j 兲
local minima at the origin and at the points ␩ˆ i , i i⫽1 i⫽1 j⫽i⫹1

⫽1, . . . ,n, but no constraints were placed on G at the points
␩ˆ i ⫹ ␩ˆ j , ␩ˆ i ⫹ ␩ˆ j ⫹ ␩ˆ k , etc. Consequently, G may be smaller
at such points that at the A and Mi local minima; i.e., non-
physical phases can appear. The relative values of G at A,
Mi , ␩ˆ i ⫹ ␩ˆ j , etc., are controlled by the parameters B and C.
Since F i j (1,1)⫽Z i j ⫹Z ji ⫺B⫹C, the elimination of minima
at ␩ˆ i ⫹ ␩ˆ j , ␩ˆ i ⫹ ␩ˆ j ⫹ ␩ˆ k , etc., can be achieved by choosing
B⭐0 and C⬎0.
Define 0̄⫽(0, . . . ,0), which corresponds to A, and ¯ ␩i
⫽(0, . . . ,0,␩ i ,0, . . . ,0). The phase transformation condi-
tions are as follows:
A→Mi :
⫹ 共 a⫺3 兲共 ␩ 2j ␧ti ⫹ ␩ 2i ␧t j 兲兴 . 共8兲
It is easily verified that ␧t satisfies all requirements: ␧t (0̄)
␩ i )⫽ ␧ti , ␧t (¯
⫽0, ␧t ( ˆ ␩ i )⫽ ␧ti ␸ (a, ␩ i ).
The thermodynamic equilibrium conditions ⳵ G/ ⳵ ␩ i ⫽0
(i⫽1, . . . ,n) have n⫹1 solutions corresponding to A and
the Mi : ␩ ⫽0̄ and ␩ ⫽ ˆ ␩ i , i⫽1, . . . ,n. There are other solu-
tions of ⳵ G/ ⳵ ␩ i ⫽0 that correspond to maxima or saddle
points. In the case of a single variant, there is, in addition to
the extrema at ␩ 1 ⫽0 and ␩ 2 ⫽1, an extremum at
␩ 3 ⫽ 冑关 A⫺a ␴:␧t ⫹a ␪ ␴:共 ␧␪ 0 ⫺ ␧␪ 1 兲 ⫹⌳ 兴 /3h,

⌳ªa ␭ ␴:共 ␭0 ⫺␭1 兲 :␴/2⫹a 3␭ ␴:„共 ␭30 ⫺␭31 兲 :␴…:␴/3

⳵ G 共 ␴, ␪ ,0̄ 兲
2
⭐0 ⫹a 4␭ ␴:共 ␴:„共 ␭ 40 ⫺␭41 兲 :␴…兲 :␴/4,
⳵ ␩ 2i a␭
⇒a ␴:␧ti ⫹a ␪ ␴:共 ␧␪ i ⫺ ␧␪ 0 兲 ⫹ ␴:共 ␭i ⫺␭0 兲 :␴
2
hªA⫺4⌬G ␪ ⫺ 共 a⫺4 兲 ␴:␧t ⫹ 共 a ␪ ⫺4 兲 ␴:共 ␧␪ 0 ⫺ ␧␪ 1 兲
a 3␭
⫹ ␴:„共 ␭3i ⫺␭30 兲 :␴…:␴ ⫹ 共 a ␭ ⫺4 兲 ␴:共 ␭0 ⫺␭1 兲 :␴/2
3
⫹ 共 a 3␭ ⫺4 兲 ␴:„共 ␭30 ⫺␭31 兲 :␴…:␴/3
a 4␭
⫹ ␴:„␴:共 ␭4i ⫺␭40 兲 :␴…:␴⭓A; ⫹ 共 a 4␭ ⫺4 兲 ␴:„␴:„共 ␭40 ⫺␭41 兲 :␴……:␴/4, 共9兲
4

134201-3
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

which corresponds to a maximum in G if ␩ 3 ⬍1. The height

of the activation barrier for the A→M PT can be calculated
by substituting ␩ 3 in Eq. 共1兲:

G 共 ␴, ␪ , ␩ 3 兲 ⫺G 共 ␴, ␪ ,0兲 ⫽ 关 2A⫺9⌬G ␪ ⫺ 共 2a⫺9 兲 ␴:␧t

⫹ 共 2a ␪ ⫺9 兲 ␴:共 ␧␪ 0 ⫺ ␧␪ 1 兲
⫹ 共 2a ␭ ⫺9 兲 ␴:共 ␭0 ⫺␭1 兲 :␴/2
⫹ 共 2a 3␭ ⫺9 兲 ␴:共 ␭30 ⫺␭31 兲 :␴:␴/3
⫹ 共 2a 4␭ ⫺9 兲 ␴:␴:共 ␭40
⫺␭41 兲 :␴:␴/4兴 ␩ 43 /3. 共10兲

The activation barrier for the M→A PT is G( ␴, ␪ , ␩ 3 )

⫺G( ␴, ␪ ,1), which can be obtained by adding G( ␴, ␪ , 0)
⫺G( ␴, ␪ , 1)⫽ ␴:␧t ⫺⌬G ␪ to Eq. 共10兲.
It is a good approximation over a modest range of tem-
peratures to take ⌬G ␪ and A to be linear functions of the
temperature,1

⌬G ␪ ⫽z 共 ␪ ⫺ ␪ e 兲 , A⫽A 0 共 ␪ ⫺ ␪ c 兲 , z⬎0, A 0 ⬎0,

共11兲
where ␪ e is the equilibrium temperature for stress-free A and
M, ⫺z is the jump in specific entropy at the equilibrium
temperature, and ␪ c is the critical temperature at which
stress-free A loses its thermodynamic stability. Then

f ⫽A 0 共 ␪ ⫺ ␪ c 兲 ␩ 2 /2⫹ 关 3z 共 ␪ ⫺ ␪ e 兲 ⫺A 0 共 ␪ ⫺ ␪ c 兲兴 ␩ 4 FIG. 1. G̃( ␩ ) for equal deviations of the critical temperatures

⫹ 关 A 0 共 ␪ ⫺ ␪ c 兲 ⫺4z 共 ␪ ⫺ ␪ e 兲兴 ␩ /2.
6
共12兲
Designating the critical temperature at which stress-free M
loses its thermodynamic stability as ¯␪ c , one obtains ¯␪ c ⫽ ␪ c
⫹6z( ␪ e ⫺ ␪ c )/(6z⫺A 0 ) with A 0 ⬍6z from Eq. 共6兲. The in-
equality A 0 ⬍6z was assumed in the derivation of the equa-
tion for ¯␪ c and it follows from the evident inequalities ¯␪ c
⬎ ␪ e ⬎ ␪ c that it is not contradictory. For equal deviations of
the critical temperatures from the equilibrium temperature
one obtains A 0 ⫽3z and further simplification of Eq. 共12兲.
For PT’s that can be treated as one dimensional, i.e., ␴:␧t
⫽ ␴ ␧ t , where ␴ and ␧ t are scalar measures of stress and
transformation strain,1 Eq. 共9兲 for ␩ 3 provides us with the
unstable equilibrium-stress–transformation-strain curve. Ne-
glecting the differences between the compliances and ther-
mal strain tensors of A and M we find
3 共 A⫺4⌬G ␪ 兲 ␩ 2 ⫺A
␴⫽ . 共13兲
␧ t 关 3 共 a⫺4 兲 ␩ 2 ⫺a 兴
from the equilibrium temperature at various stresses for 共a兲 ␪
⫽200 K and 共b兲 ␪ ⫽250 K. The same parameter values—namely
␧ t ⫽0.1, a⫽3, ␪ c ⫽100 K, ␪ e ⫽200 K, and A 0 ⫽3 MPa/K—were
used for both plots.
6 共 A 0 ⫺za 兲 ␪ ⫹za ␪ e ⫺A 0 ␪ c
H⫽ . 共15兲
␧t a 共 6⫺a 兲
This expression for H is the same as for the 2-3-4-5
potential.1 For A 0 ⬎za (A 0 ⬍za) the hysteresis grows 共de-
creases兲 with temperature and for A 0 ⫽za it is independent
of temperature.
The equation ␧⫽⫺ ⳵ G/ ⳵␴ gives the relation between
strain and order parameters for any actual transformation
path at constant ␴. Consider a single martensitic variant. In
thermodynamic equilibrium, ␧( ␴, ␩ )⫽⫺ ⳵ G/ ⳵␴ and 共9兲 for
␩ 3 ( ␴) constitute a parametric relation between strain and
order parameter with ␴ as the parameter. This ␧-␩ relation
simplifies considerably in a linear elastic material for which
Eq. 共13兲 holds:

In the approximation 共11兲, the ␴ -␩ curve depends linearly on 3 共 A⫺4⌬G ␪ 兲 ␩ 2 ⫺A

temperature, ␧⫽␧ t ␸ 共 a, ␩ 兲 ⫹␭ . 共16兲
␧ t 关 3 共 a⫺4 兲 ␩ 2 ⫺a 兴

3 关 A 0 共 ␪ ⫺ ␪ c 兲 ⫺4z 共 ␪ ⫺ ␪ e 兲兴 ␩ 2 ⫺A 0 共 ␪ ⫺ ␪ c 兲 This equation is valid for 0⭐ ␩ ⭐1, i.e., for ␭A/(␧ t a)⭐␧
␴⫽ , 共14兲 ⭐␧ t ⫹␭(A⫺6⌬G ␪ )/ 关 ␧ t (a⫺6) 兴 .
␧ t 关 3 共 a⫺4 兲 ␩ ⫺a 兴
2
In Fig. 1, we plot G̃ªG⫹ 21 ␴:␭:␴ versus ␩ for two tem-
and for the stress hysteresis Hª ␴ ( ␩ ⫽0)⫺ ␴ ( ␩ ⫽1) one peratures and various stresses in the approximation 共11兲 for
obtains A 0 ⫽3z, which corresponds to ␪ e ⫽( ␪ c ⫹¯␪ c )/2; the exten-

134201-4
THREE-DIMENSIONAL LANDAU . . . . III. . . .
sion to negative ␩ will be considered in Sec. II B. In contrast
to the 2-3-4-5 polynomial 共see Fig. 6 in Ref. 1兲, the curvature
共second derivative with respect to ␩ ) at the martensitic mini-
mum is significantly larger than at the austenitic minimum.
Indeed, the ratio 关 ⳵ 2 G(1)/ ⳵ ␩ 2 兴 / 关 ⳵ 2 G(0)/ ⳵ ␩ 2 兴 for the 2-4-6
potential is 4 times greater than for the 2-3-4-5 polynomial.
共Note that the curvatures of the A and M minima are the
same for the 2-3-4-5 potential at thermodynamic equilib-
rium.兲 This difference is not related to differences in the
elastic moduli or to the conditions for the loss of stability of
homogeneous phases. This difference in curvature leads to
significant differences between the profiles and energies of
the critical martensitic nuclei for the 2-4-6 and 2-3-4 poten-
tials 共see Sec. V兲. It also leads to convergence of the solu-
tions of the time-dependent Ginzburg-Landau equations to
different equilibrium domain wall configurations for the
same thermodynamic, initial, and boundary conditions,
which will be described in detail elsewhere.
The stress-strain curve for the 2-4-6 polynomial is similar
to that of the 2-3-4-5 polynomial, but there are differences;
see Fig. 5 in Ref. 1. First, for a⫽3, ␴ is a linear function of
␩ for the 2-3-4-5 polynomial, but ␴ is a nonlinear function
of ␩ for the 2-4-6 polynomial. The absolute value of the
tangent elastic modulus at ␩ ⫽1 is much larger for the 2-4-6
polynomial than for the 2-3-4-5 polynomial, a consequence
of the greater curvature of G( ␩ ) at ␩ ⫽1.
B. Continuation of the 2-4-6 order parameters
to negative values
Continuation of the order parameters to negative values
requires that the term C ␩ 3i ␩ 3j in F i j be replaced by
C 兩 ␩ i 兩 3 兩 ␩ j 兩 3 . Then F i j is invariant under ␩ i →⫺ ␩ i or
␩ j →⫺ ␩ j and spurious minima at ⫾ ␩ˆ i ⫾ ␩ˆ j , ⫾ ␩ˆ i ⫾ ␩ˆ j
⫾ ␩ˆ k , etc., can be eliminated by choosing B⭐0 and C⬎0,
as is the case for positive order parameters.
We start with the special case of martensitic variants with
transformation strains that differ only in sign. This can be the
case only if the transformation strain is purely deviatoric
since the volumetric strains must be of the same sign. This
condition is approximately met for the IPS variants in, for
example, CuAlNi, CuZnGa, CuZn, CuAlZn, AgCd, and NiAl
alloys.8 These pairs of IPS variants with ␧t of opposite sign,
M⫹ and M⫺ , have equal elastic compliances. The thermal
strain tensors will be treated below. Our 2-4-6 Landau poten-
tial, Eqs. 共1兲–共5兲, can be modified to describe transforma-
tions between A and n IPS variants and among n IPS variants
with n/2 order parameters. A single order parameter is asso-
ciated with M⫹ and M⫺ ; the variants are located at ⫾ ␩ˆ i for
some i. The modified form of the 2-4-6 potential is obtained
by substituting ␧ti sgn( ␩ i ) for ␧ti and n/2 for n. The sign
function sgn( ␩ i )⫽ ␩ i / 兩 ␩ i 兩 is always multiplied by ␩ m i ,
where 2⭐m⭐6. The second derivatives of the quadratic
terms include terms of the form 2 sgn( ␩ i ); therefore,
⳵ 2 G/ ⳵ ␩ 2i is discontinuous at the origin; G and ⳵ G/ ⳵ ␩ i are
continuous at the origin.
Let us consider for simplicity ␩ -independent elastic com-
pliances. For just two IPS variants M⫹ and M⫺ , the equilib-
134201-5
PHYSICAL REVIEW B 68, 134201 共2003兲
rium condition ⳵ G/ ⳵ ␩ ⫽0 gives, in addition to the roots ␩ 1
⫽0, ␩ 2 ⫽0, and ␩ 3 关see Eq. 共9兲兴, the negative roots
冑
a ␴:␧t ⫹A
␩ 4 ⫽⫺1, ␩ 5 ⫽⫺ .
3 关共 a⫺4 兲 ␴:␧t ⫺4⌬G ␪ ⫹A 兴
共17兲
The PT conditions, which follow from ␩ 3 ⫽0,1 and ␩ 5 ⫽0,
⫺1, are
A A
A→M⫹ : ␴:␧t ⭓ ; A→M⫺ : ␴:␧t ⭐⫺ ;
a a
6⌬G ␪ ⫺A
M⫹ →A: ␴:␧t ⭐ ;
6⫺a
6⌬G ␪ ⫺A
M⫺ →A: ␴:␧t ⭓⫺ . 共18兲
6⫺a
For ⌬G ␪ ⬎A(a⫺3)/3a, the M⫺ →A PT occurs at smaller
stresses than does the A→M⫹ PT; for a⫽3 this occurs at
␪ ⬎ ␪ e . In this case M⫺ first transforms to A and then, after
an increase in stress or a decrease in temperature, A trans-
forms to M⫹ . In the opposite case, M⫺ transforms directly to
M⫹ because A is unstable.
A volume preserving IPS is a simple shear in direction m
in the habit plane with normal n, ␧t ⫽ 21 ␥ t (mn⫹nm), ␴:␧t
⫽ ␶␥ t , ␶ ªm"␴"n; here ␶ and ␥ t are the shear stress and
strain. The one-dimensional treatment presented in Ref. 1 is
applicable.
In Fig. 1, the dependence of G̃ªG⫹ 21 ␴:␭:␴ on ␩ in the
interval 关 ⫺1,1兴 is presented for two temperatures and vari-
ous stresses for A 0 ⫽3z.
If part ¯␧ti of the transformation strain is the same for the
M⫹ and M⫺ variants and another part ¯␧ti is of opposite sign,
then both M⫹ and M⫺ can be described by the single order
parameter ␩ i upon substituting ¯␧ti ⫹ ¯␧ti sgn( ␩ i ) for ␧ti in Eqs.
共1兲 and 共5兲. For IPS, ¯␧ti ⫽␧ni ni and ¯␧ti ⫽ 21 ␥ t (mi ni ⫹ni mi ),
where ␧, the strain normal to the habit plane, is equal to the
volumetric strain. The thermal strain tensor may be similarly
decomposed, ␧␪ i ⫽ ␧␪dei v sgn( ␩ i )⫹ ␧␪v iol , where ␧␪dei v and ␧v␪ iol
are the deviatoric and volumetric parts of ␧␪ i .
For the cubic-orthorhombic PT 共e.g., Cu-Ni-Al and
Au-Cd alloys兲 and the cubic-monoclinic-II PT 共e.g., Cu-
Zn-Al alloys兲, ¯␧ti describes the diagonal components of ␧ti
and ¯␧ti corresponds to the only shear component.9 Three or-
der parameters are needed to describe the six martensitic
variants in the first case and six order parameters are required
to describe the 12 variants in the second case. For the cubic-
monoclinic-I PT 共e.g., Ni-Ti alloys兲, ¯␧ti is associated with
one shear and the three diagonal components of ␧ , and ¯␧ ti ti
is associated with the other two shear components.
Note that a similar procedure for decreasing the number
of degrees of freedom by a factor of 2 can be followed for
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

the 2-3-4-5 polynomial. In addition to substituting ¯␧ti
⫹ ¯¯␧ti sgn( ␩ i ) for ␧ti , one needs to substitute 兩 ␩ i 兩 for ␩ i for
all odd powers of ␩ i .
III. POTENTIALS IN HYPERSPHERICAL COORDINATES
In the n-dimensional space of order parameters, all mar-
tensitic variants are located on the unit hypersphere; thus it is
natural to construct Gibbs potentials using the hyperspherical
order parameters r and ␺ k , k⫽1, . . . ,n. Here r is the radial
coordinate in order-parameter space and ␲ ␺ k /2 is the angle
between the radius vector r and the ␩ k axis:
␸ 4 共 r 兲 ⫽ar 2 ⫹ 共 4⫺2a 兲 r 3 ⫹ 共 a⫺3 兲 r 4 , 0⬍a⬍6,
f 4 共 ␪ ,r 兲 ⫽Ar 2 ⫹ 共 4⌬G ␪ ⫺2A 兲 r 3 ⫹ 共 A⫺3⌬G ␪ 兲 r 4 , 共21兲
and for p⫽6 see Eqs. 共2兲 and 共3兲 with r substituted for ␩ k .
The functions P, Q, and q are to be determined. The require-
ment that G give the free energies of M1 and M2 at ␺ ⫽0 and
␺ ⫽1, respectively, implies the conditions
P 共 0 兲 ⫽0, P 共 1 兲 ⫽1, Q 共 0 兲 ⫽Q 共 1 兲 ⫽0. 共22兲
Without loss of generality we assume q(1)⫽1. We require
the radial derivative of G(r, ␺ ) to vanish at the origin and at
r⫽1 for 0⭐ ␺ ⭐1:

r⫽ 冉兺 冊
n

i⫽1
␩ 2i
1/2

, 0⭐ ␺ k ⫽
2
␲
␩k
cos⫺1 ⭐1,
r
⳵ G 共 0,␺ 兲 ⳵ G 共 1,␺ 兲
⳵r
⫽
⳵r
⫽0⇒
dq 共 0 兲 dq 共 1 兲
dr
⫽
dr
⫽0. 共23兲

冉 冊 共19兲
n
␲ Similarly, the ␺ derivative of G(r, ␺ ) is forced to vanish at
兺 cos2
k⫽1
␺ ⫽1.
2 k all r for ␺ ⫽0,1:

The third equation is a constraint that can be eliminated by ⳵ G 共 r,0兲 ⳵ G 共 r,1兲 d P共 0 兲 d P共 1 兲

expressing G in terms of the ␩ k . New Gibbs potentials in r ⫽ ⫽0⇒ ⫽ ⫽0,
⳵␺ ⳵␺ d␺ d␺
and ␺ k can be derived from our 2-3-4 and 2-4-6 potentials
共24兲
G(r) for a single martensitic variant by allowing for ␺ k de- dQ 共 0 兲 dQ 共 1 兲
pendence in the transformation strain and including a term ⫽ ⫽0.
that introduces ␺ k -dependent barriers between all variants. d␺ d␺
We emphasize that the potentials in hyperspherical coordi- These conditions include the equilibrium conditions
nates are not simply reparametrizations of the 2-3-4 and
␩ j )/ ⳵ ␩ i ⫽0 (i, j⫽1,2) and also impose additional con-
⳵ G( ˆ
2-4-6 potentials in Cartesian coordinates but rather new po-
straints on the form of the potential. Requiring that the in-
tentials with somewhat different physics. In contrast to the
equality
2-3-4 and 2-4-6 potentials, the Gibbs potentials in r and the
␺ k have the following desirable features.
共i兲 They have no unphysical minima and have no con-
⳵ 2 G 共 1,␺ 兲 / ⳵ ␺ 2 ⫽⫺ ␴:共 ␧t2 ⫺ ␧t1 兲 d 2 P/d ␺ 2 ⫹Ād 2 Q/d ␺ 2 ⭐0
共25兲
stants that do not appear in the phase equilibrium and trans-
formation conditions. for ␺ ⫽0 ( ␺ ⫽1) coincide with Eq. 共6兲 in part II 共Ref. 2兲—
共ii兲 The paths of minimum free energy between variants i.e., it give the same M1 →M2 (M2 →M1 ) PT criterion as Eq.
are great circles on the unit hypersphere. Variant-variant 共7兲 or 共5兲 in 共Ref. 2兲—results in
transformations occur along these paths and can be param-
etrized by a single angle. This property of a potential in ⳵ 2 P 共 0 兲 / ⳵ ␺ 2 ⫽⫺ ⳵ 2 P 共 1 兲 / ⳵ ␺ 2 ⫽6,
hyperspherical coordinates makes it possible to obtain an 共26兲
analytical kink solution of the time-independent Ginzburg- ⳵ 2 Q 共 0 兲 / ⳵ ␺ 2 ⫽ ⳵ 2 Q 共 1 兲 / ⳵ ␺ 2 ⫽2.
Landau equation that connects martensitic variants without
passing through an austenitic minimum. Similar solutions We also require limr→0 (q/ ␸ )⫽limr→0 (q/r 2 )⫽0, which
cannot be obtained for the potentials in Cartesian order pa- eliminates the barrier to the M1 ↔M2 PT in the vicinity of the
rameters. origin. Otherwise, following M1,2→A, the A remembers the
variant from which it came. Restricting our attention to 2-3-4
and 2-4-6 polynomials, we obtain the following functions
A. Two martensitic variants
which satisfy the above requirements:
We now consider the case n⫽2 for ␩ -independent elastic
compliances and thermal strain tensor. A polar coordinate Q⫽ ␺ 2 共 1⫺ ␺ 兲 2 , P⫽ ␺ 2 共 3⫺2 ␺ 兲 ,
system can be used and all derivations are quite simple. Our 共27兲
Gibbs potential is given by q 4 ⫽4r 3 ⫺3r 4 , q 6 ⫽3r 4 ⫺2r 6 .
The polynomials q, f, and ␸ must be of the same degree. If
G 共 ␴, ␪ ,r, ␺ 兲 ⫽⫺ ␴:␭:␴/2⫺ ␴:关 ␧t1 ⫹ 共 ␧t2 ⫺ ␧t1 兲 P 共 ␺ 兲兴 ␸ p 共 r 兲
this were not the situation, then additional unphysical ex-
⫹ f p 共 ␪ ,r 兲 ⫹ĀQ 共 ␺ 兲 q p 共 r 兲 , 共20兲 trema might appear and an analytical study would not be
possible. The polynomials in ␺ can be used with both fourth-
where p⫽4 corresponds to the 2-3-4 polynomial, p⫽6 cor- and sixth-degree polynomials in r. Note that Q can be written
responds to the 2-4-6 polynomial, and ␺ ⬅ ␺ 1 ⫽1⫺ ␺ 2 . For in the form Q⫽ ␺ 21 ␺ 22 . For n martensitic variants, this gener-
p⫽4 we have1 alizes to Q⫽ ␺ 21 ••• ␺ 2n ; see next section.

134201-6
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

Variant-variant transformations can be studied analyti- M j →Mi : 3 ␴:共 ␧t j ⫺ ␧ti 兲 ⫹Ā⭐0. 共32兲
cally by putting r⫽1 in the above equations. For example, a
solution of the equation ⳵ G(1,␺ )/ ⳵ ␺ ⫽0 is ␺ 3 ⫽1/2 These conditions are most easily verified by first expanding
⫺(3S)/(2Ā) with Sª ␴:( ␧t2 ⫺ ␧t1 ), which is the location of the potentials 共29兲 around the points ␩ˆ j and 0̄ to second
the barrier for the M1 →M2 PT. The corresponding activation order in the order parameters and then calculating the deriva-
barriers are tives.
In the neighborhood of ␩ˆ j the potential is
G 共 1,␺ 3 兲 ⫺G 共 1,0兲 ⫽ 共 1⫺3S/Ā 兲 3 共 Ā⫹S 兲 /16,
n
共28兲 1 4
G 共 1,␺ 3 兲 ⫺G 共 1,1兲 ⫽ 共 1⫹3S/Ā 兲 共 Ā⫺S 兲 /16.
3 G⫽⌬G ␪ ⫺ ␴:␭:␴⫺ ␴:␧t j ⫹ 2
2 ␲
兺
i⫽ j
关 3 ␴:共 ␧t j ⫺ ␧ti 兲 ⫹Ā 兴 ␩ 2i

A variant-variant kink solution of the time-independent ⫹ ␻ p 关共 6⫺a 兲 ␴:␧t j ⫹A⫺6⌬G ␪ 兴共 ␩ j ⫺1 兲 2 ⫹O 共 ␩ 3 兲 ,

Ginzburg-Landau equation for the r⫽1 potential will be ob-
tained in Sec. V. 共33兲
where ␻ 4 ⫽1 and ␻ 6 ⫽2. This always has an extremum
B. n martensitic variants
at ␩ˆ j :
We assume ␩ -independent elastic compliances and ther-
mal strain. Equations 共20兲 and 共27兲 generalize to ⳵ G 共 ␩ˆ j 兲
⫽0, i, j⫽1, . . . ,n. 共34兲
n ⳵␩i
1
G⫹ ␴:␭:␴⫽⫺ ␴:
2 k⫽1
兺
␧tk 共 1⫺3 ␺ 2k ⫹2 ␺ 3k 兲 ␸ p 共 r 兲 The mixed derivatives ⳵ 2 G/ ⳵ ␩ i ⳵ ␩ j (i⫽ j) of Eq. 共33兲 vanish
at all ␩ˆ k . It follows that the conditions ⳵ 2 G( ␩ˆ j )/ ⳵ ␩ 2j ⭐0 and
n
⳵ 2 G( ␩ˆ j )/ ⳵ ␩ 2i ⭐0 are the conditions for M j →A and M j
⫹ f p 共 ␪ ,r 兲 ⫹Ā 兿
k⫽1
␺ 2k q p 共 r 兲 . 共29兲
→Mi , respectively. The M j →A PT condition is given by

It is easy to check using ␺ 1 ⫹ ␺ 2 ⫽1 that, for n⫽2, Eq. 共29兲 ⳵ 2 G 共 ␩ˆ j 兲

reduces to Eq. 共20兲 with P and Q given by Eq. 共27兲. ⫽2 ␻ p 关共 6⫺a 兲 ␴:␧t j ⫹A⫺6⌬G ␪ 兴 ⭐0, 共35兲
We now verify the equilibrium conditions—namely, ⳵ ␩ 2j
⳵ G/ ⳵ ␩ i ⫽0—at the points 0̄⫽(0, . . . ,0) and ␩ˆ j , j which coincides with Eq. 共30兲. Similarly, the M j →Mi PT
⫽1, . . . ,n, as well as the phase instability conditions: condition is
6⌬G ␪ ⫺A ⳵ 2 G 共 ␩ˆ j 兲 8
Mi →A: ␴:␧ti ⭐ ; 共30兲 ⫽ 关 3 ␴:共 ␧t j ⫺ ␧ti 兲 ⫹Ā 兴 ⭐0, 共36兲
6⫺a
⳵ ␩ 2i ␲2
A in agreement with Eq. 共32兲.
A→Mi : ␴:␧ti ⭓ ; 共31兲
a In the neighborhood of the origin we have

FIG. 2. Level curves of G̃ for NiAl at ␪ ⫽0 and normal stresses ␴ 1 ⫽ ␴ 2 ⫽ ␴ 3 ⫽0: 共a兲 2-3-4 polynomial in polar coordinates, 共b兲 2-3-4
polynomial in Cartesian coordinates, and 共c兲 2-4-6 polynomial in Cartesian coordinates.

134201-7
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 3. Level curves of G̃ for NiAl at ␪ ⫽0 and normal stresses ␴ 1 ⫽1000, ␴ 2 ⫽⫺1000, ␴ 3 ⫽⫺3000: 共a兲 2-3-4 polynomial in polar
coordinates, 共b兲 2-3-4 polynomial in Cartesian coordinates, and 共c兲 2-4-6 polynomial in Cartesian coordinates.

1
G⫽⫺ ␴:␭:␴⫹ ␻ ⫺1
2 p ⫺a ␴: 冋
k⫽1
兺
n

␧tk 共 1⫺3 ␺ 2k ⫹2 ␺ 3k 兲 ⫹A r 2 册 A→M j :

⳵ 2 G 共 r⫽0,␺ j ⫽0,␺ i ⫽1 for i⫽ j 兲
⳵r2
共37兲
⫽ 共 2/␻ p 兲共 ⫺a ␴:␧t j ⫹A 兲 ⭐0, 共38兲

to second order in r. Obviously, G has an extremum at the
origin for any stress or temperature. Unlike the mixed de-
rivatives at the martensitic extrema, the mixed derivatives do
not vanish at the origin, but as we shall see, the A→Mi PT is
unaffected. The A instability must be determined from the
first fulfillment of the condition ⳵ 2 G/ ⳵ r 2 ⭐0 in some radial
direction. Since the minima of ⳵ 2 G(0̄)/ ⳵ r 2 are along the
coordinate axes, one gets
for the variant with the maximum value of ␴:␧t j , in agree-
ment with Eq. 共31兲.
Thus, the potential 共29兲 in hyperspherical coordinates sat-
isfies the equilibrium and PT conditions.
If ¯␧ti is the same for M⫹ and M⫺ variants and ¯␧ti is of
opposite sign, then the number of order parameters is re-
duced by a factor of 2 by substituting ¯␧ti ⫹ ¯␧ti sgn( ␩ i ) for ␧ti
in Eq. 共29兲.

FIG. 4. Level curves of G̃ for NiAl at ␪ ⫽0 and normal stresses ␴ 1 ⫽4051, ␴ 2 ⫽⫺2000, ␴ 3 ⫽⫺3000: 共a兲 2-3-4 polynomial in polar
coordinates, 共b兲 2-3-4 polynomial in Cartesian coordinates, and 共c兲 2-4-6 polynomial in Cartesian coordinates.

134201-8
THREE-DIMENSIONAL LANDAU . . . . III. . . .
IV. COMPARISON OF POTENTIALS: THE NiAl CUBIC-
TETRAGONAL PT
In part II 共Ref. 2兲 we found all material parameters of the
2-3-4-5 potential for the cubic-to-tetragonal PT in NiAl. The
relevant material constants for the 2-3-4-5 and polar 2-3-4
potentials are
␧t1 ⫽ 兵 0.215;⫺0.078;⫺0.078其 , a⫽2.980,
Ā⫽5320 MPa, ␪ e ⫽215 K,
共39兲
A 0 ⫽4.40 MPa K⫺1 , ␪ c ⫽⫺183 K, B⫽0,
D⫽500 MPa;
the tensors ␧t2 and ␧t3 can be obtained by permutation of
components. The constant D does not appear in the polar
potential. We chose C⫽⫺1000 from the condition that at
␴ ⫽⫺7000 MPa, which is far outside the region of stability
of the martensite and consequently far beyond stresses of
interest, G( ␴ ,0,1,1)ⰇG( ␴ ,0,1,0), and no unphysical
minima exist. This ensures that the same is true at smaller
compressive stresses or in tension.
Plots of G̃( ␴, ␪ , ␩ 1 , ␩ 2 )⫽G( ␴, ␪ , ␩ 1 , ␩ 2 )⫹ 21 ␴:␭:␴ for
PT’s in NiAl for zero stress and two three-dimensional stress
states at ␪ ⫽0 K for the 2-3-4-5, 2-4-6, and polar 2-3-4 po-
tentials are shown in Figs. 2– 4. The normal stresses on the
faces of the crystal are denoted ␴ i ; all stresses are in MPa.
We applied a large compressive stress ␴ 3 to suppress the
appearance of the third variant. Growth of G̃ corresponds to
variation from black to white. The driving force is orthogo-
nal to the level curves. There are no unphysical minima
present. This is particularly noteworthy for the polar 2-3-4
potential since it contains no constants that can be tuned to
eliminate unphysical minima. We also checked this for the
three-variant case by analyzing G̃ numerically.
For zero stresses 共Fig. 2兲, both M variants are stable and A
is metastable.
For ␴ 1 ⫽⫺ ␴ 2 ⫽1000 共pure shear in the 1-2 plane兲, ␴ 3
⫽⫺3000, M1 is stable, M2 is metastable, and A is unstable
共Fig. 3兲. Because of a barrier between M2 and A, A trans-
forms to M1 only.
For ␴ 1 ⫽4051, ␴ 2 ⫽⫺2000, and ␴ 3 ⫽⫺3000, M1 is
stable, and M2 is unstable only in the direction of M1 because
the barrier between M2 and A still exists 共Fig. 4兲.
The energy variation along the coordinate axes is identical
for the 2-3-4-5 potential and the polar 2-3-4 potential. The
differences between these potentials are most pronounced for
the stress-free case. Because the 2-4-6 potential has a much
larger curvature at the M minimum than at the A minimum,
there are significant differences in local values between the
2-4-6 and the other two potentials near the stable M mini-
mum. As will be shown in Sec. V, such local differences
result in different profiles for the critical M nuclei.
134201-9
PHYSICAL REVIEW B 68, 134201 共2003兲
V. CRITICAL NUCLEI AND DIFFUSE INTERFACES
In this section we will obtain and interpret some spatially
one-dimensional analytical solutions of the time-independent
Ginzburg-Landau equations for the 2-3-4, 2-4-6, and r-␺ k
potentials. As was shown in Ref. 6, some two-dimensional
problems for the cubic-rectangular PT can be treated as one
dimensional, which is the case for our models as well. Our
one-dimensional results on critical nuclei and diffuse inter-
faces hold for IPS variants—that is, for transformation
strains of the form ␧t ⫽ 21 ␥ t (mn⫹nm)sgn( ␩ )⫹␧nn, where
⫺1⭐ ␩ ⭐1 for the 2-4-6 potential and 0⭐ ␩ ⭐1 for the 2-3-4
potential. The order parameter is a function of the coordinate
x along the n axis. Despite the restriction to one spatial di-
mension, our solutions are valid for an arbitrary three-
dimensional homogeneous stress tensor ␴ in a rectangular
parallelepiped with corresponding homogeneous tractions at
its faces. In order to impose the usual boundary condition on
the order parameter 关see Eq. 共55兲兴, the faces of the parallel-
epiped must be orthogonal and parallel to n 共Fig. 5兲. In the
following subsections the parallelepiped is actually infinite in
the x direction.
A. Governing equations
We write the 2-4-6 and 2-3-4 potentials in terms of two
parameters
G̃ 6 ⫽s 1 ␩ 2 关 1⫺ 共 4⫺ P 兲 ␩ 2 /2⫹ 共 3⫺ P 兲 ␩ 4 /3兴 /2, 共40兲
G̃ 4 ⫽s 1 ␩ 2 关 1⫺ 共 6⫺ P 兲 ␩ /3⫹ 共 4⫺ P 兲 ␩ 2 /4兴 , 共41兲
s 1 ªA⫺a ␴:␧t , s 2 ª12共 ⌬G ␪ ⫺ ␴:␧t 兲 , Pªs 2 /s 1 .
共42兲
Here and later the subscripts 4 and 6 refer to 2-3-4 or 2-4-6
potentials, respectively. We have
⳵ G 6 / ⳵ ␩ ⫽⫺s 1 ␩ 共 1⫺ ␩ 2 兲关共 3⫺ P 兲 ␩ 2 ⫺1 兴 ,
⳵ G 4 / ⳵ ␩ ⫽s 1 ␩ 共 1⫺ ␩ 兲关 2⫺ 共 4⫺ P 兲 ␩ 兴 ;
⳵ 2 G 6 共 0 兲 / ⳵ ␩ 2 ⫽ 共 1/2兲 ⳵ 2 G 4 共 0 兲 / ⳵ ␩ 2 ⫽s 1 ,
⳵ 2 G 6 共 1 兲 / ⳵ ␩ 2 ⫽2 ⳵ 2 G 4 共 1 兲 / ⳵ ␩ 2 ⫽2s 1 共 2⫺ P 兲 ;
␩ 63⫽ 冑 1
3⫺ P
, G̃ 6 共 ␩ 63兲 ⫽
s 1 8⫺3 P
12 共 P⫺3 兲 2
;
共43兲
2 4 s 1 共 3⫺ P 兲
␩ 43⫽ , G̃ 4 共 ␩ 43兲 ⫽ .
4⫺ P 3 共 4⫺ P 兲 3
Here ␩ 63 and ␩ 43 correspond to the maxima of G̃, as in Eq.
共9兲.
For a single order parameter ␩ that depends only on the
coordinate x along the normal to the habit plane, the
Ginzburg-Landau energy G GL ⫽G⫹“ ␩ "␤"“ ␩ reduces to
G GL ⫽G⫹ ␤ ( ⳵ ␩ / ⳵ x) 2 , where the scalar ␤ is a linear combi-
nation of components of the second-rank tensor ␤ in the
crystal coordinates. Thus, even for a highly anisotropic A
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 5. Scheme for solution of the Landau-Ginzburg equation.

␩ (x) represents critical nuclei and kink solutions. The crystal lattice
transforms from state 1 to 2 by invariant-plane strain.

lattice, the one-dimensional case under consideration re-

quires only a single gradient parameter ␤ . The Ginzburg-
Landau energy G GL leads to the time-dependent Ginzburg-
Landau equation

⳵␩
⳵t
⫽⫺␭
␦ G GL
␦␩
⫽⫺␭
⳵G
⳵␩
⳵ 2␩
⫺2 ␤ 2 .
⳵x
冉 冊 共44兲

Here ␭⬎0 is the kinetic coefficient. We rescale variables to

cast Eq. 共44兲 in dimensionless form. The dimensionless po-
tentials and order parameters are

g 6 ⫽m 6 G̃ 6 ⫽B 6 ␰ 26 ⫺ ␰ 46 ⫹ ␰ 66 , ␰ 6 ⫽k 6 ␩ ; 共45兲

g 4 ⫽m 4 G̃ 4 ⫽B 4 ␰ 24 ⫺ ␰ 34 ⫹ ␰ 44 , ␰ 4 ⫽k 4 ␩ . 共46兲

The parameters are defined and related as follows:

B 6⫽
4 共 3⫺ P 兲
3 共 4⫺ P 兲 2
, k 6⫽ 冑冑 2
3
3⫺ P
4⫺ P
,

16共 3⫺ P 兲 2 2k 26 B 6
m 6⫽ ⫽ ;
9s 1 共 4⫺ P 兲 3 s1
共47兲
9 共 4⫺ P 兲 3 共 4⫺ P 兲
B 4⫽ , k 4⫽ ,
4 共 6⫺ P 兲 2 4 共 6⫺ P 兲 FIG. 6. Plots of 共a兲 B( P), 共b兲 k( P), and 共c兲 s 1 m( P). Solid
共dashed兲 lines correspond to the 2-4-6 共2-3-4兲 potential.

␰ 61⫽ 冑0.5共 1⫺ 冑1⫺4B 6 兲 ,

81共 4⫺ P 兲 3 k 24 B 4
m 4⫽ ⫽ . g 6 ⫽ ␰ 26 共 ␰ 26 ⫺ ␰ 61
2
兲共 ␰ 26 ⫺ ␰ 62
2
兲,
64s 1 共 6⫺ P 兲 4 s1

It is easy to check that k can be determined by the condition ␰ 62⫽ 冑0.5共 1⫹ 冑1⫺4B 6 兲 ;
dg/d ␰ ⫽0 at the martensitic minimum. Plots of B( P), k( P), 共48兲
and s 1 m( P) in the region of coexistence of A and M ( P g 4 ⫽ ␰ 24 共 ␰ 4 ⫺ ␰ 41兲共 ␰ 4 ⫺ ␰ 42兲 , ␰ 41⫽0.5共 1⫺ 冑1⫺4B 4 兲 ,
⬍2) for both potentials are presented in Fig. 6. We also
define ␰ 61 , ␰ 62 , ␰ 41 , and ␰ 42 : ␰ 42⫽0.5共 1⫹ 冑1⫺4B 4 兲 .

134201-10
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

deep potential well. For s 1 ⬍0 and P⬍2, where A and M are

unstable, there is an artificial minimum at 0⬍ ␩ 3 ⬍1.
These drawbacks do not affect the properties of the sys-
tem in the coexistence region or along instability 共PT兲 lines,
but problems can arise when thermomechanical loading
moves the point (s 1 ,s 2 ) too close to the unphysical wedge in
the s 1 -s 2 plane. If the initial phase is metastable, then rapid
loading toward the unphysical wedge can result in a trans-
formation to a spurious phase. On the other hand, if the rate
of thermomechanical loading is sufficiently low, then the
transition will be into the stable phase. In general, however,
the rate will not be low enough to ensure transformation to
the stable phase, but even if this is the case or if the initial
phase is stable, the point (s 1 ,s 2 ) must not move too close to
the unphysical wedge or else the small potential barrier sepa-
rating the stable phase from the spurious phase will be sur-
FIG. 7. Phase equilibrium and transformation diagrams in the mounted by thermal fluctuations. These shortcomings in the
s 1 -s 2 plane for the 2-3-4 and 2-4-6 potentials. potentials can be circumvented by modifying them outside
the region of phase coexistence. The only requirements for
We will analyze all results in terms of P, because it is the
such a modification are that it be continuous on the instabil-
same for both polynomials, and in terms of 0⭐B 4 ⭐9/32 and
ity lines and that it have only one minimum in the region of
0⭐B 6 ⭐1/3, which vary over narrow ranges and allow more
M instability at ␩ ⫽0 and a single minimum in the region of
vivid comparisons. It is also convenient to use the parameter
A instability at ␩ ⫽1. So, for example, for P⬎2 we can put
␣ ª P/(4⫺ P): ␣ ⫽0 for thermodynamic equilibrium, ␣ ⫽1
when M loses its stability, and ␣ ⫽⫺1 when A loses its G̃ 6 ⫽s 1 ␩ 2 共 P⫺1⫺ ␩ 2 ⫹ ␩ 4 /3兲 /2,
stability. Dimensionless order parameters and potentials are
convenient for intermediate steps in calculations but our so-
G̃ 4 ⫽s 1 ␩ 2 关 P⫺1⫺ 共 4/3兲 ␩ ⫹ ␩ 2 /2兴 , s 1 ⬎0;
lutions of the time-independent Ginzburg-Landau equations
are always presented in terms of the physical order parameter 共49兲
␩ (x) for two reasons: first, there is no direct scaling between G̃ 6 ⫽s 1 共 1⫺s 1 兲 P ␩ 4 共 ␩ 2 /3⫺1/2兲 /2,
␸ ( ␰ ) and ␸ ( ␩ ), and second, the dimensionless parameters
for the two potentials—k 4 and k 6 , for example—differ sig- G̃ 4 ⫽s 1 共 1⫺s 1 兲 P ␩ 2 共 1/3⫺ ␩ /4兲 , s 1 ⬍0.
nificantly for the same thermodynamic state. With such a modification, regions where only one phase ex-
Equilibrium and PT lines in the s 1 -s 2 plane are shown in ists can be correctly described in terms of P and B. For P
Fig. 7. The line s 2 ⫽0 ( P⫽0 and B 4 ⫽B 6 ⫽1/4) for s 1 ⭓0 is ⬎2 (0⬍B 4 ⬍9/32, 0⬍B 6 ⬍1/3), only A exists for s 1 ⬎0
the equilibrium line between A and M. For s 2 ⬎0 ( P⬎0 and and only M exists for s 1 ⬍0.
B⬎1/4) and s 1 ⬎0, A is stable 共relative to M), and for s 2 As an alternative to modifying the potential, the order
⬍0 ( P⬍0 and 0⬍B⬍1/4) and s 1 ⬎0, M is stable. Marten- parameters can be constrained in numerical simulations to
site is also stable 共relative to A) for s 1 ⬍0 and P⬎2. The the interval 关 0,1兴 or 关 ⫺1,1兴 for the extended 2-4-6 potential.
line s 1 ⫽0 for s 2 ⭐0 ( P⫽⫺⬁ and B⫽0) is the line of loss In other words, reflective boundary conditions can be im-
of stability of A; i.e., for s 1 ⭐0 and s 2 ⭐0 ( P⭓0), only M posed at the ends of these intervals.
can exist, though not everywhere in this region. The line P Introducing new spatial and time variables by the equa-
⫽2 (B 4 ⫽9/32, B 6 ⫽1/3) for s 1 ⭓0 is the line of loss of tions
stability of M; i.e., for P⬎2 and s 1 ⭓0, only A can exist.
Consequently, both phases coexist in the sector ⫺⬁⬍ P⬍2
(0⬍B 4 ⬍9/32, 0⬍B 6 ⬍1/3) for s 1 ⬎0. The sector ⫺⬁⬍ P y 6⫽
k6
冑␤ m 6
x⫽ 冑 s1
2␤B6
x⫽
冑6
4
冑 s 1 4⫺ P
␤ 冑3⫺ P
x, z 6⫽
␭k 26
m6
t;
⬍2 (0⬍B 4 ⬍9/32, 0⬍B 6 ⬍1/3) for s 1 ⬍0 is an unphysical
region because both M and A are unstable. In fact our poten- 共50兲
tials were designed to describe material behavior in the co-
existence region and at instability lines. Our potentials are
not applicable outside this region. Some of the above in-
y 4⫽
k4
冑␤ m 4
x⫽ 冑␤ s1
B4
x⫽
2
3
冑␤ 冑s 1 6⫺ P
4⫺ P
x, z 4⫽
␭k 24
m4
t,

equalities cannot be expressed in terms of B because B is a we obtain the dimensionless form of the Ginzburg-Landau
nonmonotonic function of P. equation
Austenite is stable relative to M in the wedge s 1 ⬎0, P
⬎2, but for P⬎ P Ac , where P 6c A
⫽8/3 and P 4c
A
⫽3, there is an
artificial minimum more stable than A at ␩ ⬎1 关see Eq.
共43兲兴. For s 1 ⬍0 and P⬎2, M is stable relative to A, but for
⳵␰
⳵z
⫽⫺
⳵g
⳵␰ 冉⳵ 2␰
⫺2 2 .
⳵y
冊 共51兲

2⬍ P⬍ P M c , where P 6c ⫽3 and P 4c ⫽4, a finite barrier at

M M
Henceforth we consider only time independent solutions—
␩ ⬎1 separates the martensitic minimum from an infinitely i.e., ⳵␰ / ⳵ z⫽0. The resulting equation 2d 2 ␰ /dy 2 ⫽dg/d ␰ is

134201-11
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

the equation of motion of a material point with mass equal to
2 in the potential field ⫺g( ␰ ). An energy integral reads
d ␰ /dy⫽ 冑g⫺g 0 ,
where g 0 is the integration constant. At points where
d ␰ /dy⫽0, e.g., at the center of a nucleus, g⫽g 0 . Designat-
B. Critical martensitic nucleus: M soliton in A
Let us start with the case when A exists as x→⫾⬁, hence
g 0 ⫽0. The solutions below are valid in the region of stabil-
共52兲 ity of M and metastability of A, i.e., s 1 ⬎0 and P⭐0 (0
⭐B⭐1/4). One has
␰ 41

冉 冊
ing ␰M
4 共 y 4兲⫽ ,
␰ 41
1⫹ 1⫺ sinh2 关 冑B 4 共 y 4 ⫺y 04兲 /2兴
* ⫽g GL ⫺g 0 ⫽g⫺g 0 ⫹ 共 d ␰ /dy 兲 2
g GL 共53兲 ␰ 42
共58兲
and taking into account Eq. 共52兲, one finds that the contribu- ␰ 61

冑 冉 冊
* from g⫺g 0 and the gradient of the order param-
tions to g GL ␰M
6 共 y 6兲⫽ ,
eter ␰ are the same, hence g GL* ⫽2(g⫺g 0 ). Generally, Eq. ␰ 61
2

共52兲 has periodic solutions with n diffuse interfaces. The to- 1⫹ 1⫺ sinh2 关 冑B 6 共 y 6 ⫺y 06兲兴
tal energy per unit area of n diffuse interfaces is given by ␰ 62
2

eª 冕 ⫺l
l
* dy⫽2n
g GL 冕冑 g⫺g 0 d ␰ , 共54兲
␩M
␰4 冉冑 冊 s1
␤B4
x
4 共 x 兲⫽
k4
where lª 冑s 1 /( ␤ B)L, 2L is the length of a parallelepiped in
the x direction, and the integration limits of the second inte- ⫽6 兵 6⫺ P⫹ 冑P 2 ⫺3 Pcosh关 冑s 1 / ␤ 共 x⫺x 0 兲兴 其 ⫺1 ,
gral depend on the type of interface. The energy e is finite
共59兲

冉冑 冊
even for an infinite slab. The total energy of the system is
infinite for an infinite parallelepiped in the case g 0 ⫽0. s1
␰6 x
Falk5 found periodic solutions for n domain walls in finite 2␤B6
regions, but claims that the separation between domain walls
␩M
6 共 x 兲 ⫽
k6
must be infinite in an infinite region. However, all of his
finite-l solutions depend on l only through the combination ⫽2 兵 4⫺ P⫹ 冑P 2 ⫺8 P/3cosh关 冑2s 1 / ␤ 共 x⫺x 0 兲兴 其 ⫺1/2.
l/n. Consequently, the finite-l solutions can be used for l The energies are given by
→⬁ provided n→⬁, keeping the ratio l/n, the distance be-
tween the domain walls, finite.
1⫹2 冑B 4
4 ⫽ 共 3⫺8B 4 兲 冑B 4 /6⫹ 共 B 4 ⫺1/4 兲 ln
Imposing the usual boundary conditions at the ends of the eM ,
slab 冑1⫺4B 4

d ␰ 共 ⫺l 兲 /dy⫽d ␰ 共 l 兲 /dy⫽0, 共55兲 1⫹2 冑B 6

6 ⫽ 冑B 6 /2⫹ 共 B 6 ⫺1/4 兲 ln
eM ,
one obtains 冑1⫺4B 6
共60兲

Using Eq. 共45兲, one derives

g 共 ⫺l 兲 ⫽g 共 l 兲 ⫽g 0 . 共56兲
4⫽
EM
1
k 24
冑␤ s 1 M 32 共 6⫺ P 兲
e ⫽
B 4 4 27 共 4⫺ P 兲 5/2
3
冑␤ s 1 e M4 ,

y 4共 ␰ 4 兲 ⫽ 冕冑 d␰4
B 4 ␰ 24 ⫺ ␰ 34 ⫹ ␰ 44 ⫺g 40
, 6⫽
EM
1
k 26
冑␤ s 1 M 3 冑6 共 4⫺ P 兲 2
e ⫽
2B 6 6 8 共 3⫺ P 兲 3/2
冑␤ s 1 e M6 .
共57兲
The solution ␰ M 6 (y 6 ) formally coincides with Falk’s

y 6共 ␰ 6 兲 ⫽ 冕冑 d␰6
B 6 ␰ 6 ⫺ ␰ 46 ⫹ ␰ 66 ⫺g 60
2
.
Despite the significant differences between our theory and
Falk’s5 and completely different variables and parameters,
Eq. 共57兲 2 is of the same form as Eq. 共14兲 in Ref. 5. This
means that we can use all of Falk’s formal periodic analytical
solutions and analyze them in terms of our governing param-
eters for the 2-4-6 potential. The same procedure can be
followed for the calculation of the total energy of the system.
solution.5 Falk did not provide a physical interpretation of
his solution, but we do so here. At first sight, the solution
共58兲 looks contradictory: the smaller B 共more negative P)
and the greater the stability of M, the smaller the magnitude
of the order parameter of the M nucleus. This apparent con-
tradiction disappears if we interpret the above solution as a
critical martensitic nucleus in austenite. We verified by nu-
merically solving the Ginzburg-Landau equation 共51兲 that
this solution corresponds to unstable thermodynamic equilib-
rium. An initial profile slightly larger 共or smaller兲 than

134201-12
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 8. 共a兲 Growth of a supercritical M nucleus (B⫽0.2, w⫽1.1). 共b兲 Disappearance of a subcritical M nucleus (B⫽0.24, w⫽0.9). 共c兲
Growth of a supercritical A nucleus (B⫽0.3, w⫽1.1). 共d兲 Disappearance of a subcritical A nucleus (B⫽0.26, w⫽0.9). All plots are for the
2-4-6 potential.

␰M6 (y 6 ) converges to M 共or A) everywhere; see Fig. 8. The

smaller B, the smaller the size and energy 关see Eq. 共60兲兴 of
the critical nucleus. Numerical solution of the Ginzburg-
⳵␰
⳵z
⫽⫺ 冉
⳵g共 w␰s兲
⳵␰
⳵ 2␰ s
⫺2w 2
⳵y
冊
Landau equation for the 2-3-4 potential confirms that ␰ M
4 (y 4 ) ⳵g共 w␰s兲 ⳵g共 ␰s兲
is also a critical M nucleus. ⫽⫺ ⫹w
⳵␰ ⳵␰
Only for thermodynamic equilibrium, P⫽0 or B⫽1/4,
does the magnitude of the order parameter ␰ 41⫽ ␰ 61
2
⫽1/2 cor- ⫽2 ␰ s3 w 共 w 2 ⫺1 兲关 2⫺3 ␰ s2 共 1⫹w 2 兲兴 . 共61兲
respond to complete M ( ␩ ⫽1); otherwise, it is smaller. Note
that for homogeneous stresses and temperature, there is no Consider a supercritical nucleus, i.e., w⬎1, for which we
solution for a stable M nucleus that grows with an increase in expect ⳵␰ / ⳵ z⬎0. However, this is true only when ␰ s,max
the thermodynamic driving force for the A→M PT. satisfies ␰ s,max
2
⬍2/关 3(1⫹w 2 ) 兴 ⬍1/3, which is the case for
Let us designate the solution of the stationary Ginzburg- B⬍2/9. For a small thermodynamic driving force 2/9⭐B
Landau equation as ␰ s . We will consider the 2-4-6 potential; ⭐1/4, the inequality ␰ 21 ⬍2/关 3(1⫹w 2 ) 兴 is violated in the
the derivations and results for the 2-3-4 polynomial are simi- central part of the nucleus. Consequently, the magnitude of ␰
lar. We can estimate ⳵␰ / ⳵ z at the instant when the nucleus initially decreases near the center of the nucleus while it
w ␰ s appears from Eq. 共51兲; w is a constant multiplicative grows in the rest of the nucleus. After a short initial stage,
factor: growth occurs throughout the nucleus.

134201-13
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

For a subcritical nucleus, w⬍1, ⳵␰ / ⳵ z⬍0 if the same

inequality is satisfied. In this case, the central region of the
nucleus grows during a brief initial stage, after which ␰ de-
creases throughout the nucleus until it disappears.
The above analytical estimates provide important insights
into the behavior of the nuclei but, of course, detailed, quan-
titative results on the growth or shrinkage of the nuclei can
only be obtained from numerical simulations, for example,
Fig. 8.
As was mentioned, for the 2-4-6 potential we can use
Falk’s periodic, multinuclei solutions for a finite region5 the
energy of such solutions 共the activation energy for nucle-
ation兲 is roughly proportional to the number of nuclei n/2.
However, the physical interpretation of multinuclei solutions
is unclear. Is this a stable solution or an unstable one? If it is
a metastable configuration, can it be reached under some
dynamic process and how high is the energy barrier that
separates it from a lower-energy state with a smaller number
of nuclei or from a complete M state? These problems will
be studied elsewhere.
The amplitudes of the physical order parameters of the
critical M nuclei are

␰ 41 4 共 1⫺ 冑1⫺4B 4 兲
␩ 41⫽ ⫽ ,
k4 3⫹ 冑9⫺32B 4
共62兲

␩ 61⫽
␰ 61
k6
⫽ 冑 3 1⫺ 冑1⫺4B 6
2 1⫹ 冑1⫺3B 6
.

From now on, a subscript ‘‘1’’ on an order parameter will

FIG. 9. The amplitudes of the critical M nuclei vs B: 共a兲 the
indicate the amplitude of that order parameter. The functions
function ␩ 1 (B), 共b兲 the functions ␸ 关 ␩ 1 (B) 兴 for several a. Solid
␩ 1 and ␸ ( ␩ 1 ) vs B are shown in Fig. 9. The amplitude ␩ 61 is
共dashed兲 lines correspond to the 2-4-6 共2-3-4兲 potential.
significantly larger than ␩ 41 except at the end points B⫽0
and B⫽1/4. Note that d ␩ 1 /dB is infinite at the end points
for the 2-4-6 potential and only at B⫽1/4 for the 2-3-4 poly-
nomial.
Figure 10 shows E M/ 冑␤ s 1 as a function of ␣ ⫽ P/(4 cantly greater than ␸ 4 关 ␩ 4 (x) 兴 everywhere. When P→0, the
⫺P). The energy of the 2-3-4 critical M nucleus is smaller entire region is martensite.
than the energy of the critical M nucleus for the 2-4-6 poten- The effective width W of the nucleus can be defined as
tial. The activation energy in thermodynamic equilibrium W⫽2u(⬁)/␧ ␸ ( ␩ 1 ), where u(⬁) is the displacement at in-
( ␣ ⫽0) is finite, in contrast to the infinite energy of a clas- finity and ␧ is the volumetric strain. Thus a nucleus of width
sical nucleus with a sharp interface. When A loses its stabil- W with a sharp interface and constant volumetric strain
␧ ␸ ( ␩ 1 ) produces the same displacement increment as the
6 ⫽E 4 ⫽0, while the activation energy is finite in clas-
ity, E M M

sical nucleation theory. critical nucleus. The displacement at infinity due to volumet-
ric strain can be calculated by integrating
The profiles ␩ ( 冑s 1 / ␤ x, P) and ␸ ( 冑s 1 / ␤ x, P) of the criti-
cal nuclei are shown in Fig. 11 for various P and B. The ␸
profiles are essentially strain profiles because ␧ t ␸ 关 ␩ (x) 兴 is du
the transformation strain. The ␸ and ␩ profiles are quite ␧ ␸ 关 ␩ 共 x 兲兴 ⫽ , 共63兲
dx
different, especially for small ␩ and 1⫺ ␩ where the stron-
gest nonlinearities of the function ␸ are located. The ␸ pro-
files are narrower than the ␩ profiles. For P⫽⫺1 the 2-4-6 where u(0)⫽0. The total displacement contains an addi-
and 2-3-4 nuclei are almost indistinguishable. At larger P the tional contribution due to constant elastic strain, u e
2-4-6 and 2-3-4 nuclei have the same amplitudes but the ⫽2lnn:␭:␴, which we neglect here. The effective widths of
2-3-4 nucleus is wider. At smaller P, ␸ 6 关 ␩ 6 (x) 兴 is signifi- the nuclei are

134201-14
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 10. The energy E/ 冑␤ s 1 vs ␣ ⫽ P/(4⫺ P) for the critical M nuclei, the critical A nuclei, the A-M interfaces for the 2-3-4 (AM4 ) and
2-4-6 (AM6 ) potentials, and the M-M interface for the 2-4-6 potential. Solid 共dashed兲 lines correspond to the 2-4-6 共2-3-4兲 potential.

6 共 P 兲⫽
WM
3
8 共 3⫺ P 兲 2 ␸ 6 共 ␩ 61兲
冑␤再 冑
2s 1
1
6
3
3⫺ P
tanh⫺1
1
4
冉冑 3
3⫺ P
共 4⫺ P⫺ 冑P 2 ⫺8 P/3兲 冊
⫻ 关 576⫺24共 a⫹14兲 P⫹ 共 11a⫹36兲 P 2 兴 ⫹12a⫹ 共 12⫺7a 兲 P , 冎
再 冉 冊
共64兲

4 共 P 兲⫽
WM
8
9 共 4⫺ P 兲 3 ␸ 4 共 ␩ 41兲
冑 ␤ 2
s 1 3 冑4⫺ P
tanh⫺1
6⫺ P⫺ 冑P 2 ⫺3 P
3 冑4⫺ P

⫻ 关 864⫺36共 a⫹15兲 P⫹18共 a⫹5 兲 P 2 ⫺ 共 a⫹6 兲 P 3 兴 ⫺24共 3⫺a 兲 ⫹12共 6⫺a 兲 P⫺ 共 a⫹6 兲 P 2 . 冎
We define the interface thickness as
␸共 ␩1兲
⌬ Mª , 共65兲
兩 d ␸ 关 ␩ 共 x m 兲兴 /dx 兩

where x m corresponds to the maximum of the derivative d ␩ (x)/dx. The ideal definition would be ⌬ M
ª ␸ ( ␩ 1 )/ 兩 d ␸ 关 ␩ (x) 兴 /dx 兩 max but this cannot be analyzed analytically. The interface thicknesses are

⌬M
4⫽
共 3⫹I 4 兲 5 关 12B 4 ⫺ 共 I 4 ⫺3 兲共 J 4 ⫺1 兲兴
96冑B 4 共 1⫹J 4 兲 共 1⫹I 4 ⫹2J 4 兲关 3 共 3⫹I 4 兲 ⫺32B 4 兴 冑共 I 4 ⫺3 兲共 1⫹J 4 兲 ⫹12B 4
2
冑␤ s1
,

6⫽
⌬M
共 1⫹J 6 兲 4 关 4B 6 ⫺ 共 I 6 ⫺1 兲共 J 6 ⫺1 兲兴
2 冑B 6 共 1⫹I 6 兲共 1⫹I 6 ⫹2J 6 兲共 1⫹J 6 ⫺3B 6 兲 冑共 1⫹I 6 兲共 J 6 ⫺1 兲 ⫹4B 6
冑␤ s1
, 共66兲

I 4 ª 冑9⫺32B 4 , J 4 ª 冑1⫺4B 4 , I 6 ª 冑1⫺4B 6 , J 6 ª 冑1⫺3B 6 .

The dimensionless width W M冑s 1 / ␤ and dimensionless
thickness ⌬ M冑s 1 / ␤ are plotted as functions of B in Fig. 12.
The width W M冑s 1 / ␤ tends to infinity for both potentials as
the PT equilibrium line (B⫽1/4, P⫽0) is approached, while
the interface thickness ⌬ M冑s 1 / ␤ remains finite. As B→0
(s 1 →0), W M冑s 1 / ␤ and ⌬ M冑s 1 / ␤ go to finite values; there-
fore, W M and ⌬ M tend to infinity. Excluding the neighbor-
hood of B⫽1/4, both parameters are of comparable magni-
tude; the interface is sharp only near thermodynamic
equilibrium. The width of the nucleus is larger for the 2-4-6

134201-15
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 11. Profiles of critical martensitic nuclei for various P and a⫽2.98. Solid 共dashed兲 lines are for the 2-4-6 共2-3-4兲 potential.

potential than for the 2-3-4 potential. The interface thickness ⫽g(M)⫽g 0 , i.e., g(A)⫽g(M)⫽0, which is the case when A
is significantly larger for the 2-3-4 potential for B⬎0.17 and and M are in thermodynamic equilibrium: s 2 ⫽ P⫽0, B
almost the same for both potentials for smaller B. ⫽1/4, ␰ 41⫽ ␰ 42⫽ ␰ 61 2
⫽ ␰ 62
2
⫽1/2, g 6 ⫽ ␰ 26 ( ␰ 26 ⫺1/2) 2 , g 4
The interface thicknesses ⌬ M 4 and ⌬ 6 are given to within
M
⫽ ␰ 4 ( ␰ 4 ⫺1/2) . The solutions of Eq. 共52兲 read
2 2
0.4% by the cubic polynomial approximations
⫺(y 4 ⫺y 04)/2 ⫺1
␰ AM
4 共 y 4 兲 ⫽ 关 2 共 1⫹e 兲兴 ,
⌬M
4 ⫽ 共 1.942⫹1.779B 4 ⫺12.80B 4 ⫹68.56B 4 兲
2 3
冑␤ s1
,
␰ AM ⫺(y 6 ⫺y 06) ⫺1/2
6 共 y 6 兲 ⫽ 关 2 共 1⫹e 6 ⫽3e 4 ⫽1/8;
, e AM AM
兲兴
共67兲
共68兲
⌬M
6 ⫽ 共 2.000⫹0.732B 6 ⫹0.381B 6 ⫹10.66B 6 兲
2 3
冑␤ s1
. 4 共 x 兲 ⫽ 兵 1⫹exp关 ⫺ 冑s 1 / ␤ 共 x⫺x 0 兲兴 其
␩ AM ⫺1
,

6 共 x 兲 ⫽ 兵 1⫹exp关 ⫺ 冑2s 1 / ␤ 共 x⫺x 0 兲兴 其

⫺1/2
The effective surface energy of a nucleus is defined in Ref. 7 ␩ AM .
as ⌫ Mª 关 E M⫺G( ␩ 1 )W 兴 /2, so G( ␩ 1 )W is the effective bulk
energy of the nucleus. However, G( ␩ 1 )⫽0 by definition of The solution ␩ AM
4 (x) is symmetric around x⫽x 0 but ␩ 6 is
AM

␩ 1 关see Eq. 共52兲 for d ␰ /dy⫽0 and g 0 ⫽0]; i.e., the energy of
the nucleus is localized at its surface according to this defi-
nition. For our theory, a better definition is ⌫ M
ª 兵 E M⫺G 关 ␸ ( ␩ 1 ) 兴 W 其 /2 since G 关 ␸ ( ␩ 1 ) 兴 ⫽0.
C. Kink solutions: A-M diffuse interfaces
We consider the case where the phase is A as x→⫺⬁ and
M as x→⫹⬁. Then, g(⫺⬁)⫽g(A)⫽g0⫽0 and g(⫹⬁)
not 共see Fig. 13兲 because the A and M minima of the 2-4-6
potential have different curvatures 关note, however, that
( ␩ AM 2
6 ) is symmetric around x⫽x 0 ]. In Fig. 13 we used
x 04⫽0 and x 06⫽⫺ln 3 so that ␩ 6 (0)⫽ ␩ 4 (0)⫽1/2. The
strain profiles ␸ 4 关 ␩ (x) 兴 and ␸ 6 关 ␩ (x) 兴 are very close and
exhibit smaller interface thicknesses than the ␩ (x) profiles.
A and M are in thermodynamic equilibrium, thus s 2 ⫽ P
⫽0, and consequently, ⌬G ␪ ⫽ ␴:␧t , B⫽1/4, and k 26 ⫽1/2.
The interface energy densities are

134201-16
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

3 冑2 ␤ s 1 冑 2 ␤ where G̃ 6 ( ␩ 63)⫽2s 1 /27 and G̃ 4 ( ␩ 43)⫽s 1 /16 are the energy

6 ⫽
E AM 冑2E AM
4 ⫽ ⫽ 冑A⫺ ␴:␧t a barriers between A and M at thermodynamic equilibrium.
4 4 4
Equation 共69兲 expresses the A-M interface energy in terms of
冑2 ␤ the material parameters and temperature 共or stress, which is
⫽ 冑A 0 共 ␪ ⫺ ␪ c 兲 ⫺⌬G ␪ a related to temperature via the equilibrium condition s 2
4
⫽0), the stress hysteresis, or the energy barrier between A
冑2 ␤ and M at thermodynamic equilibrium.
⫽ 冑共 A 0 ⫺za 兲 ␪ ⫹za ␪ e ⫺A 0 ␪ c The interface thickness is defined by
4

⫽
1
4
冑 a 共 6⫺a 兲 ␤ ␧ t H
3
; ⌬ AMª 冉 d ␸ 关 ␩ 共 x 兲兴
dx 冊 ⫺1
, 共70兲
max

E AM 冑
6 ⫽3 3 ␤ G̃ 6 共 ␩ 63 兲 /4,
冑
4 ⫽4 ␤ G̃ 4 共 ␩ 43 兲 /3, 共69兲
E AM which results in

⌬ AM
4 ⫽
共 21⫺5a⫹K 4 兲 5
32共 a⫺6 兲 3 关 11a 3 ⫺81共 9⫹K 4 兲 ⫺5a 2 共 24⫹K 4 兲 ⫹a 共 486⫹39K 4 兲兴
冑␤
s1
,

6 ⫽
⌬ AM
共 72⫺15a⫹K 6 兲 4
128冑2 共 a⫺6 兲 3 关 ⫺27a 2 ⫺24共 24⫹K 6 兲 ⫹5a 共 48⫹K 6 兲兴
冑␤ s1
, 共71兲

K 4 ª 冑81⫺30a⫹5a 2 , K 6 ª 冑576⫺240a⫹33a 2 .

Both ⌬ AM
4 and ⌬ 6 are complicated functions of a but they
AM
E AM 冑2 冑2
6
are accurately approximated by the polynomials ⫽s 1 6 ⌬6 ⫽
, E AM AM
p 6␤ ;
⌬ AM
6
4p6 4
共74兲
1.88⭐p 6 ⫽1.88⫹0.179a⫹0.00065a 2 ⫺0.0035a 3 ⭐2.386,
E AM
4 s1 p4
⫽ , 4 ⌬4 ⫽
E AM AM
␤.
⌬ AM
4
3p4 3
2.411⭐ p 4 ⫽⫺0.028共 a⫺3 兲 2 ⫹2.667⭐2.667. 共72兲
The differences in the A-M interface profiles and energy den-
sities between the 2-4-6 and 2-3-4 potentials are not signifi-
The interface thicknesses are given by cant.

6 ⫽
⌬ AM
p 6 AM
⌬ ⫽p 6
p4 4
冑␤ 冑 s1
⫽p 6
␤
A 0 共 ␪ ⫺ ␪ c 兲 ⫺⌬G ␪ a
D. Critical austenitic nuclei: A solitons on M
In this section we consider the case where only M exists
as x→⫾⬁. The integration constant g 0 must be a function

⫽p 6 冑 ␤
共 A 0 ⫺za 兲 ␪ ⫹za ␪ e ⫺A 0 ␪ c
of the parameters k 6 or k 4 in order to satisfy the boundary
conditions d ␰ (⫾⬁)/dy⫽0:

冑 冑
␤ g 4 ⫺g 40⫽ 共 k 4 ⫺ ␰ 4 兲共 ␰ 4 ⫺ ␰ A1 兲共 ␰ 4 ⫺ ␰ A2 兲 ,
⫽ 6p 6 ;
a 共 6⫺a 兲 ␧H
1
␰ A1 ⫽ 共 1⫺2k 4 ⫹ 冑1⫺2k 4 兲 ⭓0,

冑 冑
2
2 ␤ ␤ 共75兲
6 ⫽
⌬ AM p6 , ⌬ AM
4 ⫽p 4 .
3 冑6 G̃ 6 共 ␩ 63兲 G̃ 4 共 ␩ 43兲 1
共73兲 ␰ A2 ⫽ 共 1⫺2k 4 ⫺ 冑1⫺2k 4 兲 ⭐0,
2

If A 0 ⫽za, then the stress hysteresis, interface energy den- g 6 ⫺g 60⫽ 共 k 26 ⫺ ␰ 26 兲 2 共 ␰ 26 ⫺ ␰ A2 兲 , ␰ A ⫽ 冑1⫺2k 26 .
sity, and interface thickness are temperature independent. It
follows from Eqs. 共69兲–共73兲 that The solutions of Eq. 共52兲 are

134201-17
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 12. The dimensionless effective widths W M冑s 1 ␤ of the critical martensitic nuclei vs B for various values of a for 共a兲 the 2-4-6
potential and 共b兲 the 2-3-4 potential. 共c兲 The dimensionless thickness ⌬ M冑s 1 / ␤ of the interface of a critical martensitic nucleus and half of
the dimensionless effective width, 0.5W M冑s 1 / ␤ , of a critical martensitic nucleus, both vs B for a⫽3. Solid 共dashed兲 lines correspond to the
2-4-6 共2-3-4兲 polynomial.

␰A solutions ␰ A6 (y 6 ) and ␰ A4 (y 4 ) describe critical austenitic nu-

␰ A6 共 y 6 兲 ⫽

冑
, clei 共see Fig. 8兲. From Eq. 共76兲 we obtain the physical order
3k 26 ⫺1 parameters as functions of P:
1⫺ tanh2 共 k 6 冑3k 26 ⫺1y 兲
k 26
冑P
␩ A6 共 x 兲 ⫽

冑 冉冑 冊
,
k 4 共 2⫺4k 4 ⫹H 兲
␰ A4 共 y 4 兲 ⫽ , 共76兲 1 s1
⫺1⫹4k 4 ⫹H 2 共 3⫺ P 兲 ⫺3 共 2⫺ P 兲 tanh2 x
2 ␤
H⫽ 冑1⫺2k 4 共77兲

⫻cosh 冉 1
2
y 4 冑1⫺ 冑1⫺2k 4 ⫺4k 4 冑1⫹ 冑1⫺2k 4 ⫺4k 4 冊 .
␩ A4 共 x 兲 ⫽1

6 共 2⫺ P 兲

冉冑 冊
The parameters k 6 and k 4 are restricted to the intervals ⫺ ,
s1
关 1/冑3,1/冑2 兴 and 关 3/8,1/2兴 , respectively. It was proved nu- 4 共 3⫺ P 兲 ⫹ 冑2 P 共 6⫺ P 兲 cosh 共 2⫺ P 兲 x
merically 共as was done for the martensitic nucleus兲 that the 2␤

134201-18
THREE-DIMENSIONAL LANDAU . . . . III. . . . PHYSICAL REVIEW B 68, 134201 共2003兲

e A6 ⫽4 冕␰A
k6
共 k 26 ⫺ ␰ 26 兲 冑␰ 26 ⫺ ␰ A2 d ␰

⫽
1
2 共 4⫺ P 兲 冋冑 2
3
共 3⫺ P 兲共 2⫺ P 兲

⫺
P 共 8⫺3 P 兲
3 共 4⫺ P 兲
ln 冉 冑2 共 3⫺ P 兲 ⫹ 冑3 共 2⫺ P 兲
冑P 冊册
⯝0.248⫺0.227P⫹0.046P 2 ⫹0.003P 3 . 共78兲

As was done for the martensitic nucleus, expressions for the

width and interface thickness can be obtained for the auste-
nitic nucleus.
Figure 10 shows the energies of the critical A nuclei for
the 2-4-6 and 2-3-4 potentials versus ␣ , and Fig. 14 shows
profiles of the critical A nuclei for these two potentials. The
results for the two potentials are very similar except for the
widths in the neighborhood of P⫽0 where the 2-3-4 nucleus
is significanlty narrower than the 2-4-6 nucleus.

E. Kink solutions: M-M interfaces and barrierless nucleation

of austenite
In this section we present two M-M kink solutions of the
static Ginzburg-Landau equations. The first is a solution for
the potentials in hyperspherical coordinates, and the second
is a M⫹ -M⫺ solution of the extended ( ␩ 苸 关 ⫺1,1兴 ) 2-4-6
Landau potential that exhibits barrierless nucleation of aus-
tenite.
The minimum-energy paths between stable martensitic
variants for the potentials in hyperspherical coordinates are
great circles on the unit hypersphere, r⫽1. Kink solutions
between martensitic variants exist when both variants have
FIG. 13. Austenite-martensite diffuse interface profiles. Solid
the same energy, which implies zero stress or zero transfor-
共dashed兲 lines correspond to the 2-4-6 共2-3-4兲 polynomial. mation work. The potential along the great circle from Mi to
any other variant is G⫽Ā ␺ 2 (1⫺ ␺ ) 2 ⫹⌬G ␪ for both poten-
where 0⭐ P⭐2. The total energies per unit area of the criti- tials. Both variants are twin-related IPS variants, ␧t ( ␺ )
cal nuclei are ⫽ 12 ␥ t 关 ⫺1⫹2 ␺ 2 (3⫺2 ␺ ) 兴 (mn⫹nm)⫹␧nn 关see Eqs. 共20兲
and 共27兲兴; that is, the normal strain is constant, the shear
32 共 6⫺ P 兲 3 A strain varies from ⫺ ␥ t to ␥ t , and the PT does not proceed
E A4 ⫽ 冑␤ s 1 e4 , through A. The solution of the static Ginzburg-Landau equa-
27 共 4⫺ P 兲 5/2 tion ⳵ G/ ⳵ ␺ ⫽2 ␤ ⳵ 2 ␺ / ⳵ x 2 is

e A4 ⫽4 冕
␰ A1
k4
共 ␰ 4 ⫺k 4 兲 冑共 ␰ 4 ⫺ ␰ A1 兲共 ␰ 4 ⫺ ␰ A2 兲 d ␰ 4 ␺ ⫽ 共 1⫹exp关 ⫺ 冑Ā/ ␤ 共 x⫺x 0 兲兴 兲 ⫺1 , 共79兲

冋
which interpolates between Mi at ⫺⬁ and M j⫽i at ⫹⬁. The
1 3 solution 共79兲 coincides with Eq. 共68兲 for ␩ AM
⫽ 共 12⫺6 P⫹ P 2 兲 冑2 共 4⫺ P 兲共 2⫺ P 兲 4 (x) when s 1 is
4 共 6⫺ P 兲 3 2 substituted for Ā. For a⫽3, P( ␺ )⫽ ␸ 4 ( ␺ ); thus the thick-

冉 冊册
ness of the interface is given by Eqs. 共72兲 and 共73兲 for ⌬ AM
4 共 3⫺ P 兲 ⫹3冑2 共 4⫺ P 兲共 2⫺ P 兲
4

⫺ P 共 3⫺ P 兲共 6⫺ P 兲 ln with s 1 →Ā: ⌬ ␺MM⫽2.667冑␤ /Ā. The total kink energy per

冑2 P 共 6⫺ P 兲 unit area is E ␺MM⫽ 冑␤ Ā/3.
⯝0.083⫺0.086P⫹0.024P 2 ⫺0.001P 3 ; The static Ginzburg-Landau equation for the 2-4-6 poten-
tial admits a kink solution such that ␩ (⫹⬁)⫽⫹1, ␩ (⫺⬁)
⫽⫺1; i.e., the structure goes to M⫹ as x→⫹⬁ and to M⫺
3 共 4⫺ P 兲 2 A
E A6 ⫽ 冑6 ␤ s 1 e6 , as x→⫺⬁. There is no corresponding solution for the 2-3-4
8 共 3⫺ P 兲 3/2 potential, so we can drop the subscript ‘‘6.’’ Equation 共75兲

134201-19
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

FIG. 14. Profiles of critical austenitic nuclei for various P and a⫽2.98. Solid 共dashed兲 lines correspond to the 2-4-6 共2-3-4兲 polynomial.

remains valid but with imaginary ␰ A . It is convenient to

designate an imaginary root of the equation g⫽g 0 as ␰ A i;
then,
E MM⫽
k
1
2
冑␤ s 1 MM
2B
e ,

共82兲
g⫺g 0 ⫽ 共 k 2 ⫺ ␰ 2 兲 2 共 ␰ 2 ⫹ ␰ A2 兲 , ␰ A ⫽ 冑2k 2 ⫺1. 共80兲 e MM
⫽4 冕 k

0
共 k ⫺ ␰ 兲 冑␰
2 2 2
⫹ ␰ A2 d ␰
The solution of Eq. 共52兲 is

␰共 y 兲⫽
k sinh共 k 冑3k 2 ⫺1y 兲 ⫽
1
2 冋
k 冑3k 2 ⫺1⫹ 共 1⫺8k 2 ⫹12k 4 兲 ln
k⫹ 冑3k 2 ⫺1
冑2k 2 ⫺1 冉 冊册 .

冑
,
3k 2 ⫺1 Figure 10 shows E MM/ 冑␤ s 1 versus ␣ . Note that
⫹sinh2 共 k 冑3k 2 ⫺1y 兲
2k 2 ⫺1 E MM/ 冑␤ s 1 →⬁ when B→0 ( ␣ →⫺1); however,
s 1 /B→⫺3s 2 /4 and E MM/ 冑␤ →0.901冑⫺s 2 ⫽3.12冑⫺⌬G ␪

冉冑 冊
共81兲
in the same limit. When P→0, E MM/ 冑␤ → 冑s 1 /2⫽ 冑A/2.
s 1 共 2⫺ P 兲 x
sinh The energy of the M⫹ -M⫺ interface and the energies of the
␤ 2 critical A and M nuclei all coincide when A and M are in
␩共 x 兲⫽

冑 冉 冊 冉冑
2 s 1 共 2⫺ P 兲 x
3 1⫺ ⫹sinh2
P ␤
This solution is valid for P⭐0, 0⭐B⭐1/4. Since ␴:␧t
⫽0, it follows that P⫽12⌬G ␪ /A; hence ⌬G ␪ ⭐0 since P
⭐0, which implies stability of M and metastability of A, or
equilibrium. The profiles ␩ (x) and ␸ 关 ␩ (x) 兴 for various P
are shown in Fig. 15. The total energy per unit area of the
M⫹ -M⫺ interface is
冊
. thermodynamic equilibrium.
It is convenient to introduce the dimensionless tempera-
ture Tª( ␪ e ⫺ ␪ )/( ␪ e ⫺ ␪ c ), 0⭐T⭐1. Then s 1 ⫽A⫽A 0 ( ␪ e
2
⫺ ␪ c )(1⫺T) and P⫽12zT/ 关 A 0 (T⫺1) 兴 . Plots of
E MM/ 冑␤ A 0 ( ␪ e ⫺ ␪ c ) vs T for various values of 0⭐A 0 /12z
⭐1/2 are shown in Fig. 16.
As was mentioned in Ref. 11 and 12, soliton splitting
occurs 共Fig. 15兲 as the A-M equilibrium line is approached
(B→1/4,P→0); that is, the M⫹ -M⫺ diffuse interface splits
into M⫹ -A and A-M⫺ diffuse interfaces separated by an A

134201-20
THREE-DIMENSIONAL LANDAU . . . . III. . . .
FIG. 15. M⫺ -M⫹ interface profiles for various P.
region. As P→0 ⫺ the solution 共81兲 for ␩ (x) assumes the
form sgn(x)/ 兵 1⫹6/关 兩 P 兩 sinh2(冑s 1 /2␤ x) 兴 其 1/2, whose magni-
tude is less than or equal to ␦ for 兩 x 兩 ⭐x ␦ , where x ␦
⫽ 冑␤ /2s 1 ln(24␦ 2 / 兩 P 兩 ) for 冑兩 P 兩 Ⰶ ␦ Ⰶ1; the width of the
austenitic region grows logarithmically as the equilibrium
line is neared.
The thickness of the M⫹ -M⫺ interface can be estimated
for P⭐⫺1 by the expression
FIG. 16. M⫺ -M⫹ dimensionless interface energy
E MM/ 冑␤ A 0 ( ␪ e ⫺ ␪ c ) vs dimensionless temperature T for various
values of A 0 /12z.
134201-21
PHYSICAL REVIEW B 68, 134201 共2003兲
FIG. 17. 共a兲 The M⫺ -M⫹ dimensionless interface width
⌬ MM冑s 1 / ␤ vs B for various a. 共b兲 the M⫺ -M⫹ dimensionless inter-
face width ⌬ MM冑A 0 ( ␪ e ⫺ ␪ c )/ ␤ vs dimensionless temperature T for
a⫽2.98 and various values of A 0 /12z.
⌬ MMª2x 0.95
冉 冑 冊
4 3 共 2⫺ P 兲
⫽ 冑␤ /s 1 sinh⫺1 ␩ 0.95 ⫺ ,
冑2⫺ P 共 1⫺ ␩ 0.95
2
兲P
␩ 0.95⫽0.931⫺0.011a⫹0.003a 2 ⫺0.000 08a 3 . 共83兲
Here x 0.95 is defined by the condition ␸ 关 ␩ (x 0.95) 兴 ⫽0.95.
Equation 共83兲 1 was obtained from the condition ␩ (x 0.95)
⫽ ␩ 0.95 using Eq. 共81兲 2 . The function ␩ 0.95(a) was deter-
mined by numerical solution of the equation ␸ ( ␩ 0.95)
⫽0.95 and then approximated by the cubic polynomial, Eq.
共83兲 2 . Plots of ⌬ MM冑s 1 / ␤ vs B for various a and
⌬ MM冑A 0 ( ␪ e ⫺ ␪ c )/ ␤ vs T for a⫽2.98 共NiAl value兲 and vari-
ous values of 0⭐A 0 /12z⭐1/2 are shown in Fig. 17. For P
⭓⫺1 the splitting is so pronounced that we no longer have
an interface in the usual sense of the word. Equation 共83兲
then gives the width of the austenitic nucleus.
Near the A-M equilibrium line, 0⬍k 2 ⫺1/2Ⰶ1, and Eq.
共81兲 for the order parameter becomes
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE
␰ 共 y 兲 ⫽ 冑2 sinh共 y/2兲关 exp共 y 0 兲 ⫹4 sinh2 共 y/2兲兴 ⫺1/2,
y 0 ⫽⫺ln共 k 2 ⫺1/2兲 .
Now expand the hyperbolic sine around y 0 Ⰷ1, sinh(y/2)
⬇exp(y0/2)exp关(y⫺y0)/2兴 /2, and substitute into Eq. 共84兲:
␰ 共 y 兲 ⫽ 关 2 共 1⫹e ⫺(y⫺y 0 ) 兲兴 ⫺1/2.
This is identical to ␰ AM
6 关see Eq. 共68兲兴, the A-M diffuse inter-
face, as expected.
Austenite nucleation inside a homogeneous, thermody-
namically stable martensitic phase is suppressed by the large
activation energy 关see Eq. 共78兲兴 required to form an unstable
critical nucleus. In contrast, the formation of an A region
between the M⫹ and M⫺ occurs with no cost in energy. Soli-
ton splitting is a barrierless mechanism for A nucleation in
the region of M stability. A stable austenitic nucleus is
formed and grows as the temperature approaches the equilib-
rium temperature, and expands to infinity at the equilibrium
temperature. Note that for PⰆ0, when the A embryo is not
visible in the ␩ (x) profile in Fig. 15, it can be seen in the
␸ (x) profile; this is because d ␩ (0)/dx⫽0 but
d ␸ 关 ␩ (0) 兴 /dx⫽ 关 d ␸ (0)/d ␩ 兴关 d ␩ (0)/dx 兴 ⫽0. We believe
that such a mechanism may be observable in experiments on
three-dimensional systems when the transformation strain is
an IPS 共to avoid distortion along the interface兲. This may be
the case for the interface between two IPS variants for any
PT.
VI. NIAL CUBIC-TETRAGONAL PT: M-M AND M-A
INTERFACES
Let us estimate the parameter ␤ for NiAl alloys. The
transformation strain for a variant-variant transformation in
NiAl is an IPS and the transformation does not pass through
A, so our results for M-M interfaces in polar coordinates are
applicable to NiAl. Taking Ā⫽5320 MPa 关Eq. 共39兲兴 one ob-
tains
E ␺MM⫽2.43⫻104 冑␤ N 1/2m⫺1 ,
⌬ ␺MM⫽6.22⫻10⫺5 冑␤ N ⫺1/2
m.
High-resolution electron microscopy13 of Ni65Al35 共the data
that we used in Ref. 2 and in Sec. IV are for Ni61Al39)
brackets the width of the martensite-martensite interface be-
tween one and several interatomic distances 共see Figs. 5 and
6 in Ref. 13兲. If we assume ⌬ ␺MM⫽0.3⫻10⫺9 m, which cor-
responds to an interatomic distance, then we obtain ␤
⫽2.33⫻10⫺11 N and E ␺MM⫽0.117 J m⫺2 from Eq. 共86兲.
⫺9
This value for ␤ gives ⌬ AM
6 ⫽0.243⫻10 m from Eq. 共73兲
and E 6 ⫽0.079 J/m from Eq. 共74兲 for the stress-free case
AM 2 b共 ␩ 兲 ⫽b关 ␸ 共 ¯␩ 兲 ⫹Int共 ␩ 兲兴 , ␧p 共 ␩ 兲 ⫽bn关 ␸ 共 ¯␩ 兲 ⫹Int共 ␩ 兲兴 /d.
at ␪ ⫽300 K. If we take ⌬ ␺MM⫽10⫺9 m, then ␤ ⫽2.588
⫻10⫺10 N, E ␺MM⫽0.391 J m⫺2 , ⌬ AM 6 ⫽0.809⫻10
⫺9
m, and
E 6 ⫽0.263 J/m from Eq. 共74兲. If the surface energy is
AM 2
known, then a more precise estimate of ␤ can be made using
Eq. 共86兲 or Eqs. 共73兲 and 共74兲.
PHYSICAL REVIEW B 68, 134201 共2003兲
Note that the relatively high value of E AM 6 in comparison
to 0.01⫺0.02 Jm⫺2 for steels is related to the very high bar-
共84兲 riers for stress-induced PT, i.e., Ā and A 0 ( ␪ e ⫺ ␪ c ). Our Eq.
4 in terms of G̃ 4 ( ␩ 43), the energy barrier between
共69兲 for E AM
A and M at thermodynamic equilibrium, exactly coincides
with the corresponding equation in Ref. 7 because Eq. 共69兲 is
共85兲 independent of stress, and at zero stress and for one M vari-
ant both Gibbs potentials coincide. Our value for ⌬ AM 4 at a
⫽2.98 is 0.667 of the corresponding value in Ref. 7.
VII. PHASE FIELD THEORY OF DISLOCATIONS
In this section we discuss a serious shortcoming of the
present-day phase field theory of dislocations and then dem-
onstrate that this drawback can be eliminated by following
an approach similar to the one we used to construct our Lan-
dau potentials.
In the phase field theory of dislocations 共see, e.g., Ref. 3兲
the Burgers vector and consequently the plastic strain depend
on the applied stress even in the elastic regime—that is,
when dislocations are in stable equilibrium and no plastic
flow occurs. Such a dependence is in conflict with the defi-
nition of plastic strain in macroscopic plasticity theory. A
␴-dependent Burgers vector is inconsistent with the well-
established theory of dislocations.14 It also implies dissipa-
tion during elastic deformation of plastically deformed
material.10 The ␴ dependence of the plastic strain in the
phase field theory is analogous to the ␴ dependence of the
transformation strain in Landau theories of PT’s; see analysis
in Ref. 1.
Consider one slip plane and one slip direction—i.e., one
Burgers vector b. Then Eqs. 共3兲 and 共4兲 in Ref. 3 for the
Burgers vector and plastic strain reduce to b( ␩ )⫽b␩ and
␧p ( ␩ )⫽bn␩ /d⫽ ␧p ␩ , where ␩ is the density function
共phase field兲 for dislocations, n is the normal to the slip
plane, and d is the distance between the slip planes. The local
potential of the crystal lattice 关Eq. 共8兲 in Ref. 3兴 is f
⫽A sin2(␲␩), which leads to the Gibbs potential G⫽
⫺ ␴:␭:␴/2⫺ ␴:␧p ( ␩ )⫹A sin2(␲␩). The corresponding equa-
tion of thermodynamic equilibrium is ⳵ G/ ⳵ ␩ ⫽0⫽⫺ ␴:␧p
⫹ ␲ A sin(2␲␩); hence 2 ␲ ␩ ⫽arcsin关␴:␧p /( ␲ A) 兴 ⫹n,
共86兲 where n⫽0,1,2, . . . is the number of dislocations. There-
fore, in thermodynamic equilibrium, the order parameter ␩ ,
the Burgers vector b( ␩ ), and plastic strain ␧p ( ␩ ) depend on
the stress ␴. We can avoid this unphysical dependence on ␴
by first breaking the order parameter into an integer part
Int( ␩ ) and a fractional part ¯␩ ª ␩ ⫺Int( ␩ )苸 关 0,1兴 , and then
incorporating the dependence on ¯␩ through the 2-3-4 or
2-4-6 polynomials ␸ ( ␩ ), as was done for the transformation
strain in our Landau potentials:
共87兲
The term Int( ␩ )⫽n accounts for the presence of n disloca-
tions in the slip plane, each with Burgers vector b. The Gibbs
potential for three shear stresses is shown in Fig. 18. The
thermodynamic equilibrium condition ⳵ G/ ⳵ ␩ ⫽0 has the
134201-22
THREE-DIMENSIONAL LANDAU . . . . III. . . .
FIG. 18. Gibbs potential G̃ for dislocations for three shear
stresses.
roots ␩ ⫽n(n⫽1,2,3, . . . ) corresponding to thermodynamic
equilibrium of n dislocations. It also has the roots ¯␩ 3 of the
equation
d ␸ 共 ¯␩ 3 兲
␴:␧p ⫽ ␲ A sin共 2 ␲ ¯␩ 3 兲
d␩
corresponding to the maxima of G( ␴, ␪ , ␩ ), which represent
the activation barriers to the dislocation motion. An analyti-
cal solution for ¯␩ 3 cannot be found, but if f ⫽A sin2(␲␩) is
replaced by periodic 2-3-4 or 2-4-6 crystal potentials in ␩ ,
for example,
f 4 共 ¯␩ 兲 ⫽A ¯␩ 2 共 1⫺ ¯␩ 兲 2 , f 6 共 ¯␩ 兲 ⫽A ¯␩ 2 共 1⫺ ¯␩ 2 兲 2 /2, 共89兲
then analytic solutions can be obtained. For the crystal po-
tentials 共89兲 our Gibbs potential in ¯␩ is identical in form to
the G( ␩ ) for twinning, so we can use all of the results pre-
viously obtained for twinning. In particular, for a⫽3,
¯␩ 43⫽ 共 A⫺3 ␴:␧p 兲 /2A, ¯␩ 63⫽ 冑A⫺3 ␴:␧p / 冑3 共 A⫹ ␴:␧p 兲 .
We write ␴:␧p ⫽ ␶␥ p with ␶ ⫽b"␴"n/ 兩 b兩 the resolved shear
stress and ␥ p ⫽ 兩 b兩 /d the plastic shear strain. The equilibrium
␶ ⫺ ¯␩ curve can be obtained from Eq. 共90兲:
A A 1⫺3 ¯␩ 2
␶ 4⫽ ␩ 兲,
共 1⫺2 ¯ ␶ 6⫽ . 共91兲
3␥ 3 ␥ 1⫹ ¯␩ 2
The condition for loss of stability of a dislocation and its
barrierless motion—namely, ␶ ⭓A/3␥ 共see Fig. 18兲—is the
same for both polynomials. Reverse motion occurs for ␶ ⭐
PHYSICAL REVIEW B 68, 134201 共2003兲
⫺A/3␥ . The generalization to multiple slip systems can be
effected by means of the approach followed in Ref. 3.
VIII. CONCLUDING REMARKS
The most significant difference between the 2-3-4-5 and
2-4-6 potentials is that the ratio of the curvature of the M
minimum to the curvature of the A minimum is 4 times
greater for the 2-4-6 potential: 关 ⳵ 2 G 4 (1)/ ⳵ ␩ 2 兴 / 关 ⳵ 2 G 4 (0)/
⳵ ␩ 2 兴 ⫽1⫺ P/2 and 关 ⳵ 2 G 6 (1)/ ⳵ ␩ 2 兴 / 关 ⳵ 2 G 6 (0)/ ⳵ ␩ 2 兴
⫽4(1⫺ P/2). The critical M nucleus is particularly sensitive
to the larger curvature at the M minimum of the 2-4-6 poten-
tial. For example, its amplitude is larger for the 2-4-6 poten-
tial; especially near the M instability line, its width is larger
for the 2-3-4 potential when B⬎0.17, and its energy is
greater for the 2-4-6 potential. On the other hand, the energy
and amplitude of the critical A nucleus are nearly the same
for both potentials, and the profiles, widths, and energies of
the A-M interfaces are very similar. Despite the fact that all
phase equilibrium and transformation conditions are the
same for both potentials for a homogeneous distribution of
order parameter, the difference in curvature does affect the
energetics and relative stability of phases for a nonuniform
distribution.
Note that our potentials do not have parameters that con-
trol the relative curvatures of G( ␩ ) at the A and M minima.
共88兲
We remind the reader that these curvatures are not related to
elastic moduli, as is the case for Landau potentials with
strain-based order parameters. We plan to generalize our po-
tentials by introducing additional parameters that control the
curvatures at the A and M minima and study the effect of
variations in these parameters on critical nuclei and inter-
faces. The actual values of these parameters can be deter-
mined from the results of atomistic calculations.
In this paper we considered only homogeneous
nucleation—that is, nucleation in defect-free crystals or parts
of crystals: nanocrystals, thin films, or nanoprecipitates. Our
results can be used to approximately model the nonclassical
nucleation of an ellipsoidal region, which was carried out in
Ref. 7, but more precisely and with allowance for applied
stresses. Heterogeneous nucleation at various dislocation
configurations 共pileups, low- and high-angle boundaries兲, as
well as generation of dislocations during PT’s can be studied
using a combination of the phase field theories of PT’s and
dislocations developed in this paper.
共90兲 The free surface boundary condition d ␩ /dx⫽0 is satis-
fied at the maxima of the A and M critical nuclei. Conse-
quently, a critical nucleus at a free surface is simply half of a
critical nucleus in the bulk. Since the energy of a surface
nucleus is half of the energy of the corresponding nucleus in
bulk, the probability of its appearance is significantly higher.
We plan to obtain and compare analytical and numerical
solutions of the time-dependent Ginzburg-Landau equations
for the 2-3-4-5 and 2-4-6 potentials. We also plan to develop
a generalization of our Landau theory for large strains and
large material rotation, a very challenging problem. Such a
generalization is essential for PT’s with components of trans-
formation strain exceeding 0.1 and for loadings that are ac-
companied by finite rotations of the crystal lattice.
134201-23
VALERY I. LEVITAS, DEAN L. PRESTON, AND DONG-WOOK LEE PHYSICAL REVIEW B 68, 134201 共2003兲

ACKNOWLEDGMENTS edged. V.I.L. also acknowledges NSF support 共Grant No.

CMS-02011108兲 and D.W.L. also acknowledges support of
Support of Los Alamos National Laboratory for V.I.L. un- the ME Department of Texas Tech University. We would like
der consulting agreement No. C-8060 and for V.I.L. and to thank Dr. D. Schryvers for providing and discussing his
D.W.L. under Contract No. 52844 is gratefully acknowl- data on the thickness of the variant-variant interface in NiAl.

*FAX: 共253兲 679 8926. Electronic address: valery.levitas 8

F. Falk and P. Konopka, J. Phys.: Condens. Matter 2, 61 共1990兲.
@coe.ttu.edu 9
K. Bhattacharyya and R.V. Kohn, Acta Mater. 4, 529 共1996兲.
1
V.I. Levitas and D.L. Preston, Phys. Rev. B 66, 134206 共2002兲. 10
V.I. Levitas, Large Deformation of Materials with Complex Rheo-
2
V.I. Levitas and D.L. Preston, Phys. Rev. B 66, 134207 共2002兲. logical Properties at Normal and High Pressure 共Nova Science,
3
Y.U. Wang, Y.M. Jin, A.M. Cuitiño, and A.G. Khachaturyan, Acta New York, 1996兲.
Mater. 49, 1847 共2001兲. 11
A.E. Jacobs, Phys. Rev. B 31, 5984 共1985兲.
4
A. Artemev, Y. Jin, and A.G. Khachaturyan, Acta Mater. 49, 1165 12
J. Lajzerkowicz, Ferroelectrics 35, 219 共1981兲.
共2001兲. 13
Ph. Boullay, D. Schryvers, and J.M. Ball, Acta Mater. 51, 1421
5
F. Falk, Z. Phys. B: Condens. Matter 51, 177 共1983兲. 共2003兲.
6
A.E. Jacobs, Phys. Rev. B 46, 8080 共1992兲. 14
J. P. Hirth and J. Lothe, Theory of Dislocations 共Krieger, Malabar,
7
D.M. Haezebrouck, Ph.D. thesis, Northwestern University, 1987. FL, 1992兲.

134201-24