(Cambridge University Press) Evolution of Phase Transitions - A Continuum Theory 1st Edition by Rohan Abeyaratne (2006)
(Cambridge University Press) Evolution of Phase Transitions - A Continuum Theory 1st Edition by Rohan Abeyaratne (2006)
(Cambridge University Press) Evolution of Phase Transitions - A Continuum Theory 1st Edition by Rohan Abeyaratne (2006)
This work began with the authors’ exploration of the applicability of the
finite deformation theory of elasticity when various standard assumptions
such as convexity of the energy or ellipticity of the field equations of equi-
librium are relinquished. The finite deformation theory of elasticity turns
out to be a natural vehicle for the study of phase transitions in solids where
thermal effects can be neglected. This is a valuable work for those inter-
ested in the development and application of continuum-mechanical models
that describe the macroscopic response of materials capable of undergoing
stress- or temperature-induced transitions between two solid phases. The
focus is on the evolution of phase transitions, which may be either dynamic
or quasi-static, controlled by a kinetic relation that in the framework of clas-
sical thermomechanics represents information that is supplementary to the
usual balance principles and constitutive laws of conventional theory. The
book should be of interest to mechanicians, material scientists, geophysi-
cists, and applied mathematicians.
ROHAN ABEYARATNE
Massachusetts Institute of Technology
JAMES K. KNOWLES
California Institute of Technology
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
To the C7: Gina, Kenny, Kevin, Kristen, Liam, Linus, & Nina;
Part I Introduction
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 What this monograph is about 3
1.2 Some experiments 7
1.3 Continuum mechanics 9
1.4 Quasilinear systems 10
1.5 Outline of monograph 11
Part II Purely Mechanical Theory
2 Two-Well Potentials, Governing Equations
and Energetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Introduction 19
2.2 Two-phase nonlinearly elastic materials 20
2.3 Field equations and jump conditions 25
2.4 Energetics of motion, driving force and dissipation
inequality 27
3 Equilibrium Phase Mixtures and Quasistatic
Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1 Introduction 32
3.2 Equilibrium states 33
3.3 Variational theory of equilibrium mixtures
of phases 37
3.4 Quasistatic processes 42
3.5 Nucleation and kinetics 44
3.6 Constant elongation rate processes 47
3.7 Hysteresis 53
vii
viii CONTENTS
This monograph threads together a series of research studies carried out by the au-
thors over a period of some fifteen years or so. It is concerned with the development
and application of continuum-mechanical models that describe the macroscopic
response of materials capable of undergoing stress- or temperature-induced transi-
tions between two solid phases.
Roughly speaking, there are two types of physical settings that provide the mo-
tivation for this kind of modeling. One is that associated with slow mechanical or
thermal loading of alloys such as nickel–titanium or copper–aluminum–nickel that
exhibit the shape-memory effect. The second arises from high-speed impact exper-
iments in which metallic or ceramic targets are struck by moving projectiles; the
objective of such studies – often of interest in geophysics – is usually to determine
the response of the impacted material to very high pressures. Phase transitions are
an essential feature of the shape-memory effect, and they frequently occur in high-
speed impact experiments on solids. Those aspects of the theory presented here
that are purely phenomenological may well have broader relevance, in the sense
that they may be applicable to materials that transform between two “states,” for
example, the ordered and disordered states of a polymer.
Our development focuses on the evolution of the phase transitions modeled
here, which may be either dynamic or quasistatic. Such evolution is controlled by a
“kinetic relation,” which, in the framework of classical thermomechanics, represents
information supplementary to the usual balance principles and constitutive laws of
conventional theory. We elucidate the rather remarkable way in which the classical
theory “calls for” this kind of supplementary information when the material is
capable of changing phase, though such additional information is not called for –
indeed, cannot be imposed – in the case of a single-phase material.
The simplest context in which to illustrate the need for kinetic relations and the
role they play is that furnished by the purely mechanical theory of one-dimensional
nonlinear elasticity, with thermal effects suppressed. After the Introduction, which
comprises Part I of the monograph, we pursue the subject in this context in Part II.
Even this simplest version of the theory to be set out here has some utility, as we
show in Chapters 3 and 4. Part III presents the full three-dimensional theory, taking
xiii
xiv PREFACE
both mechanical and thermal effects into account. We specialize this theory to one
space dimension in Part IV, where we are able to make some comparisons with
experiments. In Part V, we discuss some three-dimensional problems.
The material presented here is drawn primarily from our own research over the
period from the late 1980s forward. We came to this subject as practitioners of solid
mechanics interested in exploring the range of applicability of the finite deformation
theory of elasticity when various standard assumptions such as convexity of various
energies or ellipticity of the field equations of equilibrium were relinquished. When
broadened in this way, finite elasticity is a natural vehicle for the study of those
aspects of phase transitions in solids that can be discussed with thermal effects
neglected. Nonlinear thermoelasticity, similarly unencumbered by conventional
restrictions, provides the natural framework for the study of mechanical and thermal
effects together.
Our hope is that this book will be of interest to materials scientists, engineers
and geophysicists as well as to mechanicians and applied mathematicians. The per-
fectly prepared reader would be acquainted with continuum mechanics at the level
of Chadwick’s Continuum Mechanics, Wiley, New York, 1976; with thermodynam-
ics as treated, for example, in J. L. Ericksen’s Introduction to the Thermodynamics
of Solids, Chapman and Hall, New York, 1991; with material behavior as described
by T. H. Courtney in Mechanical Behavior of Materials, McGraw-Hill, New York,
1990; with partial differential equations at the level of J. D. Logan’s An Introduction
to Nonlinear Partial Differential Equations, Wiley-Interscience, New York, 1994;
and with the elements of Cartesian tensors as discussed, for example, in Linear
Vector Spaces and Cartesian Tensors, Oxford, New York, 1998, by J. K. Knowles.
However, expecting many potential readers to be less than perfectly prepared, we
have tried to make the presentation as self-contained as is practicable, citing appro-
priate sources for those results that are used but not derived.
Although the book deals almost entirely with our own work, we have neverthe-
less had the enormous benefit of interactions with many others, and it is a pleasure
to acknowledge them all with gratitude. We would be remiss not to mention the
particular influence that Tom Ahrens, Kaushik Bhattacharya, Mort Gurtin, Rick
James, Stelios Kyriakides, Jim Rice, the late Eli Sternberg, Lev Truskinovsky, and
our former doctoral students, especially Phoebus Rosakis and Stewart Silling, have
had on our learning of this subject.
Some of the fruitful interactions alluded to above took place in small, informal
summer gatherings held at MIT’s Talbot House in South Pomfret, Vermont. We are
indebted to MIT for the use of this wonderful place, which – alas – is no longer
owned by MIT.
Special thanks go to Debbie Blanchard, who drew the figures in the early part
of the book, and then taught us how to draw the rest.
We are grateful to Olaf Weckner for a careful and constructive critical reading
of the early chapters.
We acknowledge with thanks the past financial support of the U.S. National
Science Foundation, the U.S. Army Research Office, and especially the U.S. Office
PREFACE xv
1
The subscripts s and f denote “start” and “finish” respectively, whereas M and A denote martensite,
the low-temperature phase, and austenite, the high-temperature phase. So, for example, Ms is the
temperature at which the formation of martensite starts, etc.
3
4 EVOLUTION OF PHASE TRANSITIONS
1
e
has
2
re p
6
atu
per
hase
-tem
re p
h
Hig
u
erat
3
p
4
-tem
Low
Mf As Ms Af
θ
Figure 1.2. A schematic plot of a representative linear dimension of the specimen versus temperature
θ during a cycle of cooling and reheating. The specimen transforms between a high-temperature phase
and a low-temperature phase.
ψ Phase boundary
F+
ψ+ F−
ψ−
(b)
F
F+ F−
(a)
Figure 1.3. (a) A schematic plot of the free energy ψ versus the deformation gradient F at a fixed
temperature, showing local minima at F = F+ and F = F− . (b) Two phases characterized by F+ and F−
coexisting in a body, separated by a phase boundary.
There is now a “thermodynamic driving force,” which tends to transform the high-
energy phase “plus” into the low-energy phase “minus”. Accordingly, this driving
force makes it favorable for the phase boundary to move into the region associated
with F+ , thereby transforming the high-energy phase into the low-energy phase as
it crosses the moving interface. The driving force vanishes in phase equilibrium.
How fast the phase boundary moves – the question of “kinetics” – is determined
by the microstructure underlying the two phases and the details of the mechanism
by which one lattice changes into the other.
In order to relate this discussion to the earlier one, we simply need to relate the
plus and minus phases here to what we previously called the high-temperature and
low-temperature phases. For example, if, as a function of the temperature θ, the
energies ψ(F+ , θ ) and ψ(F− , θ ) are such that ψ + > ψ − for high temperatures and
ψ + < ψ − for low temperatures, then the minus phase is energetically preferred at
high temperatures and can therefore be identified with what we previously called
the high-temperature phase; similarly the plus phase can be identified with the
low-temperature phase.
Next, consider a stress-free body composed of two coexistent phases that are
in mechanical and phase equilibrium. If the body is now subjected to some stress,
this alters the relative favorability of the two phases and will, in principle, lead to
phase boundary motion. Similarly, this will happen if the temperature is changed.
It should be noted that in the phenomenological theory of thermoelasticity, if
one is simply given a potential ψ with multiple local minima, one cannot say
much about the different phases associated with the different minima. They might,
for example, be associated with the liquid and gaseous phases of an elastic fluid,
two crystallographically distinct lattices of a crystalline solid, the disordered and
ordered states of a polymer, and so on. In order to physically identify the phases
one must consider the underlying microstructure.
This monograph explores in detail a number of the concepts referred to above.
We study questions related to the quasistatic and dynamic responses of an elas-
tic material characterized by a free energy potential with multiple local min-
ima. We pay particular attention to the notions of driving force and kinetics. The
driving force plays a central role in characterizing the dissipation rate during
INTRODUCTION 7
time-dependent processes, and we make this notion rigorous under very general
circumstances and then specialize it to various particular cases. We address both
modeling and mathematical issues. For example on the modeling front, we cal-
culate an explicit free energy function ψ for a material with cubic and tetragonal
phases. We develop various models, both phenomenological and micromechan-
ical, of the transformation kinetics. And we solve a number of boundary-initial
value problems connected to experiments. On the mathematical front, we show
that quasistatic and dynamic problems for two-phase materials, when formulated
as in classical elasticity, suffer from a severe lack of uniqueness of solution to
boundary–initial value problems. We show that the degree of nonuniqueness is
precisely filled up by a proper description of the nucleation and kinetics of the
transformation.
This monograph is not meant to provide an encyclopedic treatment of phase
transitions. Rather, it simply collects together, and retells as a single story, the
results from a series of studies mostly by the present authors on the motion of phase
boundaries in solids.
crosses this moving interface, it transforms from the high-energy phase to the low-
energy phase. The flux that is conjugate to this driving force is the rate at which
the material transforms, or equivalently, the (Lagrangian) speed of the interface.
The process of phase transformation, like any nonequilibrium process, is governed
by a kinetic law that relates the flux to the driving force and other appropriate
variables. Therefore in order to fully model the time evolution of a phase transition,
one must specify an elastic potential that characterizes each phase and a kinetic law
that characterizes the evolution of the phases. The detailed form of the kinetic law
depends on the microscale dynamic process by which one lattice transforms to the
other. Some preliminary efforts to derive kinetic models from lattice considerations
can be found in Abeyaratne and Vedantam [7], and Purohit [44].
convex nor strictly concave, and so the quasilinear system suffers from the lack
of uniqueness alluded to above, even with the entropy inequality in place. Rather
than replacing the entropy inequality with a “uniqueness-generating admissibility
criterion,” we take the point of view that the lack of uniqueness reflects a deficiency
in the physical model. The entropy inequality, being a statement of the second
law of thermodynamics, must be retained. Rather, additional information from the
physics of phase transformations needs to enter the mathematical model. Important
insight into the information that is lacking can be obtained if one can determine all
solutions to the quasilinear system and organize them in some illuminating way.
We were able to do this explicitly in a special, algebraically simple case, [3], and
thereby conclude that the nonuniqueness arises for two distinct reasons: it pertains
to (i) the question of whether or not the motion involves a phase boundary and (ii)
if the motion does involve a phase boundary, then the indeterminacy of the phase
boundary speed. This suggests, exactly as the materials scientists would tell us,
that the quasilinear system must be supplemented with a “nucleation condition”
that addresses item (i) and a “kinetic relation” that addresses item (ii). In [3] we
demonstrate that the solution to the Riemann problem for this quasilinear system is
unique when the problem statement is supplemented in this way; see also LeFleeh
[34, 35]. Each of the previously mentioned admissibility criteria can be viewed as
corresponding to a particular (mathematical) kinetic relation, [4, 5], though deriving
the kinetic law from the underlying physics is largely an open question.
REFERENCES
[1] R. Abeyaratne, C. Chu, and R.D. James, Kinetics of materials with wiggly
energies: theory and application to the evolution of twinning microstructures
in a Cu–Al–Ni shape memory alloy. Philosophical Magazine A, 73 (1996),
pp. 457–97.
[2] R. Abeyaratne and J.K. Knowles, On the driving traction acting on a surface
of strain discontinuity in a continuum. Journal of the Mechanics and Physics
of Solids, 38 (1990), pp. 345–60.
[3] R. Abeyaratne and J.K. Knowles, Kinetic relations and the propagation of
phase boundaries in solids. Archive for Rational Mechanics and Analysis,
114 (1991), pp. 119–54.
INTRODUCTION 13
[37] T.J. Lim and D.L. McDowell, Mechanical behavior of a NiTi shape memory
alloy under axial-torsional proportional and nonproportional loading. ASME
Journal of Engineering Materials and Technology, 121 (1999), pp. 9–18.
[38] M. Meyers, Dynamic Behavior of Materials. Wiley, New York, 1994.
[39] N. Nakanishi, Lattice softening and the origin of SME, in Shape-Memory
Effects in Alloys, edited by J. Perkins. Plenum Press, New York, 1975,
pp. 147–76.
[40] O. Oleinik, On the uniqueness of the generalized solution of the Cauchy
problem for a nonlinear system of equations occurring in mechanics. Uspekhi
Matematicheskii Nauk (N.S.), 12 (1957), pp. 169–76 (in Russian).
[41] O. Oleinik, Uniqueness and stability of the generalized solution of the Cauchy
problem for a quasilinear equation. Uspekhi Matematicheskii Nauk (N.S.), 14
(1959), pp. 165–70 (in Russian).
[42] K. Otsuka and C.M. Wayman, Shape Memory Materials. Cambridge Univer-
sity Press, Cambridge, 1998.
[43] R. Pego, Phase transitions in one-dimensional nonlinear viscoelasticity: ad-
missibility and stability. Archive for Rational Mechanics and Analysis, 97
(1986), pp. 353–94.
[44] P.K. Purohit, Dynamics of Phase Transitions in Strings, Beams and Atomic
Chains. Ph.D. Thesis, California Institute of Technology, 2001.
[45] L.M. Schetky, Shape-memory alloys. Scientific American, 241 (1979),
pp. 74–82.
[46] J.A. Shaw and S. Kyriakides, Thermomechanical aspects of NiTi. Journal of
the Mechanics and Physics of Solids, 43 (1995), pp. 1243–81.
[47] M. Shearer, Nonuniqueness of admissible solutions of Riemann initial value
problems for a system of conservation laws of mixed type. Archive for Rational
Mechanics and Analysis, 93 (1986), pp. 45–59.
[48] M. Slemrod, Admissibility criteria for propagating phase boundaries in a van
der Waals fluid. Archive for Rational Mechanics and Analysis, 81 (1983),
pp. 301–15.
[49] K. Thornton and P.W. Voorhees, The dynamics of interfaces in elastically
stressed solids, in Thermodynamics, Microstructures and Plasticity, edited
by A. Finel et al. Kluwer, Dordrecht, 2003, pp. 91–106.
[50] L. Truskinovsky, Equilibrium phase interfaces. Soviet Physics Doklady, 27
(1982), pp. 551–3.
[51] L. Truskinovsky, Structure of an isothermal phase discontinuity. Soviet
Physics Doklady, 30 (1985), pp. 945–8.
Part II Purely Mechanical Theory
2 Two-Well Potentials, Governing
Equations, and Energetics
2.1 Introduction
In this chapter, we assemble the basic field equations and jump conditions for a one-
dimensional, purely mechanical theory of nonlinear elasticity; although thermal
effects will be omitted, inertia will be taken into account. The theory presented
here is general enough to describe nonlinearly elastic materials that, under suitable
conditions of stress, are capable of existing in either of two phases. As we shall see,
a key feature of this theory is that the potential energy of the material, as a function
of strain at a fixed stress, has two local minima. The associated constitutive relation
between stress and strain will then necessarily be nonmonotonic, possessing a
maximum and a minimum connected by an unstable regime in which stress declines
with increasing strain.
Experiments that provide the motivation for the theory about to be developed fall
into two categories. The first of these involves slow tensile loading and unloading
of slender bars or wires composed of materials such as shape-memory alloys. The
model to be constructed to describe experiments of this kind is one of uniaxial
stress in a one-dimensional nonlinearly elastic continuum, and the processes to be
studied for this model are quasistatic. The stress-induced phase transitions in such
experiments occur in tension, so the two minima in the potential energy density
occur at positive – or extensional – values of strain, as do the extrema in the stress–
strain relation.
In the second type of experiment, the specimen is typically a circular cylindrical
disk composed of a metal or a ceramic and subjected to normal impact on one
of its plane faces by a projectile or “flyer-plate” moving at high speed. In the
model appropriate to these experiments, the disk is replaced by an infinite slab of
nonlinearly elastic material whose thickness is that of the disk, and which finds itself
in a state of uniaxial strain in response to the impact; inertial effects are important.
The materials in such experiments undergo phase changes in compression, so the
minima in the potential energy and the extrema in the stress–strain relation now
occur at negative – or compressive – values of strain.
19
20 EVOLUTION OF PHASE TRANSITIONS
If inertial effects are retained in both of the nonlinearly elastic models alluded
to above, the resulting mathematical theories are identical in the sense that the field
equations and jump conditions for both models involve a single scalar strain, a
scalar stress constitutively determined by this strain, and a scalar particle velocity.
The independent variables are the time and a single space coordinate. For the case
of impact in the slab, there are, in addition, transverse components of normal stress
that may be of interest in some circumstances, but these do not enter the governing
equations and, if needed, are to be calculated after the basic boundary–initial value
problem corresponding to a particular experiment has been solved. In general, the
state of stress in the slab will not be hydrostatic.
For definiteness, we shall develop the one-dimensional theory in this chapter in
the context of the thin bar, rather than the slab. We thus deal at first with a general
rising–falling–rising stress–strain curve whose maximum and minimum occur at
tensile strains, specializing in Chapter 3 to the case of the trilinear material for
purposes of detailed analysis of quasistatic motions in the bar. Only minor modifi-
cations will prove to be needed to model the slab, for which phase transitions occur
in compression. We shall introduce these modifications in Chapter 4, where they
will be needed in our analysis of the dynamics of impact-induced phase transitions.
The notion of a two-phase nonlinearly elastic material is introduced in the fol-
lowing section. In Section 2.3, we state the version of global balance of momentum
appropriate to the dynamics of uniaxial stress in a one-dimensional bar, and we
describe the two-well potential energy density and the associated nonmonotonic
nonlinearly elastic stress–strain relation that underlies the model used in this chap-
ter. The balance principle, and the constitutive law, together with the kinematic
assumption that displacement is everywhere continuous, provide the fundamental
field equations and jump conditions of the theory. Throughout the book, we shall
adopt the Lagrangian, or material, description of the processes considered.
In Section 2.4, we shall study the energetics of motions in the bar, accounting
for inertial effects. As in classical gas dynamics, propagating discontinuities in the
fields of strain, particle velocity, and stress affect the relation between work and
energy in a major way. Indeed, the fact that moving strain discontinuities give rise
to an imbalance between the rate of work done on a part of the body containing such
jumps and the rate of increase of the total mechanical energy contained in this part
leads to the notion of the scalar driving force acting on a strain discontinuity. This
notion plays a central role in further developments. In particular, the product of the
driving force and the referential, or Lagrangian, velocity of the moving discontinuity
represents the associated dissipation rate. The requirement that this dissipation rate
be nonnegative furnishes the analog of the second law of thermodynamics in the
purely mechanical setting of the present chapter.
W(γ)
W( P(γ, σ)
P(
(a) (b)
γM γm
γ γ
0 γM γm 0
Figure 2.1. (a) The nonconvex strain energy density W(γ ) for an elastic material that can exist in either
of two phases. (b) The associated two-well potential energy density as a function of strain at a fixed
stress for which the material can exist in either phase. The function W changes curvature at the strains
γ M and γm.
σ̂(γ)
σM
σ γT (σ)
γ
−1 0 γM γm
ΓL (σ) ΓH (σ)
P(γ, σ) P(γ, σ)
γ γ
0 0
L L H
H
P(γ, σ)
(c): σm ≤ σ < σ0
γ
0 H
L
Figure 2.3. Potential energy for a two-phase material as a function of strain for a fixed value of stress
(a) above, (b) at, and (c) below the Maxwell stress σ0 . In each case, the coordinates of the point L
corresponding to the low-strain energy well are ( L (σ ), gL (σ )), while those of H are ( H (σ ), gH (σ )); see
(2.5) and (2.7).
be the difference between these two Gibbs energies at the same stress σ . One can
show that the derivative e (σ ) is given by
e (σ ) = γT (σ ) > 0, (2.9)
so that e(σ ) is an increasing function of σ on [σm , σ M ]. Moreover e(σm ) < 0, while
e(σ M ) > 0. There is therefore a unique value σ0 of stress – the Maxwell stress – at
which the values of the two minima of P, and so the two Gibbs energies, coincide:
g L (σ0 ) = g H (σ0 ). (2.10)
For σm ≤ σ < σ0 , the low-strain minimum of P(·, σ ) lies below the high-strain
minimum, while for σ0 < σ ≤ σ M , the reverse is true. We thus speak of the low-
strain phase as stable and the high-strain phase as metastable below the Maxwell
stress and vice versa above the Maxwell stress. At the Maxwell stress, both phases
are stable. Figure 2.3 shows schematic plots of P(γ , σ ) versus γ for values of σ
above, at, and below the Maxwell stress for a material whose stress–strain curve is
as shown in Figure 2.2.
It is easy to show that (2.10) can be rewritten, after using (2.7), (2.2), and (2.1),
as
(σ0 )
H
σ̂ (γ ) dγ = σ0 ( H (σ0 ) − L (σ0 )) , (2.11)
L (σ0 )
which is an alternative equation for the Maxwell stress σ0 . This shows that the
Maxwell stress cuts lobes of equal area from the rising–falling–rising stress–strain
curve of Figure 2.2.
TWO-WELL POTENTIALS, GOVERNING EQUATIONS, AND ENERGETICS 25
σx = ρvt . (2.13)
vx = γt . (2.14)
Next, suppose that the motion involves a single strain discontinuity, whose
referential location at time t is x = s(t). Suppose that this discontinuity lies between
x1 and x2 . Then the global balance of momentum (2.12) and the assumed continuity
y = x + u(x,t)
s(t) ξ(t)
x y
0 x1 x2 L y0(t) y1(t) y2(t) yL(t)
Reference configuration
con guration Current configuration
con guration
Figure 2.4. Reference and current configurations of a bar.
26 EVOLUTION OF PHASE TRANSITIONS
According to (2.22), a discontinuity can propagate with real speed ṡ if and only
if the chord connecting the two points (γ ± , σ̂ (γ ± )) on the stress–strain curve has
nonnegative slope. In particular, no such discontinuity can correspond to two points
on the unstable branch of the curve. We call the discontinuity a shock wave if the
strains γ ± are either both in the low-strain phase or both in the high-strain phase.
If, on the other hand, γ + is in the low-strain phase while γ − is in the high-strain
phase, or vice versa, the discontinuity is called a phase boundary. It will become
clear in the next section that we need not consider discontinuities in which one of
the strains γ ± lies on the unstable branch of the stress–strain curve.
where f (t) is called the driving force acting on the strain discontinuity and is
given by f (t) = fˆ(γ + (t), γ − (t)). Here γ ± = γ ± (t) = γ (s(t)±, t) are the strains
immediately ahead of and behind the strain discontinuity and the function fˆ is
defined, for γ − > −1, γ + > −1, by
f = fˆ(γ + , γ − ) = [[W (γ )]] − (1/2) σ̂ (γ + ) + σ̂ (γ − ) [[γ ]]
γ+ (2.25)
= γ − σ̂ (γ )dγ − (1/2)(σ̂ (γ + ) + σ̂ (γ − ))(γ + − γ − ).
Since fˆ(γ + , γ − ) is defined for all strains γ ± , the stresses σ ± = σ̂ (γ ± ) need not
lie in [σm , σ M ].
Equations (2.23)–(2.25) apply at a strain discontinuity in any nonlinearly elastic
material. For a two-phase material, there are two kinds of strain discontinuities, viz.
shock waves and phase boundaries, and the notion of driving force and its relation
to the dissipation rate are relevant at both.
The concept of driving force, if not the name, was independently employed in
the setting of phase transitions by Heidug and Lehner [5], Truskinovsky [9], and
Abeyaratne and Knowles [1]. It is related to notions used earlier by Eshelby [4] and
28 EVOLUTION OF PHASE TRANSITIONS
σ̂(γ)
f = A-B
γ
0 γ− γ+
Figure 2.5. Geometric interpretation of the driving force f = fˆ (γ + , γ − ) at a strain discontinuity as the
difference of the signed areas A and B.
Rice [8] in more general settings, and by Knowles [6] for quasistatic processes in
nonlinearly elastic solids. There is also a close connection to the idea of driving
force that plays a prominent role in materials science; see, for example, Section 1.9
of [7].
In our study of mixed-phase equilibria in the next chapter, we shall see that
the static version of driving force is directly related to the energy difference e(σ )
introduced in (2.8).
For any σ̂ (γ ), one may interpret f = fˆ(γ + , γ − ) given by (2.25)3 geometrically
as the (signed) difference between the area under the stress–strain curve between
γ − and γ + and the area of the trapezoid formed by the following four points in
the γ , σ -plane: (γ − , 0), (γ − , σ̂ (γ − )), (γ + , 0) and (γ + , σ̂ (γ + )); see Figure 2.5.
It follows as an immediate consequence of this interpretation that if σ̂ (γ ) is
linear, then the driving force vanishes at every strain discontinuity; all such discon-
tinuities are therefore dissipation free for a linearly elastic material.
If there is in fact no discontinuity at x = s(t), so that [[γ ]] = 0, then it follows
from (2.25) and (2.24) that f = 0 and D = 0; hence the rate of work coincides
with the rate of change of total energy, and the rate of dissipation vanishes. This
is the usual setting of nonlinear elasticity, which is dissipation free only because
moving strain discontinuities are absent. When a genuine discontinuity is present,
the driving force – and therefore D – need not vanish, though they may do so
in the exceptional case for which the states (γ − , σ̂ (γ − )) and (γ + , σ̂ (γ + )) are such
that the areas A and B in Figure 2.5 happen to be equal. Thus moving strain
discontinuities in general produce dissipation in nonlinear elasticity theory, just as
shocks waves do when modeled as in classical inviscid gas dynamics. In the present
purely mechanical setting, the counterpart of the second law of thermodynamics
is the requirement that, at any strain discontinuity, the dissipation rate D(t) must
be nonnegative. Thus the driving force f and the Lagrangian velocity ṡ of the
TWO-WELL POTENTIALS, GOVERNING EQUATIONS, AND ENERGETICS 29
σ̂(γ)
A s1(t) s2(t)
0 L x
γ1− γ2+
γ1+
B γ2−
γ
0 γ1− γ1+ γ2− γ2+
Figure 2.6. The signed areas A and B corresponding to the driving forces acting on discontinuities at
the ends of an interval in which the strain lies in the unstable regime of the stress–strain curve.
σ̂(γ)
µH
1
µL
1
γ
0 γM γm
convex in the high-strain phase do not arise for the trilinear material. Moreover, this
material is simpler than the more general one of Figure 2.2 in that the respective
speeds cL and cH of shock waves in the low- and high-strain phases are, from (2.22),
known independently of the local states of strain at the shock. They are given by
cL = µL /ρ, cH = µH /ρ, (2.27)
where µL and µH are the constant elastic moduli of the two phases, shown in the
figure as the slopes of the two rising branches. Trilinear materials will be examined
in more detail in the next two chapters.
REFERENCES
[1] R. Abeyaratne and J.K. Knowles, On the driving traction acting on a surface of
strain discontinuity in a continuum. Journal of the Mechanics and Physics of
Solids, 38 (1990), pp. 345–60.
[2] P. Chadwick, Continuum Mechanics. Dover, New York, 1999.
[3] J.L. Ericksen, Equilibrium of bars. Journal of Elasticity, 5 (1975), pp. 191–201.
[4] J.D. Eshelby, Continuum theory of lattice defects, in Solid State Physics, edited
by F. Seitz and D. Turnbull, Volume 3. Academic Press, New York, 1956,
pp. 79–144.
[5] W. Heidug and F.K. Lehner, Thermodynamics of coherent phase transforma-
tions in non-hydrostatically stressed solids. Pure and Applied Geophysics, 123
(1985), pp. 91–8.
[6] J.K. Knowles, On the dissipation associated with equilibrium shocks in finite
elasticity. Journal of Elasticity, 9 (1979), pp. 131–58.
[7] D.A. Porter and K.E. Easterling, Phase Transformations in Metals and Alloys.
Van Nostrand-Reinhold (UK) Ltd, London, 1981.
[8] J.R. Rice, Continuum mechanics and thermodynamics of plasticity in relation
to microscale deformation mechanisms, in Constitutive Equations in Plasticity,
edited by A. S. Argon. MIT Press, Cambridge, MA, 1975, pp. 23–79.
[9] L. Truskinovsky, Structure of an isothermal discontinuity. Soviet Physics Dok-
lady, 30 (1985), pp. 945–8.
3 Equilibrium Phase Mixtures and
Quasistatic Processes
3.1 Introduction
In the present chapter we study the equilibrium and quasistatic response of a thin
bar composed of a material modeled by the stress–strain curve shown in Figure 2.2
of the preceding chapter. There is an extensive experimental literature devoted to
tensile loading and unloading of bars made of materials that are capable of under-
going displacive phase transitions; often the materials studied are technologically
important shape-memory alloys such as nickel–titanium. For a small sample of
this literature, the reader might consult the papers of Krishnan and Brown [13],
Nakanishi [17], Shaw and Kyriakides [19], and Lin et al. [14], as well as the ref-
erences cited there. The loading in such experiments is slow, in the sense that
inertia is insignificant. The objective is typically the determination of the relation
between the applied stress and the overall elongation of the bar, though in some
studies, such as that of Shaw and Kyriakides [19], local strain and temperature
measurements are made as well. The stress–elongation relation that is observed
in such experiments exhibits hysteresis, the phase transition being the primary
mechanism responsible for such dissipative behavior. For a given material, the size
and other qualitative features of the hysteresis loops depend on the loading rate
and the temperature at which the test takes place. In the model to be discussed
in this chapter, thermal effects are omitted; they will be accounted for in later
chapters.
Our interest here is twofold: first, the purely mechanical equilibrium theory
provides the simplest setting in which to show that, if unconventional material
models such as that represented by Figure 2.2 are to be studied, the standard prob-
lems of continuum mechanics are in fact indeterminate when posed in terms of
the usual balance principles, boundary conditions, and the bulk constitutive law. If
such indeterminacy is to be removed, additional information must be supplied in
formulating the basic problems. Second, we wish to assess the extent to which the
purely mechanical theory of one-dimensional bars composed of two-phase non-
linearly elastic materials can predict the experimentally observed features of the
quasistatic hysteresis loops.
32
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 33
σ
L
s
two-phase
o-phase mixture
σM
low-strain
low-strain phase high-strain phase
δ = LΓL (σ) σ0 δ = LΓH (σ)
σm
δ = δ̂(σ, s)
s = constant
0 γmL
δ
γM L
Figure 3.1. The “flag” of mixed-phase equilibrium states arising from (3.1) and (3.3), together with the
“flagpoles” of pure low-strain and high-strain equilibria and a generic curve of constant s.
The collection of equilibrium phase mixtures described by (3.1) and (3.3) may
be represented by the set of all pairs (δ, σ ) obtained from (3.3) as σ varies between
σm and σ M and s ranges over the interval [0, L]. This set is shown in the δ, σ -plane
as Figure 3.1. The figure includes the stress-elongation curves δ = L (σ )L and
δ = H (σ )L that describe the respective single-phase responses of the bar in the
low- and high-strain phases. We speak of the set described in the figure as the “flag”
of mixed-phase equilibria, together with the two “flagpoles” corresponding to the
pure phases.
In seeking to understand and resolve the indeterminacy of the displacement
boundary value problem, we shall find that the driving force plays a significant
role.
Since σ is uniform in the equilibrium phase mixture, the general representation
(2.25) for the driving force in terms of the strains γ ± specializes to
γ+
f = σ̂ (γ ) dγ − σ (γ + − γ − ) (3.4)
γ−
in the static case. Geometrically, the static driving force coincides with the difference
of the areas of the two lobes of the stress–strain curve (Figure 2.2) cut off by the
line parallel to the γ -axis through the stress σ . Clearly, one may rewrite (3.4) as
f = [[W − σ γ ]] = [[P(γ , σ )]], (3.5)
where P(γ , σ ) is the potential energy per unit reference volume. Thus for an equi-
librium mixture of phases, the driving force acting on the phase boundary at x = s
coincides with the jump in the potential energy density across this interface. Going
36 EVOLUTION OF PHASE TRANSITIONS
further, one notes from (3.5), (2.7), and (2.8) that the static driving force also co-
incides with the difference e(σ ) of the minimum values of potential energy at the
stress σ :
σ
f = f(σ ) = e(σ ) = γT (σ ) dσ , (3.6)
σ0
where σ0 is the Maxwell stress and γT (σ ) is the transformation strain at the stress
σ ; see (2.6). We see, incidentally, that the Maxwell stress may be interpreted geo-
metrically as the stress at which the two lobes cut from the stress–strain curve as
mentioned above have equal area. Since the stress in a mixed-phase equilibrium
state must lie between σm and σ M , the range of possible values of driving force is
[ f m , f M ], where, by (3.6),
σ0 σM
fm = − γT (σ ) dσ < 0, fM = γT (σ ) dσ > 0. (3.7)
σm σ0
From (3.6), it follows that the driving force acting on the interface between
the phases vanishes if the stress in the bar coincides with the Maxwell stress. In
this case, we say the bar is in phase equilibrium. Note that the bar may sustain a
mixture of phases that is in mechanical equilibrium (σ is uniform) but not in phase
equilibrium (σ = σ0 ).
The total energy E per unit cross-sectional area of the bar due to the strain field
(3.1) is given by
L
E= W (γ (x)) d x = W ( H (σ ))s + W ( L (σ ))(L − s). (3.8)
0
∂ Ē(δ, s) ∂ Ē(δ, s)
= σ, = − f. (3.9)
∂δ ∂s
The total potential energy is
L
= P(γ (x), σ ) d x = E − σ δ. (3.10)
0
From (3.16) one can conclude that this extremum is in fact a minimum. We have
therefore established the following result:
Among all mixed-phase equilibria of the form (3.1) and (3.3) with δ given,
the one for which the stress in the bar coincides with the Maxwell stress
minimizes the total energy for the given δ.
We might say that, among all the mixed-phase equilibria (3.3) with a given δ,
the “Maxwell phase mixture” just described is “most stable” in an energy sense.
We next show that the energy-minimizing property of the Maxwell equilibrium
phase mixture persists when the class of competing deformations is much larger
38 EVOLUTION OF PHASE TRANSITIONS
than that encompassing only the mixed-phase strain distributions (3.1). To do this,
we study the behavior of the total stored energy
L
E{ũ} = W (ũ (x)) d x (3.17)
0
We seek a ũ that minimizes E{ũ} over the class of admissible candidates; such
a ũ would be stable in a minimum-energy sense, not merely with respect to the
mixed-phase competitors of (3.1), but with respect to all candidates in the enlarged
class.
Suppose the minimum of E{ũ} is attained at a function ũ = u, continuous
on [0, L], whose gradient u (x) has a discontinuity at x = s, but, together with
u (x), is continuous elsewhere in [0, L]. Let γ (x) = u (x) and σ (x) = σ̂ (γ (x)),
respectively, be the strain and stress formally associated with this minimizer, and
we write γ ± = γ (s±), σ ± = σ (s±) for the respective strains and stresses on either
side of the jump at x = s.
To derive conditions on this u that are necessary if it is to be a minimizer, we study
E{ũ} for ũ’s that are close to u. Let ũ(·, ) be a one-parameter family of admissible
functions on [0, L] for which ũ(x, 0) = u(x). Assume that, for some 0 > 0, ũ is
continuous in (x, ) on [0, L]×(− 0 , 0 ), with first and second derivatives that are
continuous there except across x = s̄(), where s̄() is a smooth function of for
which s̄(0) = s. Thus in particular, ũ is close to u if || is small.
Because ũ is continuous across x = s̄(), the limiting value of ũ on this curve
is ũ(s̄(), ). Differentiating this with respect to shows that the jumps in the first
derivatives ũ x and ũ at this discontinuity must satisfy
Note that (3.19) is the precise analog of the kinematic jump condition (2.16) with
the parameter here playing the role played by the time t in that setting.
Next, let
s̄( ) L
E() = W (ũ x (x, )) d x + W (ũ x (x, )) d x (3.20)
0 s̄( )
be the one-parameter family of total energies associated with ũ(x, ). Since by
assumption, E{ũ} has a minimum at ũ = u, E() has a minimum at = 0, and
therefore E (0) = 0. By making use of integration by parts, the restriction (3.19),
and the fact that ũ(x, ) satisfies the boundary conditions imposed on u(x) for all
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 39
η(s+) + η(s−)
R = (σ + − σ − ) , (3.23)
2
+ σ+ + σ− +− −
S = W (γ ) − W (γ ) − (γ − γ ) s̄ (0), (3.24)
2
and we have set
η(x) = ũ (x, 0). (3.25)
In the expressions for Q, R, and S, the number s̄ (0) is arbitrary, as is the func-
tion η(x), except that necessarily η(0) = η(L) = 0. Since η may be taken to vanish
outside an arbitrary subinterval of (0, s) and s̄ (0) may be chosen to be zero, the
requirement E (0) = 0 implies that necessarily σ (x) is constant on (0, s). Similarly,
one shows that σ (x) is constant on (s, L). It then follows from (3.22) that Q = 0.
Continuing to choose s̄ (0) = 0, we conclude that R = 0, and it follows (since
η(s±) may be chosen arbitrarily) that σ + = σ − , so that σ (x) is indeed constant
on (0, L); thus Q = R = 0, and so from (3.21) and the vanishing of E (0), one
has S = 0 as well. Finally, the arbitrariness of s̄ (0) allows us to conclude that the
contents inside the braces on the right in (3.24) must vanish, which is to say that
the driving force acting on the phase boundary is zero. This conclusion represents
an instance of the so-called Weierstrass–Erdmann corner condition from the cal-
culus of variations, which is discussed, for example, in the book by Gelfand and
Fomin [8]; for a discussion in the setting of three-dimensional nonlinear elasticity,
see [1].
We have thus proved the following result:
(i) The associated stress must be uniform in the bar, so that mechanical
equilibrium prevails;
(ii) The associated static driving force acting on the phase boundary must
vanish, so that phase equilibrium prevails.
The fact that the static driving force vanishes implies that the stress in the bar
is the Maxwell stress σ0 . Reference to Figure 3.1 shows that a necessary condition
40 EVOLUTION OF PHASE TRANSITIONS
If it is further assumed that, in the minimizer with a single phase boundary, the
high-strain phase is on the left, we immediately recover the value of s in (3.13). If
δ is in the range (3.26), then s as given by (3.13) indeed lies in the interval [0, L].
Thus if there is a displacement field with the properties assumed above that answers
to the given δ and minimizes the total energy, the location s of its phase boundary
is fully determined. Moreover, such a minimizing displacement, if it exists, must
be given by
H (σ0 )x, 0 ≤ x ≤ s,
u(x) = (3.27)
L (σ0 )(x − L) + δ, s ≤ x ≤ L .
satisfies the necessary conditions for a local energy minimizer that leaves the bar
in mechanical equilibrium in either the low-strain phase or the high-strain phase if
either
or
γm L ≤ δ < ∞ (high-strain phase minimizer), (3.30)
respectively.
Given a δ for which both the two-phase and a single-phase solution exist, one may
compare the total stored energies to determine an energetically preferred solution.
Let E L (δ) E H (δ), and E HL (δ) be the total energies in the low-strain, high-strain,
and mixed-phase solutions, respectively. Direct calculation gives
δ/L
E L (δ) = L W (δ/L) = L W ( L (σ0 )) + L σ̂ (γ ) dγ , −1 ≤ δ/L ≤ γ M ,
L (σ0 )
(3.31)
H (σ0 )
E H (δ) = L W (δ/L) = L W ( H (σ0 )) − L σ̂ (γ ) dγ , γm ≤ δ/L < ∞,
δ/L
(3.32)
for L (σ0 ) ≤ δ/L ≤ γ M , where the pure low-strain phase solution and the mixed-
phase solution both exist. Similarly,
H (σ0 )
E HL (δ) − E H (δ) = −L σ0 H (σ0 ) − δ/L) −
( σ̂ (γ ) dγ (3.35)
δ/L
for the appropriate range of δ. By comparing the areas represented by the two terms
in the braces in (3.34) and in (3.35), one finds that the right sides of these two
equations are both negative. Thus one has
E HL (δ) ≤ E L (δ), E HL (δ) ≤ E H (δ) (3.36)
in the respective ranges of δ where the comparisons are meaningful, so that the
mixed-phase solution is energetically preferred over either single-phase solution
whenever the latter is a competitor.
In summary, the energy-minimizing configuration of the bar is as follows: for
δ/L ≤ L (σ0 ), it involves the low-strain phase only, for δ/L ≥ H (σ0 ) it involves
only the high-strain phase, and in the intermediate range L (σ0 ) ≤ δ/L ≤ H (σ0 )
it involves an equilibrium mixture of phases at the Maxwell stress.1
1
The bold curve in Figure 3.3, which we encounter in the next section, can be associated with these
configurations.
42 EVOLUTION OF PHASE TRANSITIONS
Since f (t) vanishes if γ + (t) = γ − (t), (3.38) also applies when the bar is in a
single-phase state, asserting that D(t) = 0 in such a case.
The dissipation associated with quasistatically moving surfaces of strain dis-
continuity in three-dimensional, nonlinearly elastic solids was studied in [11].
As in the discussion of dynamics in Chapter 2, it is postulated here that the
dissipation rate cannot be negative:
The quasistatic process δ = δ(t), σ = σ (t) describes a path in the flag of equi-
librium states. As shown in Figure 3.2, the dissipation inequality (3.40) restricts
the directions that may be taken by this path at each point in the flag. At a point
(δ(t), σ (t)) lying above the “Maxwell line” σ = σ0 , the driving force f (t) is positive
by (3.6); thus by (3.40), ṡ(t) ≥ 0 at such a point, so s(t) cannot decrease in time,
and the low-strain phase cannot grow. Below the Maxwell line, s(t) must be nonin-
creasing, so the high-strain phase cannot grow. Motion in either direction along a
curve s = constant is permitted by (3.40), since such a motion is dissipation free.
For the same reason, motion in either direction along the Maxwell line σ = σ0 is
also permitted. The clusters of arrows in Figure 3.2 show the admissible directions
of a path at various points in the flag of equilibrium states.
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 43
s = constant
σ0
δ
0
Figure 3.2. Arrows showing directions of paths of quasistatic processes that are permitted by the
dissipation inequality (3.40); σ0 is the Maxwell stress.
From (3.3), the elongation, stress, and phase boundary location during a qua-
sistatic process are related through
0
δ
ΓL(σ0)L ΓH(σ0)L
curve. If the bar is subsequently shortened, the unloading path is simply the reverse
of the path followed on loading.
The Maxwell model just described is completely free of dissipation, since driving
force vanishes at all times. Such a conservative theory is far too restrictive to describe
observed dissipative phenomena such as hysteresis in a cycle of loading of the kind
just sketched.
Second, when the entire bar is in one phase, a necessary condition for nucleation
of a new phase is that the second phase be associated with a lower energy well than
the existing phase. With reference to the potential energy function sketched in
Figure 2.3, the low-strain phase is associated with the energy well on the left, while
the high-strain phase is associated with the energy well on the right. The high-
strain phase has less energy than the low-strain phase whenever σ > σ0 . However,
this does not mean that the low-strain phase will necessarily transform into the
high-strain phase when σ > σ0 , because there is an energy barrier between these
two energy wells. The smaller the energy barrier, the more likely is the nucleation
of the transformation. We shall take as the nucleation condition the criterion that
a transformation will nucleate when the height of the energy barrier is at least as
small as a certain critical value; this appears to be consistent with what is done in
materials science, for example see Chapter 10 of Christian [6]. Since the height of
the barrier can be expressed as a function of stress, which in turn can be expressed
as a function of the driving force, we can equivalently state the nucleation criterion
in terms of a critical value of driving force.
Therefore we postulate that nucleation of the low-strain → high-strain phase
transition takes place when the static driving force f that would act on the incipient
phase boundary is at least as great as a certain critical value f nLH . In view of
(3.6), driving force increases monotonically with stress, so an equivalent version
of the nucleation criterion in quasistatics asserts that nucleation of the forward
transition occurs at the particle at x = 0 if the stress in the bar satisfies σ ≥ σnLH ,
where f nLH and σnLH are related through (3.6). Similarly, the reverse transition at
x = L is associated with nucleation levels f nHL and σnHL of driving force and stress,
respectively. By (3.7), necessarily 0 ≤ f nLH ≤ f M and f m ≤ f nHL ≤ 0, and therefore
the nucleation stresses must satisfy
Often our discussion will be concerned only with the low-strain → high-strain,
or forward, phase transition; when this is the case, we shall drop the superscript LH
from the symbols for the nucleation parameters, writing instead simply f n and σn .
We now turn to the kinetics of the transformation. Upon nucleation of the
low-strain → high-strain transition, a phase boundary emerges from the end x = 0
of the bar and moves to the right; it has the high-strain phase on its left. We now
assume that its referential speed ṡ(t), which represents the rate at which the material
transforms from one phase to the other as it crosses the moving phase boundary, is
determined by the local states on the two sides of the phase boundary characterized
by the pair γ + , γ − , that is, we assume that there is a “kinetic relation” of the form
ṡ = ṽ(γ + , γ − ). (3.43)
One can show that the equilibrium requirement σ̂ (γ + ) = σ̂ (γ − ) and the definition
(3.4) of driving force determine γ + and γ − in terms of f . Thus the kinetic relation
46 EVOLUTION OF PHASE TRANSITIONS
ṡ Low-strain phase →
high-strain phase transition
ṡ = Φ(f )
frHL
f
0 frLH
High-strain phase →
low-strain phase transition
Figure 3.4. Schematic plot of a kinetic response function ( f ). If the nucleation levels of driving force
fnLH and fnHL were displayed on this figure, they would lie outside the interval ( frHL , frLH ) in keeping
with the inequalities (3.47).
that adds both viscous and strain-gradient contributions to the elastic part of the
stress; by studying smooth traveling waves in such a solid in a suitable limit that
removes these augmenting effects, one finds the particular kinetic relation obtained
in [4]. The effect of inhomogeneities on the propagation of phase boundaries in one-
dimensional tensile bars has been modeled by Bhattacharya [5] and by Rosakis and
Knowles [18]; see Section 8.4.3 for further details. These models predict irregular
phase boundary motion of the kind sometimes seen in experiments [13].
Note that the dissipation-free Maxwell model governed by the requirement
f (t) = 0 is subsumed under the present theory of kinetics; it corresponds to the
choices φ = 0 in (3.45) and f nLH = f nHL = frLH = frHL = 0 in (3.47).
Since f may be expressed in terms of σ according to (3.6), one may recast the
kinetic relation in the form
ṡ = (σ
¯ ). (3.48)
By eliminating the internal variable s between (3.3) and (3.48), one finds the “master
differential equation” relating the histories of stress σ (t) and elongation δ(t) for any
loading process. It may be written as
p(σ ) − γT (σ )δ/L σ̇ + γT (σ )δ̇/L = γT2 (σ
¯ )/L , (3.49)
where
p(σ ) = L (σ ) H (σ ) − L (σ ) H (σ ), (3.50)
and γT (σ ) = H (σ ) − L (σ ) is the transformation strain. In (3.49), usually one
would either specify σ (t) and seek δ(t), corresponding to a load-controlled process,
or δ(t) would be given, with σ (t) to be determined, in a displacement-controlled
loading program. Since the latter is the more common case in experiments of interest
here, it is the one we shall consider.
σ̂(γ)
γ = ΓL (σ) µ
σM 1 Figure 3.5. Trilinear material with equal mod-
γ = ΓH (σ) uli µ in both phases and constant transforma-
tion strain γT .
σm
µ
0 1 γ
γM γ T γm
It follows that
L (σ ) = σ/µ, H (σ ) = γT + σ/µ. (3.52)
The Maxwell stress is given by σ0 = (σ M + σm )/2, and the representation (3.6) for
the static driving force becomes
f = (σ − σ0 )γT . (3.53)
As the bar elongates at the given constant rate δ̇, the entire bar is initially in the
low-strain phase and so the stress and elongation are related by
δ = L L (σ ) = Lσ/µ (loading: low-strain phase). (3.54)
The stress rises up the left flagpole and eventually reaches the critical value σnLH at
which time the high-strain phase nucleates at x = 0. The elongation at this instant
is δnLH = L L (σnLH ) = LσnLH /µ; see Figure 3.6. During the next stage of loading
the bar is in a two-phase state. As the elongation continues to increase, the phase
σ σ
σnLH σnLH
σ 0
σ0
σnHL σnHL
δ 0
δ
0 δnLH δnHL δnLH δnHL
(a) (b)
Figure 3.6. Schematic hysteresis loops (shown bold) for the trilinear material. The high-strain phase
is nucleated during loading at (δnLH , σnLH ), and the low-strain phase is nucleated during unloading at
(δnHL , σnHL ). The bar is fully transformed prior to unloading. Figures (a) and (b) correspond to, respec-
tively, sufficiently fast and sufficiently slow loading rates, and they illustrate load-rise and load-drop at
nucleation during loading.
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 49
If the nucleation stress σnLH and the resistance stress σrLH for phase boundary motion
coincide, (3.56) shows that the initial slope is positive since (σ ¯ rLH ) = 0. On the
other hand, if σnLH > σr , one has
LH
> 0 for δ̇ > γT (σ ¯ LH ) (“load rise”),
dσ n
(3.57)
dδ δ=δLH
n < 0 for δ̇ < γT (σn ) (“load drop”).
¯ LH
Thus if σnLH > σrLH , the stress may decrease after nucleation if the elongation rate
δ̇ is slow enough; if σnLH = σrLH , this cannot occur. Figures 3.6(a) and (b) depict
two different elongation rates to illustrate load rise and load drop.
The differential equation in (3.55) holds until the point (δ(t), σ (t)) reaches the
right flagpole. At this instant the phase boundary has traversed the entire length of
the specimen and the bar is now entirely in the high-strain phase. Continued loading
causes the response to follow up the right flagpole and therefore is described by
δ = H (σ )L = Lσ/µ + LγT (loading: high-strain phase). (3.58)
Neither the kinetic relation nor the nucleation condition is now relevant. If the
loading is stopped and reversed at a subsequent time, the response first follows
down the right flagpole, eventually reaching the critical value σnHL at which time
the low-strain phase nucleates at x = L. The elongation at this instant is δnHL =
H (σnHL )L = LσnLH /µ + LγT ; see Figure 3.6. During continued unloading, the
low-strain phase grows at the expense of the high-strain phase as the phase boundary
moves leftwards, the kinetic relation being in force again. By specializing the master
50 EVOLUTION OF PHASE TRANSITIONS
equation (3.49), this unloading process can be described by the following initial-
value problem:
dσ
− (µγT /δ̇L) (σ ) = µ/L
¯
dδ (two-phase unloading). (3.59)
σ = σ HL at δ = δ HL
n n
Eventually the point (δ, σ ) returns to the left flagpole, the entire bar having returned
to the low-strain phase.
Perhaps it is worth emphasizing that in the context of a uniform bar, the nucle-
ation criterion is triggered only when the bar is in a single-phase state and is seeking
to initiate a phase transition; once the bar is in a two-phase state, the nucleation
criterion is no longer relevant, and it is the kinetic relation that now controls the
evolution of the two-phase state. In the scenario described above, the nucleation
criterion is triggered twice, once during loading and once during unloading; the
kinetic relation is in force whenever a phase boundary is present in the bar.
To simplify the results even further, consider the particular kinetic relation
M( f − fr ) if f > fr ,
ṡ = ( f ) = 0 if − fr ≤ f ≤ fr , (3.60)
M( f + fr ) if f < − fr .
Here the constant M > 0 is the phase boundary mobility; its units are those of
velocity per unit stress. Moreover, we have taken frLH = − frHL = fr so that by
(3.53), the corresponding values of stress are σrLH = σ0 + fr /γT and σrHL = σ0 −
fr /γT . In terms of stress, the kinetic response function corresponding to this ( f )
is given, according to (3.53) and (3.60), by
MγT (σ − σrLH ) if σ > σrLH ,
(σ
¯ )= 0 if σrHL ≤ σ ≤ σrLH , (3.61)
MγT (σ − σrHL ) if σ < σrHL .
For the trilinear material and the kinetic relation (3.60), the initial-value problem
(3.55) corresponding to two-phase loading has the following solution:
δ − δnLH
σ = σr + µν + σn − σr − µν exp −
LH LH LH (two-phase loading),
νL
(3.62)
and the initial-value problem (3.59) for two-phase unloading has the solution
δ − δnHL
σ = σr − µν + σn − σr + µν exp
HL HL HL (two-phase unloading).
νL
(3.63)
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 51
σ0 + Σ
Figure 3.7. Rate-independent hysteresis loop
(shown bold) for the frictional kinetic relation (3.65). σ0
The bar is fully transformed prior to unloading.
σ0 − Σ
0
δ
δ0 δ0 + LγT
σ
(δ̄, σ̄)
B
σnLH A Figure 3.8. Hysteresis loop (shown bold) for the
σ = σrLH trilinear material. Unloading commences before
C
σ0 the entire bar has transformed to the high-strain
D phase.
σ = σrHL
E
0 δ
commenced from a single-phase state and therefore involved the creation of a new
interface according to the nucleation criterion.
The response of the bar from the initial instant until the time t¯ when the loading
is reversed remains the same as in the previous program. The elongation δ̄ at t = t¯,
that is, the maximum elongation, is prescribed. The corresponding stress σ̄ = σ (δ̄)
must be such that the point (δ̄, σ (δ̄)) lies in the interior of the flag and, of course,
above the Maxwell line.
The unloading process begins from t = t¯, δ = δ̄, σ = σ̄ (point B in Figure 3.8)
and, according to the master equation (3.49), is described by the following initial-
value problem:
dσ
− (µγT /δ̇L) (σ ) = µ/L
¯
dδ (unloading). (3.67)
σ = σ̄ at δ = δ̄
As δ decreases, so does σ . Recall that the kinetic response function (σ¯ ) involves
three distinct ranges σ > σr , σr ≤ σ ≤ σr , and σ < σr , corresponding to
LH HL LH HL
the three branches of ( f ) in Figure 3.4, which by the kinetic relation correspond
to ṡ > 0, ṡ = 0, and ṡ < 0 respectively. As the stress decreases from σ̄ to 0 during
unloading, it passes through each of these ranges in sequence. First, as the stress de-
creases from σ̄ to σrLH we have ṡ > 0, and so the phase boundary continues to move
to the right even though the elongation and stress are both decreasing; the corre-
sponding unloading path is the curve BC in Figure 3.8. (This feature is not peculiar
to the particular kinetic relation (3.60) since, for any kinetic law, the dissipation
inequality will not permit the phase boundary to reverse its direction of travel until
the Maxwell line is crossed.) In the next stage, as the stress decreases from σrLH to
σrHL , the phase boundary remains stationary (ṡ = 0). The corresponding portion of
the graph of σ = σ (δ) therefore lies on a line s=constant in the flag of equilibrium
states – the straight line segment CD in Figure 3.8. After the stress passes through
σrHL , the phase boundary moves to the left, and the point (δ, σ ) eventually returns
to the left flagpole to complete the hysteresis loop.
When specialized to the kinetic relation (3.60), the initial-value problem (3.67)
for unloading is readily solved. The response involves three stages as described
EQUILIBRIUM PHASE MIXTURES AND QUASISTATIC PROCESSES 53
above. Immediately following the reversal of the sign of δ̇, one has
δ − δ̄
σ = σrLH − µν + µν + σ̄ − σrLH exp ,
νL
(3.68)
δrLH ≤ δ ≤ δ̄ (unloading, stage 1);
here δrLH is the value of elongation when the stress has diminished to the value σrLH ;
it is found from
LH
δr − δ̄ µν
exp = . (3.69)
νL µν + σ̄ − σrLH
In the second stage of unloading the path in the δ, σ -plane is the s = constant
straight line segment
µ
σ = σrLH + (δ − δrLH ), δrHL ≤ δ ≤ δrLH (unloading, stage 2), (3.70)
L
where
L
δrHL = δrLH − (σrLH − σrHL ). (3.71)
µ
In the third stage of unloading, the path moves from the point (δrHL , σrHL ) to the
low-strain flagpole; one finds that
δ − δrHL
σ = σr + µν exp
HL − 1 , δfinal ≤ δ ≤ δrHL (unloading, stage 3),
νL
(3.72)
where δfinal , the value of the elongation when the flagpole is reached, is the root of
σrHL δfinal − δrHL
δfinal = L + ν exp −1 . (3.73)
µ νL
The stress–elongation curves arising from (3.62) and (3.68)–(3.73) provide the
hysteresis loop. Examples are shown in Figures 3.8 and 3.9.
Figure 3.9 refers to a trilinear material for which σm < 0, so that permanent
deformation can occur on unloading.
3.7 Hysteresis
Our objective in this section is to study the influence of the kinetic relation on the
hysteresis associated with constant-elongation rate cycles, assuming the material
to be trilinear, but allowing the kinetic response function (σ ¯ ) to be reasonably
general. Under certain assumptions, we show that, as observed in some experiments,
the hysteresis H as predicted by the model is an increasing function of the elongation
rate δ̇ for the class of kinetic relations considered. We then adopt a particular
kinetic relation and calculate the dependence of H on δ̇ numerically. Finally, for
this particular kinetic relation, we show that, as δ̇ tends to zero, the limiting value
54 EVOLUTION OF PHASE TRANSITIONS
of H is zero only under very restrictive circumstances, so that static hysteresis [9]
normally occurs.
We make several inessential concessions to simplicity:
(i) The kinetic response function ( f ) and its inverse φ(ṡ) are assumed to be odd
functions of their arguments;
(ii) For both forward and reverse phase transitions, the nucleation level of driving
force coincides with the resistance to phase boundary motion;
(iii) Loading continues until the bar is fully transformed to the high-strain phase, at
which time unloading begins and continues until the bar is fully transformed
to the low-strain phase.
Under these assumptions, the total energy H dissipated during the cycle is twice
the energy lost during loading alone, so that
τ τ τ
H =2 D(t) dt = 2 f (t)ṡ(t) dt = 2 f (t) ( f (t)) dt, (3.74)
0 0 0
where D(t) is the dissipation rate at time t, and τ is the time required to fully
transform the bar to the high-strain phase during loading. Since this is the time
required for the phase boundary to reach the far end of the bar, (3.44) yields an
equation to determine τ :
τ
L= ( f (t))dt. (3.75)
0
monotonic in t, so that one may use the differential equation (3.76)1 to rewrite
(3.75) as follows:
fτ
( f )
µγT = d f, (3.77)
fn δ̇ − γT ( f )
where f τ is the value of driving force at the time τ of arrival of the phase boundary
at x = L. Moreover, since D(t) = f (t)ṡ(t) = f (t) ( f (t)), we may also rewrite
(3.74) in the form
L fτ
f ( f )
H =2 d f. (3.78)
µγT fn δ̇ − γT ( f )
Thus for a given elongation rate δ̇, one determines f τ = f τ (δ̇) from (3.77) and
then uses this value of f τ in (3.78) to find the total dissipation associated with the
hysteresis loop.
From (3.77), one can show that
d f τ (δ̇) δ̇ − γT ( f τ ) fτ
( f )d f
= . (3.79)
d δ̇ ( f τ ) fn (δ̇ − γT ( f ))2
Since δ̇ − γT ( f τ ) and f τ − f n have the same sign in both the load-drop and load-
rise cases, it follows from (3.79) that d f τ (δ̇)/d δ̇ > 0 for both cases, so that f τ (δ̇) is
always an increasing function of δ̇. Moreover, (3.78) and (3.79) together lead to
dH L fτ
( f τ − f ) ( f )d f
= > 0 (3.80)
d δ̇ µγT fn (δ̇ − γT ( f ))2
for both the load-rise and load-drop cases. Thus according to the present model,
the hysteresis H is always an increasing function of the elongation rate δ̇.
In the experiments on NiTi shape-memory wires carried out by Lin et al. [14],
the hysteresis is found to be an increasing function of relative elongation rate
over the two decades of δ̇/L from 1% per minute to 100% per minute at each
of three different temperatures. In practice, the thermal effects associated with
phase transformation can become significant at higher elongation rates, and so the
experimentally observed increase in hysteresis is presumably due to both thermal
and kinetic effects. This will be examined further in Chapter 10.
In the case of the special kinetic relation (3.60), the integration in (3.77) can
be carried out explicitly, furnishing the following equation for the determination of
f τ = f τ (δ̇):
δ̇ f − f − δ̇/(γ M)
τ r T
fτ − fn + log + µγT2 = 0. (3.81)
γT M f n − fr − δ̇/(γT M)
56 EVOLUTION OF PHASE TRANSITIONS
H/(µLγT2 )
ν/γT
0
With the help of (3.81), one finds that the representation (3.78) for the total dissi-
pation H specializes to
δ̇ 1 2
H/(2L) = fr + + ( f − f r ) 2
− f τ (δ̇) − f r . (3.82)
γT M 2µγT2 n
Both (3.81) and (3.82) are valid for both the load-rise and load-drop cases.
To use (3.81) and (3.82) to compute the hysteresis H as a function of δ̇ for a
given material, it is helpful to rewrite both equations in terms of dimensionless
quantities in such a way as to make clear the number of independent parameters
that must be specified. To this end, we use the dimensionless elongation rate ν intro-
duced in (3.64), and we make use of the low-strain-phase strain γτ corresponding
to the stress στ in the bar at the instant t = τ when the phase boundary arrives
at x = L. With the help of the relation (3.53) between stress and driving force,
we write
γτ = στ /µ = σ0 /µ + f τ /(µγT ) = γ0 + f τ /(µγT ), (3.83)
where γ0 = σ0 /µ is the low-strain-phase Maxwell strain. Using (3.64) and (3.83)
in (3.81) and (3.82) provides the equation to determine γτ and the corresponding
expression for the hysteresis:
γτ − γr − ν
γτ − γn + ν log = −γT , (3.84)
γn − γr − ν
H
= (γτ − γn )2 − 2(γτ − γn )(γτ − γr ) + 2γT (ν + γr − γ0 ). (3.85)
µL
For conveniently chosen but otherwise meaningless values of certain material
parameters, Figure 3.10 plots the dimensionless total dissipation H/(µLγT2 ) as
a function of a dimensionless elongation rate ν/γT in the case γr = γn .
From (3.82), one can show that, as the elongation rate tends to zero, the limiting
value of total dissipation in a loading cycle is given by
Hstatic = lim H = 2Lµ γT (γr − γ0 ) + (1/2)(γn − γr )2 ; (3.86)
ν→0
this limit is the counterpart in the present circumstances of what is called static
hysteresis in [9]. It may be noted from (3.86) that, since γT > 0 and γr ≥ γ0 , the
REFERENCES 57
REFERENCES
[11] J.K. Knowles, On the dissipation associated with equilibrium shocks in finite
elasticity. Journal of Elasticity, 9 (1979), pp. 131–58.
[12] J.K. Knowles, Hysteresis in the stress-cycling of bars undergoing solid–solid
phase transitions. Quarterly Journal of Mechanics and Applied Mathematics,
55 (2002), pp. 69–91.
[13] R.V. Krishnan and L.C. Brown, Pseudo-elasticity and the strain–memory ef-
fect in an Ag-45 at. pct. Cd alloy. Metallurgical Transactions A, 4 (1973),
pp. 423–9.
[14] P. Lin, H. Tobushi, K. Tanaka, T. Hattori, and A. Ikai, Influence of strain rate
on deformation properties of TiNi shape memory alloy. JSME International
Journal A, 39 (1996), pp. 117–23.
[15] K.N. Melton and O. Mercier, Fatigue of NiTi thermoelastic martensites. Acta
Metallurgica, 27 (1979), pp. 137–44.
[16] S. Miyazaki, T. Imai, Y. Igo, and K. Otsuka, Effect of cyclic deformation
on the pseudo-elasticity of Ti–Ni alloys. Metallurgical Transactions A, 17
(1986), pp. 115–20.
[17] N. Nakanishi, Lattice softening and the origin of SME, in Shape-Memory
Effects in Alloys, edited by J. Perkins. Plenum Press, New York, 1975,
pp. 147–76.
[18] P. Rosakis and J.K. Knowles, Continuum models for irregular phase boundary
motion in solids. European Journal of Mechanics A, 18 (1999), pp. 1–16.
[19] J.A. Shaw and S. Kyriakides, Thermomechanical aspects of NiTi. Journal of
the Mechanics and Physics of Solids, 43 (1995), pp. 1243–81.
[20] K. Tanaka, T. Hayashi, Y. Itoh, and H. Tobushi, Analysis of thermomechan-
ical behavior of shape memory alloys. Mechanics of Materials, 13 (1992),
pp. 207–15.
[21] K. Tanaka, F. Nishimura, T. Hayashi, H. Tobushi, and C. Lexcellent, Phe-
nomenological analysis on subloops and cyclic behavior in shape memory
alloys under mechanical and/or thermal loads. Mechanics of Materials, 19
(1995), pp. 281–92.
[22] H. Tobushi, P.-H. Lin, T. Hattori, and M. Makita, Cyclic deformation of TiNi
shape memory alloy. JSME International Journal A, 38 (1995) pp. 59–67.
4 Impact-Induced Transitions
in Two-Phase Elastic Materials
4.1 Introduction
We next turn to the dynamics of the two-phase nonlinearly elastic materials in-
troduced in Chapter 2. As in the theory of mixed-phase equilibria and quasistatic
processes for such materials set out in Chapter 3, the notion of driving force plays a
central role in the analysis when inertial effects are taken into account. The indeter-
minacy exhibited in Chapter 3 by even the simplest static or quasistatic problems
for two-phase materials manifests itself again in the present much richer dynami-
cal context. Moreover, the continuum-mechanical interpretations of nucleation and
kinetics again serve to restore the uniqueness of solutions to the dynamic problems
to be considered here. As in the preceding chapters, thermal effects are omitted;
they will be included in the more general settings of later chapters.
The main vehicle for our study of one-dimensional dynamics of two-phase ma-
terials is the impact problem. There is an enormous body of experimental literature
pertaining to the response of solids to shock or impact loading, much of it motivated
by questions concerning the behavior of materials at extremely high pressures, as
occurs, for example, deep in the earth. The reader will find some guidance to the
experimental literature in this field of the dynamic behavior of materials in the
books by Graham [11] and Meyers [25].
In the prototypical experiment, a specimen of the material under investigation
consists of a circular cylindrical disk restrained against radial expansion and subject
to normal impact on one of its plane faces by a “flyer-plate” moving at high speed.
Care is taken to assure that the plane faces of specimen and flyer-plate are parallel
at impact. The collision generates a compressive disturbance that propagates into
the disk, ultimately reaching the rear face of the target. Propagation speeds of
moving strain discontinuities are measured, as are particle velocities at the rear
face. The time scales in such experiments can be as short as tenths of microseconds.
Depending on the material, a phase transition may occur, as, for example, in the
case of the impact-induced graphite-to-diamond transition studied experimentally
by Erskine and Nellis [9].
59
60 EVOLUTION OF PHASE TRANSITIONS
1
µL
LPP
1
µH
HPP
Figure 4.1. The stress–strain curve for a trilinear two-phase elastic material that changes phase in
compression between a low-pressure phase (LPP) and a high-pressure phase (HPP).
A= B
−γf
Figure 4.2. The equal-area strain γ = −γ f for γ
0
the trilinear material of Figure 4.1.
LPP
A B
HPP
extended LPP branch of the stress–strain curve intersects the HPP branch, as shown
in Figure 4.1. The second is the strain γ = −γ f , which, together with γ = 0,
generates a chord that cuts off triangles of equal area from the stress–strain curve;
see Figure 4.2. These special strains are given by
We assume that γm < γT < γ M , so that one has 0 < γm < γT < γ M < γ f < γ∗ < 1.
In the version of the trilinear material shown in Figure 4.1, the stress σ̂ (−γ M ) at
the local maximum has been taken to be compressive (negative), so that the reference
state γ = 0 is the only available stress-free state. If, on the other hand, σ̂ (−γ M ) > 0,
then there would be a state in the high-pressure phase at which the stress vanishes.
In the former case, the impacted body must return to the reference state in the low-
pressure phase upon removal of load, whether or not a LPP→HPP phase transition
has occurred in response to loading; thus the loading cycle is “globally reversible”
in this case. In the latter case [σ̂ (−γ M ) > 0], the body may remain in the high-
pressure phase upon unloading if a load-induced phase transition has taken place, in
which case the cycle is “globally irreversible.” The graphite-to-diamond transition
studied in [9], for example, is thought to be globally reversible, corresponding to
the case portrayed in Figure 4.1. It is assumed here that σ̂ (−γ M ) < 0, as in the
figure.
We shall also assume that the high-pressure phase is the stiffer of the two phases,
as is often the case for real materials. Thus the moduli for uniaxial deformation in
the two phases are assumed to satisfy µH > µL . We shall also require that the special
strain −γ∗ be greater than −1 for reasons that will be clear later. These requirements,
together with the observation from Figure 4.1 that σ̂ (−γm ) < σ̂ (−γ M ), lead to the
following restriction:
4.2.2 The impact problem. The target in our impact problem is assumed to be
initially at rest in the reference configuration: v(x, 0) = 0, γ (x, 0) = 0. At time
t = 0, a prescribed constant velocity v0 is given to the particles at the boundary
x = 0 of the half-space and maintained for all subsequent times, so that v(0, t) = v0 .
We wish to construct solutions γ (x, t), v(x, t) to the field equations (2.20) that
satisfy these boundary and initial conditions and fulfill the jump conditions (2.15)
and (2.16), as well as the dissipation inequality (2.26), at any strain discontinuity.
Thus we study the following boundary–initial value problem:
σ̂ (γ )γx = ρvt ,
for x ≥ 0, t ≥ 0, (4.6)
v x = γt ,
γ (x, 0) = v(x, 0) = 0, for x ≥ 0, (4.7)
v(0, t) = v0 , for t ≥ 0. (4.8)
If there is a strain discontinuity at x = s(t), the jump conditions
[[σ̂ (γ )]] = −ρ ṡ[[v]],
(4.9)
[[v]] = −ṡ[[γ ]]
and the dissipation inequality
f (t)ṡ(t) ≥ 0 (4.10)
must hold. The inequality (4.10) is the counterpart in the present purely mechanical
theory of the requirement in adiabatic, inviscid gas dynamics that entropy should
increase across a shock wave. For the trilinear material, driving force vanishes at a
shock wave in either the LPP or HPP phase, so (4.10) is trivially satisfied at shocks.
This is not the case, of course, at phase boundaries.
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 65
Because there is no natural length present, the impact problem as just formulated
is invariant under the scale change x → kx, t → kt, where k = 0 is constant; so
it is natural to seek solutions that are scale invariant in the sense that γ (kx, kt) =
γ (x, t), v(kx, kt) = v(x, t). One can show that, in such solutions, γ and v must
be functions of x/t; for the trilinear material, all solutions with this property are
piecewise constant, with discontinuities possible only across rays x/t=constant in
the x, t-plane; there are no “fans.” Thus for a scale-invariant solution of the impact
problem, the first quadrant of the x, t-plane may be divided into sectors S1 , S2 , ..., Sn
centered at the origin; we take S1 to be adjacent to the x-axis and Sn adjacent to the
t-axis. Thus on S1 , one has x > ṡ1 t, and on Si : ṡi t < x < ṡi−1 t for i = 2, ..., n,
with ṡ0 = 0.
The dissipation inequality (4.10) has significant implications for scale-invariant
solutions of the impact problem. First, it allows us to show that a strain correspond-
ing to a point on the unstable, declining branch of the trilinear stress–strain curve
cannot arise in any sector Si . If this occurred, the length of the x-interval in Si
bearing such a strain would increase with time, contradicting an implication of the
dissipation inequality established in Section 2.4. Thus no scale-invariant solution
of the impact problem can involve states on the declining branch of the trilinear
stress–strain curve.
The dissipation inequality (4.10) can also be used to prove that the solutions we
seek can involve no more than one phase boundary and no more than one shock
wave, so that the number n of sectors Si cannot exceed 3. The proof uses two
geometric properties that we observed previously. First, because of (2.22), recall
that the speed of a discontinuity is related to the slope of the chord joining the
points (γ − , σ̂ (γ − )) and (γ + , σ̂ (γ + )) on the stress–strain curve. Consequently, the
speeds of different discontinuities can be compared by comparing the slopes of
the corresponding chords. Second, we know from (2.25) that the driving force on
a discontinuity can be related to the area between this chord and the stress–strain
curve. Figure 4.3 illustrates both of these properties for a phase boundary that has
LPP on its right.
We now prove that the number n of sectors Si cannot exceed 3. To show this, one
notes first that on the sector S1 , one will have γ = v = 0, so that the leading wave
front x = ṡ1 t is either (Case 1) a LPP→HPP phase boundary, or (Case 2) a LPP
shock wave. Dealing first with Case 1, we show that there can be no discontinuities
trailing x = ṡ1 t. Suppose the contrary, and let x = ṡ2 t, with ṡ2 < ṡ1 , be the first
ray trailing the leading disturbance in the first quadrant of the x, t-plane. This
discontinuity must be either a HPP shock wave or a HPP→LPP phase boundary. But
sketching the relevant chords on the trilinear stress–strain curve shows immediately
that the speed of a HPP shock cannot be less than that of a LPP→HPP phase
boundary as required, so x = ṡ2 t must be a HPP→LPP phase boundary. But drawing
the chords connecting the relevant states on the stress–strain curve shows that
the driving force acting on this phase boundary must be negative if 0 < ṡ2 < ṡ1 ,
violating the dissipation inequality. Thus if the leading disturbance is a LPP→HPP
phase boundary (Case 1), the number of sectors Si in the first quadrant of the
x, t-plane is n = 2.
66 EVOLUTION OF PHASE TRANSITIONS
σ = σ̂(γ)
σ̂(γ +) − σ̂(γ −)
ρṡ2 =
γ+ − γ−
LPP
f =A−B
γ− γ+
γ
0
(γ +, σ̂(γ +))
A B
(γ −, σ̂(γ −)) ρṡ2
HPP LPP
1
s(t)
x
γ− γ+
HPP
Figure 4.3. A phase boundary with HPP on its left and LPP on its right. The strains on either side of the
phase boundary are γ − and γ + respectively. The slope of the chord joining (γ − , σ̂ (γ − )) to (γ + , σ̂ (γ + ))
is ρ ṡ 2 where ṡ is the propagation velocity of the phase boundary. The difference in the shaded areas
gives the driving force f = A − B.
x = s˙ t
−
γ = γ , v = v0
Figure 4.4. Structure in the x, t-plane of a two-
wave solution with a LPP→HPP phase tran-
sition for the impact problem for the trilinear γ = γ + , v = v+
material. It is necessary that γ + > −γm and x = cL t
−1 < γ − < −γ M .
HPP
LPP
γ = 0, v = 0
LPP v0
x
0
solution to the impact problem for the trilinear material would be the conventional
one for the linear wave equation corresponding to the low-pressure phase: a LPP
shock wave moves into undisturbed material. This solution is easily found to be
given by
−v0 /cL , v0 if 0 < x < cL t,
γ, v = (4.11)
0, 0 if x > cL t;
since this solution assumes the absence of a phase change, it only exists when the
strain at each particle remains in LPP at all times, that is, when −v0 /cL > −γm .
This requires that
v0 < cL γm . (4.12)
4.3.2 Solutions with a phase transition: the two-wave case. For sufficiently large
impactor velocity v0 one would expect a phase change to occur. There are in fact
two types of solutions involving a phase transition. In the first type, the target
experiences a LPP shock wave moving with speed cL , followed by a phase boundary;
the (Lagrangian) phase boundary velocity ṡ must therefore be subsonic with respect
to the LPP shock speed cL : ṡ < cL . The structure of this “two-wave” solution with
a phase transition is indicated in Figure 4.4; the unknowns to be determined are the
constant strains γ ± behind and ahead of the phase boundary, the constant particle
velocity v + ahead of the phase boundary, and the phase boundary velocity ṡ.
Since the unknown strains and particle velocities are all constants, the differen-
tial equations (4.6) are trivially satisfied where γ , v are smooth. The jump conditions
(4.9) then determine γ ± and v + in terms of the impactor velocity v0 and the still
unknown phase boundary velocity ṡ as follows:
x = s˙ t
As depicted in Figure 4.4, this solution presumes that particles in the range
ṡt < x < cL t are in LPP, while those in 0 < x < ṡt are in HPP; consequently,
the corresponding strains γ + and γ − are required to be such that
Through (4.13), the velocities v0 and ṡ are subject to restrictions arising from
the phase segregation inequalities (4.14). From (4.5) and (4.13), one can find the
driving force at the phase boundary in terms of the given impactor velocity v0 and
the as yet unknown phase boundary velocity ṡ. The result may be written in the
form
v0 2 v0
f = −P + 2Q − R, (4.15)
γT cH γT cH
where P, Q, and R are functions of ṡ defined by
cH 2 − ṡ 2 cL + ṡ
P = µH (γT2 /2)(cH 2 − cL 2 ) , (4.16)
(cH 2 + cL ṡ)2 cL − ṡ
cH 2 − ṡ 2 cL + ṡ
Q = µH (γT2 /2)cL cH , (4.17)
(cH 2 + cL ṡ)2 cL − ṡ
γf 2cL cH 2 ṡ 2 − cH 4 (cL − ṡ)
R= µH (γT2 /2) + . (4.18)
γT (cH 2 + cL ṡ)2 (cL − ṡ)
4.3.3 Solutions with a phase transition: the one-wave case. The second type of
solution involving a phase change has the structure shown in Figure 4.5.
In this case, each particle of the target jumps directly from the undisturbed LPP
reference state to the HPP state; the phase boundary is not preceded by a shock
wave. In contrast to the two-wave case just discussed, applying the jump conditions
at the phase boundary now fully determines the unknowns γ − and ṡ in terms of the
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 69
v0
E
D
f=0
γ + = −γm
C One-wave solution
F with phase change
B
One-wave solution P f =0
with no phase change A Two-wave solution
with phase change
ṡ
O cL
Figure 4.6. Pairs (ṡ, v0 ) for which the impact problem has a solution. The segment OB of the vertical
axis shown bold, defined by (4.12), corresponds to solutions without a phase transition. The shaded
region is defined by (4.13), (4.14), and driving force f > 0; it corresponds to two-wave solutions with
a phase transition. One-wave solutions with a phase transition are associated with points on the curve
FE, which is defined by (4.19) and (4.21).
impactor velocity v0 :
− 2v0 /γT
γ = −(γT /2) 1 + 1 + 4v0 /(cH γT ) ,
2 2 2
ṡ =
.
1+ 1 + 4v02 /(cH 2 γT2 )
(4.19)
Through (4.19)1 , v0 must be restricted by the phase segregation condition −1 <
γ − < −γ M .
Using the representation (4.5) with γ + = 0 and γ − given by (4.19), one can
find the driving force acting on the phase boundary for the one-wave solution. The
requirement imposed by the dissipation inequality (4.10) then becomes a further
restriction on γ − ; this restriction turns out to be γ − < −γ f , where γ f is the special
equal-area strain introduced earlier in (4.2) and Figure 4.2. In addition, γ − must
satisfy −1 < γ − < −γ M . Since γ f > γ M , these requirements reduce to
−1 < γ − < −γ f , (4.20)
which, through (4.19)1 , becomes the following restriction on the impactor velocity:
γ f (γ f − γT ) cH < v0 < 1 − γT cH . (4.21)
4.3.4 The totality of solutions. It is helpful to describe the solutions discussed above
in a Cartesian plane in which the coordinates are the phase boundary velocity ṡ and
the impactor velocity v0 , as in Figure 4.6.
70 EVOLUTION OF PHASE TRANSITIONS
We first delineate the set of points in this plane that correspond to the no-
phase-change solution (4.11) permitted by the restriction (4.12). Thinking of these
formally as special cases of the two-wave solutions of Figure 4.4 for which ṡ = 0,
we may say that each no-phase-change solution corresponds to a point in the interval
0 < v0 < cL γm of the vertical axis shown bold in Figure 4.6.
Next we describe the points (ṡ, v0 ) corresponding to the two-wave solutions
(4.13) with a phase change, subject to the phase segregation conditions (4.14) and
the dissipation inequality (4.10), with f given by (4.5). It turns out that, of these
inequalities, the decisive ones are γ + > −γm and f > 0; all others are implied
by these. Using (4.5), one finds that f = 0 corresponds to the parabola-like curve
shown schematically in Figure 4.6; for points inside this curve, one has f > 0. The
requirement γ + > −γm corresponds to points below the curve marked γ + = −γm ,
also sketched schematically in the figure. Thus each point in the shaded region
of Figure 4.6 corresponds through (4.13) to a two-wave solution satisfying all
requirements.
Finally, the points in the ṡ, v0 -plane that correspond to the one-wave, phase-
changing solutions lie on the curve represented by the second of (4.19), subject to
the restriction (4.21). A portion of this curve, not shown in Figure 4.6, lies inside
the shaded region in the figure, while the remainder lies outside this region and
is shown bold as the curve FE. The points on the curve that lie inside the shaded
area correspond to special cases of the two-wave solution (4.13) for which the
front-running shock wave happens to be absent whence γ + = 0. For this to occur,
ṡ and v0 must be suitably related; the necessary relation is precisely the second of
(4.19). Thus only the one-wave solutions associated with points on FE correspond
to solutions not already accounted for in (4.13).
The following conclusions can be drawn from Figure 4.6:
(i) For each impactor velocity v0 in the interval OA of the vertical axis,
there is a unique solution given by (4.11); there is no phase transition.
(ii) For each v0 corresponding to a point, say P, between A and B on the
vertical axis, there are two types of solution: a no-phase-change so-
lution (4.11) corresponding to the point P itself, and a one-parameter
family of two-wave solutions (4.13) corresponding to all points in the
shaded region that lie on the horizontal line through P. Each of the latter
describes a phase transition.
(iii) For each v0 corresponding to a point on the vertical axis between B
and C, there is a one-parameter family of two-wave solutions (4.13) in
each of which the target ultimately changes phase. Solutions without a
phase change cannot occur.
(iv) For each v0 corresponding to a point on the vertical axis between C and
D, there is a unique solution; it is the one-wave solution (4.19) with a
phase transition.
(v) For each v0 corresponding to a point on the vertical axis above D, there
is no solution to the impact problem.
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 71
According to (i) above, unless the impactor velocity is large enough, there will
be no phase transition.
Conclusion (ii) describes the way in which the impact problem differs drastically
from its counterpart for a single-phase material: there is a massive loss of unique-
ness. This loss manifests itself in two ways; for impactor velocities v0 in the segment
AB of the v0 -axis, the target must first “choose” between the no-phase-change so-
lution and the one-parameter family (parameter ṡ) of two-wave, phase-changing
solutions that are also available for such a v0 . As we shall see below, this choice is
made by a nucleation criterion. If the choice favors the phase transition, the body
must then select a particular two-wave solution from the family (4.13). This is done
through a kinetic relation, which – as we show below – determines ṡ in terms of v0 ,
thus selecting a solution from the one-parameter family.
If v0 corresponds to a point on the segment BC, conclusion (iii) shows that
infinitely many solutions are again available, all involving a phase transition; a
kinetic relation is again needed to choose a particular one.
If v0 is large enough to fall in the segment CD of the vertical axis, conclusion
(iv) asserts that there is a unique one-wave solution (4.19), and it involves a phase
transition. A kinetic relation is not needed and may not be prescribed. It may be
noted that the speed ṡ of the phase boundary in this case exceeds the speed cL of LPP
shock waves; the phase boundary is said to be overdriven. The distinction between
the two-wave solution, for which a kinetic relation must be prescribed, and the
one-wave solution, where kinetics may not be imposed, apparently has a parallel in
combustion theory, where deflagrations require prescription of the relevant kinetics,
while detonations do not; see [14].
Finally, the fact that there is no solution for all sufficiently large values of v0 ,
as asserted in conclusion (v), is a consequence of the fact that, according to the
predictions of the model, such impacts result in strains of magnitude greater than
unity, corresponding to infinite contraction of the target at a finite stress. This is an
artifact of the trilinear model; it would not occur for a two-phase material model in
which the stress σ̂ (γ ) in the high-pressure phase is unbounded as the strain γ tends
to −1.
fLH
n
For maximally dissipative kinetics, the critical value of f for LPP→HPP nucleation
is chosen to be f nLH = φ(0+), which is also the resistance frLH to LPP→HPP phase
boundary motion. A schematic graph of (4.25) is shown in Figure 4.7.
Since γ + = −γm for maximally dissipative kinetics in the case of positive ṡ of
interest to us in the impact problem, reference to the stress–strain curve of Figure 4.1
suggests the following interpretation of the process just described: the phase change
is not initiated until the given impact velocity is sufficient to drive the magnitude of
the compressive strain in the target to its largest possible value in the low-pressure
phase, corresponding to the local minimum at γ = −γm in Figure 4.1. Thus the
particles at which nucleation is to take place “wait as long as possible” to jump to
the new phase, corresponding to a convention sometimes called perfect delay [30].
It is clear that (4.15)–(4.18), (4.25), and (4.26) together provide a relation be-
tween impactor velocity v0 and phase boundary velocity ṡ. This relation is most
efficiently obtained by setting γ + = −γm in (4.13)2 ; the result is
ṡ
v0 = γm cL + γT cH 2 − γm (cH 2 − cL 2 ) ,
cH 2 − ṡ 2 (4.27)
0 ≤ ṡ < cL (two-wave response).
The right side of (4.27) is an increasing, and thus invertible, function of ṡ.
74 EVOLUTION OF PHASE TRANSITIONS
For the overdriven one-wave solution, the counterpart of (4.27) is (4.19)2 , which
may be inverted to give
ṡ
v0 = γT cH 2 , cL < ṡ < 1 − γT cH (one-wave response). (4.28)
cH 2 − ṡ 2
The time history of particle velocity is of interest in experiments. Assuming
that a phase change occurs and that maximally dissipative kinetics apply, one finds
v(x, t) as a function of t at fixed x from (4.13)3 and the fact that γ + = −γm :
0, for 0 < t < x/cL
v(x, t) = γm cL , for x/cL < t < x/ṡ (two-wave response), (4.29)
v0 , for t > x/ṡ
where the phase boundary speed ṡ is related to the impactor velocity v0 by (4.27).
It may be noted from (4.29) that the particle velocity behind the front-running
shock wave is independent of the impactor velocity. As we shall see below, this
artifact of maximally dissipative kinetics is observed, at least approximately, in
some experiments.
If the impactor velocity is great enough to cause overdriven response, one has
instead the simple relation
0, for 0 < t < x/ṡ
v(x, t) = (one-wave response), (4.30)
v0 , for t > x/ṡ
where now ṡ is determined in terms of v0 by (4.28).
Figure 4.8. Particle velocity versus time as measured in the experiments of Erskine and Nellis [9].
The target was pyrolytic (ZYB) graphite. The curves have been staggered horizontally on the graph
for clarity. The two-wave structure is direct evidence for a phase transition. Reprinted from [9] with
permission from the Journal of Applied Physics.
the observed history of particle velocity has the two-wave structure predicted
by the model discussed above for all but the highest impactor velocity achieved
in the experiments. For this most severe impact, the experimental result has the
one-wave, overdriven structure that also arises in our analysis. Second, the velocity
of the front-running shock wave observed in the two-wave experiments is nearly
independent of impactor velocity, being almost constant over the range of impactor
velocities that result in two-wave response. This is consistent with the fact that,
in the trilinear material, the front-running shock travels with constant velocity cL .
Third, in the experimentally observed two-wave particle velocity histories, the par-
ticle velocities observed after the arrival of the front-running shock but prior to
the arrival of the phase boundary are also independent of impactor velocity to a
good degree of approximation. Thus the experimental results in [9] are consistent
in this sense with the predictions for the impact problem of the model based on the
trilinear material and maximally dissipative kinetics.
The preceding analysis can be applied to the experiments described in [9], but
only after several essential modifications. One must account for the deformability
of the flyer-plate, tacitly assumed to be rigid in the preceding analysis, as well as
the presence of the deformable window in contact with the specimen at its rear
face. The latter modification requires that the “half-space target” in the analysis
above be replaced by a slab of finite thickness, a change that necessitates the con-
sideration of waves reflected and transmitted at the target-window interface. An
analysis incorporating these modifications was carried out in [4]. The experimen-
tally determined propagation speeds of the two waves and the observed particle
velocities as given in [9] were used to determine the parameters γm , γT , cL , and cH .
These values, together with the density ρ, allow the calculation of the particle
velocities as functions of time as predicted by the model for the impactor ve-
locities arising in [9]. The results are shown in Figure 4.9; the graphs are the
76 EVOLUTION OF PHASE TRANSITIONS
v v
3 3
2 2
1 V = 2.60 1 V = 3.1
t t
0 0.2 0.4 0 0.2 0.4
v v
3 3
2 2
counterparts for the model of those for the experiments shown in Figure 4.8.
Table 4.1 compares the numerical values of the ultimate particle velocity vmodel
as predicted by the model with the observed values vexp . The largest error between
predicted and observed values is about 5.7% of the observed value. For further de-
tails comparing the model with the experiments reported in [9], the reader is referred
to [4].
v0 vmodel vexp
2.60 1.91 1.83
3.12 2.28 2.20
3.47 2.53 2.46
3.90a 2.84a 2.70a
a
Overdriven.
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 77
physical problems by various authors. Thus Zhong [38] has discussed the effect of
an elastic wave incident upon the interface between two phases of a trilinear elastic
bar; he finds the long-time limit of the position of the phase boundary after it is
dislodged from its initial position by the incident wave.
The analysis in [3] concerns a composite bar, half of which is made of a single-
phase elastic material while the remaining half is trilinear. A wave traveling in the
single-phase material is incident on the bond between the two halves of the bar.
The question addressed is whether the phase-changing capability of the trilinear
material can be exploited to augment or diminish the strength of the wave that is
either reflected or transmitted from the bond.
Many studies of one-dimensional dynamic problems for phase-changing elastic
bodies have been limited to trilinear materials. In [22], Lin analyzes Riemann
problems for a two-phase elastic material whose bulk behavior in either phase is
nonlinear, so that the relevant stress–strain curve looks like that in Figure 2.2.
A computational scheme capable of coping with propagating phase boundaries
has been developed by Zhong, Hou, and LeFloch [39]; this scheme has been gen-
eralized and applied to the interpretation of impact experiments by Winfree [37].
The second-order, quasilinear system (4.6) for the two unknowns γ , v has been
the subject of extensive study, perhaps motivated initially by interest in gas dy-
namics [5], but often simply as a system of mathematical interest, with pioneering
contributions from Dafermos [7], Lax [18], Liu [23], Oleinik [26], and others. For
general expositions of the theory, the reader may consult the notes by Lax [19],
the textbook of Evans [10], or the recent monograph of LeFloch [21], which em-
phasizes quasilinear systems to which supplementary restrictions such as kinetic
relations must be appended.
In the purely mechanical setting of the present chapter, the dissipation inequality
(4.10) for the system (4.6) is the counterpart of the entropy inequality arising from
the second law of thermodynamics; its role in selecting a unique weak solution to
problems in gas dynamics is well known. Many authors have studied admissibility
criteria intended to play the role for general systems of conservation laws that
the entropy inequality plays in gas dynamics; see [21] for references. One of the
most widely used such criteria involves replacing the nonlinear constitutive law by
one that adds effects due to viscosity and strain gradient to the elastic part of the
stress:
The use of this criterion in the study of phase transitions for the system (4.6) was
pioneered by Slemrod [34] and Truskinovsky [36]. Another important criterion was
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 79
proposed by Lax [18, 19]. For the system (4.6), the Lax admissibility criterion is
equivalent to the following condition given by Dafermos [7].
Lax criterion:
A strain discontinuity propagating with velocity ṡ and joining the states γ + , v +
and γ − , v − is admissible if ṡ(γ + − γ − ) ≥ 0 (ṡ(γ + − γ − ) ≤ 0) and the chord
joining the points (γ + , σ̂ (γ + )) and (γ − , σ̂ (γ − )) on the stress–strain curve lies
below (above) the tangents at (γ + , σ̂ (γ + )) and (γ − , σ̂ (γ − )).
For nonlinearly elastic materials for which σ̂ (γ ) is either a strictly convex or strictly
concave function of γ – the so-called genuinely nonlinear case [19] – the geometric
interpretation of driving force allows one to show that the dissipation inequality
(4.10) is equivalent to the Lax criterion. It is known that imposing the Lax criterion
at all strain discontinuities is sufficient in the genuinely nonlinear case for the
uniqueness of the solution to the Riemann problem associated with the system
(4.6); see Chapter VI of [21]. On the other hand, for materials that are not genuinely
nonlinear such as the ones with which we are concerned here, the Riemann problem
need not have a unique solution, even with the Lax condition in force; see [21] as
well as the remarks of Dafermos [8].
Still another admissibility condition, the so-called chord criterion, has been
studied by various authors after being proposed by Oleinik [27] for first-order scalar
conservation laws. As stated by Shearer [33], the chord criterion would apply in
our setting to any stress–strain relation σ = σ̂ (γ ).
Chord criterion:
A strain discontinuity propagating with velocity ṡ and joining the states γ + , v +
and γ − , v − is admissible if either ṡ = 0, or ṡ(γ + − γ − ) ≥ 0 (ṡ(γ + − γ − ) ≤ 0)
and the chord joining the points (γ + , σ̂ (γ + )) and (γ − , σ̂ (γ − )) on the stress–strain
curve lies below (above) the curve.
The geometric interpretation of the driving force allows one to show that any
discontinuity that is admissible according to the chord criterion also satisfies the
dissipation inequality (4.10), though the converse is false. In the genuinely non-
linear case, the chord criterion and the Lax criterion are equivalent. Pego [28] has
shown that, under suitable assumptions, the chord criterion is equivalent to a vis-
cosity criterion obtained as a special case of the viscosity–strain gradient criterion
mentioned above by choosing the strain gradient coefficient λ to be zero in (4.31).
For two-phase elastic materials, the stress–strain relation σ = σ̂ (γ ) not only fails
to be genuinely nonlinear, it fails to be monotonically increasing as well; the system
(4.6) is then hyperbolic-elliptic. For an elastic material like rubber, the curvature
of the stress–strain curve changes from concave to convex as the tensile strain
increases [35], so that genuine nonlinearity fails. Monotonicity holds, however, so
that the system (4.6) remains hyperbolic in this case. For a rubberlike bar, a kinetic
relation turns out to be needed in the impact problem; see [16, 17]. Hyperbolic
80 EVOLUTION OF PHASE TRANSITIONS
systems that fail to be genuinely nonlinear are discussed in [21], where references
to other work in this area may be found.
An existence theorem in the dynamics of two-phase elastic materials was estab-
lished by LeFloch [20].
REFERENCES
[1] R. Abeyaratne and J.K. Knowles, Kinetic relations and the propagation of
phase boundaries in solids. Archive for Rational Mechanics and Analysis,
114 (1991), pp. 119–54.
[2] R. Abeyaratne and J.K. Knowles, On the propagation of maximally dissipative
phase boundaries in solids. Quarterly of Applied Mathematics, 50 (1992),
pp. 149–72.
[3] R. Abeyaratne and J.K. Knowles, Reflection and transmission of waves from
an interface with a phase-transforming solid. Journal of Intelligent Materials
Systems and Structures, 3 (1992), pp. 224–44.
[4] R. Abeyaratne and J.K. Knowles, On a shock-induced martensitic phase
transition. Journal of Applied Physics, 87 (2000), pp. 1123–34.
[5] R. Courant and K.O. Friedrichs, Supersonic Flow and Shock Waves, Inter-
science. New York, 1948.
[6] C. Dafermos, The entropy rate admissibility criterion for solutions of
hyperbolic conservation laws. Journal of Differential Equations, 14 (1973),
pp. 202–12.
[7] C. Dafermos, Hyperbolic systems of conservation laws, in Systems of Non-
linear Partial Differential Equations, edited by J.M. Ball. Reidel, Dordrecht,
1983, pp. 25–70.
[8] C. Dafermos, Discontinuous thermokinetic processes, Appendix 4B. Ratio-
nal Thermodynamics, edited by C. Truesdell. Springer-Verlag, New York,
1984.
[9] D.J. Erskine and W.J. Nellis, Shock-induced martensitic transformation of
highly oriented graphite to diamond. Journal of Applied Physics, 71 (1992),
pp. 4882–6.
[10] L.C. Evans, Partial Differential Equations. American Mathematical Society,
Providence, RI, 1998.
[11] R.A. Graham, Solids under High-Pressure Shock Compression. Springer-
Verlag, New York, 1992.
[12] H. Hattori, The Riemann problem for a van der Waals fluid with entropy rate
admissibility criterion. Isothermal case. Archive for Rational Mechanics and
Analysis, 92 (1986), pp. 247–63.
[13] H. Hattori, The Riemann problem for a van der Waals fluid. Nonisothermal
case. Journal of Differential Equations, 65 (1986), pp. 158–74.
[14] W.D. Hayes, Gas Dynamic Discontinuities. Princeton University Press,
Princeton, 1960.
IMPACT-INDUCED TRANSITIONS IN TWO-PHASE ELASTIC MATERIALS 81
[15] R.D. James, The propagation of phase boundaries in elastic bars. Archive for
Rational Mechanics and Analysis, 73 (1980), pp. 125–58.
[16] J.K. Knowles, Impact-induced tensile waves in a rubberlike material. SIAM
Journal on Applied Mathematics, 62 (2002), pp. 1153–75.
[17] J.K. Knowles, Sudden tensile loading of a rubberlike bar. Mechanics Research
Communications, 30 (2003), pp. 581–7.
[18] P. Lax, Hyperbolic systems of conservation laws II. Communications on Pure
and Applied Mathematics, 10 (1957), pp. 537–66.
[19] P. Lax, Hyperbolic Systems of Conservation Laws and the Mathematical
Theory of Shock Waves, Regional Conference Series in Applied Mathematics,
No. 11, SIAM, Philadelphia, PA, 1973.
[20] P.G. LeFloch, Propagating phase boundaries: formulation of the problem
and existence via the Glimm method. Archive for Rational Mechanics and
Analysis, 123 (1993), pp. 153–97.
[21] P.G. LeFloch, Hyperbolic Systems of Conservation laws. Birkhäuser Verlag,
Basel, 2002.
[22] Y. Lin, A Riemann problem for an elastic bar that changes phase. Quarterly
of Applied Mathematics, 53 (1995), pp. 575–600.
[23] T.P. Liu, The Riemann problem for general 2×2 conservation laws. Transac-
tions of the American Mathematical Society, 199 (1974), pp. 89–112.
[24] J. Lubliner, A maximum dissipation principle in generalized plasticity. Acta
Mechanica, 52 (1984), pp. 225–37.
[25] M. Meyers, Dynamic Behavior of Materials. Wiley, New York, 1994.
[26] O. Oleinik, On the uniqueness of the generalized solution of the Cauchy
problem for a nonlinear system of equations occurring in mechanics. Uspekhi
Matematicheskii Nauk (N.S.), 12 (1957), pp. 169–76 (in Russian).
[27] O. Oleinik, Uniqueness and stability of the generalized solution of the Cauchy
problem for a quasilinear equation. Uspekhi Matematicheskii Nauk (N.S.), 14
(1959), pp. 165–70 (in Russian).
[28] R. Pego, Phase transitions in one-dimensional nonlinear viscoelasticity:
admissibility and stability. Archive for Rational Mechanics and Analysis, 97
(1986), pp. 353–94.
[29] T.J. Pence, On the encounter of an acoustic shear pulse with a phase boundary
in an elastic material: energy and dissipation. Journal of Elasticity, 25 (1991),
pp. 31–74.
[30] T. Poston and I.N. Stewart, Catastrophe Theory and its Applications. Pitman,
London, 1978.
[31] J.R. Rice, On the structure of stress–strain relations for time dependent plastic
deformation in metals. ASME Journal of Applied Mechanics, 37 (1970),
pp. 728–37.
[32] P. Rosakis and J.K. Knowles, Unstable kinetic relations and the dynamics of
solid–solid phase transitions. Journal of the Mechanics and Physics of Solids,
45 (1997), pp. 2055–81.
82 EVOLUTION OF PHASE TRANSITIONS
[33] M. Shearer, The Riemann problem for a class of conservation laws of mixed
type. Journal of Differential Equations, 46 (1982), pp. 426–43.
[34] M. Slemrod, Admissibility criteria for propagating phase boundaries in a van
der Waals fluid. Archive for Rational Mechanics and Analysis, 81 (1983),
pp. 301–15.
[35] L.R.G. Treloar, The Physics of Rubber Elasticity, third edition. Oxford Uni-
versity Press, Oxford, 1975.
[36] L. Truskinovsky, Structure of an isothermal phase discontinuity. Soviet
Physics Doklady 30 (1985), pp. 945–8.
[37] N.A. Winfree, Impact-Induced Phase Changes in Elastic Solids: A Contin-
uum Study Including Numerical Simulations for GeO2 . Ph.D. dissertation,
California Institute of Technology, Pasadena, CA, 1999.
[38] X. Zhong, Dynamic behavior of the interface between two solid phases in an
elastic bar. Journal of Elasticity, 41 (1995), pp. 39–72.
[39] X. Zhong, T.Y. Hou, and P.G. LeFloch, Computational methods for prop-
agating phase boundaries. Journal of Computational Physics, 124 (1996),
pp. 192–216.
Part III Thermomechanical Theory
5 Multiple-Well Free Energy Potentials
5.1 Introduction
This, and the next two chapters, address the basic concepts and ingredients underly-
ing the subject of this monograph. Most of these ideas have already been introduced
within a one-dimensional setting in the preceding chapters. We now generalize that
earlier discussion in two ways: one, we turn to a three-dimensional setting and two,
we consider both mechanical and thermal effects.
The present chapter will be concerned with the energy potential. In the one-
dimensional purely mechanical setting, we discussed the local minima, or energy
wells, of the strain energy W (γ ) as a function of strain and noted that they corre-
sponded to stress-free configurations of the material. We also discussed the potential
energy, P(γ , σ ) = W (γ ) − σ γ , as a function of strain and stress, and noted that
the local minima of P(·, σ ) corresponded to stress–strain pairs related by the con-
stitutive relation σ = σ̂ (γ ) = W (γ ). In this chapter we generalize both of these
notions.
A thermoelastic material can be characterized by a suitable free energy potential.
Most of our attention in this chapter will be focused on the local minima – or
“energy wells” – of the potential, with particular interest on free energy functions
with multiple local minima. Such multiwell free energy functions can be used to
describe materials that can exist in more than one configuration at certain stress–
temperature pairs. We say that each energy well describes a distinct phase of the
material, where the term “phase” is to be interpreted broadly. For example, the dif-
ferent energy wells might represent the liquid and vapor phases of a fluid; the
austenite and martensite phases of a crystalline solid; the randomly oriented and
aligned states of a polymer; or the intact and comminuted states of glass. The
phenomenological nature of the theory allows this latitude.
In Section 5.2 we introduce the Helmholtz free energy potential for a thermoe-
lastic continuum and describe its local minima. In order to account for the effect of
stress, in Section 5.3 we introduce the notion of the potential energy function and
discuss its energy wells and relationship to the Gibbs free energy. These ideas are
85
86 EVOLUTION OF PHASE TRANSITIONS
illustrated in Section 5.4 and Section 5.5 through the respective examples of a van
der Waals fluid and a two-phase crystalline solid.
We postpone discussing thermoelasticity more broadly until Chapter 7, our
discussion here being limited to the characteristics of the two aforementioned po-
tentials.
F = RU. (5.1)
The stretch tensor U is the symmetric positive-definite square root of the right
Cauchy–Green tensor1 C:
C = U2 = FT F. (5.2)
E = 12 (C − I), (5.3)
ψ represents energy per unit mass. Moreover, the first Piola–Kirchhoff stress tensor
σ and the specific entropy per unit mass η in such a material are related to F and θ
through the constitutive relationships
F (F, θ ),
σ = ρψ θ (F, θ );
η = −ψ (5.5)
here ρ > 0 is the mass density in the reference configuration and the subscripts
indicate the gradient with respect to F and the partial derivative with
attached to ψ
respect to θ. The Helmholtz free energy ψ is related to the more familiar internal
energy (per unit mass) ε by
ψ = ε − ηθ; (5.6)
1
For a more detailed discussion of the kinematics of a continuum, see, for example, Chadwick [3].
MULTIPLE-WELL FREE ENERGY POTENTIALS 87
it is worth noting that the relationship (5.6) holds even for materials that are not
thermoelastic. For further details on thermoelastic materials, see Section 96 of
Truesdell and Noll [10].
Frame indifference requires that the value of the energy remain unchanged by a
superposed rigid rotation. This implies that ψ depends on the deformation gradient
tensor F only through the right Cauchy–Green tensor C. Thus we can write
On the other hand, material symmetry requires that the value of ψ remain
unchanged when the reference configuration is subjected to a transformation that
preserves the symmetry of the material. Thus ψ must have the further property
θ ) = ψ(FQ,
ψ(F, θ ) for all tensors F with positive determinant and all tensors Q
in the material’s symmetry group G. The group G characterizes the symmetry of
the material in the reference configuration. For example, if the material has cubic
symmetry in the reference configuration then G contains the 24 rotations that map
a cube back into itself, whereas if it is isotropic, G contains all rotations. When
combined with the implication (5.7)1 of frame indifference, material symmetry
requires that, for each symmetric tensor C,
Then one speaks of ψ(·, θo ) as having an energy well, and one refers to the tensor
Co as being at the bottom of this energy well. A homogeneous configuration of the
body with associated right Cauchy–Green tensor Co is metastable in the sense that
it corresponds to a local minimum of the free energy; it is stable if Co is a global
minimizer of ψ.
Our interest is in materials that, at some temperature θo , possess at least two
distinct, stress-free metastable equilibrium configurations that do not differ by a
rotation. If C+ and C− denote the right Cauchy–Green tensors associated with
these two configurations, then ψ(·, θo ) must have energy wells at C+ and C− . By
continuity, the function ψ(·, θ ) will continue to have two energy wells for temper-
atures close to θo . Suppose that the two energy wells have the same height at some
temperature θT : ψ(C+ , θT ) = ψ(C− , θT ); θT is referred to as the transformation
temperature. Note that both phases are stable at θ = θT . If ψ(C+ , θ ) < ψ(C− , θ )
for θ > θT , then the plus-phase is stable at temperatures above the transformation
temperature and vice versa.
88 EVOLUTION OF PHASE TRANSITIONS
C
Σ
P P
C C
θ
Figure 5.1. Schematic phase diagram for the three phases of a CuAlNi shape memory alloy. The inset
figures show sketches of the potential energy function at different points (θ, Σ) on the phase diagram.
It should be noted that even if the Helmholtz function itself has only a single
energy well, it is still possible for the potential energy function P to have multiple
energy wells. A material characterized by such a ψ exhibits a single phase when it
is unstressed, but exhibits multiple phases when suitably stressed. Whether or not
this is so depends on the convexity of ψ.
The number and character of the local minima of the potential energy P(·, θ ; Σ)
depends on (θ, Σ) and it is often useful to characterize the minima in the tempera-
ture, stress “plane”; when used for this purpose, the (θ, Σ) plane provides what is
called the phase diagram of the material. Consider, for example, an alloy of CuAlNi
at compositions where it is a shape memory material. This alloy involves an austen-
ite phase that is cubic, one martensite phase that is orthorhombic (called γ1 ), and
a second martensite phase that is monoclinic (called β1 ); in subsequent chapters
we shall have occasion to encounter all three of these phases. Depending on the
stress and temperature, a different one of these phases is energetically favorable.
Figure 5.1 shows a schematic phase diagram for this material, indicating, at each
point (θ, Σ), what the lowest-energy phase is. For example, γ1 -martensite is seen to
be energetically preferred at points below AC and to the left of AD corresponding
to (relatively) low stress and low temperature and so on.
The potential energy function P(C, θ ; Σ) is related to, though not identical
with, the Gibbs free energy function g. Since g is expressed as a function of stress
and temperature alone, in order to explore the relationship between P and g it
is necessary to eliminate the right Cauchy–Green tensor in P in favor of stress
90 EVOLUTION OF PHASE TRANSITIONS
and temperature. If the constitutive relation Σ = 2ρψC (C, θ ) for the second Piola–
Kirchhoff stress tensor Σ is invertible, then we can write C = C̃(Σ, θ ). The potential
g defined by
g(Σ, θ) = P(C̃(Σ, θ ), θ ; Σ) = ρψ(C̃(Σ, θ ), θ ) − Σ · C̃(Σ, θ ) (5.15)
represents the Gibbs free energy per unit reference volume. Observe that the value of
the potential energy at an extremum of P(·, θ ; Σ) coincides with the corresponding
value of the Gibbs free energy at that same (θ, Σ).
We now turn to two examples that illustrate the concepts outlined here and in
the previous section.
gas molecules themselves. With regard to the second assumption, the presence of
inter-molecular forces will cause an effective backwards force on a molecule as
it approaches a wall of the container, and therefore, its effect would be to reduce
the pressure in proportion to the number of interacting pairs of molecules. This
suggests that the value of p given by the ideal gas law should be decreased by a
term proportional to 1/v 2 .
These heuristic arguments suggest that a better description of a fluid is provided
by the van der Waals model
Rθ a
p = p(v, θ ) = − 2, v > b, θ > 0. (5.21)
v−b v
A more detailed discussion of the ideas leading to (5.21) can be found, for example,
in Section 20.4 of Kestin [6]. A complete characterization of a van der Waals fluid
is provided by the Helmholtz free energy function
θ a
ψ(v, θ) = −cv θ log − Rθ log (v − b) − , v > b, θ > 0, (5.22)
θo v
√
where cv , a, b, R, θo , and vo are positive constants and v = vo J = vo det C is the
specific volume; see equations (3.49) and (3.50) of Callen [2] from which (5.22) can
be derived. Expressions for other physical quantities can be derived from (5.22).
For example (5.17) can be used to calculate the pressure, leading back to (5.21),
while the specific internal energy can be determined using
= ψ − θ ∂ψ/∂θ , which
gives
a
= cv θ − . (5.23)
v
Numerical values of the parameters a, b, and cv for which the van der Waals model
reasonably describes a variety of fluids have been reported in the literature; see for
example Table 3.1 of Callen [2].
In order to examine the ability of the van der Waals model to model multiple
phases, we consider the potential energy function
θ a
P(v, θ; p) = ψ(v, θ ) + pv = −cv θ log − Rθ log (v − b) − + pv,
θo v
(5.24)
which is defined for all v > b, θ > 0, and −∞ < p < ∞. For certain values of
temperature θ and pressure p one can readily show that the function P(·, θ ; p) has
two local minima corresponding to two phases of the fluid. For other values of
θ and p it has a single minimum corresponding to one phase. Analysis of (5.24)
shows that the detailed character of P(·, θ ; p) is as depicted in Figure 5.2: P has
two local minima when the pressure and temperature correspond to a point in the
shaded region enclosed by the curves p = p M (θ ), p = pm (θ ), p = 0, and θ = 0.
Explicit expressions for the functions pm (θ ) and p M (θ ) can be determined but
92 EVOLUTION OF PHASE TRANSITIONS
P
liquid
P
P
vapor
Figure 5.2. Phase diagram for a van der Waals fluid on the temperature, pressure plane. The line θ = θcr
and the curves p = pM (θ ) and p = pm(θ ) divide this plane into three distinct regions. The number of local
minima of the potential energy function P(·, θ ; p) at points corresponding to each of these subregions
is shown in the inset figures. The local minima of P correspond to the liquid (filled circles) and vapor
(open circles) phases of the fluid.
are not displayed; the two temperatures θcr and θ∗ shown in the figure have the
values
8 a 27
θcr = , θ∗ = θcr . (5.25)
27 b R 32
Of the two local minima, the one at the smaller value of v corresponds to the
more dense phase, and it is natural therefore to refer to it as the liquid phase; the
other minimum describes the vapor phase. When the temperature and pressure are
associated with either {(θ, p) : 0 < θ < θcr , p > p M (θ )} or {(θ, p) : θ∗ < θ <
θcr , p < pm (θ )} the potential energy has a single minimum; in the former case this
corresponds to the liquid phase, while in the latter it is associated with the vapor
phase.
Consider for instance a van der Waals fluid at some fixed temperature in the
interval θ∗ < θ < θcr . If the pressure is gradually decreased from an initial value
that exceeds p M (θ ), we see from Figure 5.2 that the fluid is first in its liquid phase
and eventually in its vapor phase. This does not tell us the value of pressure at which
the fluid changes phase. All we do know in this regard is that on the intermediate
interval of pressure, pm (θ ) < p < p M (θ ), both phases are possible.
Next we examine the stability of each phase associated with a point in the
two-phase portion of the θ, p-plane. To do this, we need to determine which of
the two local minima of the potential energy P is the global minimizer. Analysis
of (5.24) shows that there is a pressure level p0 (θ ), intermediate to p M (θ ) and
pm (θ), such that the local minimum at the smaller value of v, that is, the local
minimum associated with the liquid phase, is the global minimum for p > p0 (θ ).
The vapor phase is stable for p < p0 (θ ). The pressure p0 (θ ), at which both phases
MULTIPLE-WELL FREE ENERGY POTENTIALS 93
p = pM (θ)
liquid
vapor
liquid vapor
P
P
p = p0(θ)
P
p = pm(θ)
Figure 5.3. The stable phases of a van der Waals fluid on the temperature, pressure plane. The filled
and open circles correspond respectively to the liquid and vapor phases.
are stable, is known as the Maxwell pressure. This is depicted in Figure 5.3 in
which the vapor phase is stable on the shaded region and the liquid phase is stable
on the unshaded region. Again, suppose that the temperature is fixed in the interval
θ∗ < θ < θcr and that the pressure is gradually decreased from an initial value,
which exceeds p M (θ ). Suppose further that the fluid assumes the liquid or vapor
phase according to which of these phases is the stable one. Then, the fluid begins in
the liquid phase and transforms to the vapor phase when the pressure falls below the
value p0 (θ).
Finally it is worth describing these observations with reference to the isothermal
pressure–volume curve. In contrast to an ideal gas, the p, v-curve for a van der Waals
fluid does not always decline monotonically. If the temperature lies in the range
θ > θcr , one finds that p is in fact a monotonically decreasing function of v just as
for an ideal gas; see Figure 5.4. However for temperatures in the range 0 < θ < θcr ,
one finds that p decreases with increasing v on two intervals, b < v < vm (θ ) and
v > v M (θ), but increases on the intermediate interval vm (θ ) < v < v M (θ ). Here
vm (θ) and v M (θ ) are given by the two roots (> b) of the cubic equation
4 θ
v 3 = b(v − b)2 . (5.26)
27 θcr
The corresponding values of pressure, denoted by pm (θ ) and p M (θ ) in Figure
5.4, are given by pm (θ ) = p(vm (θ ), θ ) and p M (θ ) = p(v M (θ ), θ ) where p(v, θ) is
given by (5.21). Graphs of the functions p M (θ ) and pm (θ ) were shown previously in
Figure 5.2.
Next, observe from (5.24) that if P(·, θ ; p) has an extremum at some v, then the
triplet {v, θ, p} satisfies the pressure–volume–temperature constitutive relation p =
p(v, θ) = −∂ψ(v, θ )/∂v. Moreover, since ∂ 2 P(v, θ ; p)/∂v 2 = −∂ p(v, θ )/∂v, this
94 EVOLUTION OF PHASE TRANSITIONS
liquid
vapor
Figure 5.4. Graphs of pressure versus specific volume at three different fixed values of temperature
for a van der Waals fluid. The declining branches correspond to the liquid and vapor phases of the
fluid.
extremum is a local minimum if and only if ∂ p(v, θ )/∂v < 0. Consequently each
declining branch of the isothermal p, v-curve can be associated with a local min-
imum of P and therefore corresponds to a metastable state of the fluid; the ris-
ing branch is unstable. For the van der Waals fluid, the two declining branches
can therefore be identified with the liquid and vapor phases, the branch to the
left (associated with the smaller values of v) being associated with the liquid
phase.
Finally, the Maxwell pressure p0 (θ ) can be shown to cut off equal areas of
the p, v-curve as shown in Figure 5.5. Since we know that the liquid phase is
stable for p > p0 (θ ) and the vapor phase is stable for p < p0 (θ ), the stable phases
correspond to the bold portions of the p, v-curve in Figure 5.5; the metastable
phases are associated with the light portion.
liquid
vapor
pM (θ)
p0(θ)
Figure 5.5. Graph of the pressure p versus specific volume v for a van der Waals fluid at a fixed
temperature in the range 0 < θ < θcr . The bold portions of the curve correspond to the stable phase
while the light portion is associated with the metastable phase. The dashed portion of the curve is
unstable. The liquid phase is stable above the Maxwell pressure p0 (θ ) and metastable below it; the
vapor phase is stable below the Maxwell pressure and metastable above it. The Maxwell pressure p0 (θ )
has the property that the two shaded regions in the figure have equal area.
MULTIPLE-WELL FREE ENERGY POTENTIALS 95
to a second phase of the material, martensite. Since the two lattices are different,
{ea1 , ea2 , ea3 } = {em
1 , e2 , e3 }. Thus as the temperature of such a material is changed,
m m
the lattice vectors first vary slowly and continuously due to thermal expansion, and
then suddenly, at the transformation temperature θT , they change discontinuously.
Since the change in the lattice due to thermal expansion is much smaller than
that due to phase transformation (at least in the range of temperatures we shall be
concerned with), we will neglect thermal expansion. Both sets of lattice vectors are
therefore assumed to be independent of temperature in this chapter.
Examples of such materials include the alloys gold–cadmium (cubic and or-
thorhombic phases), nickel–titanium (cubic and monoclinic phases), and copper–
aluminum–nickel (cubic, orthorhombic, and monoclinic phases). One of the sim-
plest examples of a two-phase martensitic material is one that occurs in face-
centered cubic and face-centered tetragonal phases, For example, indium–thallium,
manganese–nickel, and manganese–copper. Figure 5.6 depicts the lattices associ-
ated with such a material; for more details, see, for example Bhattacharya [1].
Let {c1 , c2 , c3 } denote fixed unit vectors in the cubic directions and let the
lattice vectors associated with the austenite lattice be denoted by {ea1 , ea2 , ea3 }. If
ao × ao × ao denotes the lattice parameters of the cubic phase then the lattice
vectors shown in Figure 5.6(a) are
ao ao ao
ea1 = (c2 + c3 ) , ea2 = (−c2 + c3 ) , ea3 = (c1 + c3 ) .(5.27)
2 2 2
Similarly for the tetragonal phase, if the lattice parameters are denoted by a × a ×
1 , e2 , e3 } can be taken to be
c, the lattice vectors {em m m
a a c a
1 =
em (c2 + c3 ) , 2 =
em (−c2 + c3 ) , 3 =
em c1 + c3 . (5.28)
2 2 2 2
The two sets of lattice vectors {ea1 , ea2 , ea3 } and {em
1 , e2 , e3 } are each linearly
m m
ao
ea3
(a)
em
2 em
1 a
a
Consequently we may regard the martensite lattice as the image of the austenite
lattice under the linear transformation U1 . A characteristic of martensitic trans-
formations is that the change of crystal structure is diffusionless, that is, there
is no rearrangement of atoms associated with the lattice change. Consequently,
we can describe the kinematics of the phase transformation completely by the
linear transformation U1 . The tensor U1 is called the Bain stretch or transfor-
mation stretch; observe that knowledge of the austenitic and martensitic lattices
determines U1 .
One can readily verify from (5.27) and (5.28) that in the present example,
U1 = η3 c1 ⊗ c1 + η1 c2 ⊗ c2 + η1 c3 ⊗ c3 (5.30)
associated with this second configuration. Let F denote the unique nonsingular
tensor that relates these two sets of lattice vectors:
ek = Feak , k = 1, 2, 3. (5.32)
Then, according to the Cauchy–Born rule, the gradient of the deformation relating
the two continuum configurations is also F. Therefore in particular, if the refer-
ence configuration is subjected to a homogeneous deformation with deformation
gradient tensor U1 , we obtain, according to the preceding discussion, a second
stress-free configuration that corresponds to the tetragonal phase (martensite). The
deformation gradient tensors F = I and F = U1 therefore each describes a stress-
free equilibrium configuration.
In the model we develop here, we express the Helmholtz free energy ψ as a
function of the Green strain E: ψ = ψ(E, θ ) where E = (1/2)(FT F − I). Since both
phases exist at the transformation temperature θT , ψ(·, θT ) must have local minima
at E = 0 and E = E1 = (1/2)(U21 − I). By continuity, ψ(·, θ ) will continue to have
local minima at these same values2 of E for temperatures close to the transformation
temperature.
Since the austenite phase is preferred over the martensite phase for temperatures
greater than θT , this implies that the cubic phase is the low-energy phase for θ > θT ,
that is, that ψ(0, θ ) < ψ(E1 , θ ) for θ > θT . Similarly, since the martensite phase
is favored for temperatures less than θT we must also have ψ(E1 , θ ) < ψ(0, θ ) for
θ > θT . By continuity, both phases must have equal energy at the transformation
temperature: ψ(E1 , θT ) = ψ(0, θT ).
We now construct an explicit free energy function ψ(E, θ ), which possesses
the properties outlined above. Since the reference configuration corresponds to
an unstressed state of austenite, the free energy function ψ(E, θ ) must possess
cubic symmetry. It is known (see, e.g., Smith and Rivlin [8], and Green and
Adkins [5]) that, to have cubic symmetry, ψ depends on E only through the “cubic
invariants”
I1 = E 11 + E 22 + E 33 ,
I2 = E 11 E 22 + E 22 E 33 + E 33 E 11 ,
I3 = E 11 E 22 E 33 ,
I4 = E 12 E 23 E 31 ,
I5 = E 12
2
+ E 23
2
+ E 31
2
, (5.33)
I6 = (E 11 + E 22 )E 12
2
+ (E 22 + E 33 )E 23 2
+ (E 33 + 2
E 11 )E 31 ,
I7 = E 12 E 23 + E 23 E 31 + E 31 E 12 ,
2 2 2 2 2 2
I8 = 2
E 11 E 12 2
E 23 + E 22 E 23
2 2
E 31 + E 33 E 31
2 2
E 12 ,
I9 = 2
E 11 E 22 E 12 + E 22 E 33 E 23
2
+ E 33 E 11 E 31
2
,
2
We are neglecting the effect of thermal expansion here.
98 EVOLUTION OF PHASE TRANSITIONS
3
Note that the constraint I1 = 0 implies that tr E1 = 0 and therefore that the lattice stretches η1 and
η3 must be related by 2η12 + η32 = 3.
MULTIPLE-WELL FREE ENERGY POTENTIALS 99
where the prime denotes differentiation with respect to the argument and θT is the
transformation temperature.
In summary, by collecting all of the preceding requirements and introducing a
more convenient pair of coefficients d(θ ) and e(θ ), the free energy function ψ can
be expressed as
I3 I2 I2 2 I2 I3
ψ = co (θ ) + d(θ ) 3 − 2 + e(θ ) −3 2 +9 3 ,
p p p2 p p (5.39)
e(θ ) > 0, 3e(θ ) > d(θ ) > −3e(θ), d (θ ) > 0, d(θT ) = 0,
where the invariants I2 and I3 are given by (5.33). Figure 5.7 shows a contour
plot of the free energy (5.39). In view of the constraints (5.34) there are only two
independent strain components, which we have taken to be E 11 and E 22 . The figure
shows the contours of ψ(E 11 , E 22 , θ ), as given by (5.39), on the E 11 , E 22 -plane. It
shows the austenitic energy well at the origin E = 0 surrounded by three martensitic
energy wells at E = E1 , E2 , and E3 . The values of the material parameters associated
with this figure were chosen solely on the basis of obtaining a fairly clear contour
plot.
Although we only required ψ to have two local minima, one at E = 0 and the
other at E = E1 , we see from Figure 5.7 that, in fact, it has four local minima. In
order to understand this, observe from (5.8) that necessarily
where Ga is the symmetry group of the reference configuration, that is, of unstressed
austenite. Since E1 yields the global minimum of the free energy for θ < θT , it
100 EVOLUTION OF PHASE TRANSITIONS
η1 = a/ao
η2 = a/ao
η3 = c/ao
Figure 5.8. Cubic austenite and three variants of tetragonal martensite. Observe that one cannot rigidly
rotate a martensite variant in such a way as to make the atoms P, Q, R, S coincide with the locations
of these same atoms in any other martensite variant.
follows from (5.40) that the tensor QT E1 Q is also a minimizer for each Q ∈ Ga . As
Q varies over the set of all tensors in Ga , the tensor QT E1 Q takes on a certain finite
number of distinct, symmetric tensor values, which we denote by E1 , E2 , . . . , E N ,
and ψ has local minima at each of these Ek s. These stretch tensors are said to
describe the N variants of martensite. Note that because of (5.40), necessarily, all
variants have the same energy:
ψ(E1 , θ ) = ψ(E2 , θ ) = · · · = ψ(E N , θ ) . (5.41)
The number N depends on the difference in material symmetry between austenite
and martensite and in general is larger the greater this difference. When the austen-
ite phase is cubic, the number of martensitic variants is, respectively, 3, 6, or 12
depending on whether the martensite is tetragonal, orthorhombic, or monoclinic.
Figure 5.8 illustrates the three variants of tetragonal martensite. Observe that there
are three possible ways in which to stretch the cubic lattice in order to obtain the
tetragonal lattice, U1 being the one where the unequal stretching occurs in the c1
direction. More generally, one can stretch the cubic lattice by the ratio η3 in the ck
direction and by η1 in the remaining two cubic directions; this leads to the three
Bain stretch tensors
Uk = η1 I + [η3 − η1 ]ck ⊗ ck , k = 1, 2, 3, (5.42)
MULTIPLE-WELL FREE ENERGY POTENTIALS 101
−1 −0.5 0.5 1
−1
Figure 5.9. A one-dimensional cross-section of the energy (5.39) along the path (5.44) in strain space.
The figure plots w(ζ ) = ψ(E(ζ ), θ ) versus strain ζ at three different temperatures. Because we have omit-
ted the effect of thermal expansion, the locations of the local minima do not change with temperature.
The material parameters underlying this plot are the same as those associated with Figure 5.7.
Observe that E(0) = 0, E(−1) = E2 , E(1) = E3 so that this path passes through
the austenite well and two of the martensite wells. Figure 5.9 shows the variation of
energy along this path at three different temperatures: the figure displays w(ζ ) =
ψ(E(ζ ), θ ) versus ζ . The three graphs correspond to three temperatures, greater
than, equal to, and less than the transformation temperature θT . In drawing this
figure, we took the coefficient c3 (θ ) in (5.35) to depend linearly on temperature
and c22 (θ) to be independent of temperature; the specific values of the various
parameters were chosen solely on the basis of obtaining a clear figure.
Due to the restrictive nature of the kinematic constraints (5.34), the free en-
ergy function (5.35) cannot provide a quantitatively accurate model, especially if
the temperature departs from the transformation temperature. The natural gener-
alization of (5.35) is therefore to relax these constraints in the usual way by using
102 EVOLUTION OF PHASE TRANSITIONS
Next, the height h between the two energy wells, h(θ ) = ψ(E1 , θ ) − ψ(0, θ ),
can be calculated from (5.45) to be
h(θ) = 1
54
[−9( p − q)2 d2 (θ ) − ( p − q)3 d3 (θ ) + 27(2 p + q)2 d6 (θ )]. (5.53)
At the transformation temperature θT , we must have h(θT ) = 0 since the wells then
have equal height. The latent heat associated with the transformation from the cubic
phase to the tetragonal phase is
λ = − ψθ − ψθ θ (5.54)
E=E1 E=0
and consequently λ(θ ) = −θ h (θ ). Integrating this and using the fact that h(θT ) = 0
gives
θ
λ(θ )
h(θ ) = − dθ. (5.55)
θT θ
It follows from this and (5.53) that
θ
54 λ(θ ) 9 (2 p + q)2
d3 (θ) = dθ − d 2 (θ ) + 27 d6 (θ ). (5.56)
( p − q)3 θT θ p−q ( p − q)3
In summary, knowledge of the lattice parameters η1 and η3 , the mass density ρ,
(a) (a) (a)
the elastic moduli C11 , C12 , C44 of the cubic phase, and the latent heat λ determines
the six coefficients d2 to d7 through (5.51), (5.47), and (5.56). James (see Klouček
and Luskin [7]) has shown that the response predicted by this generalized form of
ψ is in reasonable agreement with the observed behavior of In–Tl.
A free energy function for a material with cubic austenite and orthorhombic
martensite has been constructed similarly by Vedantam and Abeyaratne [11] and
used to characterize a CuAlNi shape memory alloy.
By constructing and analyzing the potential energy function associated with
(5.45), one can now determine the stability of the different phases at various stresses
and temperatures, and thus construct the corresponding phase diagram in (θ, Σ)-
space.
REFERENCES
[5] A.E. Green and J.E. Adkins, Large Elastic Deformations. Clarendon Press,
Oxford, 1970.
[6] J. Kestin, A Course in Thermodynamics, Volume II. McGraw-Hill, New York,
1979.
[7] P. Klouček and M. Luskin, The computation of the dynamics of the marten-
sitic transformation. Continuum Mechanics and Thermodynamics, 6 (1994),
pp. 209–40.
[8] G.F. Smith and R.S. Rivlin, The strain energy function for anisotropic elastic
materials. Transactions of the American Mathematical Society, 88 (1958),
pp. 175–93.
[9] T.C.T. Ting, Anisotropic Elasticity: Theory and Applications. Oxford Univer-
sity Press, Oxford, 1996.
[10] C. Truesdell and W. Noll, The nonlinear field theories of mechanics, in Hand-
buch der Physik, edited by S. Flügge, Volume III/3. Springer, Berlin, 1965.
[11] S. Vedantam and R. Abeyaratne, A Helmholtz free-energy function for a Cu–
Al–Ni shape memory alloy. International Journal of Nonlinear Mechanics,
40 (2005), pp. 177–93.
6 The Continuum Theory of Driving Force
6.1 Introduction
In the present chapter, we consider a continuum in which there are three-
dimensional, thermomechanical fields involving moving surfaces of discontinuity
in strain, particle velocity, and temperature. Our objective is to set out the theory of
driving force acting on such surfaces without specifying any particular constitutive
law. The theory should be applicable to physical settings ranging from the qua-
sistatic response of solids under slow thermal or mechanical loading in which heat
conduction is present, to fast adiabatic processes in which temperature need not
be continuous and inertia must be taken into account. The theory must be general
enough to accommodate such disparate settings.
We begin by stating the fundamental balance laws for momentum and energy
in their global form. These laws are then localized where the fields are smooth,
leading to the basic field equations of the theory. Localization of the global laws at
points where the thermomechanical fields suffer jump discontinuities provides the
jump conditions appropriate to such discontinuities.
Without making constitutive assumptions, we then introduce the notion of driv-
ing force1 . The driving force arises through consideration of the entropy production
rate associated with the thermomechanical fields under study; it leads to a succinct
statement of the implication of the second law of thermodynamics for moving
surfaces of discontinuity.
The theory is developed for three space dimensions in Lagrangian, or material,
form, according to which one follows the evolution of fields attached to a given
particle of the continuum. The notion of driving force in the continuum theory of
phase transformations was developed independently by Truskinovsky [11], Heidug
and Lehner [6], and Abeyaratne and Knowles [1, 2]; see also Knowles [7]. The
description below is based on [1, 2]. For a discussion of driving force in the materials
science literature, see, for example, Christian [4] or Porter and Easterling [8].
1
In earlier publications, we have used the term “driving traction” for the driving force, since it has
the units of force per unit area. We now prefer the term “driving force” so that our terminology
conforms to that in materials science.
105
106 EVOLUTION OF PHASE TRANSITIONS
d
ŷ × (σn) d A + ŷ × ρb d V = ŷ × ρv d V, (6.2)
∂D D dt D
for every bounded subregion D of R and for all times. Here n is the unit outward
normal vector on ∂D.
The thermomechanical process also involves a heat flux vector q(x, t) (measured
per unit referential surface area), a heat supply r (x, t), and an internal energy (x, t),
both measured per unit mass. Unconventionally, we take the heat flux vector q to be
such that q · n > 0 if heat flows into a body across its bounding surface on which
n is the outward unit normal vector. Moreover, q(·, t) is assumed to be piecewise
continuous with piecewise continuous gradient on R, while r (·, t) is continuous
there; (x, t) is continuous in (x, t) with piecewise continuous first and second
derivatives. The first law of thermodynamics then requires that
σn · v d A + ρb · v d V + q·n dA + ρr d V
∂D D ∂D D (6.3)
d
= (ρ + ρv · v/2) d V.
dt D
6.2.2 Localization of the balance laws. At any point x in R where the fields enter-
ing (6.1)–(6.3) are smooth, localization yields the field equations associated with
THE CONTINUUM THEORY OF DRIVING FORCE 107
Suppose that the surface of discontinuity St intersects D for some time interval of
interest, and that the fields are continuous elsewhere in D. An alternate represen-
tation for D may be obtained from (6.15) with the help of a standard calculation
that uses the divergence theorem and accounts for the discontinuity on St when
carrying out the indicated differentiation; see, for example, p. 116 of [3]. The result
may be written in the form
jump
D = Dbulk + D , (6.16)
where the contribution Dbulk due to particles in the bulk of D is defined by
D =
bulk
{ρ η̇ − Div(q/θ ) − ρr/θ } d V, (6.17)
D
jump
while the entropy production rate D due to the portion of the surface of discon-
tinuity that lies in D is
jump
D = − {ρ[[η]] Vn + [[q · n/θ]]} d A. (6.18)
St ∩D
By expanding the divergence in (6.17) and using the local energy balance field
equation (6.6), one can find the following alternate form for Dbulk :
where
ρ 1
Dloc = σ · Ḟ + θ η̇ − ˙ dV (6.20)
D θ ρ
and
1
Dcon = q · ∇θ d V. (6.21)
D θ2
ψ = − θ η. (6.22)
jump
so that the contribution D of (6.18) to the rate of entropy production may be
rewritten as
jump [[ρ]] − σ · [[F]] − θ[[ρη]]
D = Vn d A. (6.26)
St ∩D θ
In terms of the Helmholtz free energy ψ, (6.26) becomes
jump [[ρψ]] − σ · [[F]] + ρη[[θ ]]
D = Vn d A. (6.27)
St ∩D θ
6.3.2 Driving force and the second law. Motivated by (6.26) and (6.27), we define
the driving force f (x, t) acting at a point on the surface of discontinuity St by either
of the following equivalent formulas:
f = [[ρψ]] − σ · [[F]] + ρη[[θ ]] = [[ρ]] − σ · [[F]] − θ [[ρη]], (6.28)
jump
[1, 2]. By (6.27) and (6.28), the contribution D to the entropy production rate
for D may then be written as
jump f
D = Vn d A. (6.29)
D∩St θ
In its global form, the second law of thermodynamics requires the Clausius–
Duhem inequality to hold:
D (t) ≥ 0 (6.30)
for all times and for all bounded subregions D of R. By first localizing (6.30) at a
point x at which the fields are smooth, one finds that
ρ η̇ − Div(q/θ ) − ρr/θ ≥ 0 (entropy inequality in bulk) (6.31)
at all such points of smoothness. It then follows that
Dbulk ≥ 0 (6.32)
for every bounded D and for all times.
On the other hand, localizing (6.30) at a point on the surface St and using (6.16)
and (6.17) leads to the additional requirement
f Vn ≥ 0 (entropy inequality on surface of discontinuity) (6.33)
at all points on St . Thus according to the second law, the sign of Vn , and therefore
the direction of advance of the surface St , is determined by the sign of the driving
force f acting on St . A result precisely analogous to (6.33) was found to hold true
in the purely mechanical theory of Chapter 4; see (4.10).
Suppose the thermomechanical process is adiabatic, so that q = 0, r = 0. Then
by (6.13), one has
( [[ρ]] − σ · [[F]] ) Vn = 0, (6.34)
THE CONTINUUM THEORY OF DRIVING FORCE 111
6.3.3 Driving force in the case of mechanical equilibrium. Suppose that mechan-
ical equilibrium prevails, in the sense that the right-hand sides of the two conse-
quences (6.4) and (6.10) of momentum balance are replaced by null vectors, and
assume that the temperature is continuous everywhere. Equation (6.10) now asserts
that traction is continuous across the surface St of strain discontinuity: [[σn]] = 0.
With the help of the rank-one connection (6.9), one then finds that
σ · [[F]] = σn · a = (σ ± n) · a = σ ± · [[F]]. (6.38)
This result, together with continuity of temperature, may then be used to specialize
the representation (6.28)1 of the driving force acting on a strain discontinuity to the
case of mechanical equilibrium, yielding
(driving force –
f = [[ρψ]] − (σ ± n) · a = [[ρψ]] − σ ± · [[F]]
mechanical equilibrium).
(6.39)
It may be remarked that the version (6.39) of driving force appropriate to the
case of mechanical equilibrium is not in general simply related to the “resolved”
Piola–Kirchhoff shear stress (I − n ⊗ n)σn. Moreover, when the contributions due
to inertia are retained in (6.4) and (6.10), the notion of resolved shear stress on St
is itself ambiguous, since the nominal traction σn need no longer be continuous
across the discontinuity, so that resolved shear stress will in general be different on
the two sides of St .
REFERENCES
[1] R. Abeyaratne and J.K. Knowles, On the driving traction acting on a surface
of strain discontinuity in a continuum. Journal of the Mechanics and Physics
of Solids, 38 (1990), pp. 345–60.
[2] R Abeyaratne and J.K. Knowles, A note on the driving traction acting on a
propagating interface: adiabatic and non-adiabatic processes of a continuum.
ASME Journal of Applied Mechanics, 67 (2000), pp. 829–31.
[3] P. Chadwick, Continuum Mechanics. Dover, New York, 1999.
112 EVOLUTION OF PHASE TRANSITIONS
7.1 Introduction
The developments in the preceding chapter assumed nothing about the constitutive
response of the continuum; we now restrict attention to a special class of constitu-
tive laws – the so-called thermoelastic materials introduced previously in Chapter 5.
Our discussion in Chapter 5 was focused entirely on the energy wells of the char-
acterizing energy potential. Here we discuss thermoelastic materials and nonlinear
thermoelasticity in more detail.
In Section 7.2 we state the constitutive law of nonlinear thermoelasticity, in
which stress and specific entropy are specified as functions of deformation gra-
dient and absolute temperature through the Helmholtz free energy potential. An
equivalent alternate form of the constitutive law, in which stress and temperature
are given in terms of deformation gradient and specific entropy by means of the
internal energy potential, is also discussed. The expression for the driving force
is then specialized to this setting. Next we state the heat conduction law, and in
Section 7.2.3, we write out the full theory in the form of four scalar partial differ-
ential equations involving the three components of displacement and temperature.
The accompanying jump conditions are also laid out. In the final subsection we
specialize the results to a state of thermomechanical equilibrium.
Once the full theory of thermoelasticity has been assembled, in Section 7.3 we
summarize some results concerning the notion of material stability based on an
analysis, given elsewhere, of the linearized dynamic stability of a body that has
been homogeneously deformed from a given reference configuration at a given
temperature to a new configuration at a new temperature.
Finally, we specialize the theory to the one-dimensional kinematics of uniaxial
strain in Section 7.4.
113
114 EVOLUTION OF PHASE TRANSITIONS
is a Helmholtz free energy potential ψ̂(F, θ ) such that the Helmholtz free energy
field ψ(x, t) of the preceding chapter is given by ψ̂(F(x, t), θ (x, t)), and the nominal
stress tensor and the specific entropy are related to F and θ by
where, as before, ρ is the referential mass density, and the subscripts attached to
ψ̂ indicate the gradient with respect to F and the partial derivative with respect
to θ .
As discussed in Chapter 5, at each fixed temperature θ , the value ψ̂(F, θ ) of the
Helmholtz potential should be invariant under changes of observer, a requirement
that is sometimes called the principle of objectivity or of material frame indifference.
This leads to the conclusion that ψ̂(F, θ ) must depend on F only through the
stretch tensor U or the so-called right Cauchy–Green tensor C = FT F = U2 . The
proof of this result for thermoelasticity is identical with that for pure elasticity;
see Chapter IX, including Exercise 1, of [3]. It is easy to show that this special
dependence of ψ̂(F, θ ) on F implies that the local balance of angular momentum
(6.5) is automatically satisfied for a thermoelastic material, and therefore that the
true stress tensor τ of (6.14) is automatically symmetric: τ = τ T . Since we shall
always assume that ψ̂(F, θ ) is invariant under changes of observer, we shall omit
(6.5) from the list of field equations for thermoelastic materials.
If at each temperature the material is isotropic in the reference state, then there
is a further simplification in the dependence of ψ̂(F, θ ) on F:
We shall always assume that c(F, θ ) is positive. Since the absolute temperature
is also positive, (7.4)2 implies η̂(F, θ ) is an increasing function of θ at fixed F.
It follows that, for each fixed F, (7.1) can be inverted to express the temperature
θ = θ̂ (F, η) in terms of F and η. As a result, one may define an internal energy
potential through
(F,
ˆ η) = ψ̂(F, θ̂ (F, η)) + ηθ̂ (F, η); (7.5)
cf. (6.22). It follows from (7.5) that the constitutive law (7.1) can be written in the
following alternate form:
thus the two potentials ψ̂ and ˆ are Legendre transforms of one another.
THERMOELASTIC MATERIALS 115
Using (7.5) and (7.6) in the local version (6.6) of the first law leads to the
following alternate local energy equation for thermoelastic materials:
Div q + ρr = ρθ η̇. (7.7)
Thus in the thermoelastic case, the basic local balance laws for linear momentum
and energy may be written in the forms (6.4) and (7.7), respectively.
For a thermoelastic material, it is readily shown that the integrand in (6.23)
vanishes identically so that, by (6.19), the various contributions to the entropy
production rate satisfy
Dloc = 0, Dbulk = Dcon . (7.8)
It follows from (6.16) that
jump
D = Dcon + D . (7.9)
Thus thermoelastic materials produce entropy only through heat conduction and the
propagation of surfaces of discontinuity. Indeed, if the thermomechanical process
under study is adiabatic, so that Dcon = 0, entropy production arises solely through
moving discontinuities such as St .
Using (7.8) and (6.21) in (6.32) and localizing at a point of smoothness yields
q · ∇θ ≥ 0. (7.10)
When specialized to a thermoelastic material, the two equivalent representations
in (6.28) for the driving force f on St yield the following striking formulas:
f /ρ = [[ψ̂]] − ψ̂F · [[F]] − ψ̂θ [[θ ]]
(7.11)
= [[ˆ ]] − ˆF · [[F]] − ˆη [[η]].
If the thermomechanical process is adiabatic, then (6.13) with q = 0 shows that
(7.11)2 reduces to
f /ρ = −ˆη [[η]] = −θ [[η]] (adiabatic); (7.12)
cf. (6.35). For nonadiabatic processes in a thermoelastic material,
f /ρ = [[ψ̂]] − ψ̂F · [[F]] (nonadiabatic). (7.13)
Suppose that the thermoelastic body undergoes an isothermal homogeneous
motion with gradient F(t) at time t, for t1 ≤ t ≤ t2 . Let F(t1 ) = F+ , F(t2 ) = F− .
Consider an arbitrary sub-body and let D be the region it occupies in the reference
configuration. Integration of (7.7) over D, followed by integration of the result with
respect to t over (t1 , t2 ), leads to the assertion that the total heat released by the
sub-body during this isothermal process is given by λM where M is the mass of
the sub-body, and the heat released per unit mass λ is given by
λ = −θ η(F− , θ ) − η(F+ , θ ) . (7.14)
116 EVOLUTION OF PHASE TRANSITIONS
Equation (7.14) has an obvious local interpretation for isothermal processes that
involve nonhomogeneous motions. If F+ and F− correspond to two phases of the
material then λ is the latent heat per unit mass associated with the phase transfor-
mation (F+ , θ ) → (F− , θ ).
7.2.2 The heat conduction law. For a thermoelastic material we assume that the
heat conduction law has the form
σi j = ρ ψ̂ Fi j (F , θ ), η = −ψ̂θ (F , θ ), (7.17)
qi = K i j (F, θ )θ, j . (7.18)
We use the symbol A to stand for the matrix of components of the 2-tensor A in the
given frame; the summation convention for repeated subscripts such as i, j, k, ... is
in use, and a comma preceding a subscript indicates differentiation with respect to
the corresponding x-coordinate.
The local forms (6.4) and (7.7) of the balance laws yield
Using (7.17) and (7.18) in (7.19) and (7.20) and recalling that F =
Grad ŷ(x, t) = 1 + Grad u(x, t) leads to the fundamental system of partial differen-
tial equations for the three components u i (x, t) of displacement and the temperature
θ(x, t):
Respectively, ci jkl and m i j are the components of the 4-tensor C and the 2-tensor
M defined by
C = C(F, θ ) = ρ ψ̂FF (F, θ ), M = M(F, θ ) = ρ ψ̂Fθ (F, θ ); (7.23)
C(F, θ ) and M(F, θ ) are respectively called the elasticity tensor and the stress–
temperature tensor.
In component form, the kinematic jump conditions (6.8) are
[[Fi j ]]l j = 0, [[Fi j ]]n j Vn + [[vi ]] = 0; (7.24)
the associated rank-1 connection relating F± is
Fi+j − Fi−j = ai n j . (7.25)
The momentum and energy jump conditions (6.10) and (6.11) are
[[σi j ]]n j + ρ[[vi ]]Vn = 0, (7.26)
[[σi j vi ]]n j + ρ[[ + (1/2)v j v j ]]Vn + [[q j ]]n j = 0. (7.27)
The component form of the alternate version (6.13) of the energy jump condition
is
ρ[[]]Vn − σi j [[Fi j ]]Vn + [[q j ]]n j = 0, (7.28)
and finally, the representations (7.11) for the driving force may be written as
f /ρ = [[ψ̂]] − ψ̂ Fi j [[Fi j ]] − ψ̂θ [[θ ]]
(7.29)
= [[ˆ ]] − ˆ Fi j [[Fi j ]] − ˆη [[η]].
7.2.4 Thermomechanical equilibrium. Consider the case in which the body force b
and the heat supply r vanish, and suppose that (i) the fields u and θ are independent
of time and (ii) the temperature is continuous. Then v = 0, and by the thermoelastic
constitutive law (7.1) and the heat conduction law (7.15), all fields are independent
of time, so that the field equations (6.4) and (6.6) for linear momentum and energy
become
Div σ = 0 or σi j, j = 0, (7.30)
Div q = 0 or q j, j = 0. (7.31)
The jump conditions (6.10) and (6.11) specialize to
[[σ]]n = 0 or [[σi j ]]n j = 0, (7.32)
[[q]] · n = 0 or [[q j ]]n j = 0. (7.33)
Clearly mechanical equilibrium holds at present, so we may make use of (6.38)
to note that
ψ̂F · [[F]] = ψ̂F± · [[F]]. (7.34)
118 EVOLUTION OF PHASE TRANSITIONS
Let P̂(F, θ ; σ) be the potential energy per unit volume; see (5.10):
P̂(F, θ ; σ) = ρ ψ̂(F, θ ) − σ · F. (7.36)
For the case of thermomechanical equilibrium under discussion, it is clear from
(7.35) and (7.36) that the driving force coincides with the jump in potential energy
per unit volume, in the sense that
f = [[ P̂(F, θ ; ρ ψ̂F± (F, θ ))]]. (7.37)
where ci jkl (F0 , θ 0 ) are the components of the elasticity tensor at the base state (see
(7.23)1 ), and ρ is the referential mass density.
The isentropic acoustic tensor H(n) is defined by a formula like (7.38), but
with the ci jkl (F, θ ) replaced by the components of a 4-tensor defined as in (7.23)1 ,
except that ψ̂FF (F, θ ) is replaced by ˆFF (F, η); these components are then evaluated
at F0 , η0 , where η0 is the specific entropy at the base state. Thus the isentropic
acoustic tensor bears the same formal relation to the internal energy potential at
fixed η that the isothermal acoustic tensor bears to the Helmholtz free energy
120 EVOLUTION OF PHASE TRANSITIONS
Throughout this section, we assume that body force and heat supply are both
absent, so that b = 0, r = 0.
In uniaxial strain, all particles are displaced parallel to a fixed direction, say
the x1 -axis in a rectangular Cartesian frame, and the value of the x1 -component
of displacement depends only on x1 and the time t: u 1 = u 1 (x1 , t), u 2 = u 3 = 0.
Moreover, the temperature depends only on x1 and t: θ = θ(x1 , t). To simplify
notation, it is convenient to write x, u, and v instead of x1 , u 1 , and v1 , respectively,
and to indicate x- and t-derivatives by subscripts. Thus, for example, u 1,1 (x1 , t) =
u x (x, t) and v1 (x1 , t) = v(x, t) = u t (x, t).
In uniaxial strain, the deformation gradient matrix has the form
1+γ 0 0
F = 0 1 0, (7.42)
0 0 1
where γ = γ (x, t) = u x (x, t). We write
ˆ , θ ) = ψ̂(F, θ )
(γ (7.43)
when F has the form (7.42). It then follows from the constitutive law (7.17) that
ˆ γ (γ , θ ),
σ = σ̂ (γ , θ ) = ρ η = η̂(γ , θ ) = −
ˆ θ (γ , θ ) (7.44)
where we have written σ = σ11 for the component of nominal stress of primary
interest. The alternative version (7.6) of the constitutive law allows us to equivalently
write
σ = ρ ˆγ (γ , η), θ = ˆη (γ , η), (7.45)
in terms of the potential ˆ (γ , η) = (F, η) of specific internal energy expressed as
a function of strain and specific entropy.
Since all physical quantities are independent of x2 and x3 , the momentum equa-
tion (7.21) for j = 1 then reduces in the absence of body force to
(σ̂ (γ , θ))x − ρvt = 0, (7.46)
where σ̂ is the thermoelastic stress response function for uniaxial strain introduced
in (7.44).
Setting m(γ , θ ) = m 11 (F, θ ) for the relevant component of the stress–
temperature tensor, we may write (7.46) in the expanded form
σ̂γ (γ , θ )γx + m(γ , θ)θx − ρvt = 0, (7.47)
where σ̂γ stands for ∂ σ̂ /∂γ .
Where γ and v are smooth, their relation to u requires that they satisfy the
compatibility equation
vx − γt = 0. (7.48)
For (i, j) = (1, 1), the components σi j of the nominal stress tensor cannot be ex-
pressed in terms of (γ
ˆ , θ ) or its derivatives. For i = 2, 3, the differential equations
122 EVOLUTION OF PHASE TRANSITIONS
(7.21) for linear momentum, in the absence of body force and with v1 = v2 = 0,
reduce to σ21,1 = 0 and σ31,1 = 0; unless these equations are satisfied, the ther-
moelastic material under study does not sustain nontrivial states of uniaxial strain
parallel to the x1 -axis. To avoid this difficulty, we shall assume that the sym-
metry of the material is such that, in uniaxial strain, σ21 = ψ F21 (F , θ ) = 0 and
σ31 = ψ F31 (F , θ ) = 0 when F has the form (7.42), so that (7.21) is trivially satisfied
for i=2, 3. This is the case for isotropic materials, and for certain other material sym-
metries if the direction of the x1 -axis is suitably oriented. With σ21 = σ31 = 0, bal-
ance of angular momentum implies that σ12 = σ13 = 0 as well, and that σ23 = σ32 .
Thus in uniaxial strain parallel to the x1 -axis, the nominal stress matrix σ has the
form
σ 0 0
σ = 0 σ22 σ23 . (7.49)
0 σ23 σ33
From (6.14) and (7.49), we find that the true stress matrix is then given by
(1 + γ )σ 0 0
1
τ = 0 σ22 σ23 . (7.50)
1+γ
0 σ23 σ33
We set q = q11 , K (γ , θ) = K 11 (F, θ ) and infer from the heat conduction law
(7.18) that
q = K (γ , θ )θx . (7.51)
Since the x2 - and x3 -components of the heat conduction vector q depend only on
x1 = x, the energy equation (7.22) takes the following form for uniaxial strain in
the absence of heat supply:
Equations (7.47), (7.48), and (7.52) are the three basic scalar partial differential
equations for the three unknowns γ , v, and θ associated with uniaxial strain of a
thermoelastic material.
Finally we specialize the respective jump conditions (7.24)2 , (7.26), and (7.28)
to the case of uniaxial strain:
REFERENCES
8.1 Introduction
In Chapters 3 and 4 we used particular one-dimensional initial–boundary value
problems to demonstrate that, because of a massive lack of uniqueness that ex-
ists otherwise, elasticity theory must be supplemented with a kinetic law and nu-
cleation condition if it is to be used to model the emergence and evolution of
multiphase configurations. As shown there, not only is there a need for such infor-
mation, there is also room for it in the theory. A second motivation for a kinetic
law, also presented in Chapter 3, arose by casting the quasistatic problem consid-
ered there in the framework of standard internal-variable theory; the evolution law
characterizing the rate of change of the internal variable in that theory is then the
kinetic law.
In the present chapter we provide a third approach to the notion of kinetics, this
one from a thermodynamic point of view. In addition to providing a motivation,
the discussion here allows us to describe the kinetic law within a general three-
dimensional thermoelastic setting.
In Section 8.2 we present the thermodynamic formalism of irreversible processes
in a thermoelastic body. Based on this, we introduce the notion of a thermodynamic
driving force and the flux conjugate to it, and the notion of a kinetic relation then
follows naturally. In Section 8.3 we present some phenomenological examples of
kinetic relations, while in Section 8.4 we describe examples of kinetic relations
based on various underlying transformation mechanisms. Some remarks on the
nucleation condition are made in Section 8.5.
124
KINETICS AND NUCLEATION 125
in the bulk is due solely to heat conduction; further, since the process is smooth,
there are no interfaces in the body which might produce any additional entropy.
Thus recalling (6.16), (6.19), (6.21), and (7.8), the total rate of entropy production
associated with a subregion D of R is
∇θ
D (t) = q · 2 d V ≥ 0. (8.1)
D θ
Since (8.1) holds for all D ⊂ R, and since θ > 0, necessarily
q(x, t) · ∇θ (x, t) ≥ 0 for x ∈ R, t ∈ [t0 , t1 ]. (8.2)
This shows that the heat flux vector q and the temperature gradient vector ∇θ
cannot point in opposite directions. Thus one says that “the temperature gradient
∇θ drives the heat flux q” and that “∇θ is the driving force conjugate to q.”
If this body is in thermodynamic equilibrium at an instant t∗ , then D (t∗ ) = 0
for all subregions D ⊂ R. Therefore
q(x, t∗ ) · ∇θ (x, t∗ ) = 0 for x ∈ R. (8.3)
There is nothing in the description of a thermoelastic material, or the usual principles
of continuum mechanics, that provides any information about the driving force ∇θ
and the flux q when the body is not in thermodynamic equilibrium, and the notion
of a kinetic law is intimately related to nonequilibrium thermodynamic processes.
According to the theory of irreversible processes (e.g. Sections 14-2 and 14-3 of
Callen [12] or Section 14.3.1 of Kestin [18]), such information must be provided
empirically, by relating the present value of the flux q to the present (and possibly
past) value of the driving force ∇θ (and perhaps to other state variables, such as F
and θ , as well). In the present setting, a natural example of this would be a relation
of the form
q(x, t) = q̂(∇θ (x, t), F(x, t), θ (x, t)) for x ∈ R, t ∈ [t0 , t1 ], (8.4)
where q̂ is a constitutive function in the sense that, in general, it will be different
for different materials. The kinetic law (8.4) characterizes the response of the body
when it is not necessarily in thermodynamic equilibrium; in the present context, it is
no more than the heat conduction law. In view of the inequality (8.2), the function
q̂ must have the property that at each temperature θ and deformation gradient F,
q̂(g, F, θ ) · g ≥ 0 for all vectors g. (8.5)
If q̂(g, F, θ ) is continuous at g = 0, (8.5) shows that q̂(0, F, θ ) = 0, that is the
heat flux must vanish when the temperature gradient vanishes. This further justifies
the term “thermodynamic driving force” for ∇θ. The continuum theory, on its
own, provides no further information about q̂. If the heat flux depends linearly on
the temperature gradient, so that q̂(∇θ, F, θ ) = K(F, θ )∇θ, the 2-tensor K is the
conductivity tensor and (8.5) shows that K must be positive semidefinite at θ and
F: K(F, θ )g · g ≥ 0.
126 EVOLUTION OF PHASE TRANSITIONS
Heat conduction:
q(x, t) = q̂(∇θ (x, t), F(x, t), θ (x, t)) for x ∈ R − St , t ∈ [t0 , t1 ],
Phase transition: (8.7)
Vn (x, t) = ( f (x, t), θ (x, t)) for x ∈ St , t ∈ [t0 , t1 ].
Observe that the heat conduction law holds at all points in the body where the fields
are smooth, while the kinetic law for phase transition holds at all points on the
phase boundary. Because of the entropy inequality (6.33) and (8.2), the constitutive
functions q̂ and must satisfy the requirements
at each θ and F.
In both quasistatics and dynamics in one dimension, we postulated that the
propagation speed of a phase boundary depended on the local state of the material
on either side of the interface; see equations (3.43) and (4.22) respectively. A similar
KINETICS AND NUCLEATION 127
Vn
Vn = Φ(f, θ )
fmin(θ) frHL 0
f
frLH fmax(θ)
Figure 8.1. Schematic graph of a generic kinetic response function ( f, θ ) as a function of driving
force f at fixed temperature θ ; frLH and frHL are the resistances to phase boundary motion and fmax (θ )
and fmin (θ ) are the maximum and minimum values of driving force. If the nucleation levels of driving
force fnLH and fnHL were displayed on this figure they would lie outside the open interval ( frHL , frLH ).
In applying this kinetic law at a phase boundary across which the temperature is discontinuous, the
second argument of the function is taken to be the average temperature θ .
Vn = (F+ , F− , θ + , θ − ) (8.9)
for the kinetic law that includes (8.7)2 as a special case. In one dimension, we were
able to derive the form Vn = ( f ) from this assumption; in the general theory of
the present chapter this is no longer possible so that (8.9) is a true generalization
of (8.7)2 .
A kinetic relation of the general form Vn = ( f, θ ) is both homogeneous and
isotropic in the sense that the kinetic response of the material is the same at all
particles and in all directions. This can be generalized to
Vn = ( f, θ , n, x), (8.10)
Vn Vn
(a) (b)
f f
Vn Vn
(c) (d)
f f
Figure 8.2. Graphs of propagation velocity Vn versus driving force f at fixed temperature θ for various
kinetic relations Vn = ( f, θ ): (a) linear, (b) monotonic (c) lattice trapping, and (d) maximally dissipative.
some of the sorts of modeling that one might consider, we sketch out five examples
here. It should be emphasized that our goal is to illustrate some different sorts of
models rather than to develop any physically accurate models.
In each of the calculations below we assume that the motion occurs isothermally
at a fixed temperature. Therefore we shall not display dependencies on θ explicitly
(except in the model of thermally activated processes). The kinetic response func-
tions ( f, θ ) determined below are thus valid at the fixed temperature underlying
the calculations.
The simplest form of kinetic relation governing a three-dimensional motion of a
phase boundary relates the driving force on it to its normal velocity of propagation:
Vn = ( f ). Perhaps it is worth remarking that since the kinetic response function
is a function of a single scalar independent variable, one set of experiments, say
uniaxial tension tests, completely determines ; and the function thus determined
characterizes all motions of this interface (at that temperature) even if the conditions
are not those of uniaxial stress. If the deformation field is inhomogeneous, and the
phase boundary is curved, one would use this kinetic relation locally, at each point
along the interface, relating the driving force at that point to the normal velocity of
propagation of that point. Consequently, in order to determine the kinetic response
function it is sufficient to study, say, one-dimensional motions of the body. All of
the kinetic relations derived below are developed on the basis of convenient one-
dimensional models, but for the reasons just mentioned, they are applicable more
generally.
In order to keep the analysis as simple as possible, the first three examples
will be developed within the following particularly simple context: consider a one-
dimensional bar characterized by the biquadratic strain energy density
µ
γ 2, γ ≤ γ∗ ,
2
W (γ ) = µ (8.11)
(γ − γT )2 , γ ≥ γ∗ ,
2
with the associated bilinear stress response function
µγ , γ < γ∗ ,
σ̂ (γ ) = W
(γ ) = (8.12)
µ(γ − γ ), γ > γ ;
T ∗
see Figure 8.3. Here W is related to the Helmholtz free energy function ψ through
W (γ ) = ρψ(γ ). Let
γT µγT2
γ∗ = , σ∗ = µγ∗ , f∗ = . (8.13)
2 2
Two phase equilibrium states are possible when the stress lies in the range −σ∗ <
σ < σ∗ ; the strains on the two sides of a phase boundary in mechanical equilibrium
are
σ σ
γ+ = , γ − = + γT , (8.14)
µ µ
130 EVOLUTION OF PHASE TRANSITIONS
W σ
σ = µ(γ − γT )
µ σ∗
µ W = (γ − γT )2
W = γ2 2
2
γ
γ∗ γT
σ = µγ
−σ∗
γ
γ∗ γT
when the high-strain phase is on the left and the low-strain phase is on the right.
The driving force on such a phase boundary is
f = W (γ + ) − W (γ − ) − σ (γ + − γ − ) = σ γT . (8.15)
The potential energy function is
P(γ , σ ) = W (γ ) − σ γ . (8.16)
8.4.1 Viscosity–strain gradient model. In this model we augment the elastic con-
stitutive relation σ = σ̂ (γ ) by adding to it a linear viscosity term characterized by
a coefficient ν and a linear strain-gradient term characterized by a parameter λ.
The viscosity allows for dissipation and the strain gradient allows the sharp phase
boundary to be spread over a narrow transition zone. Thus the constitutive relation
for stress, when augmented with these additional effects, reads
σ = σ̂ (γ ) + ρνγt − ρλγx x , ν > 0, λ > 0, (8.17)
where σ̂ continues to be given by (8.12). We consider a quasistatic motion and so
the stress is spatially uniform in the bar. We wish to model a phase boundary of
the purely elastic theory, which is a sharp interface that travels at a speed Vn and
separates the strains γ ± . In the augmented theory, this is described by a traveling
wave in an infinite bar, propagating at the speed Vn and approaching the strains γ ±
in the far field:
γ = γ (ξ ) for − ∞ < ξ < ∞, γ (ξ ) → γ ± as ξ → ±∞, (8.18)
where
ξ = x − Vn t. (8.19)
Without loss of generality, the respective regions ξ > 0 and ξ < 0 may be associ-
ated with the low- and high-strain phases of the material. Substituting (8.18)1 into
(8.17) leads to the differential equation
ρλγ
(ξ ) + ρνVn γ
(ξ ) + σ − σ̂ (γ (ξ )) = 0, −∞ < ξ < ∞, (8.20)
where σ is the uniform stress in the bar. The traveling wave γ (ξ ) must satisfy this
equation and the remote boundary conditions (8.18)2 . In light of the smoothness
KINETICS AND NUCLEATION 131
characteristics of the stress response function (8.12), and the order of the differential
equation (8.20), the strain field γ (ξ ) must be continuous and have a continuous first
derivative. It is convenient to introduce the nondimensional propagation speed v
and the material parameters c and ω defined by
√
Vn µ λ
v= , c= , ω=2 . (8.21)
ωc ρ ν
Explicitly solving the differential equation (8.20) by using (8.12), and enforcing
the boundary conditions (8.18)2 and the continuity of γ (ξ ) at ξ = 0, shows that the
strain field is given by
+ + p2 ξ
γ + (γ∗ − γ )e , ξ ≥ 0,
2c
γ (ξ ) = − − p1 ξ where p = −v ± v 2+1 .
γ − (γ − γ∗ )e , ξ ≤ 0,
1
2 ω
(8.22)
8.4.2 Thermal activation model. In order to construct this next model, consider
a quasistatic motion of the elastic bar and consider an instant at which the stress
has the value σ . Suppose there is a phase boundary located at x = s, and the
phase boundary is propagating into the low-strain phase at a speed Vn = ṡ. The
strains γ ± on either side of the phase boundary are given by (8.14). The potential
energy function P(·, σ ) given by (8.16) has local minima at γ − and γ + and a
local maximum between them at γ∗ ; see Figure 8.4. As the phase boundary moves
through the bar, particles immediately ahead of it “jump” from the higher energy
132 EVOLUTION OF PHASE TRANSITIONS
P (γ, σ) = W (γ) − σγ
Figure 8.4. Graph of potential energy
P versus strain γ at a fixed stress σ for
the biquadratic model (8.11) and (8.16).
The energy barriers are denoted by b± .
The figure has been drawn for σ > 0 in
which case the high-strain energy well
b+ is lower than the low-strain energy well.
b− The locations γ ± of the local minima
γ−
γ are given by (8.14).
γ+ γ∗
well of P to the lower one, and an associated kinetic relation can be constructed
by viewing this process of jumping at the atomic scale. It should be noted that at
the atomic scale, some atoms from each energy well jump into the other energy
well, and it is the net effect of this process that is reflected at the macroscopic scale.
The atoms behind the phase boundary jump from the energy well at γ − to that at
γ + and have to cross over an energy barrier
(σ∗ − σ )2
b− = P(γ∗ , σ ) − P(γ − , σ ) = + σ γT (8.24)
2µ
in order to do so; see Figure 8.4. Similarly the atoms ahead of the phase boundary
jump from the energy well at γ + to that at γ − and in doing so must cross over an
energy barrier
(σ∗ − σ )2
b+ = P(γ∗ , σ ) − P(γ + , σ ) = . (8.25)
2µ
In a “thermally activated process,” the probability that the energy of an atom is
at least as great as b is exp(−b/kθ ) where k is Boltzmann’s constant and θ is
the absolute temperature. Moreover, the average rate at which atoms jump from
one minimum to the other in such a process is proportional to the probability of
exceeding the corresponding energy barrier. Thus in a thermally activated transition,
the velocity of the phase boundary, which is a macroscopic measure of the net rate
at which atoms change from the low-strain phase to the high-strain phase, is the
difference between the rates of the two atomic transitions and is therefore given by
b+ b−
Vn = R exp − − exp − . (8.26)
r kθ r kθ
Here R is related to the frequency with which an atom attempts to cross over the
relevant energy barrier and the factor r converts the units of P (and therefore b) to
those of energy per atom. Substituting for the various quantities in this expression
by using (8.24), (8.25), and (8.15) leads to the following explicit kinetic relation
between the driving force f and the propagation speed Vn for a thermally activated
transition:
f 2 + f ∗2 f
Vn = ( f ) = 2R exp − sinh − . (8.27)
4r kθ f ∗ 2r kθ
KINETICS AND NUCLEATION 133
For a more complete description of models of this sort, see, for example, [3, 6,
7, 15, 16, 22, 23].
the atoms about their initial states that propagates away from the impacted end of
the chain at the “sonic” wave speed for the low-strain phase of the potentials. For
larger values of v0 , such a small-amplitude, low-strain-phase sonic precursor is fol-
lowed by a small oscillation about a mean amplitude much greater than that of the
precursor. With “strain” defined as the ratio of the difference in displacements of
successive atoms to the initial interatomic spacing, the strain involved in the trailing
large-amplitude wave lies in the high-strain phase of the interatomic potentials. The
“phase boundary” is the propagating abrupt change in the mean amplitude from
precursor to follower. Thus the response of the chain-of-atoms model is analogous
to that of the continuum mechanical model discussed earlier in Sections 4.3.1 and
4.3.2 in that there is a weak-impact regime without a phase transition for small v0
and a two-wave regime with a phase change for stronger impacts1 .
To determine the kinetic relation implied by the dynamics of the chain of atoms,
Purohit [25] begins with the continuum-mechanical representation of the driving
force f at a phase boundary. For the trilinear material with identical moduli µ in
the two phases, this is
µγT +
f = (γ + γ − − γ M − γ m ), (8.36)
2
where γ ± are the strains on either side of the phase boundary, γ T is the transfor-
mation strain, and γ M , γ m are the strains that delimit the two phases. Making use
of the numerically calculated strains γ ± and phase boundary speed Vn for vari-
ous values of the impact velocity, Purohit calculates the relation between f and
Vn for the 300-atom system. He finds that f increases monotonically with Vn ; the
numerically calculated points in a Vn , f -plane are described quite accurately by an
analytical relation f = φ(Vn ), in which φ(Vn ) is the maximally dissipative kinetic
response function for the particular trilinear material he considered.
Purohit also explores the effect of attributing initial velocities to the atoms,
corresponding to a chain at nonzero absolute temperature.
Other one-dimensional atomistic calculations have been carried out, for exam-
ple, by Puglisi and Truskinovsky [26], who study quasistatic processes, and by
Krumhansl and Schrieffer [20], Balk, Cherkaev, and Slepyan [9, 10], and Kresse
and Truskinovsky [19], who are concerned with dynamics.
A kinetic relation for twin boundary motion has been developed through an
analysis based on lattice dynamics by Abeyaratne and Vedantam [5] and applied to
a CuAlNi alloy. A lattice-dynamical simulation of a shock-induced phase transition
in iron has been carried out by Kadau et al. [17]. Their results exhibit both the
two-wave structure for moderate shock strength and the overdriven response for
strong shocks predicted by the model described earlier in Section 4; they do not
address the issue of a kinetic relation for the transition.
1
In the particular discrete model of [25], there is no counterpart of the “overdriven” response exhibited
in the continuum case, see Section 4.3.3, because the linear force–stretch relations between atoms
in each phase have the same moduli.
136 EVOLUTION OF PHASE TRANSITIONS
High-strain Low-strain
High-strain Low-strain phase phase
phase phase
c
L
c Vn
L
Vn
Figure 8.5. The phase boundary propagates to the right with an effective velocity Vn as a result of a
ledge, or “step,” L, which propagates transversely along the phase boundary with speed c. The two
figures correspond to two instants of time, showing the ledge L in two locations.
m ü n = F − h
(u n ) + k(u n+1 − 2u n + u n−1 ), n = 0, ±1, ±2, ..., (8.37)
We wish to model the kinetics of the transformation from the low-strain phase
to the high-strain phase. Consider therefore a motion in which all particles are
initially in the low-strain energy well and steadily advance, one particle at a time,
from that energy well to the high-strain one. Specifically, suppose that the nth
particle remains in the low-strain energy well for all times less than some critical
instant nτ ; at t = nτ it moves into the high-strain energy well and then remains
there for all subsequent times. This motion continues steadily with the (n + 1)th
particle moving into the high-strain energy well at time t = (n + 1)τ and so on.
Note that at the instant t = nτ , all particles behind the nth one belong to the high-
strain phase and therefore have undergone the transformation; all particles ahead
of the nth one are still in the original phase, and so there is a “discontinuity” at
the nth particle, which is the ledge L. The speed with which this discontinuity
propagates is
c = d/τ. (8.39)
In this steady motion the displacement of the nth particle at time t equals the
displacement of the (n + m)th particle at time t + mτ : u n (t) = u n+m (t + mτ ), and
so the displacement field must admit the traveling wave representation u n (t) = u(ξ )
where ξ = nd − ct. Note that ξ < 0 is associated with the high-strain phase and
ξ > 0 is associated with the low-strain phase.
In order to develop a continuum approximation of this problem, let x be a con-
tinuous coordinate along the spring-mass chain, and consider a displacement field
u(x, t), −∞ < x < ∞, t > 0, which is such that u(nd, t) = u n (t). The equation
of motion (8.37) can then be written as
mu tt (x, t) = F − h
(u(x, t)) + k [u(x + d, t) − 2u(x, t) + u(x − d, t)] ,
(8.40)
x = nd, n = 0, ±1, ±2, . . .
Replacing u(x ± d, t) by the first five terms in its Taylor expansion, and using
the traveling wave representation u(x, t) = u(ξ ), ξ = x − ct, leads to the ordinary
differential equation
kd 4
u (ξ ) + (kd 2 − mc2 )u
(ξ ) + F − h
(u(ξ )) = 0, −∞ < ξ < ∞. (8.41)
12
In order to determine the boundary conditions to be imposed on (8.41) we note
first that in equilibrium, when the displacement field u(ξ ) is constant, this constant
value is given according to (8.41) by the roots of the equation h
(u) = F. For the
biquadratic potential (8.38) this equation has two roots corresponding to the two
phases:
F F
u+ = , u− = + d. (8.42)
κ κ
138 EVOLUTION OF PHASE TRANSITIONS
Suppose that the material far ahead of the ledge is in equilibrium (in the low-strain
phase). Then we impose the following boundary conditions:
u(ξ ) → u + = F/κ as ξ → +∞,
(8.43)
u(ξ ) ≥ d/2 as ξ → −∞.
Equation (8.43)2 states that, far behind the interface, u(ξ ) must lie in the displace-
ment range corresponding to the high-strain phase. The stricter requirement that
the material far behind the ledge be in equilibrium in the high-strain phase can be
shown to be too stringent.
One can readily solve (8.41) and (8.43) using (8.38) on each of the ranges ξ < 0
and ξ > 0 separately. Requiring u and its first three derivatives to be continuous at
ξ = 0, and enforcing the additional continuity condition u(0) = d/2, leads to the
traveling wave solution
d +
e−bξ/d + u +
2 −u
for ξ > 0,
u(ξ ) = (8.44)
d r ξ
− −u +
e bξ/d +
− 2u cos + u −
for ξ < 0,
2 d
as well as the following kinetic relation between the force F and the dimensionless
propagation speed v of the ledge:
κd 1 − v 2 2
F = (v) = 1 − , (8.45)
4 (1 − v 2 2 )2 + 2 /3
Note that for the Frenkel-Kontorowa kinetic law (8.45) there is a minimum force,
Fmin , required to propagate the ledge. By setting v = 0 in (8.45), this value is
κd 1
Fmin = 1 − . (8.48)
4 1 + 2 /3
Observe from (8.44)2 that the displacement field behind the ledge is oscilla-
tory (and therefore not in equilibrium). These oscillations are reminiscent of those
observed by Purohit [25] in his atomistic model; see Section 8.4.4.
KINETICS AND NUCLEATION 139
σ σ = µ(γ − γT )
P (γ, σ) = W (γ) − σγ
σM
σ = µγ (a)
(b)
σm σ/µ + γT
b(σ) σ/µ
γ γ
γT γ∗
γ∗
Figure 8.6. Bilinear material. (a) Stress–strain curve. (b) Potential energy at a fixed value of stress σ in
the interval σ0 < σ < σ M , where σ0 = (σ M + σm)/2 is the Maxwell stress.
8.5 Nucleation
A kinetic relation describes the propagation of an existing phase boundary. It does
not address the issue of when a new phase (or equivalently, a new phase bound-
ary) first appears – or “nucleates” – within a configuration that involves a single
phase. As demonstrated explicitly in the example problems in Chapters 3 and 4
the phenomenon of nucleation is distinct from propagation and must therefore
be described by a criterion other than the kinetic relation. We therefore require a
“nucleation criterion.”
A complete discussion and analysis of nucleation are beyond the scope of this
monograph, not to mention the abilities of its authors! A good description of the var-
ious physical effects that must be accounted for, as well as the underlying processes,
can be found in Chapter 10 of Christian [13], especially Section 52.
Here we wish to describe a nucleation criterion that elaborates and generalizes
the criterion used in Sections 3.5 and 4.4. We present the ideas in the simplest
context: that of the purely mechanical theory of equilibria in a one-dimensional bar
composed of a two-phase material described in Chapter 3. Thermal effects are not
taken into account, but the effect of surface energy at the phase boundary, omitted
from the discussion in Chapter 3, is now included and plays a significant role.
According to the criterion to be developed here, nucleation results in a portion of
the bar of nonzero length switching instantaneously from one phase to the other.
The length of this “nucleus” vanishes in the limit of zero surface energy, reducing
the theory sketched in the present section to one that is consistent with the model
used in the earlier chapters.
Consider a bar that occupies the interval 0 ≤ x ≤ L in a reference configuration.
Suppose the left end is fixed, and a given displacement δ is prescribed at the right
end. The material of the bar is assumed to have a bilinear, two-phase stress–strain
curve2 with modulus µ common to both phases, as shown in Figure 8.6(a); the
2
This is a special case of the trilinear material of Figure 2.7 for which γ M = γm and the moduli of
the high- and low-strain phases coincide.
140 EVOLUTION OF PHASE TRANSITIONS
3
We model nucleation as occurring instantaneously. In some circumstances, nucleation must be
viewed as a time-dependent process; see, for example, Section 50 of Christian [13].
4
If the elongation continues to slowly increase after nucleation, the phase boundary created at x = snuc
will propagate according to the kinetic relation.
KINETICS AND NUCLEATION 141
δnuc /L is permitted to lie anywhere in the interval (8.52). On the other hand, the
values of stress just before and just after nucleation are different as seen from (8.50)
and (8.51):
σpre = µδnuc /L , σpost = σpre − µγT snuc /L . (8.54)
Thus in this displacement-controlled process, the stress decreases discontinuously
after nucleation. For a stress-controlled process, one can show that the elonga-
tion increases discontinuously upon nucleation to accommodate the newly created
nucleus.
To begin with, we permit the pre-nucleation stress σpre to lie anywhere5 in the
interval σm ≤ σ ≤ σ M .
Consider a generic particle that changes phase upon nucleation; in the reference
configuration, such a particle lies in the segment 0 < x < snuc of the bar. The strains
at this particle just before and just after nucleation are given by
γpre = δnuc /L , γpost = γpre + γT (1 − snuc /L). (8.55)
From (8.49), (8.54), and (8.55), one can compute the values of the potential energy
of the generic particle just before and just after nucleation:
Ppre = P(γpre , σpre ) = −σpre
2
/(2µ),
(8.56)
Ppost = P(γpost , σpost ) = −σpost
2
/(2µ) − γT (σpost − σ0 ).
One can show that there is an interval of values of snuc of the form 0 < snuc <
s̄nuc on which (8.58) holds if and only if σpre > σ0 . Said differently, there is an
available nucleus of particles, each of whose potential energies would decrease
upon nucleation as required by Assumption 1, if and only if σpre > σ0 . Thus for
nucleation to occur, Assumption 1 has allowed us to conclude that
σ0 < σpre ≤ σ M . (8.59)
As a function of strain at a fixed stress whose value exceeds the Maxwell stress,
the graph of the potential energy P(γ , σ ) is shown in Figure 8.6(b). The height
5
In the theory of quasistatic processes of Chapter 3, it is shown that the dissipation inequality implies
that nucleation from the low-strain phase to the high-strain phase cannot occur at a stress less than
the Maxwell stress. We make no use of this result here.
142 EVOLUTION OF PHASE TRANSITIONS
b = b(σ ) between the low-strain phase minimum and the local maximum at γ = γ∗
is the energy barrier that a particle in the low-strain phase must surmount in order
to jump to the high-strain phase at the stress σ . In particular, at the prenucleation
stress σpre , the energy barrier can be computed with the help of (8.49), (8.54), and
(8.55) to be
1
b(σpre ) = P(γ∗ , σpre ) − P(γpre , σpre ) = (σpre − σ M )2 . (8.60)
2µ
Since σpre lies between σ0 and σ M , it follows from (8.60) that
1
0 < b(σpre ) < (σ M − σ0 )2 = µγT2 /8. (8.61)
2µ
As σpre increases from σ0 to σ M , the barrier height b(σpre ) decreases from µγT2 /8
to zero.
Assumption 2. For nucleation to occur at a generic particle, it is necessary for
the energy barrier b(σpre ) to decrease to a value not greater than a materially
determined critical value bnuc , where 0 < bnuc < µγT2 /8.
We thus take bnuc to be given; it is convenient to write bnuc = nuc
2
/(2µ), where
nuc < µγT /2, and to use nuc instead of bnuc in the discussion to follow.
From Assumption 2 and (8.60), it follows that
σ M ≥ σpre ≥ σ M − nuc = σ0 + µγT /2 − nuc > σ0 . (8.62)
In earlier chapters the nucleation criterion was phrased solely with reference to
a single particle; this was possible because the nucleus of the new phase was taken
to have vanishing size. In contrast here, the nucleus has a finite length snuc and so we
must also consider the energetics of the entire nucleus. The nucleus 0 ≤ x ≤ snuc is
itself effectively a bar; it is fixed at its left end x = 0 and is subjected to a loading
on its right hand end x = snuc by the rest of the bar. The total potential energy of the
nucleus comprises the elastic energy of the nucleus and the energy of the loading
device, the latter being equal to the elastic energy in the rest of the bar. Thus the
potential energy of the nucleus equals the total energy in the bar.
We now wish to compare the total energy per unit cross-sectional area E post
in the bar after nucleation with the total energy E pre before nucleation. In doing
so, we shall generalize the model of the bar to include a constant surface energy6
per unit area τ at the phase boundary when computing E post . The surface energy
accounts for local distortion very close to the interface between the phases7 . Such
distortion presumably occurs on a length scale that is O(τ/µ) L. The total strain
energy in the bar is found by integrating the strain energy per unit reference volume
W (γ ) = P(γ ) + σ γ over the length of the bar, making use of appropriate pre- and
6
In order to determine the length snuc we must introduce a length scale parameter into the model by
accounting for, for example, surface energy.
7
For related discussion, see Section 52 of Christian [13].
KINETICS AND NUCLEATION 143
Assumption 3. For nucleation, it is necessary that the total energy of the system
should not increase:
E post ≤ E pre . (8.64)
One easily shows that, for sufficiently small (τ/µ)/L, (8.64) holds provided
− +
snuc ≤ snuc ≤ snuc , (8.65)
∓
where snuc > 0 are defined through
∓
1/2
γT snuc /L = (σpre − σ0 )/µ ∓ [(σpre − σ0 )/µ]2 − 2(τ/µ)/L . (8.66)
8
As noted previously, the potential energy of a particle has to increase by the barrier height b(σpre )
before it can move into the high-strain phase. Since in our model all of the particles in the segment
0 < x < snuc undergo the phase transition simultaneously, they must all increase their energies by
b(σpre ) during the process of transformation. Since this increase of energy is proportional to the
length snuc of the nucleus, it is smaller the smaller snuc is. This is the motivation for requiring the
nucleus to have its smallest allowable size.
144 EVOLUTION OF PHASE TRANSITIONS
The nucleation condition can therefore be alternatively stated in terms of f nuc and
τ as follows: if the elongation δ is slowly increased from zero, nucleation occurs
at a critical value δ = δnuc = Lσpre /µ where σpre is given by (8.71)1 and results in
the emergence of a high-strain phase nucleus of size snuc given by (8.71)2 .
REFERENCES
[1] R. Abeyaratne, C. Chu, and R.D. James, Kinetics of materials with wiggly
energies: theory and application to the evolution of twinning microstructures
in a Cu–Al–Ni shape memory alloy. Philosophical Magazine A, 73 (1996),
pp. 457–97.
[2] R. Abeyaratne and J.K. Knowles, Implications of viscosity and strain-gradient
effects for the kinetics of propagating phase boundaries in solids. SIAM Jour-
nal on Applied Mathematics, 51 (1991), pp. 1205–21.
[3] R. Abeyaratne and J.K. Knowles, A continuum model of a thermoelastic solid
capable of undergoing phase transitions. Journal of the Mechanics and Physics
of Solids, 41 (1993), pp. 541–71.
[4] R. Abeyaratne and S. Vedantam, Propagation of a front by kink motion.
in IUTAM Symposium on Variations of Domains and Free-Boundary Prob-
lems in Solid Mechanics, edited by P. Argoul et al., Solid Mechanics
and its Applications, Volume 66. Kluwer, Dordrecht, 1999, pp. 77–
84.
[5] R. Abeyaratne and S. Vedantam, A lattice-based model of the kinetics of twin
boundary motion. Journal of the Mechanics and Physics of Solids, 51 (2003),
pp. 1675–700.
[6] M. Achenbach, A model for an alloy with shape memory. International Journal
of Plasticity, 5 (1989), pp. 371–95.
[7] M. Achenbach and I. Müller, Simulation of material behavior of alloys with
shape memory. Archive for Mechanics, 37 (1985), pp. 573–85.
[8] W. Atkinson and N. Cabrera, Motion of a Frenkel–Kontorowa dislocation in
a one–dimensional crystal. Physical Review A, 138 (1965), pp. 763–6.
KINETICS AND NUCLEATION 145
[9] A.M. Balk, A.V. Cherkaev, and L.I. Slepyan, Dynamics of chains with non-
monotone stress–strain relations. I. Model and numerical experiments. Journal
of the Mechanics and Physics of Solids, 49 (2001), pp. 131–48.
[10] A.M. Balk, A.V. Cherkaev, and L.I. Slepyan, Dynamics of chains with non-
monotone stress–strain relations. II. Nonlinear waves and waves of phase
transition. Journal of the Mechanics and Physics of Solids, 49 (2001),
pp. 149–71.
[11] K. Bhattacharya, Phase boundary propagation in heterogeneous bodies.
Proceedings of the Royal Society of London A, 455 (1999), pp. 757–66.
[12] H.B. Callen, Thermodynamics and an Introduction to Thermostatistics,
Chapter 14, Second edition. Wiley, New York, 1985.
[13] J.W. Christian, The Theory of Transformations in Metals and Alloys, Part 1:
Equilibrium and General Kinetic Theory, Chapter 10, second edition. Perga-
mon Press, Oxford, 1975.
[14] J. Frenkel and T. Kontorowa, On the theory of plastic deformation and twin-
ning. Physikalische Zeitschrift der Sowjeturion, 13 (1938), pp. 1–10.
[15] M. Grujicic, G.B. Olson, and W.S. Owen, Mobility of martensite interfaces.
Metallurgical Transactions, 16A (1985), pp. 1713–22.
[16] M. Grujicic, G.B. Olson, and W.S. Owen, Mobility of the β1 − γ1 marten-
sitic interface in CuAlNi. Part I. Experimental measurements. Metallurgical
Transactions, 16A (1985), pp. 1723–34.
[17] K. Kadau, T.C. Germann, P.S. Lomdahl, and B.L. Holian, Microscopic view
of structural phase transitions induced by shock waves. Science, 296 (2002),
pp. 1681–4.
[18] J. Kestin, A Course in Thermodynamics, Volume II, Chapter 14, McGraw-Hill,
New York, 1979.
[19] O. Kresse and L. Truskinovsky, Mobility of lattice defects: discrete and contin-
uum approaches. Journal of the Mechanics and Physics of Solids, 51 (2003),
pp. 1305–32.
[20] J.A. Krumhansl and J.R. Schrieffer, Dynamics and statistical mechanics of a
one-dimensional model Hamiltonian for structural phase transitions. Physical
Review B, 11 (1975), pp. 3535–45.
[21] P.G. LeFloch, Hyperbolic Systems of Conservation Laws. Birkhäuser Verlag,
Basel, 2002.
[22] I. Müller and K. Wilmansky, Memory alloys - phenomenology and Ersatz-
model, in Continuum Models of Discrete Systems, edited by O. Brulin and
R.K.T. Hsieh. North-Holland, Amsterdam, 1981, pp. 495–509.
[23] D.A. Porter and K.E. Easterling, Phase Transformations in Metals and Alloys.
van Nostrand-Reinhold, New York, 1981.
[24] J. Pouget, Dynamics of patterns in ferroelastic–martensitic transformations.
I. Lattice model. Physical Review B, 43 (1991), pp. 3575–81.
[25] P.K. Purohit, Dynamics of Phase Transitions in Strings, Beams and Atomic
Chains. Ph.D. dissertation, California Institute of Technology, Pasadena, CA,
November 2001.
146 EVOLUTION OF PHASE TRANSITIONS
9.1 Preliminaries
Our objective in this chapter is to describe certain explicit and useful special consti-
tutive models for two-phase thermoelastic materials in a one-dimensional setting.
These models are formally applicable to either uniaxial stress in a bar or uniaxial
strain in a slab. In the former case, one considers a thermoelastic bar occupying the
interval 0 ≤ x ≤ L of the x-axis in a reference configuration that is stress free at the
absolute temperature θ0 ; in a motion, the particle at x in the reference configuration
is carried at time t to x + u(x, t). In the latter case, the slab occupies the region
0 ≤ x ≤ L , −∞ < y, z < ∞ in the reference configuration, which is stress free
at the temperature θ0 ; the particle at x, y, z is carried to x + u(x, t), y, z during a
motion. In either case, we write
γ = ux , v = ut (9.1)
for the strain and particle velocity, respectively. For a bar, the thermoelastic mate-
rial is characterized by an internal energy per unit mass, or specific internal energy,
ˆ (γ , η), where η is the entropy per unit mass. For a thermoelastic slab, the internal
energy per unit mass ˆ (F, η) for three-dimensional motions with deformation gra-
dient tensor F and entropy per unit mass η reduces in uniaxial strain to a function
of γ and η only. For the bar or the slab, the fundamental thermoelastic constitutive
law yields
σ = ρ ˆ γ (γ , η), θ = ˆ η (γ , η) > 0, (9.2)
where ρ is the (constant) mass per unit reference volume, and subscripts γ and η
indicate partial derivatives with respect to these quantities. In (9.2)1 , σ is the axial
Piola–Kirchhoff normal stress, and θ is the absolute temperature.
In uniaxial strain, there are also transverse normal components of the Piola–
Kirchhoff stress tensor, which in general do not coincide with σ and which cannot
be calculated from the restriction of ˆ (F, η) to uniaxial strain. It is assumed that the
symmetry of the material at the reference configuration is such as to admit uniaxial
strain; this is true, for example, if the material is isotropic in the reference state.
149
150 EVOLUTION OF PHASE TRANSITIONS
Throughout this chapter, we shall for definiteness use the language of uniaxial
stress in a bar, and in two-phase materials we shall assume the phase transition oc-
curs in tension. Obvious modifications can be made, and would be needed, to deal
with compression-induced phase changes, which are more natural from the exper-
imental viewpoint. Useful background references for the material of this chapter
are the article on thermoelasticity by Carlson [4] and the paper on shock waves in
thermoelastic materials by Dunn and Fosdick [6].
For a thermoelastic material, the specific heat at constant strain c(γ , η) and
a measure of thermomechanical coupling called the Grüneisen parameter1 are
defined by
ˆ η ˆ γ η
c= , β=− . (9.3)
ˆ ηη ˆ η
We assume throughout that the specific heat c is positive, so that it follows from
(9.3)1 that (9.2)2 is invertible to give
η = η̂(γ , θ ). (9.4)
The Helmholtz free energy per unit mass is then
1
In the setting of uniaxial strain, the entity β defined by (9.3)2 is often called the modified Grüneisen
parameter, the unmodified version being (1 + γ )β.
MODELS FOR TWO-PHASE THERMOELASTIC MATERIALS IN ONE DIMENSION 151
here c, β, and µ are the respective values of the specific heat, the Grüneisen pa-
rameter, and the isothermal modulus at the base state.
Since the reference configuration is assumed to be unstressed at the temperature
θ0 , one has
ˆ γ (0, η0 ) = 0, ˆ η (0, η0 ) = θ0 , or ψ̂γ (0, θ 0 ) = 0, ψ̂θ (0, θ 0 ) = −η0 , (9.10)
where η0 is the specific entropy in the reference configuration at the temperature θ0 .
Without loss of generality, we may normalize the internal energy density by speci-
fying its value at the reference configuration at the temperature θ0 ; it is convenient
to set
ˆ (0, η0 ) = c(0, η0 )θ0 , (9.11)
or equivalently ψ̂(0, θ0 ) = c(0, η0 )θ0 − η0 θ0 .
Positivity of both temperature θ and specific heat c implies that H (η) is in-
vertible; it then follows from (9.16)2 that η = K (θ / p(γ )), where K is the inverse
of H . From this, together with (9.5) and (9.12), one finds that the Helmholtz free
energy density for a Mie–Grüneisen material must be of the form
θ
ψ̂(γ , θ) = p(γ ) L + q(γ ), (9.17)
p(γ )
for a suitable function L. We note in passing that the function of γ and θ obtained
by adding an arbitrary linear function of temperature to the right side of (9.17)
continues to have the form of a Helmholtz free energy density for a Mie–Grüneisen
material.
That the Grüneisen parameter β(γ , η) be independent of η is not only necessary
for a thermoelastic material to be of Mie–Grüneisen type, it is also sufficient, as is
easily shown. A thermoelastic material whose specific heat is constant, such as the
ideal gas or the van der Waals fluid [10, 17], is a Mie–Grüneisen material.
In the special case of a Mie–Grüneisen material for which the specific heat at
constant strain c and the modified Grüneisen parameter β are both constants, it is
readily shown that (9.12) specializes to
ˆ (γ , η) = W (γ ) + cβ θ0 γ + cθ0 exp[(η − η0 )/c − βγ )], (9.18)
where we have set W (γ ) = q(γ ) − cβ θ0 γ . Using (9.10), (9.11), one finds that
W (γ ) must satisfy
W (0) = W (0) = 0, (9.19)
so that one has W (γ ) = O(γ 2 ) as γ → 0. The constitutive expressions for stress
and temperature that follow from (9.2) and (9.18) are
σ = ρW (γ ) + ρcβ θ0 {1 − exp[(η − η0 )/c − βγ ]},
(9.20)
θ = θ0 exp[(η − η0 )/c − βγ ].
Using (9.20)2 , one may write η = η̂(γ , θ) = η0 + cβ γ + c log(θ/θ0 ); by adding
a suitable constant to η if necessary, we may assume that η0 = c. The Helmholtz
free energy density is then found according to (9.5) as
ψ̂(γ , θ) = W (γ ) − cβ(θ − θ0 )γ − cθ log(θ/θ0 ). (9.21)
The associated representation of the constitutive law in terms of strain and temper-
ature is
σ = ρW (γ ) − ρcβ (θ − θ0 ), η = c[1 + βγ + log(θ/θ0 )]. (9.22)
Upon linearization about the reference configuration, (9.22)1 becomes
σ = µ0 γ − ρcβ(θ − θ0 ), (9.23)
where µ0 = ρ ψ̂γ γ (0, θ0 ) = ρW (0) is the referential isothermal elastic modu-
lus; (9.23) is the classical stress–strain–temperature relation of one-dimensional
linear thermoelasticity [4]. The coefficient of thermal expansion of the material is
α = ρcβ/µ0 .
MODELS FOR TWO-PHASE THERMOELASTIC MATERIALS IN ONE DIMENSION 153
θ
Figure 9.1. The regions, shown shaded,
(γ∗, θ∗)
θ∗ in the γ , θ -plane corresponding to the
γ = γM(θ) γ = γm(θ) low-strain phase PL and the high-strain
phase PH of the Mie–Grüneisen material
θ0
PL PH characterized by the Helmholtz free en-
γ ergy density ψ̂(γ , θ ) of (9.27), and (9.28).
−1 0 γ∗
through its Helmholtz free energy density per unit mass, and we write
W (γ ) − cβ(θ − θ 0 )γ − cθ log(θ /θ 0 ) for (γ , θ ) ∈ PL ,
ψ̂(γ , θ) = W (γ − γ ) − cβ(θ − θ )(γ − γ ) − cθ log(θ /θ ) + λ (θ − θ )/θ
T 0 T 0 T 0 0
for (γ , θ) ∈ PH ,
(9.27)
where the regions PL and PH of the γ , θ -plane are associated with the low-strain and
high-strain phases respectively. Here the constants γT > 0 and λT stand respectively
for the transformation strain and the latent heat of transformation from the low-
strain phase to the high-strain phase at the reference temperature θ 0 . Moreover
PL = {(γ , θ)| − 1 < γ < γ M (θ ), 0 < θ < θ ∗ },
(9.28)
PH = {(γ , θ)| γ > γ m (θ ), 0 < θ < θ ∗ },
where µ = µ0 > 0, γ ∗ > 0, θ ∗ > 0, and M and m are given material constants;
in particular, the curves γ = γ M (θ ) and γ = γ m (θ ) shown in Figure 9.1 are now
straight lines intersecting at the critical point γ ∗ , θ ∗ .
MODELS FOR TWO-PHASE THERMOELASTIC MATERIALS IN ONE DIMENSION 155
σ σ = σ̂(γ, θ)
γT
σM(θ)
1
µ
1
µ
−1
γm(θ)
γ
0 γM(θ)
σm(θ)
Figure 9.2. The isothermal trilinear stress–strain curve. The declining portion of the curve represents
the unstable regime of the material.
At a fixed temperature θ , one has σ̂ (-1, θ ) < σ < σ M (θ ) in the low-strain phase
PL , and σ > σ m (θ ) in the high-strain phase PH , where
σ M (θ ) = µγ M (θ ) − ρcβ(θ − θ T ),
(9.31)
σ m (θ ) = µ(γ m (θ ) − γ T ) − ρcβ(θ − θ T ),
and γ M (θ) and γ m (θ ) are given by (9.29)2,3 . We require that −1 < γ M (θ ) < γ m (θ )
and σ M (θ) > σ m (θ ); these conditions hold for 0 < θ < θ ∗ if and only if the fol-
lowing inequalities are satisfied:
By using (9.27), (9.29), and (9.31), one can show that (9.33) holds for all θ ∈
(0, θ ∗ ) if and only if the following two relations are imposed on the various material
156 EVOLUTION OF PHASE TRANSITIONS
constants:
γ ∗ = γ T /2 + (M + m)(θ ∗ − θ T )/2,
(9.34)
λT = (µγ T θ T /ρ) [(M + m)/2 − α] .
The relation between specific entropy, strain, and temperature that follows from
(9.6), (9.27), and (9.29) is
cβ γ + c[1 + log(θ /θ T )] on PL ,
η = η̂(γ , θ ) = (9.35)
cβ (γ − γ T ) + c[1 + log(θ /θ T )] − λT /θ T on PH .
The latent heat of transformation λ(θ , σ ) from PL to PH at the temperature θ
and the stress σ is given by λ = θ [η̂L (θ , σ ) − η̂H (θ , σ )] where η̂L and η̂H are the
specific entropies associated with the two phases. Thus
θ
λ = λ(θ , σ ) = λT . (9.36)
θT
For simplicity and explicitness of results, the analysis in this and the next chapter
will be carried out for the trilinear material subject to the restrictions (9.32) and
(9.34); it will be clear, however, that the theory could be readily extended so as to
apply to the more general two-phase Mie–Grüneisen material characterized by the
Helmholtz free energy density (9.27), with W (γ ) assumed to be convex, though
not quadratic, and to satisfy (9.19). The practical advantage of the trilinear material
model arises from the fact that nonlinear effects are minimized.
9.3.2 Stability of phases of the trilinear material. Within each phase, the relation
(9.30) can be inverted to give strain in terms of stress and temperature:
σ
γ = L (θ , σ ) = + α(θ − θ T ) on
L ,
µ
σ (9.37)
γ = H (θ , σ ) = γ T + + α(θ − θ T ) on
H ,
µ
where
α = ρcβ/µ (9.38)
is the coefficient of thermal expansion of the material, and the regions
L and
H
in the θ, σ -plane are the respective sets of states θ , σ at which the low-strain and
high-strain phases can exist; they are defined by
L = {(θ , σ )| σ̂ (−1, θ ) < σ < σ M (θ ), 0 < θ < θ ∗ },
(9.39)
H = {(θ , σ )| σ > σ m (θ ), 0 < θ < θ ∗ }.
The two phases can coexist on the region
LH =
L ∩
H . The regions
L ,
H ,
and
LH in the θ , σ -plane are shown in Figure 9.3.
MODELS FOR TWO-PHASE THERMOELASTIC MATERIALS IN ONE DIMENSION 157
σ σ
(b)
(a)
σ = σM(θ)
ΠH
0 0
θ∗ θ θ∗ θ
ΠL
σ = σm(θ)
σ = σ̂(−1, θ)
σ
(c)
σ = σM(θ)
0
ΠLH θ0
θ∗
θ
σ = σ0(θ)
σ = σm(θ)
Figure 9.3. The low- and high-strain phases as represented by the respective regions
L (a) and
H
(b) in the θ , σ -plane;
LH (c) is the region where the two phases may coexist. The Maxwell line σ =
σ 0 (θ) shown in (c) goes through the states (γ , θ ) for which the Gibbs free energies of the two phases
coincide; see (9.43). The slope of the Maxwell line is positive, as is the case in the figure, when λT > 0;
see (9.44).
We define the Gibbs free energy per unit reference volume for each phase by
gL (θ , σ ) = ρ ψ̂(L (θ , σ ), θ ) − σ L (θ , σ ),
on
LH . (9.40)
gH (θ , σ ) = ρ ψ̂(H (θ , σ ), θ ) − σ H (θ , σ ),
Now consider the potential energy P(γ ; θ , σ ) of (9.24) specialized through (9.27),
and (9.29) to the two-phase trilinear material. Let (θ, σ ) be fixed in
L , and seek
the extrema of P(γ ; θ , σ ) as γ varies over the interval (−1, γ M (θ )). At such an
extremum, (9.25)1 shows that γ = L (θ , σ ) must hold; since W is convex, there is
exactly one extremum, and it is a minimum. By (9.24) and (9.40), the value of P
at the minimum is precisely the Gibbs free energy density for the low-strain phase:
for (θ , σ ) ∈
L ,
gH (θ , σ ) = − 2µ
1
[σ +µα(θ −θ T )]2 − σ γT −ρcθ log(θ /θ T ) + ρλT θ −θ T on
H .
θ T
(9.42)
gL (θ , σ ) − gH (θ , σ ) = γ T [σ − σ 0 (θ )] for (θ , σ ) ∈ LH , (9.43)
where
ρ λT θ − θT
σ 0 (θ ) = . (9.44)
γT θT
The stress σ 0 (θ ) at which the Gibbs free energies of the two phases coincide is the
Maxwell stress at the temperature θ . The formula (9.43) may be compared with its
counterpart for the trilinear, two-phase elastic material in the purely mechanical
theory of mixed phase equilibria; see equations (3.5) and (3.53). By appealing to
(9.31), (9.34), and (9.29)2,3 , one can show that
σ 0 (θ ) = 12 [σ M (θ ) + σ m (θ )], (9.45)
σ σ
(a) (b)
γ γ
0 0
Figure 9.4. Isothermal and isentropic stress–strain curves for a two-phase Mie–Grüneisen model based
on (9.22)1 and (9.20)1 , with W(γ ) of the form (9.46). (a) An isothermal stress–strain curve at the referential
temperature θ = θ T . (b) An isentropic stress–strain curve at the referential specific entropy η = η0 .
ψ̂ (γ, θ)
in
cr
ea
sin
g
θ
γ
0
Figure 9.5. Helmholtz free energy ψ̂(γ , θ ) of (9.47) as a function of strain γ at various temperatures θ ,
with temperature increasing in the indicated direction. For temperatures lower than the transformation
temperature θ T , the unsheared state has minimum energy, while for θ > θ T , the sheared states are
preferred.
In their studies of propagating phase boundaries, Turteltaub [18] and Ngan and
Truskinovsky [14] make use of Helmholtz free energies of the Mie–Grüneisen form
(9.21). In [18], the two-phase thermoelastic material is trilinear, whereas in [14],
W (γ ) is a quartic polynomial that characterizes a two-phase material. In both of
these works, the associated thermoelastic constitutive law is augmented by means
of viscosity and strain gradient effects, and the kinetic relation inherited by the
limiting sharp-interface theory through this augmentation is determined.
REFERENCES
10.1 Preliminaries
We begin by setting out the one-dimensional version of the theory of thermoelastic-
ity given in Chapter 7. Consider a tensile bar that occupies the interval [0, L] of the
x-axis in a reference configuration. Let the bar undergo a longitudinal motion, so
that the particle at x in the reference configuration is carried to the point x + u(x, t)
at time t and bears the temperature θ (x, t). It is assumed that u is continuous in
(x, t) with piecewise continuous first and second partial derivatives for all x in
[0, L] and for all times t of interest, and that θ(·, t) is piecewise continuous with
piecewise continuous partial derivative on [0, L] . The strain and particle velocity at
the particle x at time t are given by γ (x, t) = u x (x, t) and v(x, t) = u t (x, t) where
the subscripts denote partial differentiation.
163
164 EVOLUTION OF PHASE TRANSITIONS
For a thermoelastic material in one dimension, the nominal stress σ (x, t) and
the specific entropy η(x, t) are given by the thermoelastic constitutive law
σ = ρ ψγ (γ , θ), η = −ψθ (γ , θ ), (10.1)
where ψ(γ , θ ) is the Helmholtz free energy per unit mass, and ρ is the referential
mass density. The heat flux q(x, t) is taken to be given by the heat conduction law
q = K (γ , θ ) θ x , (10.2)
where K (γ , θ ) is the thermal conductivity. It is assumed that the material of the bar
is homogeneous in the reference state, so that ψ, ρ, and K do not depend explicitly
on x.
Wherever the fields are smooth, kinematic compatibility and the one-
dimensional versions of the local balance laws (6.4) and (6.6) require that
vx − γt = 0, (10.3)
σx = ρvt , (10.4)
qx = ρθηt , (10.5)
where we have neglected the effects of body force and heat supply. If there is a
discontinuity in the fields at x = s(t) consistent with the assumed smoothness, the
jump conditions (7.53)–(7.55) require that
[[γ ]] ṡ + [[v]] = 0, (10.6)
[[σ ]] + ρ[[v]] ṡ = 0, (10.7)
(ρ[[]] − σ [[γ ]])ṡ + [[q]] = 0. (10.8)
Finally, the entropy inequality is satisfied by taking K > 0 and requiring that
f ṡ ≥ 0 where f = [[ρψ]] − σ [[γ ]] + ρη [[θ ]] (10.9)
is the driving force; see (6.28) and (6.33).
Now consider the special case of thermomechanical equilibrium and suppose
that (i) the fields u and θ are independent of time and (ii) the temperature is con-
tinuous. Then v = 0, and by the thermoelastic constitutive law (10.1) and the heat
conduction law (10.2), all fields are independent of time. The field equations and
interface conditions (10.3)–(10.9) now require that the stress σ and the heat flux q
be uniform through the bar, and the driving force may be expressed as
f = [[ρψ]] − σ [[γ ]]. (10.10)
where the regions PL and PH in the γ , θ -plane correspond to the low- and high-
strain phases respectively; in the special case of a trilinear material these regions
are defined by (9.28). The corresponding stress and entropy are given by
σL (γ , θ) for (γ , θ ) ∈ PL ,
σ = ρψγ (γ , θ ) = (10.15)
σH (γ , θ ) for (γ , θ) ∈ PH ,
ηL (γ , θ ) for (γ , θ) ∈ PL ,
η = −ψθ (γ , θ) = (10.16)
η (γ , θ) for (γ , θ) ∈ P .
H H
For notational simplicity, we now drop the overbars from the prescribed elonga-
tion δ̄ and temperature θ̄ . Suppose that the prescribed temperature θ , and the stress
σ associated with a solution to the equilibrium problem, turn out to correspond to
a point in the coexistence region
LH . We now seek a mixed-phase solution of the
problem of the form
γ − for 0 < x < s,
γ (x) = θ(x) = θ for 0 < x < L , (10.18)
γ + for s < x < L ,
where the kinetic response function ( f, θ ) depends on the material. The entropy
inequality requires that
( f, θ ) f ≥ 0. (10.24)
Among the four quantities s, θ, σ , and δ entering (10.25), two will be prescribed
as functions of time (θ and δ, or θ and σ ), while the remaining two are to be
determined1 . In an isothermal constant-elongation-rate process of the kind of-
ten carried out in the laboratory, one specifies constant values for θ (t) and δ̇(t),
and σ (t) is to be found. On the other hand, in thermal cycling experiments, one
specifies constant values for θ̇ (t) and σ (t) (usually σ = 0), and one wishes to
find δ(t).
Finally we turn to nucleation. In the quasistatic processes to be encountered here,
the bar will be in the low-strain phase at the beginning of the thermomechanical
loading process, and will remain there until nucleation of the transition to the high-
strain phase takes place. As in the corresponding theory of quasistatic processes for
elastic bars in Chapter 3, we assume that the end x = 0 of the bar is the nucleation
site, and that the nucleation criterion specifies a critical value of driving force
f nLH (θ) ≥ 0:
After nucleation, the phase boundary moves to the right under the control of the
kinetic relation; if and when it arrives at the end x = L, the bar is fully transformed
to the high-strain phase. If the loading is subsequently reversed, nucleation of the
reverse transition occurs if f ≤ f nHL (θ ) ≤ 0; the two nucleation values of driving
force depend, in general, on the material and the temperature.
1
One can eliminate s from (10.25) to obtain a single master differential equation involving δ, σ , and
θ only.
168 EVOLUTION OF PHASE TRANSITIONS
gH (θ , σ ) = − 2µ
1
[σ +µα(θ − θ T )]2−σ γT − ρcθ log(θ /θ T )+ρλT (θ − θ T )/θ T ,
(10.28)
where γ T , λT are the transformation strain and latent heat, respectively, at the tem-
perature θ T . Observe that the two phases have equal Gibbs free energies when the
stress vanishes and the temperature is θ T : it follows that θ T denotes the transfor-
mation temperature. Moreover, since L (θ T , 0) = 0, the reference configuration is
stress free at the transformation temperature. From (10.22) and (10.28) we find that
the driving force has the simple expression
f = γ T [σ − σ 0 (θ )] , (10.29)
where the Maxwell stress σ 0 (θ ) is given by
ρ λT θ − θ T
σ 0 (θ ) = . (10.30)
γT θT
By using (10.29) and (10.27), one can specialize the system (10.25) to the case
of the trilinear material:
ṡ = (γ T (σ − σ 0 (θ )), θ ) ,
(10.31)
δ = γ T s + σ L/µ + α (θ − θ T )L
from which we may eliminate s to get a single master differential equation:
δ̇ = γ T (γ T (σ − σ 0 (θ )), θ ) + L σ̇ /µ + α L θ̇ . (10.32)
We shall discuss initial conditions for (10.32) for the various special loading pro-
grams to be described below.
Finally turning to nucleation, one may use (10.29) to express the nucleation
thresholds f nLH (θ ) and f nHL (θ ) in terms of critical levels of stress σ LH n
(θ ) and
σ n (θ), thinking of the temperature θ as given, or alternatively as critical lev-
HL
els of temperature θ LH n
(σ ) and θ HLn
(σ ), thinking of the stress σ as given. For
example
f nLH ρ λT θ − θ T f nLH
σ LH (θ ) = σ 0 (θ ) + = + , (10.33)
n γT γT θT γT
γ T σ − f nLH
θ n (σ ) = θ T 1 +
LH , (10.34)
ρλT
with similar formulas for σ HL
n
(θ ) and θ HL
n
(σ ).
QUASISTATIC HYSTERESIS IN TWO-PHASE THERMOELASTIC TENSILE BARS 169
Suppose for a moment that the latent heat λT is positive. Then according to
our earlier discussion, the low-strain phase is austenite and the high-strain phase is
martensite. Assume that the bar is in the stress-free reference configuration at the
transformation temperature θ T . With f nLH > 0, the nucleation temperature at zero
stress θ LH
n
(0) is necessarily less than θ T , so that cooling the bar sufficiently at zero
stress will cause martensite to nucleate.
ṡ
ṡ = Φ(f, θ)
θ = 0.4 θT θ = θT
For these choices of the material constants, which are consistent with the restrictions
imposed on the trilinear material by equations (9.32) and (9.34), the latent heat λT
is positive.
Figure 10.1, in which dimensionless phase boundary velocity ds /dt is plotted
against dimensionless driving force according to the kinetic relation (10.43) for
thermal activation, shows the effect of temperature on the kinetic response function
. According to the figure, increasing the temperature at a fixed level of driving
force decreases the phase boundary velocity, although changing the parameters
M and m can produce the opposite effect. In reporting experiments involving
172 EVOLUTION OF PHASE TRANSITIONS
σ σ
θ = 0.6
θ = 0.3
0
δ 0 δ
(a) (b)
σ
θ = 0
0 δ
(c)
Figure 10.2. Stress-cycling ofthe bar at three dimensionless temperatures: (a) θ = 0.6, (b) θ = 0.3, (c)
θ = 0, each cycle corresponding to the dimensionless elongation rate δ̇ = 0.7. The nucleation values
fnLH = 0.14 and fnHL = −0.14 were used.
1500
Stress (MPa)
1000
500
The shapes of the stress–elongation curves shown in Figure 10.2 are quali-
tatively similar to some – but not all – of those determined experimentally for a
variety of alloys; see, for example, Grujicic, Olson and Owen [4], Krishnan and
Brown [7], Otsuka, Sakamoto and Shimizu [11], Otsuka and Shimizu [12], Pops
[13], and Müller and Xu [10]. Permanent deformation of the kind illustrated by
Figure 10.2(c) is found to occur at low temperatures but found to be absent at high
temperatures in experiments on AgCd [7] and NiTi [12], but some experiments on
CuZnSn indicate that permanent deformation may appear at higher temperatures,
having been absent at lower ones [13]. That increasing temperature shifts the hys-
teresis loop in the direction of increasing stress with little effect on the magnitude
of the hysteresis itself seems to be consistent with experimental observations on
CuZnSi (Pops [13]), CuAlNi (Müller and Xu [10]), and NiTi (Otsuka and Shimizu
[12]), but inconsistent with experiments on CuZnSn (Pops [13]). Experiments of
Otsuka and Shimizu [12] suggest that the magnitude of the hysteresis may in-
crease or decrease with increasing temperature, depending on the type of martensite
involved.
Significant serrations of the kind that are commonly observed in experimental
hysteresis loops – see, for example [6] – are not present in those based on the
present model and shown in Figure 10.2. Serrations are associated with irregular
phase boundary motion and have been modeled by incorporating inhomogeneous
behavior in the kinetic relation [2, 14] or by utilizing nonmonotonic kinetic response
functions [14].
∆
σ = 0 ∆
−Θ 0 Θ
θ σ = 0.5
(a)
∆ −Θ 0 Θ
θ
(b)
σ = 1.0
−Θ 0 Θ
θ
(c)
Figure 10.4. Temperature-cycling of the bar at three different dimensionless values of stress
σ = 0, 0.5, 1.0, each cycle arising from the dimensionless cooling/heating rates θ̇ = ∓ 0.1.
period of cooling, the bar remains in the low-strain phase until the temperature
reaches the nucleation value θ LHn
(σ ) for the forward transition. Upon nucleation,
a phase boundary emerges at x = 0 and moves into the bar under the control of
the master equation (10.32), which now reduces to a differential equation relating
θ and , with the given stress and the cooling rate entering as parameters. The
cooling process is continued until somewhat beyond the time at which the bar is
fully transformed; heating is then commenced, with the stress remaining at the value
σ throughout the experiment. The reverse phase transition ultimately nucleates, the
bar eventually returns entirely to the low-strain phase, and the temperature returns
to its initial value.
Figure 10.4 results from numerical solutions of the initial-value problems for
the master equation that correspond to the thermal loading cycles just described;
the dimensionless variables (10.44) and material constants (10.45) were used, with
numerical values chosen as they were for Figures 10.1 and 10.2. Three values of
the given dimensionless stress are considered, and all three cases involve the same
cooling/heating rate. In each case, cooling begins at the point θ = , = 0 in
the θ , -plane and ends when the dimensionless temperature reaches the value
θ = −; the heating portion of the cycle ultimately returns the bar to the state with
QUASISTATIC HYSTERESIS IN TWO-PHASE THERMOELASTIC TENSILE BARS 175
σ
B
Figure 10.5. The shape-memory effect. The initial A
temperature is θ = 0; elongation takes place at the
initial temperature at the rate δ̇ = 1.0 along O AB, fol- O C
lowed by shortening at the rate δ̇ = −1.0 along BC. ∆
Unstressed heating at the rate θ̇ = 0.2 then occurs
along C D E, followed by unstressed cooling at the D
rate θ̇ = −0.2 along E O.
E
θ
nucleation levels of driving force have been chosen as f nLH = 0.14 and
f nHL = −0.14. The figure is qualitatively similar to the schematic description of
the shape-memory effect given by Krishnan et al. [8] in their Figure 11.
For the purpose of qualitative comparison, we show in Figure 10.7 the counter-
part of the experimental strain versus relative elongation graphs of Figure 10.6 as
calculated from the trilinear thermoelastic model with thermally activated kinetics.
The plot refers to the dimensionless quantities introduced in (10.44) and (10.45)
and uses the numerical values (10.46).
178 EVOLUTION OF PHASE TRANSITIONS
γ/γT
0
∆/L
In the experiment for which the evolution of the phase transition is described
in detail in Sections 4.3 of [16], Shaw and Kyriakides conclude that the forward
transition from austenite to martensite nucleates at both ends of the specimen. The
experimental setup permits the approximate tracking of moving phase boundaries;
the data are consistent with the assumption that the phase boundaries move with
constant velocities of 0.25 mm sec−1 for the forward transition and 0.32 mm sec−1
for the reverse transition from martensite to austenite. These speeds are an order of
magnitude faster than the velocity of the moving end grip.
It is reported in [15] that the number of experimentally observed nucleation
sites for the austenite-to-martensite transition increases as the elongation rate is
increased, with nucleation occurring at a single site only for very slow isothermal
experiments. The stresses required to nucleate both forward and reverse transitions
are reported as functions of temperature over the range 15–55◦ C in Figure 7 of
[16]. The observed variation of each of these nucleation stresses is consistent with
linear dependence on temperature, as predicted by the trilinear thermoelastic model
with temperature-independent nucleation levels f n of driving force; see (10.33) in
Section 10.4.
2
It may be noted that in a quasistatic process in the preceding sense, the energy equation (10.5) is not
satisfied since its left-hand side vanishes because of (10.2) but its right-hand side does not vanish
in general.
QUASISTATIC HYSTERESIS IN TWO-PHASE THERMOELASTIC TENSILE BARS 179
(K θx )x = ρcθt − mθ γt , (10.47)
where c = −θ ψθ θ is the specific heat at constant strain and m = ρψγ θ is the stress–
temperature coefficient; cf. (7.23). The related jump condition (10.8) can be written,
after using (10.2), (10.9), = ψ + θ η and [[θ ]] = 0, as
where λ = θ[[η]] is the latent heat associated with the transition from the phase on
the right of the phase boundary to the phase on the left; see (7.14). Observe that
the jump condition (10.48) describes a moving point source of heat, the magnitude
of which depends on the latent heat λ and the driving force f , and whose current
location is x = s(t). The basic problem can therefore be viewed as one involving a
moving heat source whose magnitude and propagation speed are not known a priori
but must be found by using the kinetic relation.
Suppose that the stress σ (t) is prescribed. The strain field can then be expressed
in terms of the unknown temperature field by using the constitutive relation (10.17);
the partial differential equation (10.47) can thus be written in a form that involves
the temperature field as the only unknown. Given suitable initial and boundary
conditions, this, together with the jump condition (10.48), can now be solved for
θ(x, t) and s(t), keeping in mind the kinetic relation (10.23) as well as (10.22).
In the uniaxial tension experiments on NiTi carried out by Leo, Shield, and
Bruno [9] and Shaw and Kyriakides [15, 16], they observed local heating near the
phase boundary when the extension rate was moderately fast. In order to model
such experiments, where the effects of thermal diffusion are important but inertial
effects are still negligible, one must use the theory outlined above, for example, see
Kim and Abeyaratne [5].
REFERENCES
11.1 Introduction
In Chapter 4 we studied one-dimensional models of dynamic phase transitions in
the purely mechanical theory of elastic materials. Our objective here is to extend the
ideas of that chapter to the dynamics of two-phase thermoelastic materials. Much
of the analysis will be directed to the materials of Mie–Grüneisen type introduced
in Chapter 9, with special emphasis on the trilinear thermoelastic material. As
in Chapter 4, we shall study an impact-induced phase transition that occurs in
compression, rather than in tension; this will require some minor modifications of
the details of the constitutive models presented in Chapter 9.
The subject of nonlinear wave propagation in solids has an enormous literature
encompassing both experimental and theoretical work. For a sample of background
references representing a variety of viewpoints in this field, the reader might consult
the classic work on gas dynamics of Courant and Friedrichs [4], the extensive review
article of Menikoff and Plohr [7], the discussion by Ahrens [2] of the experimental
determination of the “equation of state” of condensed materials, the work of Swegle
[9] on phase transitions in materials of geologic interest, and the theory of shock
waves in thermoelastic materials presented by Dunn and Fosdick [5].
In the next section, we set out the basic field equations and jump conditions
of the dynamical theory of thermoelasticity when the kinematics are those of uni-
axial strain and the processes are adiabatic. We specialize these results to materi-
als of Mie–Grüneisen type, focusing on two-phase trilinear materials involving a
low-pressure phase (LPP) and a high-pressure phase (HPP). In Section 11.3, we
formulate the impact problem for a half-space composed of one of these materials,
and we discuss the form of its scale-invariant solutions. For a trilinear material, we
first construct solutions corresponding to the absence of a phase transition. These
involve a shock wave when the impact is compressive, which is the usual case;
for sudden tensile loading, the response involves a fan. When the impact causes a
phase transition, we first study solutions in the degenerate special case for which the
Grüneisen parameter β vanishes. We then briefly discuss the qualitative departures
from these solutions that arise when β is small.
181
182 EVOLUTION OF PHASE TRANSITIONS
vx − γ t = 0, (11.1)
σ x − ρvt = 0, (11.2)
σ γ t − ρ t = 0. (11.3)
ηt ≥ 0 (11.7)
DYNAMICS OF PHASE TRANSITIONS IN UNIAXIALLY STRAINED 183
where the fields are smooth; according to (6.35), the driving force at a strain dis-
continuity in an adiabatic process is
f = −ρ θ [[η]], (11.8)
f ṡ ≥ 0 (11.9)
θ∗
γ = γM(θ) γ = γm(θ)
PH PL
γ
−1 γ∗ 0
Figure 11.1. The regions in the γ , θ -plane corresponding to the low- and high-pressure phases, LPP
and HPP, of the trilinear material associated with (11.11) and (11.12). The regions PL and PH are defined
by (9.28) and the functions γ M (θ ) and γ m(θ ) are given by (9.29)2 .
axial component of stress and the specific entropy are now given by
μγ − ρcβ(θ − θ T ) on PL ,
σ = (11.11)
μ(γ + γ T ) − ρcβ(θ − θ T ) on PH ,
βcγ + c (1 + log(θ /θ T )) on PL ,
η= (11.12)
βc(γ + γ T ) + c (1 + log(θ /θ T )) − λT /θ T on PH ,
where the Grüneisen parameter β is defined in terms of the coefficient of ther-
mal expansion α, the modulus μ, the mass density ρ, and the specific heat c by
β = αμ/(ρc), cf. (9.30), (9.35), and (9.9). The regions PL and PH of the strain,
temperature plane corresponding to the low-pressure-phase (LPP) and the high-
pressure-phase (HPP) are shown in Figure 11.1. The corresponding specific internal
energy = ψ + ηθ is given by
2
μγ /(2ρ) + cβθ T γ + c θ on PL ,
= (γ
¯ , θ) = (11.13)
μ(γ + γ T )2 /(2ρ) + cβθ T (γ + γ T ) + c θ on PH .
In order to determine the potential ˆ (γ , η) one would solve (11.12) for θ as a
function of γ and η and substitute the result into (11.13).
A moving strain discontinuity at x = s(t) is a shock wave if the states on both
sides belong to the same material phase, so that the points (γ + (t), θ + (t)) and
(γ − (t), θ − (t)) are either both in the region PL or both in PH . If the points are
in different regions, the discontinuity is a phase boundary. For the trilinear mate-
rial, explicit formulas for the driving force at a strain discontinuity can be readily
found by using (11.12) to specialize (11.8). This leads, for a shock wave, to
f = −ρcθ β(γ + − γ − ) + log(θ + /θ − ) (11.14)
vx − γ t = 0, (11.16)
a 2 γ x − cβθ x − vt = 0, (11.17)
βγ t + θ t /θ = 0, (11.18)
By (11.14), the entropy inequality (11.9) at a shock wave in either phase becomes
At a phase boundary with LPP on the right, the jump conditions (11.4)–(11.6)
take the forms
[[v]] + ṡ[[γ ]] = 0, (11.26)
a 2 ([[γ ]] − γ T ) − cβ[[θ ]] + ṡ[[v]] = 0, (11.27)
−γ T a (γ + γ T /2) + cβ θ [[γ ]] + c[[θ ]] + λT − cβγ T θ T ṡ = 0, (11.28)
2
11.3.2 Solutions with no phase transition. For sufficiently small impactor veloci-
ties v0 , one expects the target to remain in LPP. We accordingly seek first a scale-
invariant solution of the impact problem in which a shock wave moves into the
unstrained material at rest at the temperature θ T . Thus we write
⎧ − ⎧ ⎧ −
⎨ γ , 0 < x < ṡt, ⎨ v0 , 0 < x < ṡt, ⎨ θ , 0 < x < ṡt,
γ = , v= , θ=
⎩ ⎩ ⎩
0, x > ṡt, 0, x > ṡt, θ T , x > ṡt,
(11.33)
DYNAMICS OF PHASE TRANSITIONS IN UNIAXIALLY STRAINED 187
F(ξ) F(ξ)
1 1
2 γT βw0 2 γT βw0
0 0 ξ
ξ
1 (a) 1 (b)
Figure 11.2. The function F (ξ ) that determines the dimensionless shock wave speed ξ from (11.36).
(a) w0 < 2/(γ T β) and (b) w0 > 2/(γ T β).
and seek to determine the three unknowns γ − , θ − , and ṡ from the three jump
conditions (11.22)–(11.24) with the entropy inequality (11.25) enforced. It is useful
to note first that the three homogeneous equations (11.22)–(11.24) for the jumps
[[γ ]], [[v]], and [[θ ]] have nontrivial solutions if and only if
ṡ = a 1 + (cβ 2 /a 2 )θ , (11.34)
so that necessarily the shock wave speed ṡ exceeds the isothermal wave speed
a. Eliminating θ − and γ − from the three jump conditions provides an equation
for the shock wave speed ṡ. To describe the results, it is convenient to introduce
dimensionless quantities by writing
c v0 ṡ
T = θ, w0 = , ξ= . (11.35)
γ 2T a 2 γTa a
Then one finds that the dimensionless shock wave speed ξ is given by the roots of
the equation
(ξ 2 − 1)(ξ − 12 γ T β w0 )
F(ξ ) = γ 2T β 2 TT where F(ξ ) = ; (11.36)
ξ
as depicted in Figure 11.2, the function F(ξ ) in (11.36)2 is convex for ξ > 0. Once
ξ has been found from (11.36), the strain γ − and the dimensionless temperature
T − behind the shock are given by
w0 ξ + 12 γ T β w0
γ − = −γ T , T− = TT . (11.37)
ξ ξ − 12 γ T β w0
Note that, since the dimensionless impactor velocity w0 is positive, (11.37) shows
that the strain γ − is compressive, and the temperature behind the shock is higher
than that in front. √
For small dimensionless impactor velocity w0 , (11.36) gives ξ ∼ 1 + 2TT ,
which, in dimensionless terms, is the wavespeed of linearized isentropic thermoe-
lastic waves in uniaxial strain.
Graphs of the relation (11.36) between dimensionless phase boundary speed
ξ = ṡ/a and dimensionless impactor velocity w0 = v0 /(γ T a) for three different
dimensionless initial temperatures TT = cθ T /(γ 2T a 2 ) are shown in Figure 11.3.
188 EVOLUTION OF PHASE TRANSITIONS
w0
0 ξ
For future purposes, we record the form taken by the shock-wave solution de-
scribed in (11.33) and (11.35)–(11.37) in the special case β = α = 0:
ξ = ṡ/a = 1, γ − = −γ T w0 = −v0 /a, T − = TT , (β = 0). (11.38)
To assure that the solution to the impact problem just constructed involves only
the LPP of the material, one must require that the point (γ − , θ − ) lies in the region
PL in the strain–temperature plane, so that
γ − > γ m (θ − ) = γ ∗ + m(θ − − θ ∗ ), 0 < θ − < θ ∗; (11.39)
see (9.29)2 and Figure 11.1. Ultimately, this requirement imposes an upper limit
on the impactor velocity, which depends on the many material parameters entering
the theory. The form taken by this upper limit in the special case β = α = 0 is
γT 1
v0 /a < − (M − m)(θ ∗ − θ T ). (11.40)
2 2
When the impactor velocity v0 is negative, producing tension, there is still a
formal solution consisting of a single shock wave, but it can be shown to violate the
entropy inequality. The proper solution in this case can be constructed from the fan
(11.31) and (11.32); such a construction turns out to be possible only when v0 < 0.
11.3.3 Solutions with a phase transition. When the impact problem results in a
phase transition, one typically confronts a propagating phase boundary, at which
the jump conditions (11.26)–(11.28) and the entropy inequality (11.29) must hold,
and a shock wave, at which (11.22)–(11.24) and (11.25) must be satisfied. Ana-
lytical construction of a piecewise constant, closed-form solution fulfilling these
requirements as well as the initial and boundary conditions is not practical, but
most of the qualitative features of such a solution can be appreciated by looking at
the special case in which the Grüneisen parameter β = αa 2 /c is small, which we
study by perturbing about the special case β = α = 0.
DYNAMICS OF PHASE TRANSITIONS IN UNIAXIALLY STRAINED 189
a 2 γ x − vt = 0, vx − γ t = 0, (11.41)
θ t = 0, (11.42)
where the fields are smooth. At a shock wave, the jump conditions (11.22) and
(11.23) for compatibility and momentum with β = 0 are
[[θ ]] ṡ = 0; (11.44)
for compatibility and momentum, and the jump condition (11.28) for energy balance
becomes
c[[θ ]] − γ T a 2 [γ + γ T /2] + λT ṡ = 0. (11.47)
In this special case β = 0, the field equations and jump conditions that arise
from momentum balance and compatibility - (11.41), (11.43), and (11.46) - are
precisely those for a trilinear elastic material in the purely
√ mechanical setting of
Chapter 4, with shock waves traveling at the speed a = μ/ρ. The consequences
(11.42) and (11.44) of energy balance, along with the entropy inequality (11.45) at
a shock, show that, when β = 0, temperature enters only trivially unless a phase
boundary is present. If there is a phase boundary, the associated energy balance
(11.47) does involve coupling of thermal and mechanical quantities.
When a phase transition is present, we seek a solution in which the leading
disturbance is a shock wave propagating into undisturbed material, followed by a
phase boundary. Thus we take
⎧ −
⎪
⎪ γ , v0 , θ − for 0 ≤ x < ṡ t,
⎨
γ , v, θ = γ + , v + , θ T for ṡ < x < a t, (11.49)
⎪
⎪
⎩
0, 0, θ T for x > a t,
190 EVOLUTION OF PHASE TRANSITIONS
x = ṡt
HPP
γ = γ −, v = v0 , LPP
θ = θ−
x = at
γ = γ + , v = v + , θ = θT
LPP
γ = 0, v = 0, θ = θT
x
0
Figure 11.4. The x, t-plane for a solution of the impact problem in which a phase transition occurs
(special case β = 0).
where ṡ is the phase boundary velocity; see Figure 11.4. Upon enforcing the jump
conditions (11.43), (11.46), and (11.47), one finds
v0 a
γ− = − − γT , (11.50)
a a + ṡ
λT aγ T a 2 γ 2T ṡ 2 − 2a ṡ
θ− = θT + + v0 + , (11.51)
c c 2c a 2 − ṡ 2
v0 a ṡ
γ+ = − + γT 2 , (11.52)
a a − ṡ 2
a 2 ṡ
v + = v0 − γ T 2 , (11.53)
a − ṡ 2
where ṡ is the as yet undetermined phase boundary velocity.
The entropy inequality (11.29) becomes
λT λT γ T a2 v0 ṡ 2 − 2a ṡ
− log 1 + + 2 +γT 2 ≤ 0. (11.54)
cθ T cθ T 2cθ T a a − ṡ 2
Using (11.50)–(11.53) in (11.49) furnishes a one-parameter family of phase-
changing solutions to the impact problem in the case β = 0. The as yet undetermined
phase boundary velocity ṡ and the given impactor velocity v0 must not only satisfy
the entropy inequality (11.54), but they are subject to the inequalities that follow
from the phase segregation requirements and the limits on the range of θ − :
γ + > γ m (θ T ) = γ ∗ + m(θ T − θ ∗ ),
−1 < γ − < γ M (θ − ) = γ ∗ + M(θ − − θ ∗ ), (11.55)
0 < θ − < θ ∗.
DYNAMICS OF PHASE TRANSITIONS IN UNIAXIALLY STRAINED 191
v0/(aγT)
When (11.50)–(11.53) are used in (11.55), the resulting inequalities restrict the
dimensionless velocities v0 /(γ T a) and ṡ/a and involve dimensionless parameters
given by
The restrictions may be described in a plane in which ṡ/a and v0 /(γ T a) are
the coordinates, as indicated, for example, in Figure 11.5. In constructing the
curves in the figure, the four dimensionless parameters in (11.56) have been
given the values N = 1, n = −3, TT = 0.01, and T∗ = 0.2. The shaded region
in the figure represents, in their dimensionless versions, the pairs ṡ, v0 that fulfill
the three requirements γ + > γ m (θ T ), θ − < θ ∗ and the entropy inequality f ṡ ≥ 0.
The limitations imposed by the additional inequalities γ − > γ M (θ − ) and θ − > 0
require admissible points to lie above two curves, not shown in the figure, that
themselves lie everywhere below curve A ( f = 0) that forms the lower boundary
of the shaded region; therefore these additional inequalities impose no relevant
restrictions. The curve (not shown) in the plane of Figure 11.5 on which γ − = −1
actually cuts off the shaded region at very large values of v0 , corresponding to
extreme compression, a regime in which the trilinear model is likely to be wholly
unrealistic. A similar caveat applies to the limitation imposed by the portion of
the upper boundary of the shaded region that lies above the point of intersection
of curves B and C. This limitation arises when the temperature θ − behind the
phase boundary approaches the critical temperature θ ∗ , another regime in which
the trilinear model is not to be trusted.
192 EVOLUTION OF PHASE TRANSITIONS
Let us return for a moment to the solution (11.38) to the impact problem without
a phase transition in the special case β = α = 0. The restriction (11.40) on the
impactor velocity v0 assuring that the state behind the shock wave is in the LPP
corresponds precisely to the portion of the vertical axis lying below the point Q and
shown bold in Figure 11.5.
As the given impactor velocity v0 is increased from zero, the shock-wave solu-
tion with the specimen remaining in the LPP is the only possible response to impact
until v0 reaches the value corresponding to the point P in Figure 11.5. For values
of v0 between P and Q, the no-phase-change solution and a one-parameter family
of solutions with a phase boundary are both possible. As in the purely mechan-
ical case, the nucleation criterion chooses between these two types of solutions.
When it determines that a phase transition must occur, the kinetic relation then
selects a particular member of the one-parameter family of solutions with a phase
transition. Based on the general discussion in Section 8.2, one form for a kinetic
relation is
ṡ = ( f /θ), (11.57)
where f is the driving force acting on the phase boundary, θ is the average of the
temperatures on either side, and the kinetic response function depends on the
material. To assure that the second law of thermodynamics is satisfied at the phase
boundary, one requires that satisfy (z)z ≥ 0 for all real z. If (z) increases
monotonically with z, (11.57) can be inverted to give the kinetic relation in the
form
f = θ φ(ṡ). (11.58)
By inserting the representations (11.50)–(11.52) for γ − , γ + , and θ − into
(11.15), one can find the driving force on the phase boundary. Using this and
(11.51) in (11.58) yields the relation between the given impactor velocity v0 and
the unknown phase boundary velocity ṡ needed to complete the solution of the
impact problem when a phase transition occurs:
v0 cθ T
= G(ṡ) + exp(l) {exp [φ(ṡ)/(ρc)] − 1} (β = 0), (11.59)
a γ T a2
where l = λT /(cθ T ) is the dimensionless latent heat, and
cθ T l γ
G(ṡ) = (e − 1 − l) + T (2a ṡ − ṡ 2 )/(a 2 − ṡ 2 ). (11.60)
γTa 2 2
The curve A that forms the lower boundary of the shaded admissible region in
Figure 11.5 is described by the equation v0 /(aγ T ) = G(ṡ). It can be shown that
the right side of (11.59) increases monotonically with ṡ; one must of course re-
strict φ(ṡ) so that the curve represented by (11.59) lies in the shaded region of the
figure. Specifying a value for the impactor velocity v0 will then lead to a uniquely
determined phase boundary speed ṡ.
DYNAMICS OF PHASE TRANSITIONS IN UNIAXIALLY STRAINED 193
REFERENCES
[7] R. Menikoff and B.J. Plohr, The Riemann problem for flow of real materials.
Reviews of Modern Physics, 61 (1989), pp. 75–130.
[8] M.A. Meyers, Dynamic Behavior of Materials. Wiley-Interscience, New York,
1994.
[9] J.W. Swegle, Irreversible phase transitions and wave propagation in silicate
geologic materials. Journal of Applied Physics, 68 (1990), pp. 1563–79.
Part V Higher Dimensional
Problems
12 Statics: Geometric Compatibility
12.1 Preliminaries
In this chapter we consider the simplest possible static problem in three dimensions
that involves a phase boundary, that is, a continuous, piecewise homogeneous,
two-phase deformation. Despite its simplicity, this problem serves to illustrate a
number of phenomena that do not occur in one dimension. In particular, we will
demonstrate through a series of examples that the existence of such deformations in
a crystalline solid depends sensitively on the lattice parameters. In some materials
such deformations only exist for special values of the lattice parameters and in
others, they cannot exist at all. The reader may consult the book by Bhattacharya
[4] for a complete discussion of the material in this Chapter.
Consider a thermoelastic body that occupies all of R3 , and suppose that it is
subjected to a piecewise homogeneous deformation
+
F x for x · n ≥ 0,
y= (12.1)
F− x for x · n ≤ 0,
where F± are constant tensors belonging to two distinct phases or variants. The
planar interface S = {x : x · n = 0} corresponds to the referential image of a phase
or twin boundary and the unit vector n is normal to this interface.
Suppose that the temperature θ of the body is uniform. Suppose further that
the deformation gradient tensors F± lie at the bottoms of two energy wells of the
potential ψ(F, θ ). Then the corresponding stresses vanish: σ ± = ρψF (F± , θ ) = 0.
Finally, suppose that the body is at the transformation temperature, θ = θT , so that
ψ(F+ , θT ) = ψ(F− , θT ). Then, the driving force on the phase boundary also van-
ishes: f = ρ(ψ(F+ , θT ) − ψ(F− , θT )) − σ ± · (F+ − F− ) = 0. Consequently the
body is in thermal, mechanical, and phase equilibria.
All that remains is to enforce geometric compatibility across the phase boundary.
The deformation gradient tensor suffers a jump discontinuity across S but suppose
that the deformation itself is continuous. This requires that
F+ = F− (12.2)
197
198 EVOLUTION OF PHASE TRANSITIONS
for all vectors that are tangent to the phase boundary, which in turn holds if and
only if there is a nonzero vector a such that
F+ = F− + a ⊗ n, a = 0; (12.3)
(i) under what conditions on U± does the compatibility condition (12.3) hold for
some vectors a, n, and some rotation tensors R± with F± = R± U± ? and
(ii) when (12.3) does hold, find a, n, and R± .
It is convenient to first rephrase these issues in forms that do not involve the
rotation tensors. Observe that if a, n, R+ , R− satisfy (12.3) for some U+ and U− ,
then so do Qa, n, QR+ , QR− for any rotation Q. Thus with no loss of generality
we can pick one of the rotation tensors R± arbitrarily and so we take R− = I. Next,
let D, b, and m be defined by
−1 −T −T −T
D = F+ (F− ) , b = |(F− ) n|a, m = (F− ) n/|(F− ) n|. (12.4)
The two issues presented above can now be posed in the following equivalent form:
given two distinct stretch tensors U+ and U− , define the symmetric positive definite
tensor C by (12.5) and then (i) find conditions on C under which (12.6) holds for
some nonnull vectors b and m and (ii) when these conditions hold, find b and m.
It can be readily verified that, once b and m have been found, R+ determined from
−1
D = F+ (F− )−1 = R+ U+ (U− ) = I + b ⊗ m, together with F+ = R+ U+ , F− =
R− U− , and R− = I, allow n and a to be determined from (12.4)3,2 and that the
resulting set {F+ , F− , a, n} satisfies the original compatibility condition (12.3).
The following proposition answers the rephrased questions just posed:
Proposition: Ball and James [2]. Let 0 < λ1 ≤ λ2 ≤ λ3 be the ordered eigenvalues
of C (= I) and let e1 , e2 , e3 be a set of corresponding orthonormal eigenvectors.
Then a necessary and sufficient condition for there to exist vectors b (= 0) and m
satisfying (12.6) is that
λ1 ≤ 1, λ2 = 1, λ3 ≥ 1. (12.7)
STATICS: GEOMETRIC COMPATIBILITY 199
ao
ea3
(a)
ea2 ea1
ao
ao
c
em
3
(b)
em
2 em
1 a
a
Figure 12.1. (a) Face-centered cubic lattice with lattice parameters ao × ao × ao and lattice vectors
{ea1 , ea2 , ea3 } and (b) face-centered tetragonal lattice with lattice parameters a × a × c and lattice vectors
{em1 , e2 , e3 }. The unit vectors {c1 , c2 , c3 } are associated with the cubic directions. Solely for clarity,
m m
the atoms at the vertices are depicted by (small) filled circles while those at the centers of the faces
are shown by open circles.
When (12.7) holds, there are two pairs of vectors b, m for which (12.6) holds and
they are given by
λ (1 − λ ) λ1 (λ3 − 1)
b =d
3 1
e1 + κ e3 ,
λ3 − λ1 λ3 − λ1
√ √ (12.8)
1 λ3 − λ1
m= √ − 1 − λ1 e1 + κ λ3 − 1 e3 ,
d λ3 − λ1
where d is chosen to make m a unit vector and κ = ±1. For a proof of these results
the reader is referred to1 [2].
We now apply this result to a number of examples, all of which except the
final one concern a material possessing a cubic austenite phase and a tetragonal
martensite phase (as considered previously in Section 5.5). Suppose that the cubic
lattice has lattice parameters ao × ao × ao , and let {c1 , c2 , c3 } be a set of orthonormal
vectors in the cubic directions; see Figure 12.1. The tetragonal lattice, with lattice
1
The analysis of this proposition in [2] is framed within the larger setting of sequences of energy
minimizing deformations. For a purely geometric analysis, see Bowles and Mackenzie [5], and
Wechsler Lieberman and Read, [11].
200 EVOLUTION OF PHASE TRANSITIONS
There are three distinct ways in which to carry out this transformation corresponding
to three variants of martensite characterized by stretch tensors U1 , U2 , U3 :
Since the reference configuration has been taken to coincide with unstressed austen-
ite, it corresponds to the deformation gradient tensor I.
In all of the examples below, with the exception of the last one, we shall not
explicitly write out the solutions b, m, a, n, R+ . We will focus instead on the im-
plications of the condition (12.7) for piecewise homogeneous deformations. In
particular, the examples will illustrate how (12.7) restricts the possibilities, in some
cases disallowing such deformations, in others allowing them always, and in yet
others, allowing them under special circumstances.
12.2 Examples
Example 1: Two variants of martensite. First consider a piecewise homogeneous
deformation involving two of the three martensitic variants, say U1 and U2 . Now
F+ = R+ U2 and F− = U1 and so from (12.5), C = U−1 2 −1
1 U2 U1 . The components
of C are found using (12.11) to be
2 2
η1 /η3 0 0
C = 0 η32 /η12 0 , (12.12)
0 0 1
whence the ordered eigenvalues of C are η12 /η32 , 1, η32 /η12 (if η1 < η3 ), or
η32 /η12 , 1, η12 /η32 (if η3 < η1 ). Thus the requirements (12.7) are automatically
satisfied thereby guaranteeing the existence of a piecewise homogeneous defor-
mation involving two variants. A deformation involving two martensitic variants
is called a twin or twinning deformation; the interface between the variants is a
twin boundary.
Example 2: Austenite and one variant of martensite. Next consider the possi-
bility of a piecewise homogeneous deformation involving austenite on one side
STATICS: GEOMETRIC COMPATIBILITY 201
of the interface and one variant of martensite, say U1 , on the other side. Thus
we take F+ = R+ U1 and F− = I and find from (12.5) that C = U21 . Thus from
(12.11) we have
2
η3 0 0
0 η2 0
C =
1 .
(12.13)
0 0 η12
Example 3: Austenite and twinned martensite. (Ball and James [2]) Let us now
consider the deformation described schematically in Figure 12.2 in which there is
an “interface” which separates austenite from martensite. The martensitic half-
space m · x < 0 involves bands of martensite, which alternate between variants
−1 and −2. Thus the martensitic region is twinned, and Example 1 describes
its kinematics. The deformation gradient tensors in the martensitic layers are
202 EVOLUTION OF PHASE TRANSITIONS
H
N
[h(η32 − η12 ) − η32 ]2 + η12 η32 [h(η32 − η12 ) + η12 ]2 + η12 η32
C= c 1 ⊗ c 1 + c2 ⊗ c2
η12 + η32 η12 + η32
(12.19)
(η2 − η12 )2
± 32 h(1 − h) (c 1 ⊗ c 2 + c 2 ⊗ c 1 ) + η 2
1 3c ⊗ c 3 ,
η3 + η12
which expresses C in terms of the lattice parameters η1 , η3 , the orthonormal
cubic basis vectors {c1 , c2 , c3 }, and the unknown volume fraction h; the ± here
arises from the fact that the twinning equation (12.17)1 has two solutions. Setting
det(C − I) = 0 now leads to a quadratic equation for the volume fraction h whose
roots are
2 2 2
1 2 η1 − 1 η3 − 1 η3 + η1 2
h = 1 ± 1 + 2 2 . (12.20)
2 η3 − η12
Note that the sum of these two roots is unity. When (12.20) holds, the eigenvalues
of C are found to be 1, η12 , η12 η32 . The necessary and sufficient condition (12.7),
together with the inequality that ensures that (12.20) yields real roots in the
interval [0, 1], can now be shown to be equivalent to
either η1 < 1 < η3 and η1−2 + η3−2 ≤ 2,
(12.21)
or η3 < 1 < η1 and η12 + η32 ≤ 2.
Thus in summary, an austenite-twinned martensite deformation of the assumed
form (Figure 12.2) exists provided that the lattice parameters obey (12.21) and the
variant volume fraction is given by (12.20). Table 7.1 of Bhattacharya [4] com-
pares these results with experimental observations for a number of alloys possess-
ing cubic austenite and tetragonal martensite; the agreement is seen to be good.
There are five types of interfaces to be considered: two sets of twin boundaries,
the two austenite–martensite boundaries, and the mid-rib. Enforcing geometric
compatibility across these interfaces leads to the following five conditions:
martensite/martensite : A − B = a1 ⊗ n1 ,
martensite/martensite : C − D = a2 ⊗ n2 ,
The first two of these are twinning equations and we know from Example 1
that they are solvable without the need for any restrictions. The next two equa-
tions are precisely of the type analyzed in Example 3. We know that these are
solvable provided the volume fractions h 1 and h 2 of the variants are suitably cho-
sen (and the lattice parameters satisfy some mild inequalities). This completely
determines all of the free parameters in the problem. The solvability of the final
equation therefore imposes a relationship between the lattice parameters, which
Bhattacharya [3] has shown to be
2
1 − η32 + 4η32 1 + η32
η1 =
2
2 . (12.23)
1 − η32 + 8η34
STATICS: GEOMETRIC COMPATIBILITY 205
(i) (ii)
(iii)
Figure 12.4. The cubic lattice (i), transformed into an orthorhombic lattice (ii), which in turn is
transformed into a monoclinic lattice (iii).
Suppose that the cubic austenite and the monoclinic martensite coexist with a
planar interface between them. Then we know from our preliminary discussion
that there must exist a rotation tensor R and vectors b and m such that
RKU = I + b ⊗ m. (12.27)
Given the stretch tensor U and the shear tensor K, one can find a rotation tensor
R and vectors b and m if and only if the eigenvalues λ1 , λ2 , λ3 of (KU)T KU obey
(12.7). It can be readily shown that the condition λ2 = 1 in (12.7) is satisfied
only if the amount of shear k in (12.26) is related to the stretches ηi in (12.25)
by
η12 − 1 η32 − 1
k= . (12.28)
η32
Thus if a piecewise homogeneous two-phase deformation is to be possible, the
condition (12.28), which is a restriction on the geometry of the monoclinic lattice,
must hold. If in addition to (12.28) the transformation stretches are such that
either η2 ≤ 1, η1 η3 ≥ 1, or η2 ≥ 1, η1 η3 ≤ 1, (12.29)
one can show that all requirements in (12.7) then hold. Thus, a necessary and
sufficient condition for homogeneously deformed cubic austenite and monoclinic
martensite to coexist across a planar interface is that (12.28) and (12.29) hold.
REFERENCES 207
According to (12.8), (12.30), and (12.31), the unit normal m to the austenite–
martensite interface is given by
1 − η22 η12 η32 − 1
m= p − κ p, (12.32)
η12 η32 − η22 1 η12 η32 − η22 3
where κ = ±1.
REFERENCES
[5] J.S. Bowles and J.K. Mackenzie, The crystallography of martensite transfor-
mations. Acta Metallurgica Materialia, 2 (1954), pp. 129–47.
[6] K.A. Bywater and J.W. Christian, Martensitic transformations in titanium–
tantalum alloys. Philosophical Magazine, 25 (1972), pp. 1249–73.
[7] J.C. Escobar, Plate Impact Induced Phase Transformations in Cu–Al–Ni
Single Crystals. Ph.D. dissertation, Brown University, Providence, RI, 1995.
[8] J.C. Escobar and R.J. Clifton, On pressure–shear plate impact for studying
the kinetics of stress-induced phase transformations. Journal of Materials
Science and Engineering, A170 (1993), pp. 125–42.
[9] J.C. Escobar and R.J. Clifton, Pressure–shear impact-induced phase transi-
tions in Cu-14.4 Al-4.19 Ni single crystals. SPIE, 2427 (1995), pp. 186–97.
[10] R.D. James and K.F. Hane, Martensitic transformations and shape memory
materials. Acta Materialia, 48 (2000), pp. 197–222.
[11] M.S. Wechsler, D.S. Lieberman, and T.A. Read, On the theory of the forma-
tion of martensite. Transactions of the AIME Journal of Metals, 197 (1953),
pp. 1503–15.
13 Dynamics: Impact-Induced Transition
in a CuAlNi Single Crystal
13.1 Introduction
In this chapter we determine the kinetic law for a particular material directly from
experimental data. The material is a single crystal of a CuAlNi shape memory
alloy and the experiments were conducted by Escobar and Clifton [3–5] using
impact loading to induce a transformation from cubic austenite to monoclinic β1 -
martensite.
In their experiments Escobar and Clifton used thin platelike specimens, which
were subjected to impact loading on one face with the resulting particle velocity
measured on the opposite face. The orientation of the lattice relative to the specimen
was carefully arranged so as to lead to the propagation of a phase boundary that
was parallel to the faces of the specimen. In addition, the magnitude and direction
of impact, which involved both normal and shear components, were chosen so as
to induce the desired transformation.
The material tested was a single crystal of Cu–14.44Al–4.19Ni (wt%). This
shape memory alloy involves an austenite phase that is cubic, one martensite phase
(called γ1 ) that is orthorhombic, and a second martensite phase (called β1 ) that is
monoclinic; see Example 5 in Section 12.2. Depending on the stress and tempera-
ture, a different one of these phases is energetically favorable. A schematic phase
diagram for this material showing the preferred phases at different temperatures
and stresses was shown previously in Figure 5.1. In the experiments described in the
present chapter the material transforms from austenite to β1 -martensite. In Chap-
ter 14 we will be concerned with some quasistatic experiments at very low stress
on (essentially) this same alloy, this time involving γ1 -martensite.
In order to determine the kinetic law we have to determine the driving force
and the propagation speed of the phase boundary. However the impact generates
three elastic waves, which propagate well ahead of the phase boundary. Since the
measurement of particle velocity during the experiment could only be made for
some small time after impact, the measurements in fact ended well before the
phase boundary reached the back face of the specimen. Thus the measurements do
not allow one to directly determine the arrival time of the phase boundary at the
209
210 EVOLUTION OF PHASE TRANSITIONS
back face of the specimen or the particle velocity after its passage. Both of these
quantities are needed in order to determine the phase boundary propagation speed
and the driving force. Consequently an analysis needs to be carried out in order to
infer these quantities from the measured data.
In an unstressed state, austenite and martensite are associated with the bottoms
of their respective energy wells. In our analysis we assume that during the mo-
tion, the deformation gradient tensor at any point in the specimen does not depart
significantly from this state. The deformation gradient tensor is permitted to jump
from one energy well to the other, though once it is associated with a particular
energy well it is assumed to be near its bottom. This allows us to linearize the field
equations for each phase, though the mathematical problem is still nonlinear since
the two domains occupied by the two phases are separated by a moving interface at
which the appropriate jump conditions hold. This perturbation analysis is valid for
impact velocities that are small compared with the smallest acoustic speed asso-
ciated with wave propagation normal to the specimen face. Since the temperature
changes during the experiment are thought to be relatively unimportant, the analysis
is carried out in an isothermal setting.
Details concerning the experiments can be found in Escobar [3], Escobar and
Clifton [4, 5]. They also develop a visco-plastic model that captures many features
of their observations. The material in the present chapter is an abbreviated version
of the work described in Abeyaratne and Knowles [2].
In Section 13.2 we set out some preliminary information pertaining to the ma-
terial and the motion of the specimen. The loading is characterized by a velocity
v0 = εw where w is a vector of constant (but not unit) magnitude that charac-
terizes the direction of impact, and the nondimensional parameter ε denotes the
ratio between the speed of impact and the smallest relevant acoustic speed. In or-
der to describe some of the mathematical details in the simplest possible setting,
in Section 13.3 we formulate and solve the impact problem when the loading does
not induce a phase transformation. Then in Section 13.4 we analyze the problem
with phase transformation. The analysis up to this point is not limited to cubic
austenite and monoclinic martensite. In the final section of this chapter we special-
ize and apply the results to Escobar and Clifton’s experiments and finally determine
the kinetic relation associated with the austenite → β1 -martensite transformation
in the material they tested. Many details are not described here and the reader is
referred to [2] for these. We omit, for example, issues such as how the analysis for
a half-space is used to model a thin plate; how the experiment, which involved the
impact of one CuAlNi plate on another, can be modeled by the application of a
constant velocity on a single plate; and so on.
13.2 Preliminaries
Consider a two-phase material and take the reference configuration to be one com-
posed of unstressed austenite, which is thus characterized by F = I. Let F = F0 de-
note one particular variant of martensite, also in an unstressed state. By assumption,
IMPACT-INDUCED TRANSITION IN A CuAlNi SINGLE CRYSTAL 211
austenite and martensite both correspond to local minima of the Helmholtz free en-
ergy function ψ(F) where, since we are modeling the problem in an isothermal
setting, we have suppressed ψ’s dependency on temperature. Therefore
ψF (I) = ψF (F0 ) = 0, (13.1)
and the two elasticity tensors
C+ = ρψFF (I), C− = ρψFF (F0 ) (13.2)
are both positive definite. The tensors I and F0 lie at the respective bottoms of the
austenitic and one martensitic wells. As noted in Section 12.2, if there is to be a
phase boundary separating austenite with deformation gradient I from martensite
with deformation gradient F0 , their difference, F0 − I, must be a rank-1 tensor. As
discussed in Chapter 12, this requirement typically imposes certain restrictions on
the lattice parameters, and for most alloys, these conditions do not hold. However in
certain alloys, the composition can be selected so as to force these conditions to be
satisfied, in which event it is possible to have a piecewise homogeneous two-phase
deformation. The alloy tested by Escobar and Clifton is believed to be special in
this way and so we assume that F0 − I is rank-1, and therefore that there are vectors
b and m such that
F0 − I = b ⊗ m; (13.3)
m is normal to the phase boundary. Finally we assume that austenite is the more
stable phase at the temperature underlying the isothermal calculation and so
ψ(I) < ψ(F0 ). (13.4)
Consider a half-space x1 > 0 composed of austenite, which is initially unde-
formed and at rest. Its free surface x1 = 0 is subjected to a constant velocity v0 = εw
for all times t > 0. Here ε is a nondimensional parameter and w has the dimension
of velocity. In light of the homogeneity of the material and loading, it is natural to
assume that the body undergoes a motion that depends solely on the single spatial
coordinate x1 and time. Thus it is natural to set x = x1 and to consider motions of
the form
y = y(x, t) = x + u(x, t) for x > 0, t > 0 (13.5)
subject to the initial conditions
u(x, 0) = 0, ut (x, 0) = 0 for x > 0 (13.6)
and the boundary condition
ut (0, t) = v0 = εw for t > 0. (13.7)
The vector m in (13.3) is normal to the phase boundary. If we want the phase
boundary to be parallel to the planar surface of the specimen, then we must have
m = e1 where e1 is a unit vector that is normal to the plane face of the half-space.
By cutting the specimen at the appropriate orientation, one can arrange for the
212 EVOLUTION OF PHASE TRANSITIONS
crystal lattice to be oriented relative to the specimen in such a way that this is true.
When the lattice is oriented this way we have
F0 − I = b ⊗ e1 . (13.8)
It is now possible to have a two-phase equilibrium configuration of the body involv-
ing a planar phase boundary that is parallel to the free surface of the half space; the
body is then unstressed and is composed of austenite and martensite at deformation
gradient tensors I and F0 respectively. The vector b is determined by the underly-
ing crystallography as in Example 5 of Section 12.2; since b is a measure of the
difference between the austenite tensor I and the martensite tensor F0 we refer to it
as the transformation strain vector.
Figure 13.1. The x, t-plane. The three rays x = αi t, i = 1, 2, 3, correspond to three shock waves.
R−
ε (t) R+
ε (t)
denote the position of the phase boundary at time t ≥ 0; based on the preceding
analysis, one expects three shock waves to run ahead of the phase boundary in the
austenitic region. The regions R+ −
ε (t) and Rε (t), ahead of and behind the phase
boundary, are characterized by
R+
ε (t) = {x | x > s(t; ε)}, R−
ε (t) = {x | 0 < x < s(t; ε)}. (13.25)
Suppose that the impact parameter ε is small. We assume that in the resulting
motion the deformation gradient tensor field on the austenitic region does not depart
significantly from the identity I, and that on the martensitic region it remains close
to the tensor F0 ; recall that the respective tensors I and F0 characterize austenite
and the relevant martensite variant, both unstressed. Consequently we are assuming
that during the motion the deformation gradient tensor at any point in the body does
not depart significantly from the bottoms of the two associated energy wells even
though it may jump between these two wells. More precisely, we assume that the
motion is of the form
x + εu+ (x, t) + c(t) + O(ε 2 ) on R+ ε (t),
y= (13.26)
F0 x + εu− (x, t) + O(ε 2 ) on R− (t),
ε
216 EVOLUTION OF PHASE TRANSITIONS
where c(t) denotes a rigid translation. Moreover, we suppose that the phase boundary
location can also be expanded in a series as
The strategy is to linearize the problem on each phase, and to then connect the
two regions across the phase boundary using the jump conditions, also appropriately
approximated. We omit the details of this calculation and refer the interested reader
to [2]. One first finds that s0 (t) = 0 and c(t) = 0 for t ≥ 0; and then to O(ε),
the basic problem to be solved can be described solely in terms of the austenitic
displacement field u+ (x, t) as follows: we are to determine u+ (x, t) by solving the
partial differential equation
A+ u+ +
x x = utt for x > 0, t > 0, (13.28)
u+ = u+
t = 0 for x > 0, t = 0, (13.29)
u+
t = w + ṡ1 b for x = 0, t > 0, (13.30)
1
βk = − [w + ṡ1 b] · a(k) , k = 1, 2, 3. (13.31)
αk
This provides the solution to the impact problem in terms of the unknown phase
boundary speed, which, to leading order, is ε ṡ1 .
If the kinetic relation of the phase transformation was known, then we could
complete the solution by using the kinetic law to determine ṡ1 . However in this
chapter we take the point of view that the kinetic relation is not known but that
certain experimental data are available instead. In the next section we show how
experimental measurements can be used to calculate ṡ1 , thus completing the solu-
tion. The kinetic law relating the propagation speed Vn = ε ṡ1 to the driving force
f on the phase boundary is also determined using the data.
IMPACT-INDUCED TRANSITION IN A CuAlNi SINGLE CRYSTAL 217
Table 13.1. Measured data from Table 4.1 of Escobar [3]. Reproduced with
permission from JoAnne Escobar.
Propagation speed
Experiment ṡ (m s−1 )
1 19.21
2 18.63
3 80.42
4 137.92
13.5.2 Phase boundary speed. We now calculate the phase boundary speed
ṡ = ε ṡ1 . Note first from (13.21)3 and (13.31) that v(3) = w + ṡ1 b. Thus ṡ1 can
be calculated by taking the dot product of this equation with any vector that is not
orthogonal to b. Since the transverse particle velocity εv(3) · e2 was measured, it is
natural to take the dot product of this with e2 leading to
3 (13.36)
= ρψ(I) − ρψ(F0 ) + ερ αk2 βk a(k) · b + O(ε 2 ),
k=1
where we have also used (13.16)1 , (13.19)4 , (13.20)3 , and (13.18) in obtaining the
second expression.
In order to estimate the first two terms let
where θ is the test temperature. Since f o (θ) represents the height between the two
energy wells at the temperature θ , it must vanish at the temperature θT , be positive
for θ > θT , and be negative for θ < θT . If we approximate f o (θ ) by the first two
terms of its Taylor series and use the fact that f o (θT ) = 0, we get
Next, recall that the specific entropy per unit mass, η, is related to ψ, the Helmholtz
free energy per unit reference volume, by η(F, θ ) = −ψθ (F, θ ); recall also that
the latent heat per unit mass of the austenite → martensite transformation, at the
temperature θT , is given by λT = −θT [η(F0 , θT ) − η(I, θT )]. On combining the
IMPACT-INDUCED TRANSITION IN A CuAlNi SINGLE CRYSTAL 219
θ
f o ≈ ρλT −1 . (13.39)
θT
Thus the value of the driving force in each test can be calculated from (13.36),
(13.37), and (13.39). The results are shown in Table 13.3.
13.5.4 Kinetic law. In Figure 13.3, we plot the four pairs ( f, ṡ) given in Tables 13.2
and 13.3 that correspond to the experiments. These points should lie on or near the
curve corresponding to the kinetic relation ṡ =
( f ) characterizing the austenite
to β1 -martensite transformation. The solid curve in Figure 13.3 is the graph of the
function
0 for f ≤ f ∗ ,
( f ) =
M f 2 − f 2 for f ≥ f ,
(13.40)
∗ ∗
with M = 7.07 m MPa−1 s−1 and f ∗ = 25.26 MPa. Kinetic response functions of
this form have been shown in [1] to describe the kinetics of twinning of γ1 -martensite
in the same CuAlNi alloy used by Escobar and Clifton. For further discussion, see
Chapter 14 of this monograph or Abeyaratne, Chu and James [1].
Phase boundary speed (m/s)
Driving force (M P a)
Figure 13.3. Kinetic law. The four points in the figure correspond to the data in Tables 13.2 and 13.3.
The curve is described by equation (13.40) with M = 7.07 m MPa−1 s−1 and f∗ = 25.26 MPa.
220 EVOLUTION OF PHASE TRANSITIONS
REFERENCES
[1] R. Abeyaratne, C. Chu, and R.D. James, Kinetics of materials with wiggly
energies: theory and application to the evolution of twinning microstructures
in a Cu–Al–Ni shape memory alloy. Philosophical Magazine A, 73 (1996),
pp. 457–97.
[2] R. Abeyaratne and J.K. Knowles, On the kinetics of an austenite→martensite
phase transformation induced by impact in a Cu–Al–Ni shape memory alloy.
Acta Materialia, 45 (1997), pp. 1671–83.
[3] J.C. Escobar, Plate Impact Induced Phase Transformations in Cu–Al–Ni Single
Crystals. Ph.D. dissertation, Brown University, Providence, RI, 1995.
[4] J.C. Escobar and R.J. Clifton, On pressure-shear plate impact for studying the
kinetics of stress-induced phase transformations. Journal of Materials Science
and Engineering, A170 (1993), pp. 125–42.
[5] J.C. Escobar and R.J. Clifton, Pressure-shear impact-induced phase transitions
in Cu–14.4 Al–4.19 Ni single crystals. SPIE, 2427 (1995), pp. 186–97.
14 Quasistatics: Kinetics of Martensitic
Twinning
14.1 Introduction
In the preceding chapter we determined the kinetics of a certain phase transforma-
tion using experiments that involved fast loading in which inertia was important.
In the present chapter we determine the kinetics of a different transformation using
data from quasistatic experiments. The transformation studied here is a twinning
deformation, not a phase transformation, a twin boundary being an interface that
separates two variants of martensite; see Example 1 in Section 12.2. The change in
lattice orientation across a twin boundary makes it analogous, in certain ways, to a
phase boundary, and in particular, the motion of a twin boundary is governed by a
kinetic relation.
As we have seen, the simplest form of kinetic relation governing the isothermal
motion of an interface relates the driving force on it to its normal velocity of
propagation: Vn = ( f ). Since the kinetic response function here is a function
of a single scalar independent variable, one set of experiments, say uniaxial tension
tests, completely determines ; and the function thus determined characterizes
all motions of this interface such as, say, in biaxial conditions. If the deformation
field is inhomogeneous, and the phase or twin boundary is curved, one would use this
same kinetic relation locally, at each point along the interface, relating the driving
force at that point to the normal velocity of propagation of that point. Consequently
once the kinetic relation Vn = ( f ) is known, one ought to be able to, at least in
principle, calculate the motion of an interface under very general conditions.
One of the goals of the present chapter is to point out that the behavior of
interfaces is not always that simple. For example, sometimes an interface can get
“pinned” at some point along its surface resulting in that point being held back
while the rest of the interface propagates according to the kinetic law. In order to
understand the motion of an interface in such circumstances, one must be able to
characterize the process of pinning/unpinning, in addition to knowing the kinetic
relation for normal propagation.
Often, one is interested only in the macroscopic behavior of the material at a
scale that homogenizes the motion of individual interfaces and phenomena such as
221
222 EVOLUTION OF PHASE TRANSITIONS
σ2
p2
σ2
p1
pinning/unpinning. This chapter illustrates one way to get at this effective kinetic
behavior directly.
σ1 ≥ 0 and σ2 ≥ 0 are the two stress components applied by the testing machine.
The plate was made of a Cu–14.44Al–4.19Ni (wt%) shape memory alloy, an
alloy that we have encountered in previous chapters of this monograph. In partic-
ular, the schematic phase diagram in Figure 5.1 indicates that this alloy has three
phases and that a different phase is preferred under different conditions of stress
and temperature. For example the high stresses generated in the impact experiment
described in Chapter 13 caused the specimen there to transform into the monoclinic
phase β1 -martensite. The experiments relevant to the present discussion were con-
ducted under conditions where the loaded specimen remained in the orthorhombic
phase γ1 -martensite at all times.
The γ1 -martensite lattice can be obtained from the cubic austenite lattice as
follows: let {c1 , c2 , c3 } be a set of orthonormal vectors in the cubic directions.
Stretch the cube by a stretch ratio η3 in, say, the c3 direction, and stretch it by η1 and
η2 along the diagonals of the face normal to c3 . The resulting mapping from the
cubic to the orthorhombic phase is x → U1 x where U1 , the matrix of components
QUASISTATICS: KINETICS OF MARTENSITIC TWINNING 223
of U1 , is given by
⎛ η1 +η2 η1 −η2 ⎞
2 2
0
⎜ η1 −η2 ⎟
U1 = ⎜
⎝ 2
η1 +η2
2
0⎟ ⎠ (variant − 1). (14.2)
0 0 η3
Here and henceforth, all components of vectors and tensors are taken with respect
to the cubic basis {c1 , c2 , c3 }. The stretch tensor U1 describes one of the martensitic
variants. There are six such variants associated with the three different cube edges
and the two associated cube-face diagonals. The second variant associated with
stretching the cube by η3 in the c3 direction is described by
⎛ η1 +η2 η2 −η1 ⎞
2 2
0
⎜ η2 −η1 ⎟
U2 = ⎜
⎝ 2
η1 +η2
2
0⎟ ⎠ (variant − 2). (14.3)
0 0 η3
In the experiments described in [2, 4] one edge, say c3 , of the cubic lattice was
normal to the face of the specimen, and the other two were rotated by π/4 with
respect to the specimen edges as shown in Figure 14.1. For this orientation, the two
variants (14.2) and (14.3) were energetically preferred over the four other variants
and so the others were not encountered during any of the experiments. Therefore
we shall not display the matrices U i , i = 3, . . . , 6, associated with the remaining
four variants.
The Helmholtz free energy function ψ(F) has an energy well at every tensor
F that has the representation F = QUi for any i = 1, 2, . . . , 6 and every proper
orthogonal tensor Q. Without loss of generality we can take the Helmholtz free
energy to vanish at the bottom of each of the martensite energy wells: ψ(QUi ) = 0
where i = 1, 2, . . . , 6 and Q is proper orthogonal.
14.3 Observations
As noted already, for the particular specimen orientation used in the experiments,
variants-1 and -2 were energetically preferred. When the applied stress σ1 σ2 ,
the entire specimen was composed of variant-1; when σ2 σ1 it was composed
of variant-2. During a typical test, the biaxial stress histories σ1 (t), σ2 (t) were
prescribed and the specimen transformed between these two variants.
During the process of transformation, the specimen typically involves a mixture
of the two variants in the form of a “laminate” – a family of alternating bands of
variant-1 and variant-2 as shown schematically in Figure 14.2; the dark bands in
the figure correspond to variant-1 and the white bands between them correspond to
variant-2. The interfaces between adjacent bands, that is, the boundaries separating
one variant from the next, are “twin boundaries.”
224 EVOLUTION OF PHASE TRANSITIONS
vh
(1 − v)h
Variant-1
Variant-2 Twin boundaries
Variant-1
Figure 14.2. A single laminate: dark bands of variant-1 and white bands of variant-2. The volume
fraction of variant-1 is v.
As the applied loading varies, the twin boundaries move normal to themselves,
and so the bands of one variant get thicker while those of the other variant get
narrower. Thus the volume fractions of the variants change. During a typical test,
the volume fraction history v(t) of variant-1 arising in the center of a laminate was
measured. The response to various loading histories σ1 (t), σ2 (t) was studied.
Typically, the specimen involves a patchwork of many laminates as shown
schematically in Figure 14.3, where each laminate consists of a family of alter-
nating bands of variant-1 and variant-2 as in Figure 14.2. One laminate is separated
from an adjacent laminate by a “transition layer” Th . The bands of one laminate
terminate in the transition layer when they meet the bands of an adjacent laminate.
Th
Shown enlarged
in Figure 14.4
Variant-1
Variant-2
vh
(1 − v)h
Th
Variant-2 Variant-1
v∗ h ∗
(1 − v∗)h∗
We shall not describe the numerous experimental observations of Chu and James
but simply remark on one of them. Suppose that variant-1 is, theoretically, the en-
ergetically preferred variant at some stress level (σ1 , σ2 ) = (σ1∗ , σ2∗ ). During the
experiments, subjecting the specimen to (σ1∗ , σ2∗ ) sometimes yielded a configura-
tion involving a mixture of both variants rather than the energetically favored one
involving only the single preferred variant. Even more significantly, suppose that
the specimen is subjected to a gradually varying cyclic loading with the stresses
(σ1 (t), σ2 (t)) oscillating around the value (σ1∗ , σ2∗ ); specifically, suppose that the am-
plitude of the oscillation decreases monotonically and that it eventually goes to zero.
Even for this loading, the specimen continued to yield a two-variant microstruc-
ture rather than one composed entirely of the preferred variant. This is typical of
many shape memory alloys and is indicative of the system getting “stuck” in some
metastable state that prevents it from achieving its minimum energy configuration.
This phenomenon manifested itself in various ways in the different experiments
described in [2, 4]. These observations suggest that a naive model of the kinetics
might not adequately describe the response of the system.
The goal of the present chapter, which summarizes the work reported in [1, 2], is
to understand the appropriate kinetics for this problem and to develop a model that
predicts the evolution of the volume fraction v(t) corresponding to a given loading
history (σ1 (t), σ2 (t)).
1
E= ρψ(∇y(x))dx − σn · ydx (14.4)
vol(R) R ∂R
where R is the region occupied by the specimen in the reference configuration.
Given the stress (σ1 , σ2 ) applied by the loading device, suppose that the specimen
attains a certain configuration involving variants-1 and -2, with the volume fraction
of variant-1 being v. Then the total energy of the system associated with this par-
ticular microstructure can, in principle, be calculated using (14.4) and the resulting
energy will have the form
E = E(v, σ1 , σ2 ). (14.5)
The volume fraction v therefore plays the role of an internal variable and the driving
force f conjugate to v is
∂E
f =− . (14.6)
∂v
226 EVOLUTION OF PHASE TRANSITIONS
In the formalism of internal variable theory, the evolution of the system is governed
by a kinetic relation relating, for example, the rate of change of the internal variable
to the driving force: v̇ = ( f ). For small values of driving force we can replace it
by its linearized approximation
∂E
v̇ = M f = −M , (14.7)
∂v
as long as the mobility M = (0) = 0. Equation (14.7) is often referred to as a
gradient flow of the energy E. We are thus led to the following initial value problem
for determining v(t) corresponding to given σ1 (t), σ2 (t):
14.5.1 Elastic energy of the specimen. The first term in (14.4) represents the en-
ergy E elastic associated with the specimen. In calculating it, it is convenient to
consider the transition layers Th and the rest of the specimen R − Th separately.
First consider the transition layers Th . It has been shown by Ball and James [3] that
the volume of the transition layers becomes vanishingly small as the bands become
narrower: vol(Th ) → 0 as h → 0. Thus, keeping in mind that the deformation gra-
dient tensor field is uniformly bounded, we can neglect the effect of the transition
layers in calculating the elastic energy of the specimen provided that the martensitic
bands are very fine.
Next consider the region R − Th . Here the twin boundaries are essentially
straight and parallel as sketched in Figure 14.3 and the deformation is more or
less piecewise homogeneous. The strain at a generic particle is due, in part, to
elastic deformation and, in part, to transformation. If a particle x is in variant-i
and its deformation gradient tensor is F(x), the elastic strain of this particle is the
“distance” in deformation gradient space of F(x) from the energy well associ-
ated with variant-i. On the other hand the transformation strain is the distance be-
tween the two energy wells |U1 − U2 |. From a knowledge of the lattice parameters
η1 , η2 , η3 the latter can be estimated to be |U1 − U2 | = [ tr (U1 − U2 )2 ]1/2 ≈ 0.055.
On the other hand since the material has a modulus of about 50 GPa and the speci-
men encounters a typical stress level of about 15 MPa in the experiments, the elastic
strain is of the order of 15/50 × 10−3 ≈ 0.0003. Therefore the elastic strains are
200 times smaller than the transformation strain.
QUASISTATICS: KINETICS OF MARTENSITIC TWINNING 227
14.5.2 Loading device energy. The second term in (14.4) represents the energy
E loading associated with the dead loading device. In view of the continuity of the
deformation and the uniform traction boundary condition, this term can be written
as
1
E loading = − σ· ∇y(x)dx = −σ · F̄ (14.9)
vol(R) R
where F̄ is the average deformation gradient in the specimen and the biaxial stress
σ is prescribed by (14.1).
In order to calculate F̄, we consider the transition layers Th and the rest of the
specimen R − Th separately. First consider the transition layers. As noted above,
vol(Th ) → 0 as h → 0, and the deformation gradient tensor field is uniformly
bounded. Thus, provided that the martensitic bands are very fine, one can neglect
the effect of the transition layers in calculating the average deformation gradient in
the specimen.
Therefore in calculating the average deformation gradient tensor F̄ we may ig-
nore Th and treat the specimen as a collection of laminates, where each laminate
involves a family of parallel bands of variants-1 and -2 as sketched in Figure 14.3.
The deformation within each band is homogeneous. In any one laminate, the defor-
mation gradient tensor has the form vR1 U1 + (1 − v)R2 U2 where R1 and R2 are
rotations and v is the volume fraction of variant-1 in that laminate (Figure 14.3).
Continuity of the deformation across a twin boundary within that laminate requires
that R1 U1 − R2 U2 be a rank-1 tensor and this, effectively, allows one of the ro-
tations, say R2 , to be determined; see the proposition by Ball and James and the
surrounding discussion in Section 12.1. In an adjacent laminate, since the same vari-
ants are involved, the stretch tensors are again U1 and U2 ; however, the rotations
R3 , R4 and the volume fraction v∗ may be different. Continuity of the deformation
within this laminate requires that R3 U1 − R4 U2 be a rank-1 tensor and this again ef-
fectively determines one of these R’s, say R4 . Finally, continuity of the deformation
228 EVOLUTION OF PHASE TRANSITIONS
between laminates requires that the average deformation gradient tensors in the two
laminates, vR1 U1 + (1 − v)R2 U2 and v∗ R3 U1 + (1 − v∗ )R4 U2 , differ by a rank-1
tensor. This allows one to determine one of the two unknown rotations, say R3 , and
one of the volume fractions, say v∗ . No matter how many laminates are involved,
one can proceed systematically in this way and determine all of the rotations and
volume fractions in terms of a single volume fraction v in a particular laminate and
a single rotation R.
Therefore the loading device energy takes the form E loading = −σ · F̄ = −σ ·
RD(v) where σ is prescribed and D is known in terms of v. The two unknowns in
this formula for E loading are the overall rotation of the specimen R and the volume
fraction v. During a loading process, the former is determined by the rigid body
dynamics of the specimen and the latter by the transformation kinetics. Assuming
that the macroscopic dynamics takes place much faster than the kinetics as was true
in the experiments, we choose R such that it minimizes E loading at fixed v and σ.
This leads to the following explicit expression for E loading in terms of the applied
stress (σ1 , σ2 ), the volume fraction v, and the lattice parameters η1 , η2 , η3 :
2 2
2 η1 − η2
2
E loading (v, σ1 , σ2 ) = − v 2
σ12 + σ22
η1 + η2 2
2
12
η1 − η22 2 2
+ 2v 2
σ1 η2 − σ22 η12 + (σ1 η2 + σ2 η1 )2 .
η1 + η22
(14.10)
The details of the calculations leading to this result, together with a careful descrip-
tion of the underlying assumptions, can be found in [2].
1
The relative importance of these two energetic contributions was examined in [2], where it was found
that the loading device energy (14.10) is about 400 times as large as the nominal elastic energy in
the specimen, 1/2(stress2 /modulus), which justifies the approximation E elastic E loading .
QUASISTATICS: KINETICS OF MARTENSITIC TWINNING 229
Th
Twin boundary
Variant-1
Variant-1
Twin boundary
Figure 14.4. An enlarged view of the region enclosed by the circle in Figure 14.3. The figure shows a
time sequence of sketches of a single band of one variant in one laminate meeting a band of the same
variant in a neighboring laminate. The motion of the band is constrained by the pinning of one or more
tips. As the width of the band increases, the tips split and their number increases.
These tips are pinned: they constrain the motion of the twin boundary because, as
the parallel portions of the twin interfaces move normal to themselves and the band
gets wider, the tips remain more or less stationary. As the band gets wider, at some
critical band width, the tip splits into two or more tips, and the band width increases
discontinuously. This process repeats itself with the width continuing to increase
with accompanying bursts of additional tip-splitting as described by the sequence
of schematic diagrams progressing from (a) to (d) in Figure 14.4.
The phenomenon of tip-splitting is particularly striking during a creep test.
Suppose that the specimen settles into an equilibrium configuration. Consider a
single band, which has a certain width and some number of tips. It is found that
even a very slight disturbance affects this microstructure. For example, lightly
tapping the specimen with a pencil causes additional splitting, and the number of
tips and the width of the band changes, even though the applied stresses σ1 , σ2 have
not changed. This can be repeated by further tapping.
A configuration with a certain number of tips corresponds to a particular
metastable state of the band. This same band, at the same load, has other metastable
states available to it, each corresponding to a certain number of tips. The aforemen-
tioned phenomenon of getting stuck in metastable states is intimately connected to
the availability of these different configurations. Thus in order to properly charac-
terize the evolution of the volume fraction in the specimen, one must account for
both the kinetics of the normal motion of the twin interfaces and the process of
tip-splitting.
In calculating the total energy of the system, we must therefore account for the
energy associated with the transition layers Th in which the tip-splitting occurs.
Based on the preceding discussion, this energy should have a large number of local
minima, where each minimum corresponds to a particular configuration of a band
involving a certain number of tips. The energy barriers between the minima should
be small, since the observations show that it is easy to pass from one metastable state,
corresponding to a certain number of tips, to another, associated with a different
number of tips.
The relative importance of the various energetic terms was examined in [2],
where it was found that the loading device energy (14.10) is about 50 times as
large as the elastic energy as given by (14.12). However even though the elastic
energy in the transition layer is 50 times smaller than the loading device energy,
it plays a crucial role in light of the fact that it affects the convexity of the total
energy. In particular, for suitable values of the stress, the function E(·, σ1 , σ2 ) =
E elastic (·, σ1 , σ2 ) + E loading (·, σ1 , σ2 ) is minimized by values of volume fraction v,
which are between 0 and 1 corresponding to microstructures involving a mixture
of variants; this is in contrast to the loading device energy E loading , which is always
minimized by v = 0 or 1.
As observed in Section 14.6, the energy associated with the phenomenon of
tip-splitting must have a large number of local minima, each corresponding to a
particular configuration of a band with a certain number of tips. The energy barriers
between the minima should be small since the observations show that it is easy to
increase the number of tips by additional splitting, that is, it is easy to pass from one
metastable state to another. Moreover, the distance between two adjacent minima
must be small since the change in volume fraction accompanying the appearance
of a new tip is small. These conditions suggest that we model the energy associated
with the tips by an expression such as
v
E tips (v, σ1 , σ2 ) = aε cos , (14.13)
ε
where a and ε are constants with ε 1. The height of the energy barriers is 2aε
and the distance between adjacent minima is 2πε.
14.8 Kinetics
The total energy of the system is now given by
where the energies E elastic , E loading , and E tips are given by (14.12), (14.10), and
(14.13) respectively. The kinetic relation is now a gradient flow based on (14.14)
and this leads to the initial-value problem
for determining the volume fraction vε (t) corresponding to given σ1 (t), σ2 (t); here
E ε = ∂ E ε /∂v. The existence of multiple local minima of E ε (·, σ1 , σ2 ) suggests
that this evolution law does have the capability for describing the phenomenon of
getting stuck in metastable states.
Since we are interested in the macroscopic response of the specimen and not
in the detailed behavior of each individual twin band, we need to homogenize the
evolution law (14.15) by passing to the limit ε → 0. The mathematical analysis of
the homogenization of (14.15) will not be presented here; this can be found in [2].
The analysis in [2] proves that the solution vε (t) of (14.15) converges uniformly to
232 EVOLUTION OF PHASE TRANSITIONS
Vn
Here f = −∂ Ê/∂v is the driving force calculated from the energy Ê = E elastic +
E loading , which does not include tip-splitting effects; and f ∗ = a where aε is the
amplitude of the tip-splitting energy (14.13). Equation (14.16) shows that for small
values of driving force, | f | < f ∗ , the volume fraction is “stuck” and does not
evolve and that it begins to change only when the driving force is sufficiently large.
We note in passing that this phenomenon of getting stuck (often called “lattice
trapping”) is common to other models of kinetics as well; see, for example, the
Frenkel–Kontorowa model in Section 8.4.5.
The model constructed here involves four unknown parameters c1 , c2 , M, and
f ∗ , the lattice parameters η1 , η2 , η3 having been measured independently. One set
of numerical values of these four parameters was used in [2] to show that the quan-
titative predictions of this kinetic law compare reasonably well with the numerous
responses measured by Chu and James; see [2].
REFERENCES
[1] R. Abeyaratne, C. Chu, and R.D. James, Kinetics and hysteresis in marten-
sitic single crystals, in Proceedings of Symposium on the Mechanics of Phase
Transformations and Shape Memory Alloys. American Society of Mechanical
Engineers Applied Mechanics Division, 1994, Vol. 189, pp. 85–98.
[2] R. Abeyaratne, C. Chu, and R.D. James, Kinetics of materials with wiggly
energies: theory and application to the evolution of twinning microstructures
in a Cu–Al–Ni shape memory alloy. Philosophical Magazine A, 73 (1996),
pp. 457–97.
[3] J.M. Ball and R.D. James, Fine phase mixtures as minimizers of energy. Archive
for Rational Mechanics and Analysis, 100 (1987), pp. 13–52.
QUASISTATICS: KINETICS OF MARTENSITIC TWINNING 233
Abeyaratne, R., 9, 10, 12, 13, 27, 31, 57, 80, Clifton, R.J., 8, 14, 159, 161, 208, 209, 210,
103, 104, 105, 111, 112, 123, 134, 135, 144, 211, 219, 220
161, 179, 193, 207, 210, 219, 220, 232 Courant, R., 80, 181, 193
Achenbach, M., 13, 144
Ahrens, T.J., xiii, 181, 193 Dafermos, C., 10, 13, 77, 78, 79,
Artemev, A., 13 80
Atkins, J.E., 97, 98, 104 Dein, J.L., 159, 162
Atkinson, W., 144 Delaey, L., 180
Duerig, T., 7, 13
Balk, A.M., 135, 145 Dunn, J.E., 150, 161, 181, 193
Ball, J.M., 201, 207, 226, 227, 232
Bazant, Z., 13 Easterling, K.E., 31, 105, 112, 145
Bhattacharya, K., xiii, 9, 13, 47, 57, 95, Ericksen, J.L., xiii, 9, 10, 13, 22, 31, 37, 57, 95,
103, 134, 145, 179, 197, 201, 203, 204, 98, 103
205, 207 Erskine, D.J., 8, 14, 59, 61, 75, 76, 80
Bless, S.J., 161 Escobar, J.C., 8, 14, 208, 209, 210, 211, 217,
Bo, Z.H., 13 219, 220
Bourne, N.K., 123 Eshelby, J.D., 9, 14, 27, 31
Bowles, J.S., 199, 208 Evans, L.C., 14, 78, 80
Brar, N.S., 161
Brinson, L.C., 7, 8, 9, 13, 14 Falk, F., 160, 162
Brocca, M., 13 Fomin, S.V., 39, 57
Broner, F., 57 Fosdick, R.L., 150, 161, 181, 193
Brown, L.C., 7, 14, 32, 58, 173, 180 Frenkel, J. 136, 138, 145, 232
Bruno, O., 14, 179, 180 Fried, E., 9, 14
Burkart, M.W., 175, 179 Friedrichs, K.O., 80, 181, 193
Bywater, K.A., 8, 13, 201, 208 Fritz, J.N., 162
Funakubo, H., 7, 14
Cabrera, N., 144
Callen, H.B., 91, 103, 125, 126, 145, 161 Gao, X., 8, 14
Carlson, D.E., 150, 161 Gelfand, I.M., 39, 57
Carter, W.J., 162 Germann, T.C., 145
Chadwick, P., xiii, 25, 31, 86, 103, 111 Goldsztein, G.H., 57
Chen, Y.-C., 182, 193 Graham, R.A., 59, 80
Cherkaev, A.V., 135, 145 Green, A.E., 97, 98, 104
Christian, J.W., 7, 8, 13, 45, 57, 105, 112, 140, Grujicic, M., 145, 172, 173, 179
142, 145, 201, 208 Grüneisen, E., 162
Chu, C., 8, 12, 13, 134, 144, 219, 220, 222, 225, Gurtin, M.E., 9, 14, 112, 123,
232, 233 146
235
236 AUTHOR INDEX
Strogatz, S., 57 Vedantam, S., 10, 13, 103, 104, 135, 144
Swegle, J.W., 181, 194 Voorhees, P.W., 9, 15
238
SUBJECT INDEX 239
Elasticity tensor, 117, 211 local heating in slow cycling of NiTi wires,
Elastic moduli 179
for thermoelastic material in uniaxial pressure-shear impact on CuAlNi, 209ff
strain quasistatic hysteresis in NiTi wires, 173
isentropic modulus, 150 (Figure 10.3), 176ff
isothermal modulus, 150
for trilinear elastic material Failure wave, 159
in uniaxial strain, 62 Fan, 60
in uniaxial stress, 47 First law of thermodynamics. See Laws of
Energy thermodynamics
barrier, 45, 132, 158, 170 Flux
free, 4 conjugate to driving force, 10, 125
Gibbs free energy, 23, 85, 90, 153, of heat, 106
157, 166 Flyer plate, 59ff
Helmholtz free energy, 11, 85ff, 97, 109, Free energy. See Energy
113ff, 150, 152, 154, 160, 164, 211 Frenkel-Kontorowa model for twinning,
for cubic tetragonal crystal, 95 136
for ideal gas, 90
for van der Waals fluid, 91 Gibbs free energy. See Energy
Helmholtz free energy potential, 114 Graphite-to-diamond phase transition,
internal energy, 86, 106 74ff
for ideal gas, 90 Grüneisen parameter
for van der Waals fluid, 91 modified, 150
internal energy potential, 114 related to coefficient of thermal expansion,
of laminated variants of martensite, 226ff 150
potential energy, 19, 21 (Figure 2.1(a)), unmodified, 150 (footnote)
24 (Figure 2.3), 35, 45, 88ff,
118, 132, 153, 157 Heat conduction, 109
for bilinear two-phase material, 130, law for thermoelastic materials, 116
139 (Figure 8.6(b)) as a kinetic law, 125
for van der Waals fluid, 91 Helmholtz free energy. See Energy
strain energy, 20, 38, 85, 129 High-pressure phase (HPP), 62, 184
convex, 21 Hysteresis, 4, 53ff, 163ff
surface energy, 40, 133, 139, 142 experimental hysteresis loops, 173
wells, 21, 45, 85, 87, 99 (Figure 5.7), (Figure 10.3), 177 (Figure 10.6)
132, 197, 210, 223 loops in purely mechanical theory, 48
Entropy (Figure 3.6), 51 (Figure 3.7), 52
inequality, 10, 78, 110, 111, 164, 167, (Figure 3.8)
183 loops in thermoelastic material, 172
per unit mass, 108, 114 (Figure 10.2), 174 (Figure 10.4)
production rate, 105, 108ff for shape-memory cycle, 175 (Figure 10.5)
for thermoelastic materials, 115, 125, static, 56 (Figure 3.10)
126
specific, 108, 114 Impact, 8
Equilibrium experiments, 59
mechanical, 5, 36, 39, 40, 43, 111 graphite-to-diamond experiments, 74ff
phase, 5, 36, 39, 40, 43, 126, 158 comparison with model, 76 (Table 4.1,
thermal, 126 Figure 4.9)
thermodynamic, 125 experimental data, 75 (Figure 4.8), 76
thermomechanical, 117, 164ff (Table 4.1)
Experiments induced phase transition, 59ff, 181ff
biaxial quasistatic loading in CuAlNi, in CuAlNi single crystal, 209ff
222 problem, 59, 185
creep induced by phase transition, 159 loss of uniqueness in, 71
impact-induced graphite-to-diamond nucleation and kinetics for, 71ff
transition, 74ff one-wave phase-changing solution for, 68
240 SUBJECT INDEX