Scripta Materialia 50 (2004) 181–186
www.actamat-journals.com
Kinetics of martensitic transitions in shape-memory alloys
~osa
Antoni Planes *, Francisco-Jose Perez-Reche, Eduard Vives, Lluıs Man
Departament d’Estructura i Constituents de la Materia, Facultat de Fısica, Universitat de Barcelona, Diagonal, 647, E-08028 Barcelona,
Catalonia, Spain
Accepted 4 September 2003
Abstract
The role of thermal fluctuations in driving the martensitic transition in shape-memory alloys is studied. Experiments show that
both athermal and thermally activated kinetics can be obtained in different alloys. We establish that this depends upon the relative
importance of the characteristic times associated with the driving field and nucleation.
2003 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Athermal; Thermally activated; Nucleation; Martensite; Shape-memory
1. Introduction
When an external parameter is varied during a firstorder phase transition, the system evolves from a given
state (typically metastable) towards a more stable state
which corresponds to a minimum of a thermodynamic
potential. Such a transition involves overcoming an
energy barrier separating the minima corresponding to
each phase. Thermal fluctuations are expected to be
relevant for the system for it to overcome the energy
barrier. Within this scenario––which is the most accepted conventional description of first-order phase
transitions––the value of the external parameter (temperature, stress, magnetic field. . .) at which the transition occurs is not always the same upon repeating the
experiment, but rather the values at the transition are
spread out within a certain range. Such a spread reflects
the stochasticity of thermal fluctuations. The phase
transitions conforming to the above description are referred to as thermally activated. Furthermore, because
these transitions can occur at constant values of all external parameters, they are often called isothermal. In
isothermal transitions the amount of resulting phase
depends on the value of the external parameter and also
depends explicitly on time. The transition exhibits hysteresis and it is rate dependent. Examples of these kinds
of transitions are found among the most common first*
Corresponding author. Tel.: +34-93-402-1181; fax: +34-93-4021174.
E-mail address: toni@ecm.ub.es (A. Planes).
order phase transitions as, for instance, the freezing of
water on cooling.
In nature there are systems which undergo first-order
phase transitions that, at least from an experimental
point of view, do not conform to the description given
above. In these systems, thermal fluctuations do not
seem to play any role in the kinetics of the transition.
They can only take place while the external parameter is
changing, the transition appears to be instantaneous on
practical time scales, and the amount of transformed
material does not explicitly depend on time and is only a
function of the value of the external parameter. These
transitions are referred to as athermal [1]. Upon repeating the experiment, they always occur at the same
value of the external parameter. Although these transitions also exhibit hysteresis, they do not appear to be
rate dependent (at least for the rates available in typical
experiments). Actually, the path followed by the system
is influenced by the existence of disorder (dislocations,
grain boundaries, composition, atomic configuration,
etc.) which controls the actual distribution of energy
barriers. As the system evolves, it passes through a sequence of metastable states and the kinetics is characterized by jumps (avalanches) from one metastable state
to another which occur during very short times [2].
Typical examples of athermal transitions are the magnetization reversal in ferromagnetic systems [3] and
structural phase transitions in many solids [4]. A common feature of all these systems is the existence of
long-range interactions which arise from compatibility
constraints [5].
1359-6462/$ - see front matter 2003 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.scriptamat.2003.09.021
182
A. Planes et al. / Scripta Materialia 50 (2004) 181–186
The present paper focuses on the study of the martensitic transition in shape-memory alloys. The martensitic transition in these alloys has traditionally been
regarded as athermal. However, recent experiments have
questioned the athermal character of these transitions,
and much debate has arisen. Kakeshita et al. [6] performed electrical resistance measurements under isothermal conditions on Cu–Al–Ni alloys. They found
that after a certain incubation time at a temperature
above the transition temperature (determined from
continuous cooling) a sudden change in the resistance
was observed, which was associated with the occurrence
of the martensitic transition. After these results, Aspelmeyer et al. [7] investigated the Ni–Al alloy system by
means of a reflectivity technique. They observed changes
in the reflected intensity of a laser beam caused by the
appearance of a surface relief at a constant temperature.
Very recently, however, Otsuka et al. [8] have pointed to
the possibility that the apparently isothermal character
found in these alloys could be due to the occurrence of a
certain diffusion process. In order to check for this
possibility, they investigated the Ti–Ni alloy system
which is well known for the absence of significant difussion processes. They performed isothermal electrical
resistance experiments similar to those conducted by
Kakeshita et al. [6], and did not find any indication of
isothermal martensite growth. They concluded that the
martensitic transition was not time dependent if diffusion processes were suppressed.
It is the aim of the present paper to contribute to the
debate of whether the martensitic transition in shapememory alloys is athermal or isothermal by presenting
accurate experiments on several Cu-based shape-memory alloys, and giving a simple model which show us up
to what extent a transition can be classified as athermal.
The model furnishes a unified description of the experimental results showing athermal as well as isothermal
characteristics.
2. Experimental results
2.1. Experimental procedures
The usual experimental procedure to assess the athermal or isothermal character of martensitic transitions has
been the following: the sample is held at a constant temperature while some physical property (electrical resistance, reflectivity, etc.) is continuously monitored. The
holding temperature is located slightly above the temperature at which the transition has been detected to start
on a continuous cooling, which is assumed to be the socalled martensitic start temperature Ms . Such a procedure
has the inconvenience that the temperature where the
transition starts during continuous cooling may depend
on the cooling rate. Furthermore, as will be shown below,
the values may differ significantly from one cycle to
another even if they are performed at the same rate.
Therefore, the concept of Ms may be ill-defined in some
cases and, as a consequence, there is an intrinsic uncertainty in determining how far the holding temperature is from Ms . We have recently proposed [9] the two
following different procedures which enable these
problems to be circumvented.
2.1.1. Stepwise cooling
This procedure resembles the usual process of isothermal holding, but in this case the sample is cooled
down in a stepwise manner. Each step consists of an
isothermal plateau lasting a time Dt, followed by cooling
down to the next plateau at a given rate. It is convenient
to test different values for Dt and for the temperature
difference between consecutive plateaux (DT ). This
procedure is advantageous in that it does not require an
a priori determination of Ms because the stepwise cooling is continued until the sample transforms (either
during the isothermal plateau or during the short cooling ramp), and hence, for large enough Dt it determines
the transition temperature unambiguously with a maximum error of DT . It is worth remarking that by making
DT small, it is possible to achieve a very fine tuning for
the detection of the isothermal growth of martensite.
2.1.2. Scaling
A procedure complementary to that described above
consists of cooling the sample through the whole transition, while continuously monitoring some physical
parameter. This cooling has to be repeated at different
well-controlled cooling rates. Under continuous cooling,
athermal and isothermal systems will behave differently
as explained below.
Let W be a physical quantity which is supposed to
depend on the amount of transformed material. If
thermal fluctuations are irrelevant, W will depend on
temperature but it will not explicitly depend on time
WðT Þ. By contrast, for a thermally activated process, the
amount of transformed material will also depend on
time WðT ; tÞ. The time derivative for the athermal case
reads:
dW dW dT
¼
;
dt
dT dt
ð1Þ
while for the isothermal case:
dW oW dT oW
¼
þ
:
dt
oT dt
ot
ð2Þ
Consider now that dW=dt is measured at different
cooling rates and that the ratio ½ðdW=dtÞ=ðdT =dtÞ ¼
_ =T_ is plotted as a function of T . It is clear from Eqs.
W
(1) and (2) that for the athermal case all data points will
collapse on a single curve (scaling), while no such collapse or scaling will be observed in the isothermal case.
183
A. Planes et al. / Scripta Materialia 50 (2004) 181–186
2.2. Results
Ms (K)
TH
255
250
TL
245
3
σ (K)
Among all the experimental techniques commonly
used to study martensitic transitions, it is worth choosing one being very sensitive to the transformation of a
small amount of material. This is the case of the acoustic
emission (AE) technique which we have used for the
present study. This technique is probably the best suited
to monitor the early stages of the transformation. Full
details of the experimental system are given elsewhere
[9,10].
We have investigated several Cu-based and Ni-based
single crystals. In this paper we will present results
corresponding to Cu–Al–Ni and Cu–Zn–Al crystals,
which transform towards 2H and 18R martensitic
structures, respectively. The samples were annealed at
850 C and quenched into iced-water. After this heat
treatment, all samples were annealed for several days at
room temperature. By doing this we were sure that all
diffusion processes had finished and that the vacancy
concentration had reached its equilibrium value [11–13].
In addition, the samples were subjected to a large
number of cycles (more than 50) before the actual experiments started, so that the state of the sample does
not significantly evolve from one cycle to the next during
the experiment [14].
An example of the results obtained is shown in Fig. 1.
For Cu–Zn–Al, we have not found AE under isothermal
conditions (Fig. 1(a)) in any of the experiments, including long-temperature plateaux (Dt > 40 h) separated
by step intervals of DT ¼ 0:2 K. By contrast, for Cu–Al–
Ni, no isothermal AE is observed for plateaux with
DT > 0:8 K, but for DT < 0:8 K, AE was systematically
observed at constant temperature (Fig. 1(b)). From
these results it could be inferred that thermal fluctuations appear to be irrelevant in Cu–Zn–Al, while they
260
2
1
0
0
1
2
3
4
5
6
dT/dt (K/min)
Fig. 2. hMs i and r as a function of dT =dtð¼ T_ Þ. The continuous lines
are fits of the proposed model. The temperatures TH and TL obtained
from the fit are indicated.
are operative in the nucleation of Cu–Al–Ni. The results
from continuous cooling are consistent with these
statements; good scaling is observed for Cu–Zn–Al (Fig.
1(c)) while there is no scaling at all for Cu–Al–Ni (Fig.
1(d)). Furthermore, for Cu–Al–Ni the transition does
not seem reproducible from one cycle to another (performed at the same rate), but rather, the values obtained
for Ms spread over a temperature range and exhibit a
stochastic character. We have thus measured this
quantity for a large number of cycles which enables us to
perform a statistical analysis of the results by computing
the average value hMs i and its standard deviation r ¼
ðhMs2 i hMs i2 Þ1=2 . The results obtained are shown in
Fig. 2 (solid symbols) as a function of the cooling rate.
We would like to point out that any possible fake dependence due to ageing phenomena was taken into account [9]. The mean value of the transition temperature
seems to slightly decrease on increasing the cooling rate.
More evident is the increase in the standard deviation of
the values as the cooling rate increases.
3. Model
Fig. 1. AE during stepwise cooling. (a) Cu–Zn–Al, (b) Cu–Al–Ni.
Scaling corresponding to measurements at different T_ . (c) Cu–Zn–Al,
(d) Cu–Al–Ni. Notice that in the scaled representation the scatter increases for smaller T_ since the relative error in jdN =dT j is proportional
to T_ 1 .
The measured values of the transition temperature
are spread over a certain range due to thermal fluctuations. A suitable theory to explain the behaviour of hMs i
and r is based on the analysis of the mean first-passage
time [15] between the parent phase and the martensitic
phase. Here we briefly summarise this theoretical approach. The probability P ðtÞ for the system to remain in
the metastable phase after time t is given by: dP ðtÞ=dt ¼
kðtÞP ðtÞ, where kðtÞ is the transition probability per
unit time. As the transition seems to be athermal for
certain systems, while it is not in others, our main assumption in reproducing the experimental observations
is that thermal fluctuations are only active within two
characteristic limiting temperatures: when T > TH , k ¼ 0
184
A. Planes et al. / Scripta Materialia 50 (2004) 181–186
so that no transition is possible, and when T 6 TL ,
k ¼ 1 so that the transition occurs with absolute certainty. This assumption leads to the following expressions for the average temperature transition hMs i and its
fluctuations r2 :
hMs i ¼ TH T_ hti;
ð3Þ
2
r2 ¼ T_ 2 ht2 i þ TH2 2T_ TH hti hMs i :
ð4Þ
where hti and ht2 i can be obtained once the transition
probability is known. Actually, the transition probability is related to the energy barrier between parent and
product phases DF. A reasonable assumption is
0
for T > TH
k¼ m 1
ð5Þ
for T 6 T
0 DFðT Þ
H
where m0 is a characteristic energy per unit time related
to thermal fluctuations.
The energy barrier can be evaluated from a freeenergy functional which is adequate to describe the
martensitic transitions in shape-memory alloys. It has
been shown [16–18] that a good candidate is a Landau
free energy with two anharmonically coupled order parameters: a shear homogeneous strain e and the amplitude g of an inhomogeneous shear strain or shuffle which
corresponds to an incipiently unstable phonon. These
two order parameters are suitable to describe the crystallographic transition mechanism [19]. Symmetry considerations allow the following free-energy expansion:
1
1
1
1
F ¼ m x2 g2 þ bg4 þ cg6 þ C 0 e2 jeg2 ;
2
4
6
2
ð6Þ
where C 0 is the shear elastic constant and x, the frequency of the incipiently unstable phonon. The remaining constants are assumed to be positive. Minimization
with respect to e leads to an effective free energy, which
after suitable rescaling can be expressed in the following
simple form:
2
f ¼ ða 1Þg2e þ g2e ðg2e 1Þ ;
ð7Þ
where ge is the reduced order parameter. The preceding
equation contains a single dimensionless control parameter a, which is temperature dependent
a¼
16m x2 c
3ðb
2j2 2
Þ
C0
¼
T TL
:
TH TL
ð8Þ
Notice that the term ðb 2j2 =C 0 Þ must become negative
for the first-order martensitic transition to take place. A
typical feature of martensitic materials is the temperature softening for shearing {1 1 0} planes along Æ1 1 0æ
directions, made evident by the decrease of C 0 and x2 on
reducing temperature. In this model, such a softening
shows up by a decrease of a when the temperature is
reduced. Within the context of the present study, this
free energy has to be considered as a local free energy
adequate to account for the onset of the martensitic
Fig. 3. Landau free-energy as a function of a.
transition at Ms . From this viewpoint, the parameter a
should be considered as an effective parameter related to
renormalized values of C 0 and x in the spirit of the local
soft-mode theory as proposed by Guenin and Clapp
[20].
We have plotted the reduced free energy as a function
of the reduced order parameter for selected values of a
in Fig. 3. For a > 43 there is only a minimum, corresponding to the parent phase; as a decreases, two symmetric minima develop and for a ¼ 1 all three minima
have the same energy value and the two phases are
equally stable. This corresponds to the situation of local
equilibrium between parent and product phases. At this
point, if thermal fluctuations are operative, they can
drive the system from the parent phase to the product
phase. This situation corresponds to the temperature TH .
Upon further cooling the softening of C 0 and x lead
to a reduction of a so that the minima corresponding to
the product phase become deeper and the energy barrier
decreases, thus increasing the probability of transformation in the isothermal case (see Eq. (5)). In the
athermal limit the existence of the energy barrier prevents the transformation to the product phase, and
hence, the transformation can only take place when
there is no energy barrier; this corresponds to a ¼ 0, and
the minimum corresponding to the parent phase no
longer exists. This situation is the limit of stability of the
parent phase (temperature TL ) that is, below this temperature the system is sure to be in the product phase.
For a 6 1 (T 6 TH ), the probability that the system
transforms to the product phase is obtained from Eq. (5)
and from the expression for the barrier within the proposed model. It is given by
m
:
ð9Þ
9a þ ð4 aÞ3=2 8
pffiffiffi
where m ¼ m0 ð1 þ 3 3Þ=DFðTH Þ is a characteristic frequency for nucleation. The dependence of the transition
probability on a is depicted in Fig. 4. The discontinuity
at a ¼ 1 reflects the fact that the nucleation of the
product phase is not possible for T > TH while its
probability is finite at T ¼ TH .
kðaÞ ¼
A. Planes et al. / Scripta Materialia 50 (2004) 181–186
185
small. However, if this time becomes comparable to sdr
rate-dependent effects will occur, which are not associated with thermal fluctuations.
From the three time scales introduced, the following
scenarios can be contemplated:
Fig. 4. Transition probability k divided by m as a function of the reduced temperature a for the model presented in Section 3.
Within the preceding model a transition must be
classified as (ideally) athermal if it always occurs at TL ,
independently of the rate at which temperature is swept.
This means that fluctuations are not operative in this
limit. In any other case, the transition is, strictly
speaking, thermally activated. In this case, the model
described above enables one to compute the mean first
passage time which, through Eqs. (3) and (4), leads to
values of hMs i and r2 as functions of T_ =mðTH TL Þ. We
have used these expressions to perform a simultaneous
non-linear minimum-v2 fitting to the experimental data
shown in Fig. 2 for Cu–Al–Ni. The values obtained are
TH ¼ 256 1 K, TL ¼ 248 1 K and m ’ 5 102 s1 .
The fit (shown as a continuous line in Fig. 2) agrees
remarkably well with experimental data, and reproduces
the increase of r with T_ and also the slightly decreasing
behaviour of Ms .
4. Discussion
From the results presented above it can be concluded
that, as already mentioned, athermal transitions do not
occur strictly at finite temperatures. Nevertheless, in
systems that evolve by processes of avalanches from one
metastable state to another, there are three characteristic
time scales involved in the transition whose relative
importance controls the kinetics observed in these systems. The typical time scale for thermal fluctuations is
given by sfl ¼ m1 and is related to the passage probability per unit time. On the other hand, the rate of
change of the external field (temperature in our case) is
characterized by the following characteristic time:
sdr ¼ ðTH TL Þ=T_ . Notice that when sfl sdr , fluctuations become irrelevant, and the transition exhibits an
athermal character. For Cu–Al–Ni we have estimated
that this occurs for cooling rates above 30 K/min, while
for Cu–Zn–Al, rates above 0.1 K/min are sufficient.
Finally, once the energy barrier is overcome, the
system takes a given characteristic time to decay to the
subsequent metastable state (sav ). This time is not given
by the proposed model, but it is expected to be very
• When sav sdr sfl , the effect of thermal fluctuations will be irrelevant, and the transition will exhibit
metastability, an athermal character and rate-independent properties. This is, for instance, the case of
the studied Cu–Zn–Al alloy, for which TH TL is very
small [9].
• When sav sdr ’ sfl the effect of thermal fluctuations
will be observable and the transition will exhibit
metastability and an isothermal character. This is
the case, for instance, of the Cu–Al–Ni alloy for the
studied range of cooling rates.
• When sav sfl sdr the transition is expected to follow an equilibrium path.
An important feature which remains open is the following; how do the different time scales, and particularly
sdr which depends on TH TL , relate to physical properties of a given system? The different observed behaviour of systems transforming to different martensitic
structures suggests that relevant time scales are intimately related to the different accommodation mechanisms of the shape-change necessary to minimize the
stored elastic energy during the transformation. Actually, microgliding and microtwinning are the operative
mechanisms in Cu-based shape-memory alloys transforming to the 18R and 2H phases respectively [21].
Finally, we would like to point out that the relaxation
from one metastable state to another metastable state
both in the athermal as well as in the thermally activated
cases is an irreversible process with associated energy
dissipation. It is therefore interesting to consider briefly
the dissipation occurring during the martensitic transformation, which is directly related to the hysteresis of
the transition. This problem has been very controversial
in the past [22]. In any case, for thermally induced
transitions it has been well established that the dominant mechanism of hysteresis is the relaxation of elastic
strain energy. With this idea in mind, it has been conjectured that all energy losses can be considered, in
thermodynamic language, as mechanical work that the
system dissipates without entropy production [23]. This
is supported by the fact that elastic energy is released in
the form of elastic waves which are detected as AE. It
must be stressed that, once these elastic waves are absorbed in the neighbourhood of the system, the dissipative work finally gives rise to entropy production. The
preceding interpretation seems to be more adequate for
transitions displaying an athermal character which appear to be independent of cooling rate. Regarding the
results presented in the present paper, this interpretation
186
A. Planes et al. / Scripta Materialia 50 (2004) 181–186
is therefore more adequate for the transition to the 18R
structure (in which thermal fluctuations are in practice
not operative) than for the 2H . Whether or not this is
the case still remains an open issue. Very precise calorimetric measurements in samples under very wellcontrolled thermomechanical treatments are needed to
answer this important question.
Acknowledgements
This work has received financial support from the
CICyT (Spain), project MAT2001-3251, and from the
CIRIT (Catalonia), project 2001SGR00066. F.J.P acknowledge DGCyT for a Ph.D. Grant.
References
[1] Cao W, Krumhansl JA, Gooding RJ. Phys Rev B 1990;41:11319.
[2] Vives E, Ortın J, Ma~
nosa Ll, Rafols I, Perez-Magrane R, Planes
A. Phys Rev Lett 1994;72:1694.
[3] Bertotti G. Hysteresis in magnetism. New York: Academic Press;
1998.
[4] Nishiyama Z. Martensitic transformations. New York: Academic
Press; 1978.
[5] Rasmussen KØ, Lookman T, Saxena A, Bishop AR, Albers RC,
Shenoy SR. Phys Rev Lett 2001;87:055704.
[6] Kakeshita T, Takeguchi T, Fukuda T, Saburi T. Mater Trans JIM
1996;37:229.
[7] Aspelmeyer M, Klemradt U, Wood LT, Moss SC, Peisl J. Phys
Stat Sol (a) 1999;174:R9.
[8] Otsuka K, Ren X, Takeda T. Scr Mater 2001;45:145.
[9] Perez-Reche FJ, Vives E, Ma~
nosa Ll, Planes A. Phys Rev Lett
2001;87:195701.
[10] Perez-Reche FJ, Vives E, Ma~
nosa Ll, Planes A. J Phys IV 2003, in
press.
[11] Ahlers M. Prog Mater Sci 1986;30:135.
[12] Hurtado I, Segers D, Van Humbeeck J, Dorikens-Vanpraet L,
Dauwe C. Scr Metall Mater 1995;33:741.
[13] Ma~
nosa Ll, Jurado M, Gonzalez-Comas A, Obrad
o E, Planes A,
Zarestky J, et al. Acta Mater 1998;46:1045.
[14] Carrillo Ll, Ma~
nosa Ll, Ortın J, Planes A, Vives E. Phys Rev Lett
1998;81:1889.
[15] Mahato MC, Shenoy SR. J Stat Phys 1993;73:123.
[16] Lindg
ard PA, Mouritsen OG. Phys Rev Lett 1986;57:2458.
[17] Krumhansl JA. Solid State Comm 1992;84:251.
[18] Planes A, Ma~
nosa Ll. Solid State Phys 2001;55:159.
[19] Ahlers M. Philos Mag A 2002;82:1093.
[20] Guenin G, Clapp PC. Proceedings of the International Conference on Martensitic Transformations. Nara: The Japan Institute
of Metals; 1986. p. 171.
[21] Pelegrina JL, Ahlers M. Acta Metall Mater 1990;38:293.
[22] Lovey FC, Torra V. Prog Mater Sci 1999;44:189.
[23] Ortın J, Planes A. Acta Metall 1988;36:1873.