4 Statistical Theories of Phase Transitions

Kurt Binder

Institut für Physik, Johannes Gutenberg-Universität Mainz,

Mainz, Federal Republic of Germany

List of Symbols and Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . 240

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
4.2 Phenomenological Concepts . . . . . . . . . . . . . . . . . . . . . . . . 246
4.2.1 Order Parameters and the Landau Symmetry Classification . . . . . . . . . 246
4.2.2 Second-Order Transitions and Concepts about Critical Phenomena
(Critical Exponents, Scaling Laws, etc.) . . . . . . . . . . . . . . . . . . . 260
4.2.3 Second-Order Versus First-Order Transitions; Tricritical
and Other Multicritical Phenomena . . . . . . . . . . . . . . . . . . . . . 269
4.2.4 Dynamics of Fluctuations at Phase Transitions . . . . . . . . . . . . . . . 277
4.2.5 Effects of Surfaces and of Quenched Disorder on Phase Transitions:
a Brief Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
4.3 Computational Methods Dealing with the Statistical Mechanics
of Phase Transitions and Phase Diagrams . . . . . . . . . . . . . . . . 283
4.3.1 Models for Order – Disorder Phenomena in Alloys . . . . . . . . . . . . . 283
4.3.2 Molecular Field Theory and its Generalization
(Cluster Variation Method, etc.) . . . . . . . . . . . . . . . . . . . . . . . 288
4.3.3 Computer Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . 293
4.4 Concepts About Metastability . . . . . . . . . . . . . . . . . . . . . . . 298
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
4.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

Phase Transformations in Materials. Edited by Gernot Kostorz

Copyright © 2001 WILEY-VCH Verlag GmbH, Weinheim
ISBN: 3-527-30256-5
240 4 Statistical Theories of Phase Transitions

List of Symbols and Abbreviations

A prefactor of Friedel interaction
a lattice spacing
ai nearest-neighbor distance
Â, ¢ critical amplitude of specific heats
B critical amplitude of the order parameter
c concentration
C specific heat
C phenomenological coefficient
c lattice spacing in z-direction
c cluster of geometric configuration
Dc concentration difference
Ĉ, Ĉ¢ critical amplitudes of ordering susceptibilities
ccoex concentration at coexistence curve
ci concentration of lattice site i
cijk … elastic constants
d dimensionality
D̂ critical amplitude at the critical isotherm
d* marginal dimensionality
E electric field
ei random vectors
el (k, x) phonon polarization vector
F [F (x)] Helmholtz energy function in operator form
f factor of order unity
F Helmholtz energy
DF* Helmholtz energy barrier
F (j ) Helmholtz energy functional
F̃ scaling function of the Helmholtz energy
f, f I, …, f IV, f coefficients
F0 background Helmholtz energy of disordered phase
fcg coarse-grained Helmholtz energy
fm n quadrupole moment tensor
Freg background term of the Helmholtz energy
g constant factor in the structure function
G reciprocal lattice vector
G (x) order parameter correlation function
G̃ (x / x ) scaling function of the correlation function
Ĝ critical amplitude of the correlation function
G (x , x ) order parameter correlation function
g nc multispin correlation function
H ordering field
H magnetic field
Ᏼ Hamiltonian
H̃ scaled ordering field
 List of Symbols and Abbreviations 241

h, hc uniform magnetic field, critical value

HR random field (due to impurities)
i number and index for different lattice sites
J energy gained when two like atoms occupy neighboring sites
J exchange interaction
J̃ Fourier transform of the exchange interaction
Jij exchange interaction between spins at sites i and j (in cases where it is ran-
dom)
Jm strength of the magnetic interaction
Jnn nearest-neighbor interaction
Jnnn next-nearest-neighbor interaction
k wavevector of phonon
K1 , K2 phenomenological coefficients
kB Boltzmann constant
kF Fermi wavenumber
L coarse-graining length
ᏸ Lagrangian
M period (in Fig. 4-10)
M magnetization
Ml mass of an atom
mm sublattice magnetization
Ms spontaneous magnetization
N degree of polymerization
na lattice sites in state a
p pressure
P polarization
p dielectric polarization
P (J) statistical distribution of random variable J
P ({Si }) probability that a configuration {Si } occurs
p+ probability for spin up
p– probability for spin down
pb parameter in Landau equation (bicritical points)
Q normal coordinate
qa primitive vectors of the reciprocal lattice
qmax value of q where c (q) is at a maximum
q* nonzero wavenumber
qEA ordering parameter for the Ising spin glass
r phenomenological coefficient in the Landau expansion
R effective range of interaction
R ratio of Jnn and Jnnn
R* critical radius of a droplet
R il site of atom l in unit cell i
r, r ¢ coefficient in the Landau equation
rc radius of polymer coil
S entropy
242 4 Statistical Theories of Phase Transitions

S (q) structure factor

S (q, w ) scattering function
Si , Sj unit vectors in direction of the magnetic moment
t time
t reduced temperature 1 – T/T0 (in Sec. 4.2.2)
T0 temperature where r changes sign
T1 temperature where metastable ordered phase vanishes
Tb bicritical point
Tc critical temperature
tcr reduced temperature at crossover
TL Lifshitz point
Tl (x) transition temperature of superfluid 4He
Tt temperature of tricritical point
u phenomenological coefficient
u coefficient in Landau equation
U enthalpy
ul displacement vector
Ud surface of a d-dimensional sphere
v coefficient in the Landau equation
v ( |x |) Friedel potential
V total volume of the thermodynamic system
V (xi – xj ) pairwise interaction of two particles
Vab interaction between atoms A and B
w coefficient in the Landau equation
W action in classical mechanics
x, xn , xm , xa position vector and its components
z dynamical critical exponent
z coordinate (of position vector)
z coordination number
Z partition function
ᒗ scaling variable

a1m , a2 notation for phases in phase diagram

am magnetic phase
an nonmagnetic phase
a (x) short-range order parameter for metal alloys
a, a ¢ critical exponents related to specific heat
b exponent related to the order parameter
Ĝ critical amplitude of the susceptibility
g, g ¢ critical exponents related to the susceptibility
G0 phenomenological rate factor
d critical exponent at the critical isotherm
dmn Kronecker symbol
e binding energy of a particle
 List of Symbols and Abbreviations 243

e ik strain tensor
h exponent related to the decay of correlations at Tc
l wavelength
l label of the phonon branch
L factor changing the length scale
m chemical potential
m magnetic moment per spin
Dm chemical potential difference
m coex chemical potential at phase coexistence
n exponent related to the correlation length
x correlation length
xi concentration difference (Eq. (4-126))
r density
r (x) charge density distribution function
t characteristic time
j phase shift in Friedel potential
F order parameter
F (x) order parameter density
Fi concentration difference between two sublattices
F ms order parameter of metastable state
Fs spinodal curve
c (q) wavevector-dependent susceptibility
c el dielectric tensor
cT response function of the order parameter
cT isothermal susceptibility
cT staggered susceptibility
y concentration difference between two sublattices
y I, II ordering parameter for D03 structure
ya amplitude of mass density waves
w characteristic frequency (for fluctuations)
w̃ scaling function

AF antiferromagnetic
cg coarse-grained (as index)
coex coexistent (as index)
CV cluster variation method
DAG dysprosium aluminum garnet
ESR electron spin resonance
f.c.c. face-centered cubic
LRO long-range order
MC Monte Carlo (method)
MD molecular dynamics
MFA mean field approximation
MFA molecular field approximation
244 4 Statistical Theories of Phase Transitions

ms metastable state (as index)

NMR nuclear magnetic resonance
P paramagnetic
sc simple cubic
SRO short-range order
 4.1 Introduction 245

4.1 Introduction (Tsakalakos, 1984; Gonis and Stocks, 1989;

Turchi and Gonis, 1994), magnetic materi-
This chapter deals with the general theo- als (De Jongh and Miedema, 1974; Ausloos
retical concepts of phase transitions of ma- and Elliott, 1983), ferroelectrics (Blinc and
terials and their basis in statistical thermo- Zeks, 1974; Jona and Shirane, 1962), liquid
dynamics. This field has seen extensive crystals (Pershan, 1988; De Gennes, 1974;
scientific research over many decades, and Chandrasekhar, 1992), polymeric materials
a vast amount of literature exists, e.g., the (Flory, 1953; De Gennes, 1979; Riste and
statistical theory of critical point phenom- Sherrington, 1989; Binder, 1994; Baus et
ena fills a series of monographs that so far al., 1995), etc., or on particular types of
consists of 17 volumes (Domb and Green, phase transitions, such as commensurate –
1972 – 1976; Domb and Lebowitz, 1983 – incommensurate transitions (Blinc and Le-
1997). Introductory texts require whole vanyuk, 1986; Fujimoto, 1997), multicriti-
books (Stanley, 1971; Ma, 1976; Patashin- cal transitions (Pynn and Skjeltorp, 1983),
skii and Pokrovskii, 1979; Yeomans, 1992), first-order transitions (Binder, 1987 a),
and so does a comprehensive account of martensitic transformations (Nishiyama,
the Landau theory of phase transitions 1979; Salje, 1990), glass transitions
(Tolédano and Tolédano, 1987). (Jäckle, 1986; Angell and Goldstein, 1986;
This chapter therefore cannot give an ex- Cusack, 1987; Zallen, 1983; Zarzycki,
haustive description of the subject; instead 1991), percolation transitions (Stauffer and
what is intended is a tutorial overview, Aharony, 1992), melting (Baus, 1987),
which gives the flavor of the main ideas, wetting transitions (Dietrich, 1988; Sulli-
methods, and results, with emphasis on the van and Telo da Gama, 1985), metal – insu-
aspects which are particularly relevant for lator transitions (Mott, 1974; Friedmann
materials science, and hence may provide a and Tunstall, 1978), etc. Other approaches
useful background for other chapters in this focus on particular ways of studying phase
book. Hence this chapter will contain transitions: from the random-phase approx-
hardly anything new for the specialist; for imation (Brout, 1965) and the effective-
the non-specialist it will give an initial or- field theory (Smart, 1966) to advanced
ientation and a guide for further reading. techniques such as field-theoretical (Amit,
To create a coherent and understandable 1984) or real space renormalization (Burk-
text, a necessarily arbitrary selection of hardt and van Leeuwen, 1982), and com-
material has been made which reflects the puter simulation studies of phase transi-
author’s interests and knowledge. Reviews tions applying Monte Carlo methods
containing complementary material have (Binder, 1979, 1984 a; Mouritsen 1984;
been written by De Fontaine (1979), who Binder and Heermann, 1988; Binder and
emphasizes the configurational thermody- Ciccotti, 1996) or the molecular dynamics
namics of solid solutions, and by Khacha- techniques (Ciccotti et al., 1987; Hockney
turyan (1983), who emphasizes the theory and Eastwood, 1988; Hoover, 1987; Binder
of structural transformations in solids and Ciccotti, 1996), etc. For particular
(which, from a different perspective, are models of statistical mechanics, exact de-
discussed by Bruce and Cowley, 1981). scriptions of their phase transitions are
There are also many texts which either con- available (Baxter, 1982). Of particular
centrate on phase transformations of par- interest also are the dynamics of critical
ticular types of materials, such as alloys phenomena (Enz, 1979) and the dynamics
246 4 Statistical Theories of Phase Transitions

of first-order phase transitions (Gunton are ubiquitous, e.g., diamond is the meta-
et al., 1983; Koch, 1984; Binder, 1984 b, stable modification of solid carbon whereas
1989; Haasen et al., 1984; Zettlemoyer, graphite is the stable phase, some com-
1969; Abraham, 1974; Gunton and Droz, ments about the statistical mechanics of
1983; Herrmann et al., 1992). The latter metastability are made in Sec. 4.4.
subject is treated in two other chapters
in this book (see the Chapters by Wagner
et al. (2001) and Binder and Fratzl (2001)) 4.2 Phenomenological Concepts
and will not be considered here.
In this chapter, we shall briefly discuss In this section, the main facts of the the-
the statistical thermodynamics of phase ory of phase transitions are summarized,
transitions on a phenomenological macro- and the appropriate terminology intro-
scopic level (Sec. 4.2), i.e., the Landau the- duced. Rather than aiming at completeness,
ory of first- and second-order transitions, examples which illustrate the spirit of the
critical and multicritical phenomena, and main approaches will be discussed, includ-
also the dynamics of fluctuations at phase ing a discussion of critical phenomena and
transitions, and the effects of quenched scaling laws.
disorder in solids. The second part (Sec.
4.3) is devoted to the complementary “mi-
4.2.1 Order Parameters and the Landau
croscopic” approach, where we start from
Symmetry Classification
“model Hamiltonians” for the system under
consideration. The statistical mechanics of Table 4-1 lists condensed matter systems
such models naturally leads to the consider- that can exist in several phases, depending
ation of some computational methods, i.e., on external thermodynamics parameters
methods based on the molecular field the- such as pressure, p, temperature, T, electric
ory and its generalizations, and methods or magnetic fields (E, H). We assume that
based on computer simulation techniques. an extensive thermodynamic variable can
Since in solid materials metastable phases be identified (i.e., one that is proportional

Table 4-1. Order parameters for phase transitions in various systems.

System Transition Order parameter

Liquid – gas Condensation/evaporation Density difference Dr = r l – rg

(2) (1)
Binary liquid mixture Unmixing Composition difference Dc = c coex – c coex
Nematic liquid Orientational ordering –12 ·3 (cos2q ) – 1Ò
Quantum liquid Normal fluid a suprafluid ·y Ò (y : wavefunction)
Liquid – solid Melting/crystallization rG (G = reciprocal lattice vector)

Magnetic solid Ferromagnetic (Tc ) Spontaneous magnetization M

Antiferromagnetic (TN ) Sublattice magnetization Ms
(2) (1)
Solid binary mixture Unmixing Dc = c coex – c coex
AB Sublattice ordering y = (Dc – DcI )/2
II

Dielectric solid Ferroelectric (Tc) Polarization P

Antiferroelectric (TN ) Sublattice polarization Ps
Molecular crystal Orientational ordering Ylm (J, j)
 4.2 Phenomenological Concepts 247

to the volume of the system) which distin-

guishes between these phases (examples
are also given in Table 4-1), called the “or-
der parameter”. We shall denote the order
parameter as F, and the conjugate thermo-
dynamic variable, the “ordering field”, as
H. Using the thermodynamic potential, F,
which has as “natural variables” a field, H
(which is an “intensive” thermodynamic
variable, i.e., independent of the volume),
and the temperature, T, we have
F = – (∂F/∂H )T (4-1)
the other derivative being the entropy,
S = – (∂F/∂T )H (4-2)
As examples, consider a ferromagnet, where
the order parameter is the magnetization,
M, or a ferroelectric, where the order pa-
rameter is the dielectric polarization, P, Figure 4-1. The fluid-magnet analogy. On varying
the chemical potential m at m (T )
coex , the density r jumps
M = – (∂F/∂H)T (4-3 a) from the value at the gas branch of the gas – liquid co-
(1)
existence curve (rgas = r coex ) to the value at the liquid
P = – (∂F/∂E)T (4-3 b) branch of the coexistence curve (r liquid = r coex(2)
) (top
left). Similarly, on varying the (internal) magnetic
It is clear that such thermodynamic rela- field H, the magnetization M jumps from the nega-
tions can be written for any material, but a tive value of the spontaneous magnetization (– Ms ) to
quantity qualifies as an “order parameter” its positive value (top right). While this first-order
when a particular value of the “ordering liquid – gas transition occurs at a curve m coex (T ) in
the m – T-plane ending in a critical point (m c , Tc )
field” exists where the order parameter ex- where the transition is then of second order, the curve
hibits a jump singularity between two dis- where phases with positive and negative spontaneous
tinct values (Fig. 4-1). This means that for magnetization can coexist simply is H = 0 (T < Tc )
these values of the ordering field a first-or- (middle). The order parameter (density difference
der phase transition occurs, where a first Dr, or spontaneous magnetization Ms ) vanishes ac-
cording to a power law near Tc (bottom).
derivative of the thermodynamic potential
F exhibits a singularity. At this transition,
two phases can coexist; e.g., at the liq- be characterized by the same critical expo-
uid – gas transition for a chemical potential nents (see below), there is also an impor-
m = m coex (T ), two phases with different tant distinction that in the magnetic prob-
density coexist; and in a ferromagnet at lem, the Hamiltonian possesses a symme-
zero magnetic field, phases with opposite try with respect to the change of sign of the
sign of spontaneous magnetization can co- magnetic field; reversing this sign and also
exist. Although the fluid-magnet analogy reversing the sign of the magnetization
(Fig. 4-1) goes further, since the first-order leave the Hamiltonian invariant. Because
lines in the ( m , T ) or (H, T ) plane in both of this symmetry, the transition line must
cases end in critical points which may even occur at H = 0. Conversely, if the system
248 4 Statistical Theories of Phase Transitions

at H = 0 is in a monodomain state with cannot be measured directly, they never-

either positive or negative spontaneous theless provide a useful conceptual frame-
magnetization, this symmetry is violated; work.
this is described as “spontaneous symmetry Another problem which obscures the
breaking”. No such obvious symmetry, on analogy between different phase transitions
the other hand, is identified for the liq- is the fact that we do not always wish to
uid – gas transition of fluids. Consequently, work with the corresponding statistical en-
the curve m coex (T ) is a nontrivial function sembles. For a liquid – gas transition, we
in the m – T-plane, and no simple symmetry can control the chemical potential via the
operation acting on the gas phase atoms ex- fluid pressure, and thus a grand-canonical
ists that would transform this phase into a ensemble description makes sense. How-
liquid, or vice versa. Similar lack of sym- ever, for the binary mixture (AB), the or-
metry between the phases is noted for un- dering field would be a chemical potential
mixing transitions in binary fluid or solid difference D m = mA – m B between the spe-
mixtures, where the order parameter is a cies. In an ensemble where D m is the exter-
(2) (1)
concentration difference, Dc = c coex – c coex , nally controlled variable A atoms could
cf. Table 4-1, whereas solid binary mix- transform into B, and vice versa. Experi-
tures which exhibit sublattice ordering do mentally, of course, we do not usually have
possess a symmetry. The order parameter in mixtures in contact with a reservoir with
an alloy such as brass (b-CuZn) is the dif- which they can exchange particles freely,
ference between the relative concentration instead mixtures are kept at a fixed relative
of the two sublattices, y = (Dc II – Dc I )/2. concentration c. (Such an equilibrium with
However, the two sublattices physically are a gas reservoir can only be realized for
completely equivalent; therefore, the Ha- interstitial alloys such as metal hydrides or
miltonian possesses a symmetry against the oxides.) Thus the experiment is described
interchange of the two sublattices, which by a canonical ensemble description, with
implies that y changes sign, just as the (T, c) being the independent variables, al-
(idealized!) ferromagnet does for H = 0 though the grand-canonical ensemble of
(Fig. 4-1). In this example of an alloy mixtures is still useful for analytical theo-
which undergoes an order – disorder transi- ries and computer simulation. Now, as is
tion where the permutation symmetry obvious from Fig. 4-1, in a canonical en-
between the two sublattices is spontane- semble description of a fluid, the first-
ously broken, the “ordering field” conju- order transition shows up as a two-phase
gate to the order parameter is a chemical coexistence region. Here, at a given den-
potential difference between the two sub- sity, r , with rgas < r < r liquid , the relative
lattices, and hence this ordering field is amounts of the two coexisting phases are
not directly obtainable in the laboratory. given by the lever rule if interfacial contri-
The situation is comparable to the case butions to the thermodynamic potential can
of simple antiferromagnets, the order pa- be neglected. The same is true for a binary
rameter being the “staggered magnetiza- mixture with concentration c inside the two-
(1) (2)
tion” (= magnetization difference between phase coexistence region, c coex < c < ccoex .
the sublattices), and the conjugate ordering These different statistical ensembles also
field would change sign from one sublat- have pronounced consequences on the dy-
tice to the other (“staggered field”). Al- namic properties of the considered sys-
though the action of such fields usually tems: in binary mixtures at constant con-
 4.2 Phenomenological Concepts 249

centration, and also in fluids at constant

density, the order parameter is a conserved
quantity, whereas for a fluid at constant
chemical potential, the order parameter is
not conserved. The latter situation occurs,
for instance, for fluid – gas transitions in
physisorbed monolayers at surfaces, which
can exchange molecules with the surround-
ing gas phase which is in equilibrium with
this adsorbed layer.
An important distinction to which we
turn next is the order of a phase transition.
In the examples shown in the upper part of
Fig. 4-1, a first derivative of the appropri-
ate thermodynamic potential has a jump
singularity and therefore such transitions
are called first-order transitions. However,
if we cool a ferromagnet down from the
paramagnetic phase in zero magnetic field,
the spontaneous magnetization sets in con-
tinuously at the critical temperature Tc
(lower part of Fig. 4-1). Similarly, on cool-
ing the alloy b-CuZn from the state where Figure 4-2. Schematic variation with temperature T
it is a disordered solid solution (in the plotted for several quantities near a critical point Tc :
body-centered cubic phase), sublattice or- specific heat CH (top), ordering “susceptibility” c T
(middle part), and correlation length x of order pa-
dering sets in continuously at the critical
rameter fluctuations (bottom). The power laws which
temperature (Tc = 741 K, see Als-Nielsen hold asymptotically in the close vicinity of Tc are in-
(1976)). Whereas the first derivatives of dicated.
the thermodynamic potential at these con-
tinuous phase transitions are smooth, the
second derivatives are singular, and there- In this context (see also Fig. 4-1), a, a ¢, b ,
fore these transitions are also called sec- g , and g ¢ are critical exponents, while Â, ¢,
ond-order transitions. For example, in a B̂, Ĉ, and Ĉ¢ are called critical amplitudes.
ferromagnet the isothermal susceptibility Note that B̂ and b refer to the spontaneous
c T and the specific heat typically have magnetization, the order parameter (Fig. 4-1)
power law singularities (Fig. 4-2)
Ms = Bˆ (1 − T / Tc )b (4-6)
c T ≡ − ( ∂ F/ ∂H )T
2 2 T → Tc

⎧Cˆ ( T / Tc − 1) − g , T > Tc Behavior of the specific heat such as de-

= ⎨ −g′
(4-4) scribed by Eq. (4-5) immediately carries
ˆ
⎩ ′ (1 − Tc / T ) , T < Tc
T → Tc C
over to systems other than ferromagnets
CH ≡ − T ( ∂ 2 F/ ∂T 2 ) H = 0 such as antiferromagnets, the liquid –gas
system near its critical point, and brass
⎧ Aˆ ( T / Tc − 1) − a , T > Tc near its order – disorder transition; we must
= ⎨ −a′
(4-5)
ˆ
⎩ ′ (1 − T / Tc ) , T < Tc
T → Tc A remember, however, that H then means the
250 4 Statistical Theories of Phase Transitions

appropriate ordering field. In fact, this is parameter F as scalar quantities; although

also true for Eq. (4-4), but then the physical this is correct for the gas – fluid transition
significance of c T changes. For a two-sub- and for the unmixing transition of binary
lattice antiferromagnet, the ordering field mixtures, the uniaxial ferro- or antiferro-
is a “staggered field”, which changes sign magnets, uniaxial ferro- or antiferroelec-
between the two sublattices, and hence is trics, etc., and for order – disorder transi-
thermodynamically conjugate to the order tions in alloys or mixed crystals (solid so-
parameter of the antiferromagnet. Al- lutions), when only two sublatties need to
though such a field normally cannot be ap- be considered, there are also cases where
plied in the laboratory, the second deriva- the order parameter must have a vector or
tive, c T (in this case it is called “staggered tensor character. Obviously, for an iso-
susceptibility”) is experimentally access- tropic ferromagnet or isotropic ferroelec-
ible via diffuse magnetic neutron scatter- tric the order parameter in the three-dimen-
ing, as will be discussed below. Similarly, sional space is a three-component vector
for the ordering alloy b-CuZn, H stands for (see Eq. (4-3)). It also makes sense to con-
a chemical potential difference between the sider systems where a planar anisotropy is
two sublattices; the response function c T is present, such that M or E (Eq. (4-3)) must
again physically meaningful and measures lie in a plane and hence a two-component
the peak intensity of the diffuse scattering vector applies as an order parameter. How-
of X-rays or neutrons. (This scattering ever, for describing antiferromagnetic or-
peak occurs at the superlattice Bragg spot dering and for order – disorder transitions
characteristic of the sublattice ordering with many sublattices multicomponent or-
considered.) Finally, in the gas – liquid der parameters are also needed, and the
transition considered in Fig. 4-1, H is the number of components of the order param-
chemical potential and c T the isothermal eter, the so-called “order parameter dimen-
compressibility of the system. sionality”, is dictated by the complexity of
As will be discussed in more detail in the structure, and has nothing to do with
Sec. 4.2.2, the divergences of second deriv- the spatial dimension. This is best under-
atives of the thermodynamic potential at a stood by considering specific examples.
critical point (Eqs. (4-4), (4-5), Fig. 4-2) Consider, for example, the ordering of
are linked to a diverging correlation length Fe – Al alloys (Fig. 4-3): whereas in the dis-
of order-parameter fluctuations (Fig. 4-2). ordered A2 phase Fe and Al atoms are ran-
Hence any discussion of phase transitions domly distributed over the available lattice
must start with a discussion of the order pa- sites, consistent with the considered con-
rameter. The Landau theory (Tolédano and centration (although there may be some
Tolédano, 1987) which attempts to expand short-range order), in the ordered B2 phase
the thermodynamic potential in powers of (the FeAl structure) the bcc lattice is split
the order parameter, gives a first clue to the into two inter-penetrating simple cubic (sc)
question of whether a transition is of sec- sublattices, one taken preferentially by Fe,
ond or first order (see Sec. 4.2.3). the other by Al atoms. This is the same
We first identify the possible types of or- (one-component) ordering as in b-CuZn.
der parameters, since this will distinctly af- What is of interest here is the further “sym-
fect the nature of the Landau expansion. In metry breaking” which occurs for the D03
Eqs. (4-1) and (4-4) to (4-6) we have structure (realized in the Fe3 Al phase):
treated the ordering field H and the order here four face-centered cubic (f.c.c.) sub-
 4.2 Phenomenological Concepts 251

d) is preferentially occupied by Al, all

other sublattices being preferentially occu-
pied by iron. However, the role of the sub-
lattices (a, c) and (b, d) can be inter-
changed; hence a two-component order pa-
B2 rameter is needed to describe the structure,
namely (see for example, Dünweg and
Binder (1987))

y I = ma − mc + mb − md
y II = − ma + mc + mb − md
mm ≡ (1 / N ) ∑ (2 ci − 1) (4-7)
iŒm

where ci is the concentration of lattice site i

and mm the sublattice “magnetization” in
pseudo-spin language. As can example of
an ordering with m = 8 components, we can
consider the f.c.c. antiferromagnet MnO:
D03 as Fig. 4-4 shows, the magnetic structure
consists of an antiferromagnet arrangement
of ferromagnetically ordered planes. If the
ordering were uniaxially anisotropic, we

Figure 4-3. Body-centered cubic lattice showing the

B2 structure (top) and D03 structure (bottom). The
top part shows assignments of four sublattices, a, b,
c, and d. In the A2 structure, the average concentra-
tions of A and B atoms are identical on all four sub-
lattices whereas in the B2 structure the concentra-
tions at the b and d sublattices are the same (example:
stoichiometric FeAl, with Fe on sublattices a and c;
Al on sublattices b and d). In the D03 structure, the
concentrations at the a and c sublattice are still the
same, whereas the concentration at sublattice b dif-
fers from the concentration at sublattice d (example:
stoichiometric Fe3Al, with Al on sublattice b; all
other sublattices taken by Fe). From Dünweg and
Binder (1987). Figure 4-4. Schematic view of the MnO structure,
showing the decomposition of the fcc lattice into two
ferromagnetic sublattices with opposite orientation
lattices a, b, c, d must be distinguished. Al- of the magnetization (indicated by full and open cir-
though the concentrations on sublattices a cles, respectively). Each sublattice is a stack of par-
allel close-packed planes (atoms form a triangular
and c are still equivalent, a further symme- lattice in each plane). Note that there are four equiv-
try breaking occurs between the sublattices alent ways in which these planes can be oriented in
b and d, such that only one sublattice (e.g., an fcc crystal.
252 4 Statistical Theories of Phase Transitions

would have m = 4 because of the four pos-

sible ways of orienting the planes (and
there could be eight kinds of different do-
mains coexisting in the system). Since for
MnO the order parameter can take any or-
ientation is a plane, it has two independent
components, and hence the total number of
order parameter components is m = 8 (Mu-
Figure 4-5. Schematic view of a lattice plane of a
kamel and Krinsky, 1976). An even larger para-H2 crystal, indicating the orientational order of
number of components is needed to de- the ellipsoid H2 molecules.
scribe the ordering of solid 3He. In view of
this, it is clear that even m Æ •, which is
also called the spherical model (Berlin and see Grazhdankina (1969)). In this example
Kac, 1952; Joyce, 1972) is a useful limit to it is clear that the spontaneous magnetiza-
consider from the theoretical point of view. tion, M, is the “primary order parameter”
Apart from this m-vector model for the whereas the tetragonal distortion (c/a – 1),
order parameter, Table 4-1 illustrates the where c is the lattice spacing in the lattice
need to consider order parameters of tenso- direction where the magnetization direc-
rial character. This happens in molecular tion occurs and a is the lattice spacing
crystals such as para-H2 (Fig. 4-5), N2 , O2 perpendicular to it, is a “secondary order
and KCN, in addition to liquid crystals. parameter”. For purely structural phase
Whereas the atomic degrees of freedom transitions where all considered degrees
considered for ferro- or antiferromagnetic of freedom are atomic displacements, the
order are magnetic dipole moments, for proper distinction between primary and
ferro- or antiferroelectric order electric di- secondary order parameters is much more
pole moments, the degree of freedom of the subtle (Tolédano and Tolédano, 1987).
molecules in Fig. 4-5 which now matters is We first formulate Landau’s theory for
their electric quadrupole moment tensor: the simplest case, a scalar order parameter
density F (x). This density is assumed to
⎛ 1 3 ⎞ be small near the phase transition and
fmn = ∫ dx r ( x ) ⎜ xm xn − ∑ xl2 dmn ⎟ (4-8)
⎝ 3 l =1 ⎠ slowly varying in space. It can be obtained
by averaging a microscopic variable over a
where r (x) is the charge density distribu- suitable coarse-graining volume Ld in d-di-
tion function of a molecule, x = (x1 , x2 , x3 ), mensional space. For example, in an aniso-
and dmn is the Kronecker symbol. tropic ferromagnet the microscopic vari-
Proper identification of the order param- able is the spin variable F i = ±1 pointing in
eter of a particular system therefore needs a the direction of the magnetic moment at
detailed physical insight, and often is com- lattice site i; in an ordering alloy such as b-
plicated because of coupling between dif- CuZn the microscopic variable is the dif-
ferent degrees of freedom, e.g., a ferromag- ference in concentration of Cu between the
netic material which is cubic in the para- two sublattices I and II in unit cell i, F i =
magnetic phase may become tetragonal as ciII – ciI ; and in an unmixing alloy such as
the spontaneous magnetization develops, Al – Zn, the microscopic variable is the
owing to magnetostrictive couplings (for a concentration difference between the local
more detailed discussion of this situation, Al concentration at lattice site i (ci = 1 if it
 4.2 Phenomenological Concepts 253

is taken by Al and ci = 0 otherwise) and the the change of sign of the order parameter
critical concentration, F i = ci – ccrit . In all for H = 0 and thus odd powers such as
of these cases F (x) is considered to be an F 3 (x) do not occur; this is true for magnets
order parameter field defined in continuum (no direction of the magnetization is pre-
space: ferred without magnetic field in a ferro-
magnet) and for sublattice ordering of al-
F ( x ) = ∑ Fi / Ld (4-9)
i ΠLd
loys such as b-CuZn (since whether the Cu
atoms preferentially occupy sublattices a, c
x being the center of gravity of the “vol-
in Fig. 4-3 or sublattices b, d is equivalent),
ume” Ld . The appropriate magnitude of the
but it is not true in general (e.g., third-order
linear dimension L of the coarse-graining
terms do occur in the description of the
cell will be discussed later; obviously it
Cu3Au structure, or in the ordering of rare
must be much larger than the lattice spac-
gas monolayers adsorbed on graphite in
ing in order for the continuum description –
the ÷3 structure, as will be discussed be-
to make sense. Then a Helmholtz energy
low).
functional F [F (x)] is assumed:
In order to understand the meaning and
1 F use of Eq. (4-10), consider first the fully
F [F ( x )] = 0 (4-10)
kB T kB T homogeneous case, —F (x) ∫ 0, F (x) ∫ F 0 ;
then F [F ] is the standard Helmholtz en-
+ ∫ dx ⎧⎨ r F 2 ( x ) + uF 4 ( x )
1 1
ergy function of thermodynamics, which
⎩2 4
needs to be minimized with respect to F in
H 1 ⎫
− F (x) + [ R ∇F ( x )]2 ⎬ order to determine the thermal equilibrium
kB T 2d ⎭ state, and Ú dx = V the total volume of the
where F0 is the background Helmholtz en- system. Thus,
ergy of the disordered phase, and r, u, and (4-12)
1 ⎛ ∂F ⎞
R are phenomenological constants. (In fact, ⎜ ⎟ = r F0 + u F0 = 0
3

R can be interpreted as the effective range kB T V ⎝ ∂F0 ⎠ T

H =0
of interaction between the atomic degrees
which is solved by
of freedom F i , as will be seen below.) Ob-
viously, Eq. (4-10) is the Taylor series-type F0 = 0 , T > Tc
expansion of F [F (x)] in powers of F (x) F 0 = ± (– r/u) 1/2
(4-13)
and —F (x), where just the lowest order = ± (r¢/kB u)1/2 (Tc /T – 1)1/2 , T < Tc
terms were kept. This makes sense if both
the coefficients u and R 2 are positive con- Hence Eqs. (4-10) and (4-12) indeed yield
stants at Tc , whereas the essential assump- a second-order transition as T is lowered
tion which defines F as playing the role of through Tc . For T < Tc , a first-order transi-
an order parameter of a second-order phase tion as a function of H occurs, since F 0
transition is that the coefficient r changes jumps from (– r/u)1/2 to – (– r/u)1/2 as H
its sign at the transition as the variable of changes sign. This behavior is exactly that
interest (the temperature in the present shown schematically in Fig. 4-1, with
case) is varied, b = 1/2, B̂ = (r¢/kB u)1/2 and F 0 (H = 0) = MS .
If u < 0 in Eq. (4-10), however, we must
kB Tr = r¢ (T – Tc ) (4-11) not stop the expansion at fourth order but
Note also that in Eq. (4-10) we have as- rather must include a term –16 v F 6 (x) (as-
sumed a symmetry in the problem against suming now v > 0). Whereas in the second-
254 4 Statistical Theories of Phase Transitions

Figure 4-6. Schematic variation of the Helmholtz energy in the Landau model at transitions of (a) second or-
der and (b) first order as a function of the (scalar) order parameter F. Cases (a) and (b) assume a symmetry
around F = 0, whereas case (c) allows a cubic term.

order case F (F ) has two minima for T < Tc this value, i.e.,
which continuously merge as T Æ Tc and
only one minimum at F = 0 remains for [ F (F0 ) − F ( 0 )]/(V kB T )

= F02 ⎛ + F02 + F04 ⎞ = 0

T > Tc (Fig. 4-6 a) F [F ] now has three min- r u v
(4-15)
ima for T0 < T < Tc , and the temperature T0 ⎝2 4 6 ⎠ T = Tc
where r changes sign (r = r¢ (T – T0 ) now) With some simple algebra Eqs. (4-14) and
differs from the phase transition tempera- (4-15) yield
ture Tc where the order parameter jumps
discontinuously from zero for T > Tc to Tc = T0 + 3 u2/(32 r¢v) (4-16)
(F 0 )Tc = ± (3 u/4 v)1/2, see Fig. 4-6 b. and the “stability limit”, where the mini-
These results are found by analogy with mum describing the metastable ordered
Eq. (4-12) from phase in the disordered phase above Tc dis-
1 ⎛ ∂F ⎞ appears, is given by
⎜ ⎟
kB T V ⎝ ∂F0 ⎠ T T1 = T0 + u2/(8 r¢v) (4-17)
H =0
= F0 ( r + uF02 + v F04 ) = 0 (4-14) The significance of such metastable states
as described by the Landau theory for first-
which is solved by order transitions and the associated stabil-
ity limits T0 (for the disordered phase at
F02 = − u /( 2 v ) + (u / 2 v )2 − r / v
T < Tc ) and T1 will be discussed further in
(choosing the minus sign of the square root Sec. 4.4.
would yield the maxima rather than the The alternative mechanism by which a
minima in Fig. 4-6 b). On the other hand, first-order transition arises in the Landau
we know that [F – F (0)]/(V kB T ) = 0 in the theory with a scalar order parameter is
disordered phase, and Tc can be found (Fig. the lack of symmetry of F against a sign
4-6 b) from the requirement that the free change of F. Then we may add a term
energy of the ordered phase then is equal to –13 w F 3 to Eq. (4-10), with another pheno-
 4.2 Phenomenological Concepts 255

menological coefficient, w. For u > 0, F (F ) only in mean-field theory, but lacks any
may have two minima (Fig. 4-6 c); again fundamental justification in statistical me-
the transition occurs when the minima are chanics. Schemes such as those shown in
equally deep. For r = r¢ (T – T0 ) this hap- Fig. 4-6 make sense for a local “coarse-
pens when grained free energy function” only (which
depends on the length scale L introduced in
Tc = T0 + 8 w 2/(81 u r¢) (4-18)
Eq. (4-9)), see Sec. 4.2.2, but not for the
the order parameter jumping there from global free energy.
F 0 = – 9 r/w to F 0 = 0. Again a stability After this digression we return to the
limit of ordered state in the disordered generalization of the Landau expansion,
phase occurs, i.e. Eq. (4-10), the case where the order param-
eter has vector or tensor character. How
T1 = T0 + w 2/(4 u r¢) (4-19)
can we find which kinds of term appear in
At this point, an important caveat should be the expansion?
emphasized: free energy curves involving Basically, there are two answers to this
several minima and maxima as drawn in question. A general method, which follows
Fig. 4-6 are used so often in the literature below, is a symmetry classification based
that many researchers believe these con- on group theory techniques (Landau and
cepts to be essentially rigorous. However, Lifshitz, 1958; Tolédano and Tolédano,
general principles of thermodynamics indi- 1987). A very straightforward alternative
cate that in thermal equilibrium the ther- approach is possible if we consider a par-
modynamic potentials are convex func- ticular model Hamiltonian Ᏼ (for a brief
tions of their variables. In fact, F (F 0 ) discussion of some of the most useful mod-
should thus be convex as a function of F 0 , els of statistical mechanics for studying
which excludes multiple minima! For Fig. phase transitions, see Sec. 4.3.1). We can
4-6 a and b this means that for T < Tc in then formulate a microscopic mean-field
states with – F 0 < F < F 0 (where F 0 is the approximation (MFA), such as the effec-
solution of Eqs. (4-13) or (4-14), respec- tive field approximation of magnetism
tively) the thermal equilibrium state is (Smart, 1966) or the Bragg-Williams ap-
not a pure homogeneous phase: rather, the proximation for order – disorder phenomena
minimum free energy state is given by the in alloys (De Fontaine, 1979), where we
double-tangent construction to F (F ) and then expand the MFA Helmholtz energy di-
this corresponds to a mixed phase state (the rectly.
relative amounts of the coexisting phases As an example of this approach, we con-
are given by the well known lever rule). sider a specific model of a ternary alloy
Now it is standard practice, dating back to where each lattice site i may be taken by ei-
van der Waals’ interpretation of this equa- ther an A, a B, or a C atom, assuming con-
tion of state for fluids, to interpret the part centrations cA = cB = cC = 1/3, and assuming
of F (F ) in Fig. 4-6 which lies above the that an energy J is achieved if two neigh-
F (F ) given by the double-tangent con- boring sites are taken by the same kind of
struction as a metastable state provided atom. This is a special case of the q-state
that c T = (∂2 F/ ∂F 2 )T < 0, whereas states Potts model (Potts, 1952; Wu, 1982), the
with c T < 0 are considered as intrinsically Hamiltonian being
unstable states. As will be seen in Sec. 4.4,
Ᏼ Potts = − ∑ J dSi S j , Si = {1, 2, …, q} (4-20)
this notion is intrinsically a concept valid 〈 ij 〉
256 4 Statistical Theories of Phase Transitions

The states 1, 2, and 3 of the Potts spin stand in the expansion:

for the three kinds of atomic species A, B,
and C here and ·ij Ò denotes a sum over F zJ
=− − ln 3
nearest neighbor pairs in the lattice. In the V kB T 6 kB T
MFA, we construct the Helmholtz energy ⎛ zJ ⎞ 2
F = U – T S simply by expressing both en- + 3 ⎜1 − ⎟ (F1 + F2 + F1 F2 )
2
⎝ 3 kB T ⎠
thalpy U and entropy S in terms of the frac- 9
tions na of lattice sites in states a . The en- + (F12 F2 + F1 F22 ) + … (4-24 )
2
tropy is simply the entropy of randomly
mixing these species, and, using Stirlings As expected, there is a temperature T0
equation, this yields the standard expres- (= z J/3 kB ) where the coefficient of the
sion (see elementary textbooks on statisti- quadratic term changes sign.
cal thermodynamics): Of course, for many phase transitions a
3 specific model description is not available,
S=−V ∑ na ln na (4-21) and even if a description in terms of a
a =1 model Hamiltonian is possible, for compli-
In the enthalpy term, MFA neglects correla- cated models the approach analogous to
tions in the occupation probability of neigh- Eqs. (4-20) and (4-24) requires tedious cal-
boring sites. Hence the probability of find- culation. Clearly the elegant but abstract
ing a nearest neighbor pair in a state is sim- Landau approach based on symmetry prin-
ply na2 , and in a lattice with coordination ciples is preferable when constructing the
number z there are z /2 pairs per site. Hence Landau expansion. This approach starts
from the observation that usually the disor-
3
zJV
U=− ∑ na2 (4-22) dered phase at high temperatures is more
2 a =1 “symmetric” than the ordered phase(s) oc-
and thus curring at lower temperature. Recalling
the example described in Fig. 4-3, referring
3 3
F zJ to Fe – Al alloys: in the high temperature
=− ∑ na2 + ∑ na ln na (4-23)
V kB T 2 kB T a =1 a =1 A2 phase, all four sublattices a, b, c, and d
are completely equivalent. This permuta-
We could directly minimize F with respect tion symmetry among sublattices is broken
3
in the B2 phase (FeAl structure) where the
to na , subject to the constraint ∑ na = 1,
a =1 concentration on sublattices a, c differs
since each site should be occupied. In order from the concentration on sublattices b, d.
to make contact with the Landau expan- A further symmetry breaking occurs when
sion, however, we rather expand F in terms we go from the B2 phase to the D03 phase
of the two order parameter components (Fe3Al structure), where the concentration
F1 = n1 – 1/3 and F2 = n2 – 1/3 (note that all on sublattice b differs from that on sublat-
ni = 1/3 in the disordered phase). Hence we tice d. In such cases the appropriate struc-
recognize that the model has a two-compo- ture of the Landau expansion for F in terms
nent order parameter and there is no sym- of the order parameter F is found from the
metry between F i and – F i . So cubic terms principle that F must be invariant against
in the expansion of F are expected and do all symmetry operations of the symmetry
occur, whereas for a properly defined order group G0 describing the disordered phase.
parameter, there cannot be any linear term In the ordered phase, some symmetry ele-
 4.2 Phenomenological Concepts 257

ments of G0 fall away (spontaneously This approach also carries over to cases
broken symmetry); the remaining symme- where the order parameter is a tensor. For
try elements form a subgroup G of G0 . example, for elastic phase transitions
Now the invariance of F must hold separ- (Cowley, 1976; Folk et al., 1976) the order
ately for terms F k of any order k and this parameter is the strain tensor {e ik}. Apply-
requirements fixes the character of the ing the summation convention (indices oc-
terms that may be present. curring twice in an expression are summed
Rather than formulating this approach over), the Landau expansion is
systematically, which would require a
F [e ik (x)]
lengthy and very mathematical exposi-
tion (Tolédano and Tolédano, 1987), we = F0 + Ú dx ( –12 ciklm e ik e lm
rather illustrate it with a simple example. + –13 ciklmrs
(3)
e ik e lm e rs
Suppose a cubic crystal exhibits a transi- + –14 ciklmrsuv
(4)
e ik e lm e rs e uv
tion from a para-electric to a ferroelectric
+ … + gradient terms) (4-27)
phase, where a spontaneous polarization
P = (P1 , …, Pn ), n = 3, appears. F is then Here the ciklm are elastic constants and
(3) (4)
given as follows: ciklmrs and ciklmrsuv analogous coefficients
of higher order (“anharmonic”) terms. For
= 0 + ∫ dx ⎡⎢ c el−1 P 2
F [ P ( x )] F 1
most elastic transitions symmetry permits
kB T kB T ⎣2 some nonzero ciklmrs (3)
and hence leads to
1 ⎛ n 4 n ⎞ first-order transitions. Examples of such
+ ⎜ u ∑ Pm + u ′ ∑ Pm2 Pj2 ⎟ + … systems are the “martensitic transitions” in
4 ⎝ m =1 m < j =1 ⎠
Nb3Sn and V3Si; there the elastic distortion
R2 n ⎤ jumps at the transition from zero to a very
+ ∑ (∇Pm )2 + …⎥ (4-25)
small value (e ik ≈ 10– 4 ) and a Landau ex-
2d m =1 ⎦
pansion makes sense. Note that the Landau
Whereas the quadratic term of a general theory is not very useful quantitatively for
dielectric medium would involve the in- transitions which are very strongly first or-
verse of the dielectric tensor, ∑ ( c el−1 )ij Pi Pj , der, since in that case high-order terms in the
ij
this term is completely isotropic for cubic Taylor expansion are not negligible. More
crystals. Inversion symmetry requires in- about martensitic phase transformations can
variance against P Æ – P and hence no be found in the Chapter by Delaey (2001).
third-order term occurs. The fourth-order Just as in a ferromagnetic transition the
term now contains the two “cubic invari- inverse magnetic susceptibility c –1 van-

( )
ants” ∑ Pi4 and ∑ Pi2 . Here we invoke
2 ishes at Tc (Eq. (4-4)) and in a ferroelectric
i i transition the inverse dielectric susceptibil-
the principle that all terms allowed by sym- ity c el–1 vanishes at Tc (Eq. (4-25)), in a
metry will actually occur. Now Eq. (4-25) “ferroelastic” transition one of the elastic
leads, in the framework of the Landau the- constants ciklm vanishes at T0 . For example,
ory, to a second-order transition if both in KCN such a softening is observed for the
elastic constant c44 . Note that this material
u>0 and u + u¢ > 0 (4-26)
has already been mentioned as an example
whereas otherwise we have a first-order of a phase transition where the order pa-
transition (then terms of sixth order are rameter is the electric quadrupole moment
needed in Eq. (4-25) to ensure stability). tensor (Eq. (4-8)), describing the orienta-
258 4 Statistical Theories of Phase Transitions

tion of the dumbbell-shaped CN–-ions: (mediated via acoustic phonons) is more

at the same time, it can be considered as important than their direct electric quadru-
an example for an elastic transition, where pole – quadrupole interaction. The elastic
the order parameter is the strain tensor character of the phase transition is therefore
(emn (x))! Such an ambiguity is typical of an intrinsic phenomenon for this material.
many first-order transitions, because vari- We briefly discuss the appropriate con-
ous dynamical variables (such as a local struction of the order parameter (compo-
dielectric polarization P (x), a local quadru- nents) for order – disorder transitions. Fig.
pole moment fmn (x) and the strain emn (x) 4-3 (and Eq. (4-7)) illustrated how we can
may occur simultaneously in a crystal and visualize the ordered structure in real
are coupled together. In a Landau expan- space, and apply suitable symmetry opera-
sion, such couplings are typically of the bi- tors of the point group. However, it is often
quadratic energy – energy coupling type, more convenient to carry through a corre-
i.e., various types of mixed fourth-order sponding discussion of the ordering in re-
terms occur, such as P 2 (x) emn em ¢ n ¢ umn m ¢ n ¢ . ciprocal space rather than in real space (re-
Although this is also true for second-order member that the ordering shows up in
transitions, there the “primary order pa- superlattice Bragg spots appearing in the
rameter” is distinct from the fact that the reciprocal lattice in addition to the Bragg
associate inverse response function c T–1 spots of the disordered phase). For exam-
vanishes at Tc . For the “secondary order ple, consider rare gas monolayers adsorbed
parameters” the corresponding inverse re- on graphite: at low temperatures and pres-
–
sponse function stays finite. With first- sures, the adatoms form a ÷3 structure
order transitions, the inverse response commensurate with the graphite lattice.
–
function of the primary order parameter This ÷3 structure can be viewed as a trian-
would vanish at the hypothetical tempera- gular lattice decomposed into three sublat-
ture T0 (stability limit of the disordered tices, such that the adatoms preferentially
phase), which, however, cannot normally occupy one sublattice. Mass density waves
be reached in a real experiment, see Sec. are taken as an order parameter (Bak et al.,
4.4. K + CN– is a good example of such a 1979; Schick, 1981):
complicated situation with three local or- 3
der parameters coupled together, since the r ( x ) = ∑ [ya exp( i q ⋅ x )
CN– ion also has a dipole moment. After a =1
+ y − a exp( − qa ⋅ x )] (4-28)
the first-order transition at T = 110 K from
the cubic “plastic crystal” phase (where Here the qa are the three primitive vectors
there is no long-range orientational order, associated with the reciprocal lattice of the
the CN – dumbbells can rotate) to the tet- rare gas monolayer (a being the lattice
ragonally distorted, orientationally ordered spacing of the triangular lattice)
phase, a second first-order transition oc-
curs at T = 80 K to an antiferroelectric 2p ⎛ 1 ⎞
q1 = ⎜ 0, ⎟
phase where the dipole moments order. In a ⎝ 3⎠
this system, only a microscopic theory 2p ⎛ 1 1 ⎞
(Michel and Naudts, 1977, 1978; De Raedt q2 = ⎜− , − ⎟
a ⎝ 2 2 3⎠
et al., 1981; Lynden-Bell and Michel, 1994)
could clarify that the elastic interaction 2p ⎛ 1 1 ⎞
q3 = ⎜ ,− ⎟ (4-29)
between the quadrupole moments fmn (x) a ⎝ 2 2 3⎠
 4.2 Phenomenological Concepts 259

and ya and y– a are the (complex!) order structures in terms of „spin density waves“
parameter amplitudes. In constructing the (Villain, 1959; Kaplan, 1959). We shall
free energy expansion with the help of Eq. outline the connection between this ap-
(4-28), note that the periodicity of the proach and the MFA, generalizing the ap-
underlying graphite lattice allows invariant proach of Eqs. (4-21) – (4-23) slightly, in
“umklapp” terms (the phase factors of Sec. 4.3. Here we only mention that these
third-order terms add up to a reciprocal lat- concepts are closely related to the descrip-
tice vector of the graphite lattice). Keeping tion of structural transitions in solids,
only those terms in the Landau expansion, where the order parameter can often be
–
and which are nonzero in the ÷3 structure, considered as a “frozen phonon”, i.e., a dis-
with the real order parameter components placement vector wave (Bruce and Cowley,
––
(i = ÷–1) 1981). Note also that the approach of con-
3 centration waves is not restricted to solids,
F1 = ∑ (ya − y − a ) /( 2 3 ) but can also be used to describe meso-
a =1 phases in fluid block-copolymer melts
3
(Leibler, 1980; Fredrickson and Helfand,
F2 = ∑ (ya − y − a ) /( 2 i 3 ) (4-30)
a =1 1987; Fredrickson and Binder, 1989; Bin-
der, 1994), see Fig. 4-7, in liquid crystal-
the expression line polymers, etc.
F 1 As a final remark in this section we men-
= − r (F12 + F22 ) (4-31) tion that not for all phase transitions in sol-
V kB T 2
ids there does exist a group – subgroup rela-
1 1
+ w (F13 − 3F1 F22 ) + u (F12 + F22 )2 tionship between the two groups G1 and G2
3 4 describing the symmetry of the phases co-
is obtained. This is equivalent to the result existing at the transition. These phases can-
for the three-state Potts model, Eq. (4-24) not be distinguished by an order parameter
(Alexander, 1975), after the quadratic form which is zero in one phase and becomes
is diagonalized. A three-dimensional ana- nonzero in the other. Such transitions must
log of this order (where the planes exhibit- be of first order. Examples of this situation
–
ing ÷3 structure are stacked together to are well known for structural phase transi-
form a hexagonal lattice) occurs in the tions, e.g., the tetragonal – orthorhombic
intercalated compound C6Li (Guerard and transition of BaTiO3 or the “reconstruc-
Herold, 1975; Bak and Domany, 1979). tive” transition from calcite to aragonite
Similarly to the description of the den- (Guymont, 1981). For the so-called “non-
sity modulation in the superstructure of disruptive transitions” (Guymont, 1981),
adsorbed layers (in two dimensions) or the new structure can still be described in
interstitial compounds (in three dimen- the framework of the old structure (i.e., its
sions) in terms of mass density waves of symmetry elements can be specified, the
the adsorbate (or interstitial, respectively), Wyckoff positions can be located, etc.).
the superstructure ordering in binary alloys Landau-type symmetry arguments still
can be described in terms of concentration yield information on the domain structures
waves (Khachaturyan, 1962, 1963, 1973, arising in such phase transitions (Guymont,
1983; De Fontaine, 1975, 1979). The same 1978, 1981). The tetragonal – orthorhombic
concept was used even earlier to describe transition of BaTiO3 is considered to be an
the magnetic ordering of helimagnetic spin example of such a non-disruptive transi-
260 4 Statistical Theories of Phase Transitions

ond-order transition vanishes continuously

(Fig. 4-1). We consider the accompanying
critical singularities (Fig. 4-2), in the
framework of Landau theory (Eq. (4-10)).
For a proper understanding of the critical
fluctuations of the order parameter, we
must no longer restrict the treatment to the
homogeneous case —F (x) ∫ 0 as was done
in Eqs. (4-12) and (4-13); we now wish to
consider, for example, the response to an
inhomogeneous, wavevector-dependent or-
dering field H (x):

H (x) = Hq exp (i q · x) (4-32)

Although for a magnetic transition such a

Figure 4-7. (a) Chemical architecture of a diblock field cannot be directly applied in the la-
copolymer. A diblock copolymer consists of a poly- boratory, the action of such fields can be
merized sequence of A monomers (A-block) cova-
indirectly probed by appropriate scattering
lently attached to a similar sequence of B monomers.
(b) The microphase separation transition occurs measurements, e.g., magnetic neutron scat-
when a compositionally disordered melt of copoly- tering. This is because the scattering can
mers, which are in random coil configurations (right) be viewed as being due to the inhomogene-
transforms to a spatially periodic, compositionally ous (dipolar) field that the magnetic mo-
inhomogeneous phase (left) on lowering the temper-
ment of the neutron exerts on the probe.
ature. For nearly symmetric copolymers the ordered
phase has the lamellar structure shown. Since the Therefore, the scattering intensity for a
wavelength of the concentration wave here is of the scattering vector q is related to the wave-
order of the coil gyration radius, which may be of the vector-dependent susceptibility (see Kittel
order of 100 Å for high molecular weight, the order- (1967)). Similarly, by the scattering of
ing occurs on a “mesoscopic” rather than micro-
X-rays the wavevector-dependent response
scopic scale (Å) and hence such phases are called
“mesophases”. From Fredrickson and Binder (1989). function to the sublattice ordering field can
be measured in ordering alloys such as b-
CuZn (see Als-Nielsen (1976) for a discus-
tion, while “reconstructive transitions” and sion of such scattering experiments in vari-
martensitic transformations of the type of ous systems).
the f.c.c. – b.c.t. transition in Fe are “disrup- Hence in order to deal with Eq. (4-10)
tive” (see the Chapter by Delaey (2001)). we now have to minimize a functional
We shall not go into detail about such prob- rather than a simple function as was done
lems here, however. in Eq. (4-12). In order to do so, we note
that the problem of minimizing Ú dx f (y ,
—y ) in Eq. (4-10) is analogous to the prob-
4.2.2 Second-Order Transitions
lem of minimizing the action in classical
and Concepts about Critical Phenomena
mechanics, W = Ú dt ᏸ (x, ẋ ), where ᏸ is the
(Critical Exponents, Scaling Laws, etc.)
Lagrangian and ẋ = dx/dt, the velocity of
We now return to the case where at a a particle at point x (t = time). Just as in
critical point the order parameter of a sec- classical mechanics, where this problem
 4.2 Phenomenological Concepts 261

is solved in terms of the Euler – Lagrange For T < Tc , using Eq. (4-13) the term 5 v F04
equation near Tc is still negligible, and hence

( ∂ᏸ / ∂x ) −
d
[ ∂ᏸ / ∂ ( dx / dt )] = 0 −1 1
cT = = (4-37)
dt 2 kB T r 2 r ′ ( Tc − T )
we conclude here that x = R2 /( − 2 rd ) = ( R / 2 r ′ d ) (T − Tc ) −1/ 2
( ∂ f / ∂y ) − ∇ [ ∂ f / ∂ (∇y )] = 0 Eqs. (4-36) and (4-37) simply imply the
in order that the Helmholtz energy func- well known Curie – Weiss law, i.e., the ex-
tional is a minimum. Eq. (4-10) thus yields ponents g, g ¢, defined in Eq. (4-4), are, in
the framework of the Landau theory,
r F ( x ) + uF 3 ( x ) + v F 5 ( x ) g = g¢ = 1 (4-38)
R2 2 H (x)
− ∇ F (x) = (4-33) Comparing Eqs. (4-35) and (4-37) we also
d kB T
conclude that
Using
n = n ¢ = 1/2 (4-39)
F (x) = F 0 + DF (x) = F 0 + DFq exp (i q · x) The physical meaning of the correlation
Eqs. (4-32) and (4-33) yield, on linearizing length x is easily recognized if the well
in DFq for small Hq , the wavevector-de- known fluctuation relationship
pendent susceptibility 1
c (q) = ∑ exp( i q ⋅ x )
DFq kB T x
c (q) ≡ × [ 〈F ( 0 ) F ( x )〉 − F02 ] (4-40)
Hq
−1 is expanded to second order in q as
1 ⎡ R2 2 ⎤
⎢ r + 3 u F0 + 5 v F0 + d q ⎥
2 4
=
kB T ⎣ ⎦ 1 ⎧
c (q) ≈ ⎨∑ [ 〈F ( 0 ) F ( x )〉 − F0 ]
2
cT kB T ⎩ x
= (4-34 )
1 + x 2 q2 1 ⎫
− ∑ ( q ⋅ x )2 [ 〈F ( 0 ) F ( x )〉 − F02 ]⎬ (4-41)
From Eq. (4-34) we can simply read off
2 x ⎭
the temperature dependence of the suscep- which can be written as
tibility c T (Eq. (4-4)) and the correlation
1
length x of order parameter fluctuations, c (q) = ∑ [〈F ( 0 ) F ( x )〉 − F02 ]
which is generally expected to behave as kB T x

⎧ q2
⎧xˆ ( T / Tc − 1) − n , T > Tc × ⎨1 − ∑ x 2 [〈F ( 0 ) F ( x )〉 − F02 ]
x =⎨ˆ −n′
(4-35) ⎩ 2d x
⎩x ′ (1 − Tc / T ) , T < Tc ⎫
∑ [〈F ( 0 ) F ( x )〉 − F02 ]⎬ (4-42)
Above Tc , the terms 3 u F02 + 5 v F04 = 0, x ⎭
hence Comparing Eqs. (4-34) and (4-42), we find
1 1 that in terms of the order parameter corre-
cT = = (4-36 )
kB T r r ′ ( T − Tc ) lation function G (x),
x = R2 /( rd ) = ( R / r ′ d ) (Tc − T ) −1/ 2 G( x ) = 〈F ( 0 ) F ( x )〉 − F02 (4-43)
262 4 Statistical Theories of Phase Transitions

and c T and x 2 are related to the zeroth and Tc . For T < Tc , we have instead, from Eqs.
second moment of the correlation function: (4-10) and (4-13)
F − F0
= F02 ⎛ + F02 ⎞
1 r u
cT = ∑ G (x) (4-44)
⎝2 4 ⎠
kB T x kB T V
− r2 (1 − T / Tc )2
1 ≈ − r′2 (4-52)
2
x = ∑ x G (x) ∑ G (x)
2
(4-45) =
4u 4 kB2 u
2d x x
which implies C = r¢ 2 (T /Tc ) / (4 kB u) for
Eq. (4-44) clearly shows that the critical di-
T < Tc . This jump singularity of the specific
vergence of c T occurs because the correla-
heat at Tc instead of the power law in Eq.
tions G (x) become long ranged: whereas
(4-5) is formally associated with vanishing
off Tc the correlation function for large | x|
critical exponents:
decays exponentially,
a = a¢ = 0 (4-53)
ln G (x) Æ – | x|/ x , | x| Æ • (4-46)
The question must now be asked whether
at Tc we have a power-law decay of the cor- this description of critical phenomena is
relation function: accurate in terms of the Landau theory. The
G (x) = Ĝ | x| – (d – 2 + h ) , T = Tc (4-47) free energy functional F [F (x)] in Eq.
(4-10) should be considered as an effective
The exponent describing this critical decay Hamiltonian from which a partition func-
has been defined such that tion Z can be obtained so that the true
h=0 (4-48) Helmholtz energy becomes:

in the Landau theory, whereas in general F = – kB T ln Z (4-54)

h ⫽ 0, as will be discussed below. = – kB T ln Ú d [F (x)] exp {– F [F (x)]/kB T}
In order to make our collection of critical It is natural to expect that the main contri-
exponents complete, we also consider the bution to the functional integral comes
critical isotherm from the region where the integrand is larg-
F0 = D̂ H 1/d , T = Tc (4-49) est, i.e., the vicinity of the point where
F [F (x)]/kB T has its minimum. If we as-
Using Eq. (4-33) for a uniform field we sume that the distribution over which the
conclude that average in Eq. (4-54) is a delta function at
F0 = (kB Tu)–1/3 H 1/3 (4-50) the value F 0 yielding the minimum, i.e.,
we assume that fluctuations of the order
i.e., D̂ = (kB Tu)–1/3 and parameter make a negligible contribution
d = 1/3 (4-51) to the functional integral, then the mini-
mum of F is equivalent to the minimum of
in the framework of the Landau theory. F (F ). In general, however, this is not true,
Finally, we turn to the specific heat, Eq. as the effects of fluctuations modify the
(4-5), which behaves rather pathologically critical behavior drastically, and the Lan-
in the Landau theory: for T > Tc and H = 0, dau theory does not hold for systems such
Eq. (4-10) just implies F = F0 , i.e., the as b-CuZn.
specific heat associated with the ordering A first hint of the conditions for which
described by F is identically zero above the Landau theory is valid can be obtained
 4.2 Phenomenological Concepts 263

by considering the effect of fluctuations (4-46). Replacing the right inequality in

within the context of the Landau theory it- Eq. (4-57) by the equality, L = x , as done in
self. The criterion Ginzburg (1960) sug- Eq. (4-55), gives therefore the maximum
gests is that the Landau theory is valid if permissible choice for L, for which the
the mean-square fluctuation of the order fluctuations in Eq. (4-55) are smallest. Even
parameter in a correlation volume is small then, however, it turns out that Eq. (4-55)
in comparison with the order parameter typically is not fulfilled for T close to Tc . In
square itself: order to see this, we write Eq. (4-55) in
terms of the local variable F i and denote
·[F (x) – F 0 ]2 ÒL = x O F 02 (4-55)
the number of lattice sites i contained
Here L is the coarse-graining length intro- Ld for the choice L = x as N (t), where
duced in Eq. (4-9) in order to transform the t ∫ 1 – T /Tc . Then Eq. (4-55) becomes
microscopic lattice description of ordering
[ ∑ Fi − 〈Fi 〉 ] ∑ Fi
2 2
phenomena in solids to a continuum de- O
i ΠVx ( x ) i ΠVx ( x )
scription. On the lattice level, the local or-
(4-58)
der parameter F i shows a rapid variation
from one lattice site to the next, and the where we have denoted the volume of size
mean-square fluctuation is very large. For L = x centered at x as Vx (x). Making use of
example, for an Ising model of an aniso- the translational invariance of correlation
tropic ferromagnet, functions and ·F i Ò = Ms (t), the spontane-
(4-56) ous order parameter which is independent
Ᏼ Ising = − J ∑ Si S j − H ∑ Si , Si = ± 1 of i, Eq. (4-58) becomes
〈 i, j 〉 i

Where J is the exchange interaction, the N (t ) ∑ [〈Fi = 0 Fi 〉 − Ms2 (t )]

i ΠVx ( x )
sum ·i, j Ò extends once over all nearest
O N 2 (t ) Ms2 (t ) (4-59)
neighbor pairs, and Si points in the direc-
tion of the local magnetic moment at lattice If the sum over correlations in Eq. (4-59)
site i, we have F i = Si , i.e., F i2 ∫ 1. In this were to extend over all space, it would sim-
case ·[F i – ·F i Ò]2 Ò = 1 – Ms2 is never small ply be the “susceptibility” kB T c (t) (see
in comparison with the square of the spon- Eq. (4-42)). Since the sum does contain
taneous magnetization Ms2 near Tc . Carry- just the volume region over which the F i s
ing out the local averaging defined in Eq. are strongly correlated with each other, the
(4-9) reduces this local fluctuation in a sum clearly is of the order of f kB T c (T ),
“volume” Ld. This averaging should, in where f < 1 is a factor of order unity, which
principle, be carried out over a length scale should have no critical (vanishing or diver-
L that is much larger than the lattice spac- gent) temperature dependence as T Æ Tc.
ing a but much smaller than the correlation Hence in the inequality Eq. (4-59) this fac-
length x , tor may also be omitted, and we conclude
(Als-Nielsen and Laursen, 1980) that
aOLOx (4-57)
Whereas for a study of critical phenomena
c (t) O N (t) O Ms2 (t) (4-60 a)
it is permissible to average out local effects Making use of N (t) = [x (t)/a]d , a being the
on the scale of a lattice spacing, relevant lattice spacing, we obtain:
spatial variations do occur on the scale of
x , as we have seen in Eqs. (4-34) and const. O [x (t)]d Ms2 (t) c –1 (t) (4-60 b)
264 4 Statistical Theories of Phase Transitions

where in suitable units the constant on the and Laursen, 1980), the essential part of
left-hand side of the equality is of order the magnetic Hamiltonian is a magneto-
unity. Using Eqs. (4-4), (4-6), and (4-35) static dipole – dipole interaction,
we obtain (anticipating g = g ¢ and n = n ¢)
3 z2 − x2
– n d +2 b + g Ᏼ = − m2 ∑ Si S j , Si = ± 1 (4-64)
const. O t (4-61) i≠ j | x |5
Inserting the Landau values for the critical where m is the magnetic moment per spin
exponents b = 1/2, n = 1/2, g = 1 and know- and z the coordinate of x in the direction of
ing the critical amplitude of x ~ R, the inter- the uniaxial anisotropy. Unlike the iso-
action range, Eq. (4-60) implies tropic Ising Hamiltonian in Eq. (4-56), here
the sign of the interaction depends on the
const. O (R/a)d t (d – 4)/2 (4-62) direction in the lattice. Therefore, fluctua-
This condition for the validity of the Lan- tions display an essential anisotropy: in-
dau theory for d < 4 always breaks down stead of the isotropic result (Eq. (4-34)) of
as t Æ 0 (T Æ Tc ). In fact, for d < 4 close the well known Ornstein-Zernike type,
enough to Tc a regime occurs where fluctu- the wavevector-dependent susceptibility
ations dominate the functional integral (Eq. c (q) = S (q)/kB T (S (q) = structure factor)
(4-54)). The “crossover” from the mean- becomes anisotropic:
field regime (where the Landau description S (q) ~ [x – 2 + q2 + g (qz /q)2 ]–1 (4-65 a)
is essentially appropriate) to the non-mean-
field regime occurs at a reduced tempera- where qz is the z-component of q and g is a
ture distance t = tcr , where Eq. (4-62) is constant. Whereas Eq. (4-32) implies that
treated as an equality: for d = 3, S (q x = 1) = S (q = 0)/2, the equation q x = 1
which for Eq. (4-32) defines a sphere of ra-
tcr ~ (R/a)– 6 (4-63) dius x –1 in q-space now defines an aniso-
tropic surface. From
Hence systems with a large but finite range
of interaction behave in an essentially Lan- x – 2 = q2 + g (qz /q)2 (4-65 b)
dau-like manner. An example of such a be-
we obtain an object having the shape of a
havior is the unmixing critical point in
flat disc, with radius 1/x in the qx – qy plane
polymer fluid mixtures with a high degree
but maximum extension of order 1/x 2 in
of polymerization N : since each coil has a
the qz direction. This object, defined by
radius rc ~ N 1/2, the monomer density in-
S (q) = –12 max {S (q)}, can be interpreted as
side the sphere taken by a polymer coil is
the Fourier transform of the correlation
only of order r ~ N/rc3 ~ N – 1/2, which im-
volume. This implies that the correlation
plies that each coil interacts with N 1/2
volume in real space is a long ellipsoid,
neighbor coils, and the effective interaction
with linear dimensions x in the qx , qy di-
volume (R/a)d in Eqs. (4-62) and (4-63)
rection but with linear dimension x 2 in the
should be taken as N 1/2, i.e., tcr ~ N – 1 (De
qz direction. In d-dimensions, this argu-
Gennes, 1979; Binder 1984 c, 1994).
ment suggests for uniaxial dipolar systems:
At this point, we emphasize that using
N (t) = [x (t)/a]d in Eq. (4-60) is not valid in N (t) = [x (t)/a]d +1 (4-66)
systems with long-range anisotropic inter-
which yields in Eq. (4-60)
actions, such as uniaxial dipolar magnets.
For a system such as LiTbF4 (Als-Nielsen const. O t – n (d +1) + 2 b + g = t – (d – 3)/2 (4-67)
 4.2 Phenomenological Concepts 265

where in the last equality, the Landau ex- well over a wide temperature range (Salje,
ponents have been used. Comparing Eqs. 1990).
(4-61) and (4-67) shows that d-dimensional For all other systems we do not expect
uniaxial dipolar systems somehow corre- the Landau critical exponents to be an ac-
spond to (d + 1)-dimensional systems with curate description of the critical singular-
isotropic short-range forces. Therefore the ities. This is borne out well by the exact so-
marginal dimension d*, above which the lution for the two-dimensional Ising model
Landau theory predicts the values of criti- (Onsager, 1944; McCoy and Wu, 1973),
cal exponents correctly, is d* = 3 for uniax- where the critical exponents have the values
ial dipolar systems, unlike the standard iso-
a = 0 (log), b = 1/8, g = 7/4,
tropic short-range case where d* = 4.
For elastic phase transitions in cubic d = 15, n = 1, h = 1/4 (4-71)
crystals where the combination of elastic Obviously, these numbers are a long way
constants c11 – c12 softens as the critical from the predictions of the Landau theory.
point is appproached, such as the system Note that a = 0 in Eq. (4-71) has the mean-
PrAlO3 , the structure factor S (q) has the ing C ~ | log | t ||, unlike the jump singular-
form: (4-68) ity of the Landau theory.
S (q) ~ [x – 2 + q2 + A (qz /q)2 + B (q^ /q)2 ]– 1 Even in the marginal case of uniaxial di-
polar ferromagnets (or uniaxial ferro-
where q = (qz , q^ ) and A and B are con- electrics; see, e.g., Binder et al. (1976) for
stants. In this case the correlation vol- a discussion), the Landau theory is not
ume defined from the condition S (q) = completely correct; it turns out (Larkin and
–12 max {S (q)} in real space is a flat disk Khmelnitskii, 1969; Aharony, 1976) that
with radius x 2 and diameter x , i.e., for elas- the power laws have the Landau form but
tic systems: are modified with logarithmic correction
N (t) = [x (t)/a]d +2 (4-69) factors:

In this case, the Ginzburg criterion, Eq. cT = Ĝ± t – 1 | ln t |1/3, tÆ0 (4-72 a)
(4-60), yields 1/3
C = ± | ln t | , tÆ0 (4-72 b)
– n (d +2) + 2 b + g – (d – 2)/2
const. O t =t (4-70) Ms = B̂ (– t)1/2 | ln (– t)|1/3, t Æ 0 (4-72 c)
instead of Eqs. (4-62) or (4-67), respec- Ms | Tc = D̂ H 1/3 1/3
| ln H | , HÆ0 (4-72 d)
tively. Since the marginal dimensionality
for such systems is d* = 2, three-dimen- Although for uniaxial ferroelectrics the ex-
sional systems should be accurately de- perimental evidence in favor of Eq. (4-72)
scribed by the Landau theory, and this is is still scarce, convincing experimental ev-
what is found experimentally for PrAlO3. idence does exist for dipolar ferromagnets
Examples for elastic phase transitions such as LiTbF4 (see, e.g., Als-Nielsen and
which are of second order are scarce, as is Laursen (1980) for a review).
understandable from the symmetry consid- Eq. (4-72) results from including fluctu-
erations which lead to first-order transi- ation contributions to the functional inte-
tions in most cases (Cowley, 1976; Folk gral in Eq. (4-54) systematically, which can
et al., 1976). However, for many first-order be done in a most powerful way be renor-
ferroelastic transitions the Landau descrip- malization group theory (Aharony, 1973,
tion also fits the experimental data very 1976; Fisher, 1974; Wilson and Kogut,
266 4 Statistical Theories of Phase Transitions

1974; Ma, 1976; Amit, 1984; Yeomans, “universality”. Again we wish to convey
1992). A description of this theory is be- only the flavor of the idea to the reader,
yond the scope of this chapter; we only rather than to present a thorough discus-
mention that it is this method which yields sion. Consider, for example, the decay of
highly accurate approximations for the val- the correlation function of order parameter
ues of critical exponents of three-dimen- fluctuations at the critical point, Eq. (4-47):
sional systems. For example, for the uniax- changing the length scale from r to r¢ = L r
ial magnets (as described by the Ising Ha- would change the prefactor Ĝ but leave the
miltonian, Eq. (4-56), experimental exam- power-law invariant. The physical inter-
ples being the antiferomagnets MnF2 or pretation of this fact is the “fractal struc-
FeF2) or the ordering alloy b-brass, the ex- ture” of critical correlations (Mandelbrot,
ponents are predicted to be (LeGuillou and 1982): just as a fractal geometric object
Zinn-Justin, 1980): looks the same on every length scale, the
pattern of critical fluctuations looks the
a ≈ 0.110, b ≈ 0.325, g ≈ 1.240,
same, irrespective of the scale. Slightly off
d ≈ 4.82, n ≈ 0.63, h ≈ 0.032 (4-73) Tc , then, there should be only one relevant
Whereas within the Landau theory the “or- length scale in the problem, the correlation
der parameter dimensionality” n does not length x , which diverges as T Æ Tc , and de-
matter, as far as the values of the critical tails such as the precise behavior of the
exponents are concerned, n does matter if correlation functions for interatomic dis-
we go beyond the Landau theory. For ex- tances should not matter. As a result, it is
ample, for isotropic magnets as described plausible that the correlation function
by the well known Heisenberg model of G (x, x ) should depend only on the ratio of
magnetism, the two lengths x and x , apart from a scale
factor:
Ᏼ Heis = − ∑ Jij Si ⋅ S j − m H ∑ Siz (4-74)
i≠ j G (x, x ) = Ĝ x – (d – 2 + h ) G̃ (x /x ) (4-76)
Si being a unit vector in the direction of the Obviously, the condition G̃ (0) = 1 for the
magnetic moment at lattice site i, we have “scaling function” G̃ (x /x ) ensures that Eq.
(LeGuillou and Zinn-Justin, 1980): (4-76) crosses over smoothly to Eq. (4-47)
a ≈ – 0.116, b ≈ 0.365, g ≈ 1.391, as T Æ Tc , where x Æ ∞.
d ≈ 4.82, n ≈ 0.707, h ≈ 0.034 (4-75) From the “scaling hypothesis” in Eq.
(4-76) we immediately derive a “scaling
Again these results are in fair agreement law” relating the exponents g , n and h. Us-
with available experimental data, such as ing Eq. (4-44) and denoting the surface of
the isotropic antiferromagnet RbMnF3 (see a d-dimensional unit sphere as Ud (Ud = 4 p
Als-Nielsen (1976) for a review of experi- in d = 3) we obtain
ments on critical point phenomena). Note
1 G˜
that the negative value of a in Eq. (4-75) cT = ∑ G (x) ≈ ∫ G ( x , x ) dx
implies that the specific heat has a cusp of kB T x kB Tc
finite height at Tc . ∞
Ud Gˆ 1−h
The renormalization group theory also = ∫x G˜ ( x / x ) dx
kB Tc 0
provides a unifying framework and justifi-
∞
cation for two important concepts about 2 −h Ud Gˆ 1−h
=x ∫ᒗ G˜ (ᒗ) dᒗ (4-77)
critical phenomena, namely “scaling” and kB Tc 0
 4.2 Phenomenological Concepts 267

Since the integral over ᒗ = x /x in Eq. (4-77) cancel out. On the other hand, this also
yields a constant, we conclude yields
c T ~ x (2 – h ) ~ | t | – n (2– h ) ~ | t | – g F | T = Tc ~ H 1/(1 + g /b ) = H1/d ,
cf. Eqs. (4-4) and (4-35). Thus the scaling i.e., d = 1 + g /b (4-83)
relation results in
The scaling assumptions also imply that
g = n (2 – h ) (4-78) exponents above and below Tc are equal,
i.e, a = a ¢, g = g¢, and n = n¢. Now there is
Eq. (4-76) indicates that G (x, x ) depends
still another scaling law which results from
on the two variables x and x in a rather spe-
considering all degrees of freedom inside a
cial form, namely, it is a homogeneous func-
correlation volume x d to be highly corre-
tion. A similar homogeneity assumption is
lated with each other, while different corre-
also true for the singular part of the free
lation volumes can be considered as essen-
energy, (4-79) tially independent: this argument suggests
F = Freg + t 2 – a F̃ (H̃ ), H̃ = H Ĝ t – g – b/B̂ that the singular part of the Helmholtz en-
ergy per degree of freedom can be written
where the singular temperature dependence
as Fsing ≈ [1/Ns (t)] ¥ const., since indepen-
for H̃ = 0 is chosen to be compatible with
dent degrees of freedom do not contribute
Eq. (4-5), Freg is a background term which
any singular free energy. Since
is analytic in both T and H even for T = Tc ,
while F̃ (H̃ ) is another “scaling function”. Ns (t) ~ x d ~ |t | – dn
At this point, we present Eq. (4-79) as a
postulate, but it should be emphasized that we conclude, by comparison with Eq.
both Eqs. (4-76) and (4-79) can be justified (4-79), that
from Eq. (4-54) by the renormalization dn = 2 – a (4-84)
group approach.
Combining Eqs. (4-1) and (4-79) we ob- Note that this so-called “hyperscaling rela-
tain: tion” with the Landau-theory exponents is
(4-80)
Gˆ ( 2 − a − g − b ) ˜ ˜ only true at the marginal dimension d* = 4,
F = − ( ∂F / ∂H )T = − t F′ ( H ) whereas the other scaling relationships ob-
Bˆ
viously are fulfilled by Landau exponents
Defining – (Ĝ /B̂) F̃ ¢ (H̃ ) ∫ B̂ M̃ (H̃ ) with independent of the system dimensionality.
M̃ (0) = 1, we obtain: However, all scaling relationships (includ-
ing Eq. (4-84)) are satisfied for the two-
F = B̂t b M̃ (H̃), b = 2 – a – g – b (4-81)
and three-dimensional Ising model (Eqs.
since for H̃= 0, Eq. (4-80) must reduce to (4-71) and (4-73)) and the three-dimen-
Eq. (4-6). Taking one more derivative we sional Heisenberg model, Eq. (4-75).
find Let us consider the two-dimensional
cT = (∂F /∂H )T = Ĝ t – g M̃¢(H̃) (4-82) Heisenberg model. For this model the ef-
fects of statistical fluctuations are so strong
hence Eq. (4-79) is compatible with Eq. that they destabilize the ordering alto-
(4-4), as it should be. The condition that gether: no spontaneous ordering exists in
Eq. (4-81) reduces to the critical isotherm d = 2 for any system with order parameter
for t Æ 0 requires that M̃ (H̃ Æ •) ~ dimensionality n ≥ 2, and a critical point
H̃ b /(g + b ), in order that the powers of t occurs only at zero temperature, Tc = 0, for
268 4 Statistical Theories of Phase Transitions

n > 2. Thus d = 2 is the “lower critical di- model of magnetism, Eq. (4-74), since the
mensionality” d1 for n ≥ 2, whereas for ferroelectric ordering would also be truly
n = 1 we have d1 = 1: quasi-one-dimensional isotropic. However, in the presence of a
orderings are always unstable. The case nonvanishing “cubic anisotropy”, u¢ ≠ 0, in
n = 2, d = 2 is very special: a phase transi- general a different universality class re-
tion still occurs at a nonzero Tc , the so- sults. The effect of such higher order invar-
called “Kosterlitz – Thouless” transition iants in the Landau expansion are particu-
(Kosterlitz and Thouless, 1973); for T < Tc larly drastic again in systems with reduced
we have Ms (T ) ∫ 0 while at the same time dimensionality; e.g., the model with n = 2
x (T ) = •, and the correlation function but cubic anisotropy no longer exhibits a
shows an algebraic decay, Eq. (4-47), with Kosterlitz – Thouless transition, but rather
a temperature-dependent exponent h . At a nonzero order parameter is stabilized
first sight, this behavior may appear fairly again. Owing to the “marginal” character
esoteric, but in fact it is closely related of the cubic anisotropy in d = 2, however,
to the “roughening transition” of crystal the behavior is not Ising-like but rather the
surfaces (Weeks, 1980). Such roughening pathological case of a “universality class”
transitions have been observed for high-in- with “nonuniversal” critical exponents oc-
dex crystal faces of various metals (Sala- curs. The latter then do depend on “micro-
non et al., 1988). The power-law decay of scopic” details, such as the ratio of the
correlations, Eq. (4-47), can be related to interaction strengths between nearest and
the behavior of the height – height correla- next nearest neighbors in the lattice (Krin-
tion function of a crystal surface in the sky and Mukamel, 1977; Domany et al.,
rough state (Weeks, 1980). As is well 1978; Swendsen and Krinsky, 1979; Lan-
known, the roughness or flatness of crystal dau and Binder, 1985). Again such prob-
surfaces has a profound effect on their ad- lems are not purely academic, but relevant
sorption behavior, crystal growth kinetics, for order – disorder transitions in chemi-
etc. (Müller-Krumbhaar, 1977). sorbed layers on metal surfaces, such as O
The other important concept about criti- on W (110) (see Binder and Landau (1989)
cal phenomena is “universality”: since the for a review of the theoretical modeling of
only important length scale near a critical such systems).
point is provided by the correlation length One important consequence of univer-
which is much larger than all “micro- sality is that all gas – fluid critical points in
scopic” lengths such as lattice spacing and three-dimensional systems, all critical
interaction range, it is plausible that “de- points associated with unmixing of fluid or
tails” on the atomic scale do not matter, and solid binary mixtures, and all order – disor-
systems near a critical point behave in the der critical points involving a single-com-
same way provided that they fall in the ponent order parameter (such as b-brass)
same “universality class”. It turns out that belong to the same universality class as the
universality classes (for systems with three-dimensional Ising model. All these
short-range interactions!) are determined systems not only have the same critical ex-
by both spatial dimensionality d and order ponents, but also the scaling functions
parameter dimensionality n, and in addi- M̃ (H̃ ), G̃ (ᒗ) etc. are all universal. Also,
tion, the symmetry properties of the prob- critical amplitude ratios are universal,
lem: e.g., Eq. (4-25) for n = 3 and u¢ = 0 such as Ĉ/Ĉ +, (Eq. (4-4), Â/¢ (Eq. (4-5)),
falls in the same class as the Heisenberg or more complicated quantities such as
 4.2 Phenomenological Concepts 269

D̂ Ĝ B̂d –1. The theoretical calculation and netization (due to the coupling F 2 (x) M (x)
experimental estimation of such critical in the Hamiltonian, as mentioned above).
amplitude ratios are discussed by Privman However, if we now consider the variation
et al. (1991). of F with M, combining both laws we get
An interesting problem concerns the (assuming a > 0)
critical singularities associated with nonor-
dering “fields”. These singularities occur
F ~ [Mc (T ) – M ]b /(1– a ) (4-85)
because the ordering field F (x) considered Considering now the phase transition in the
only in Eqs. (4-10) and (4-54) couples to space of {M, T} as independent thermody-
another quantity: e.g., for an antiferromag- namic parameters, we then find that on
net in a magnetic field the order parameter crossing the critical line Mc (T ) the expo-
F (the sublattice magnetization) couples to nent describing the vanishing of the order
the uniform magnetization M, to the lattice parameter is b /(1 – a ) rather than b. This
parameters, to the electron density, etc. effect generally occurs when the critical
Since the Hamiltonian is invariant against line depends on extensive (rather than in-
interchange of the sublattices which im- tensive) thermodynamic variables and is
plies a sign change of F , this coupling is called “Fisher renormalization” (Fisher,
quadratic in F 2. Whereas in mean field 1968). This is very common for order – dis-
theory the resulting critical behavior of the order phenomena in adsorbed layers at
nonordering “field” is then proportional to fixed coverage, or in alloys at fixed con-
·F Ò2 ~ t 2 b , taking fluctuations into account, centration, for unmixing critical points in
the actual critical behavior is proportional ternary mixtures, etc. Other exponents also
to ·F 2 Ò ~ t (1– a ), i.e., an energy-like singu- become “renormalized” similarly, e.g., g is
larity. Such energy-like singularities are replaced by g /(1 – a ), etc. A detailed anal-
predicted for the electrical resistivity r el at ysis (Fisher, 1968) shows that no such
various phase transitions (Fisher and “renormalization” of critical exponents oc-
Langer, 1968; Binder and Stauffer, 1976 a), curs for systems with a fixed concentration
for the refractive index n r (Gehring and c if the slope of the critical line vanishes,
Gehring, 1975; Gehring, 1977), etc. Look- dTc (c)/dc = 0, for the considered concen-
ing for anomalies in the temperature deriv- tration. This is approximately true for b-
ative dr el (T )/dT, dn r (T )/dT etc., which brass, for instance.
should exhibit specific-heat-like singular-
ities, is often a more convenient tool for lo-
4.2.3 Second-Order Versus First-Order
cating such phase transitions than measure-
Transitions; Tricritical and Other
ments of the specific heat itself.
Multicritical Phenomena
An important phenomenon occurs when
a nonordering field which is coupled to the An important question for any phase
order parameter in this way, such as the transition is deciding a priori whether it
magnetization in an antiferromagnet in an should be a second- or first-order transi-
external field, is held fixed. Suppose we tion; general principles are sought in an-
first study the approach to criticality by let- swering this question, such that there
ting the uniform magnetic field h tend to its would be no need for specific experimental
critical value hc (T ): we then have the law data.
F ~ [hc (T ) – h]b for the order parameter In the Landau theory, general symme-
and M – Mc (T ) ~ [hc (T ) – h]1– a for the mag- try conditions exist which allow second-
270 4 Statistical Theories of Phase Transitions

order transitions (Lifshitz, 1942). We here by Baxter (1973), these models have sec-
merely state them, without even attempting ond-order transitions. The critical expo-
to explain the group-theoretical language nents for these models are now believed
(see, e.g., Tolédano and Tolédano, 1987, to be known exactly, e.g., (q = 3) a = 1/3,
for a thorough treatment): b = 1/9, g = 13/9 and (q = 4) a = 2/3,
b = 1/12, g = 7/6 (Den Nijs, 1979; Nienhuis
(i) The order parameters F transform as
et al., 1980). As mentioned in Sec. 4.2.1, an
a basis of a single irreducible repre-
experimental example for the three-state
sentation X of the group G0 character- –
Potts model is the ÷3 superstructure of var-
izing the symmetry properties of the
ious adsorbates on graphite. Convincing
disordered phase.
experimental evidence for a specific heat
(ii) The symmetric part of the representa-
divergence of He4 on grafoil described by
tion X3 should not contain the unit
a = 1/3 was presented by Bretz (1977). The
representation.
system O on Ru (001) in the p (2 ¥ 2) struc-
(iii) If the antisymmetric part of X2 has a
ture (Piercy and Pfnür, 1987) falls into the
representation, the wavevector q asso-
class of the four-state Potts models and
ciated with X is not determined by
again the data are in reasonable agreement
symmetry. In this case q is expected
with the predicted critical exponents.
to vary continuously in the ordered
In d = 3 dimensions, however, the three-
phase.
states Potts model has a weak first-order
If these conditions are met, a transition transition (Blöte and Swendsen, 1979) so
can nevertheless be first order, because a that the Landau rule (ii) is not violated.
fourth-order term can be negative (see Sec. This is of relevance for metallurgical sys-
4.2.1). If they are not met, the transition tems such as the CuAu ordering (Fig. 4-8)
must be first order, according to the Lan- on the f.c.c. lattice, which, according to
dau rules. Domany et al. (1982), belongs to the class
The first of these rules essentially says of the three-states Potts model, while the
that if in the ordered phase two quantities Cu3Au structure belongs to the class of the
essentially independent of each other (not four-state Potts model. Another example of
related by any symmetry operation, etc.) the three-state Potts model is the structural
play the role of order parameter compo- transition of SrTiO3 stressed in the [111]
nents F1 , F2 , there is no reason why in the direction (Bruce and Aharony, 1975). The
quadratic term r1 (T ) F12 + r2 (T ) F22 of the order – disorder transition for all systems
Landau expansion the coefficients r1 (T ), described by these types of ordering are al-
r2 (T ) should change their sign at the same ways of first order, so there is no contradic-
temperature. Thus, if F1 , F2 are “primary” tion with this Landau rule, except for the
order parameter components, they should transition to a charge density wave state in
appear via a first-order transition. 2H-TaSe2 (Moncton et al., 1977) which ap-
The second condition essentially implies parently is second order although there is a
the absence of third-order terms (F 3 ) in the third-order invariant (Bak and Mukamel,
Landau expansion. It turns out, however, 1979). It is possible, of course, that the
that in d = 2 dimensions there are well- transition is very weakly first order so that
known counter-examples to these rules, experimentally it could not be distin-
namely the Potts model (Potts, 1952) with guished from second order. This remark
q = 3 and q = 4 (see Eq. (4-20)); as shown also holds for other examples of apparent
 4.2 Phenomenological Concepts 271

alternating magnetic moment direction be-

tween adjacent sheets (cf. Figs. 4-4 and
4-8 a). Experimentally first-order transi-
tions are known for FeO (Roth, 1958), TbP
(Kötzler et al., 1979) (these systems have
order parameter dimensionality n = 4), UO2
(Frazer et al., 1965; this is an example with
n = 6), MnO (Bloch and Mauri, 1973), NiO
(Kleemann et al., 1980) (examples with
n = 8), etc. The standard phenomenological
understanding of first-order transitions in
these materials invoked magnetostrictive
couplings (Bean and Rodbell, 1962) or
crystal field effects (in the case of UO2 ; see
Blume, 1966), which make the coefficient
u in Eq. (4-10) negative and thus produce a
free energy of the type shown in Fig. 4-6 b.
It has been suggested that the first-order
character of the phase transition in these
materials is a fundamental property due
to the large number n of order-parameter
components and the symmetry of the
Hamiltonian: renormalization group ex-
Figure 4-8. Five crystallographic superstructures on
the face-centered cubic (fcc) lattice and their equiva- pansions discussed by Mukamel et al.
lent antiferromagnetic superstructures if in a binary (1976 a, b) and others suggested that all
alloy A-atoms (full circles) are represented as “spin such transitions are “fluctuation-induced
up” and B atoms (open circles) are represented as first-order transitions”, i.e., all antiferro-
“spin down”. Case (a) refers to the CuAuI structure,
magnets with n ≥ 4 order parameters must
(b) to the A2B2 structure, (c) to the A3B structure of
Cu3Au type, (d) to the A3B structure of Al3Ti type, have first-order transitions. However, there
and (e) the A2B structure as it occurs in Ni2V or seem to be many counter-examples of cu-
Ni2Cr. In case (a) the labels 1, 2, 3, and 4 indicate the bic n ≥ 4 antiferromagnets with phase tran-
decomposition of the fcc lattice into four interpene- sitions of apparently second order, such as
trating simple cubic (sc) sublattices.
BiTb (Kötzler, 1984), GdTb (n = 4; Hulli-
ger and Siegrist, 1979), and GdSb (n = 8;
McGuire et al., 1969). The reasons for
violations of the Landau rules discussed by this discrepancy between renormalization
Tolédano and Pascoli (1980), Tolédano group predictions and experiments are not
(1981), and Tolédano and Tolédano (1987). clear.
There are many transitions which the There are various other cases where
above Landau symmetry criteria would “fluctuation-induced first-order transitions”
permit to be of second order but which are occur, as reviewed in detail by Binder
actually observed to be first order. Exam- (1987 a). Here we mention only the exam-
ples are type I or type II antiferromagnetic ple of the phase transition from disordered
structures, consisting of ferromagnetic block copolymer melts to the lamellar mes-
(100) or (111) sheets, respectively, with an ophase (Fig. 4-7 b). For this system the
272 4 Statistical Theories of Phase Transitions

free-energy functional does not attain its which implies a specific heat divergence
minimum at wavevector q = 0 in reciprocal C ~ (– t)–1/2, i.e., a t = 1/2. We note that this
space but on a surface given by the equa- set of exponents also satisfies the scaling
tion | q | = q* = 2 p/l , l being the wave- laws in Eqs. (4-78), (4-81), (4-83), and
length of the lamellar pattern. This reflects (4-84). Moreover, using this set of expo-
the degeneracy that there is no preferred di- nents in the Ginzburg criterion, Eqs. (4-58)
rection and hence the lamellae may be to (4-61), we find that, with the Landau ex-
oriented in any direction. Owing to the ponents for a tricritical point, this condi-
large “phase space” in q-space where the tion is marginally fulfilled, unlike the case
inverse of the wavevector-dependent sus- of ordinary critical points (Eq. (4-62)). It
ceptibility goes “soft” (c –1 ( | q | = q*) Æ 0 turns out that d* = 3 is the marginal dimen-
as T Æ Tc ), the mean-square amplitude of sion for tricritical points, and the mean-
the local fluctuation of the order parameter field power laws are modified by logarith-
·F 2 Ò – ·F Ò2 would diverge as T Æ Tc , mic correction terms similar to those noted
which is physically impossible. It can be for dipolar systems (Eq. (4-72)). Experi-
shown that this difficulty is avoided be- mental examples for tricritical points in-
cause the strong local order parameter fluc- clude strongly anisotropic antiferromagnets
tuations turn u negative near Tc (Brazov- in a uniform magnetic field (e.g., FeCl2 ,
skii, 1975; Fredrickson and Helfand, 1987; (Dillon et al., 1978), and dysprosium alu-
Fredrickson and Binder, 1989), producing minum garnet (DAG) (Giordano and Wolf,
a free energy, as shown in Fig. 4-6 b, for the 1975), see Fig. 4-9 for schematic phase
case of “symmetric” diblock copolymers diagrams), systems undergoing structural
where mean field theory would predict a phase transitions under suitable applied pres-
second-order transition. Experiments seem sure, such as NH 4Cl which has a tricritical
to confirm the first-order character of the point at Tt = 250 K and pt = 128 bar (Yelon et
transition (Bates et al., 1988). al., 1974) or the ferroelectric KDP at Tt =113
For certain systems, for which u > 0 and K and pt = 2.4 kbar (Schmidt, 1978), etc. A
thus a second-order transition occurs, it is model system which has been particularly
possible by variation of a non-ordering carefully studied is He3 –He4 mixtures: the
field to change the sign of the coefficient u transition temperature Tl (x) of the normal
at a particular point. This temperature Tt , fluid–superfluid He4 is depressed with in-
where u vanishes in Eq. (4-14), is called a creasing relative concentration x until, at a
tricritical point. From Eq. (4-14) we then tricritical point Tt = Tl (x t ), the transition be-
immediately find comes first order. This then implies a phase
(4-86)
separation between a superfluid phase with
F0 = (– r /v)1/4 = (r ¢/v)1/4 (Tt /T – 1)1/4 a lower He3 content and a normal fluid He3-
Hence the tricritical order parameter expo- rich phase. Most common, of course, are
nent b t = 1/4, while Eqs. (4-32) to (4-48) re- tricritical phenomena in fluid binary mix-
main unchanged and thus g t = 1, n t = 1/2, tures arising from the competition of gas –
h t = 0. The critical isotherm, however, be- liquid transitions and fluid – fluid phase
comes v F 05 = H/kB T, i.e., d t = 5. From Eq. separations in these systems (Scott, 1987).
(4-15) we finally find Just as the vanishing coefficient u in Eq.
(4-10) leads to a multicritical point, the tri-
F (F0 ) − F ( 0 ) t − t 1/ 2
=− ⎛ ⎞
critical point, another multicritical point is
(4-87)
V kB T 3⎝ v ⎠ associated with the vanishing of the coeffi-
 4.2 Phenomenological Concepts 273

The peak height of c (q) for |q | = qmax is

then described by
Gˆ
c ( qmax ) = (4-90)
t0 − K12 /( 4 K2 )
which implies that the actual critical point
occurs at Tc = TL [1 + K12/(4 K2 )]. An exam-
ple of such an ordering characterized by a
Figure 4-9. Schematic phase diagrams of antiferro- nonzero wavenumber q*= qmax is the meso-
magnets with uniaxial anisotropy in an applied uni- phase formation in block copolymer melts
form magnetic field H|| applied parallel to the easy (Fig. 4-7) (see Leibler (1980)). Related phe-
axis: Case (a) shows the case of weak anisotropy,
nomena also occur in crystals exhibiting
case (b) the case of intermediate anisotropy, where in
addition to the antiferromagnetic ordering of the spin “incommensurate superstructures”. Usu-
components in the direction of the easy axis, a spin- ally c (q) is then not isotropic in q-space; in
flop ordering of the transverse components also oc- uniaxial systems, Eqs. (4-88) to (4-90) then
curs. In case (a) both transitions T|| (H|| ), and T^ (H|| ) apply only if q is parallel to this preferred
are of second order and meet in a bicritical point. For
direction. The superstructure described by
intermediate strength of the anisotropy the line
T^ (H|| ) does not end at the bicritical point, but rather the wavelength l = 2 p/qmax is determined
in a critical end-point at the first-order transition line. by the coefficients K1 and K2 and thus in
Then a tricritical point also appears where the anti- general it is not a simple multiple of the lat-
ferromagnetic transition T|| (H|| ) becomes first order. tice spacing a, but (l /a) is an irrational
For very strong anisotropy, the spin-flop phase disap-
number. This is what is meant by the term
pears altogether.
incommensurate. Examples of such incom-
mensurate structures are “helimagnetic
cient of the [—F (x)]2 term; since this term structures” such as VF2 , MnAu2 , Eu, Ho
was also introduced as the lowest-order and Dy (Tolédano and Tolédano, 1987).
term of a systematic expansion, higher-or- Note that in these systems the reason for
der terms such as [—2 F (x)]2 must then be the incommensurate structure is not the
included in Eq. (4-10). This problem is existence of a negative coefficient in front
most conveniently discussed in the frame- of the [—F (x)]2 term but the existence of
work of the wavevector-dependent suscep-
⎛ ∂F − ∂F + ⎞
tibility c (q), Eq. (4-34), which including terms such as ⎜F + − F− ⎟ , where
the contribution resulting from a term ⎝ ∂z ∂z ⎠
[—2 F (x)]2 can be written as (t0 = 1 – TL /T ) F + = F 0 cos (p z /c), F – = F 0 sin (p z /c), c
being the lattice spacing in the z-direction.
Gˆ0 Such terms involving linear terms in —F (x)
c (q) = (4-88)
t0 + K1 q 2 + K2 q 4 are allowed by symmetry in certain space
where K1 and K2 are phenomenological co- groups and are called “Lifshitz invariants”.
efficients. If K1 < 0, the divergence of c (q) Moving from lattice plane to lattice plane
does not occur for t0 = 0, q = 0 but at a value in the z-direction in such a helimagnetic
qmax found from: spin structure, the spin vector (pointing
perpendicular to z) describes a spiral struc-
d ture like a spiral staircase. Of course, such
(t0 + K1 q 2 + K2 q 4 ) = 0
dq incommensurate superstructures (also
⇒ qmax = ( − K1 / 2 K2 )1/ 2 (4-89) called “modulated phases”) are not restrict-
274 4 Statistical Theories of Phase Transitions

ed to magnetic systems, and many exam- such as pressure, magnetic field (for mag-
ples of dielectric incommensurate phases netic structures), concentration (for alloys;
have been identified (e.g., NaNO2 (Tani- see Fig. 4-10), etc.). Theoretically under
saki, 1961), Rb2ZnCl 4 , BaMnF4 , thiourea certain conditions even a staircase with an
[SC(NH2 )2 ], see Tolédano and Tolédano infinite number (mostly extremely small!)
(1987)). Whereas Rb2ZnCl 4 and BaMnF4 of steps can be expected (“devil’s stair-
also involve the existence of Lifshitz invar- case”, see Selke (1988, 1989, 1992) for a
iants, thiourea is an example of the case discussion and further references). We also
discussed in Eqs. (4-88) to (4-90). Various emphasize that the modulation need not in-
examples of modulated superstructures ex- volve only one direction in space (one
ist in metallic alloys; see Selke (1988, wavevector), but can involve several wave-
1989, 1992) and De Fontaine and Kulik
(1985) for reviews of the pertinent theory
and experimental examples such as Al3Ti
(Loiseau et al., 1985) and Cu3Pd (Broddin
et al., 1986).
Whereas for TeTc the ordering of the
incommensurate phase can be described in
terms of a sinusoidal variation of the local
order parameter density F (x), the nonlin-
ear terms present in Eq. (4-10) at lower
temperatures imply that higher-order har-
monics become increasingly important.
Rather than a sinusoidal variation, the
structure is then better described in terms
of a periodic pattern of domain walls (or
“solitons”), and we talk about a “multi-
soliton lattice” or “soliton staircase”. An-
other important fact about modulated
phases is that the wavevector characteriz-
ing the modulation period is not fixed at
qmax , but varies with temperature or other
parameters of the problem. This is ex-
pected from the third of the general Landau
rules formulated at the beginning of Sec.
4.2.3. Usually this variation of q stops at Figure 4-10. Period M (in units of the lattice spacing
some commensurate value where a “lock- in the modulation direction) for the alloy Al3 – xTi1+ x
as a function of the annealing temperature, as ob-
in transition” of the incommensurate struc- tained from visual inspection of high-resolution elec-
tures occurs. In the incommensurate re- tron microscopic images. Each of the fifteen different
gime rational values of the modulation commensurate superlattices observed is composed of
(“long-period superstructures”) may have antiphase domains of length one or two (1-bands or
an extended regime of stability, leading to 2-bands, as indicated by sequences ·211Ò, ·21Ò etc.)
based on the L12 structure. Very long annealing times
an irregular staircase-like behavior of the were needed to produce these long-period super-
modulation wavelength l as a function of structures supposedly at equilibrium. From Loiseau
temperature (or other control parameters et al. (1985).
 4.2 Phenomenological Concepts 275

vectors. An example of the case of a two- While experimental examples for Lif-
rather than one-dimensional modulation is shitz points are extremely scarce, e.g., the
LiKSO4 (Pimenta et al., 1989). structural transition in RbCaF3 under (100)
Here we discuss only the special case stress (Buzaré et al., 1979), more common
of Eqs. (4-88) – (4-90) briefly where, by multicritical points are bicritical points
changing external control parameters, it where two different second-order transi-
is possible to reach a multicritical point tion lines meet. Consider, for example, the
where K1 vanishes, the so-called “Lifshitz generalization of Eq. (4-10) to an n-compo-
point” (Hornreich et al., 1975). From Eq. nent order-parameter field, but with differ-
(4-88) we then find that for T = Tc (= TL ) we ent coefficient ri for each order-parameter
have c (q) ~ q – 4, and comparing this with component F i :
the result following from a Fourier trans-
1 F
formation of Eq. (4-47), c (q)Tc ~ q – 2 + h , Ᏺ [F ( x )] = 0 (4-92 )
we conclude that h L = – 2 for the Lifshitz kB T kB T

+ ∫ dx ⎧⎨ r1 F12 ( x ) + r2 F22 ( x ) + …
point, whereas for T > Tc we still have 1 1
c (q = 0) = Ĝ0 t0–1, i.e., g L = 1. Since for ⎩ 2 2
K1 = 0 Eq. (4-88) can be rewritten as 1 u
+ rn Fn2 ( x ) + [F 2 ( x )]2
2 4
c (q) = Ĝ0 (1 – TL /T )–1/ [1 + x 4 q 4 ] ⎫
R2
x = [K2 /(1 – TL /T )]1/4 (4-91) + [(∇F1 )2 + … + (∇Fn )2 ]⎬
2d ⎭
we conclude that nL = 1/4 for the correla- For simplicity, the fourth-order term has
tion length exponent at the Lifshitz point. been taken as fully isotropic. If r1 = r2 = …
Also for the Lifshitz point we can ask = rn , we would have the isotropic n-vector
whether the Landau description presented model discussed in Secs. 4.2.2 and 4.2.3.
here is accurate, and whether or not statis- We now consider the case where some of
tical fluctuations modify the picture. The these coefficients differ from each other,
result is that the critical fluctuations are but a parameter p exists on which these co-
very important. A marginal dimensionality efficients depend in addition to the temper-
d* = 8 results in the case of an “isotropic ature. The nature of the ordering will be
Lifshitz point”, where in determined by the term ri F i2 , for which
K1x (∂F /∂x)2 + K1y (∂F /∂y)2 + K1z (∂F /∂z)2 the coefficient ri changes sign at the high-
est temperature. Suppose this is the case
we have simultaneous vanishing of all co- where i = 1 for p < pb , then we have a one-
efficients K1x , K1y , and K1z . Note that the component Ising-type transition at Tc1 ( p)
more common case is the “uniaxial Lifshitz given by r1 ( p, T ) = 0 (the other compo-
point”, where only K1z vanishes while K1x nents F i being “secondary order parame-
and K1y remain nonzero. In this case the ters” in this phase). If, however, for p > pb
multicritical behavior is anisotropic, and the coefficient r2 ( p, T ) = 0 at the highest
we must distinguish between the correla- temperature Tc2 ( p), it is the component F 2
tion length x || describing the correlation which drives the transition as the primary
function decay in the z-direction and the order parameter. The point p = pb , Tc1 ( p) =
correlation length x ^ in the other direc- Tc2 ( p) = Tb is then called a bicritical point.
tions. Eq. (4-91) then holds for x || only, An example of this behavior is found for
whereas Eq. (4-39) still holds for x ^. weakly uniaxial antiferromagnets in a uni-
276 4 Statistical Theories of Phase Transitions

form magnetic field H|| (Fig. 4-9). A model order parameter, due to the perpendicu-
Hamiltonian for such a system can be writ- lar components Msx and Msy of the stag-
ten as gered magnetization). In the framework of
Eq. (4-92), this would mean n = 3 and
Ᏼ = − ∑ {J (1 − D ) [ Six S jx + Siy S jy ]
〈 i, j 〉
r2 ∫ r3 ≠ r1 , with H|| being the parameter p.
+ J Siz S jz } − H|| ∑ Siz (4-93) Both lines TN|| (H|| ) and TN^ (H|| ) and the
i
first-order line between the two antiferro-
where we assume J < 0 (antiferromagne- magnetic structures join in a bicritical
tism) and for D > 0 the easy axis is the z- point. A well known example of such a be-
axis. For small H|| , we have a uniaxial anti- havior is GdAlO3 with Tb = 3.1242 K (Roh-
ferromagnetic structure and the order pa- rer, 1975). Another example, in our opin-
rameter is the z-component of the stag- ion, is the Fe – Al system where the ferro-
gered magnetization Msz . For stronger paramagnetic critical line joins the second-
fields H|| , however, we have a transition order A2 – B2 transition (Fig. 4-11). In the
to a spin-flop structure (two-component metallurgical literature, the magnetic tran-

Figure 4-11. The Fe – Al phase diagram, as obtained from a mean-field calculation by Semenovskaya (1974)
(left), and experimentally by Swann et al. (1972) (right). The ferro-paramagnetic transition is shown by the
dash-dotted line. The crystallographically disordered (A2) phase is denoted as a n in the nonmagnetic and a m in
the ferromagnetic state. The ordered FeAl phase having the B2 structure (Fig. 4-3) is denoted as a 2 , and the or-
dered Fe3Al phase having the D03 structure (Fig. 4-3) is denoted as a 1n or a 1m , depending on whether it is non-
magnetic or ferromagnetically ordered, respectively. Note that first-order transitions in this phase diagram are
associated with two-phase regions, since the abscissa variable is the density of an extensive variable, unlike the
ordinate variable in Fig. 4-9. From Semenovskaya (1974).
 4.2 Phenomenological Concepts 277

sition of Fe – Al alloys is often disregarded some characteristic time t which diverges

and then the multicritical point where the as T Æ Tc :
A2 – B2 transitions change from second to
t ~ x z ~ |1 – T /Tc | – n z (4-94)
first order is interpreted as a tricritical
point. Bicritical phenomena for structural Many of the concepts developed for static
phase transitions are also known, such as critical phenomena (scaling laws, univer-
SrTiO3 under stress in the [001] direction sality, etc.) can be carried over to dynamic
(Müller and Berlinger, 1975). critical phenomena: “dynamic scaling”
The structure of the phase diagram near (Halperin and Hohenberg, 1967; Ferrell et
bicritical points has been analyzed in detail al., 1967; Hohenberg and Halperin, 1977),
by renormalization group methods (Fisher “dynamic universality classes” (Halperin
and Nelson, 1974; Fisher, 1975 a, b; Muka- et al., 1974, 1976; Hohenberg and Halpe-
mel et al., 1976 a, b), experiment (Rohrer rin, 1977), etc. However, the situation is
and Gerber, 1977) and Monte Carlo simu- much more complicated because the static
lations (Landau and Binder, 1978). Apart universality classes are split up in the dy-
from the fact that the exponents introduced namic case, depending on the nature of the
so far take different values as T Æ Tb , there conservation laws, mode coupling terms in
is one additional exponent, the “crossover the basic dynamic equations, etc. For ex-
exponent” F, describing the singular ap- ample, we have emphasized that aniso-
proach of phase-transition lines towards tropic magnets such as RbMnF3 , ordering
the multicritical point. This behavior, and alloys such as CuZn, unmixing solid mix-
also other more complicated phenomena tures such as Al – Zn, unmixing fluid mix-
(tetracritical points, tricritical Lifshitz tures such as lutidine – water, and the gas –
points, etc.) are beyond the scope of this fluid critical point all belong to the same
chapter. For an excellent introduction to Ising-like static universality class, but each
the field of multicritical phenomena we re- of these examples belongs to a different dy-
fer to Gebhardt and Krey (1979) and for a namic universality class! Thus, in the an-
comprehensive review to Pynn and Skjel- isotropic antiferromagnet, no conservation
torp (1983). law needs to be considered, whereas the
conservation of concentration matters for
all mixtures (where it means the order pa-
4.2.4 Dynamics of Fluctuations rameter is conserved ) and for ordering al-
at Phase Transitions loys (where the order parameter is not con-
served but coupled to a conserved “nonor-
Second-order phase transitions also dering density”). Whereas in solid mix-
show up in the “critical slowing down” of tures the local concentration relaxes simply
the critical fluctuations. In structural phase by diffusion, in fluid mixtures hydrody-
transitions, we speak about “soft phonon namic flow effects matter, and also play a
modes”; in isotropic magnets magnon role at the liquid – gas critical point. In the
modes soften as T approaches Tc from be- latter instance, conservation of energy
low; near the critical point of mixtures the needs to be considered; it does not play a
interdiffusion is slowed down; etc. This role for phase transitions in solid mixtures,
critical behavior of the dynamics of fluctu- of course, where the phonons act as a “heat
ations is again characterized by a dynamic bath” to the considered configurational de-
critical exponent z; we expect there to be grees of freedom.
278 4 Statistical Theories of Phase Transitions

We do not give a detailed account of crit- to establish its precise value (Wansleben
ical dynamics here, but discuss only the and Landau, 1987).
simplest phenomenological approach (van It is important to realize that not all fluc-
Hove, 1954) for non-conserved order pa- tuations slow down as a critical point is ap-
rameters. We consider a deviation of the or- proached, but only those associated with
der parameter DF (x, t) from equilibrium at long-wavelength order parameter varia-
space x and time t, and ask how this devia- tions. This clearly expressed in terms of the
tion decays back to equilibrium. The stan- dynamic scaling relationship of the charac-
dard assumption of irreversible thermody- teristic frequency:
namics is a relaxation assumption; we put
w (q) = q z w̃ ( ᒗ), ᒗ = qx (4-99)
∂ ∂Ᏺ
DF ( x, t ) = − G0 (4-95) where w̃ (ᒗ) is a scaling function with the
∂t ∂DF ( x ) properties w̃ (•) = const. and w̃ (ᒗ O 1) ~ ᒗ– z ,
where G0 is a phenomenological rate factor. and, hence, is consistent with Eq. (4-94).
Using Eq. (4-10) and linearizing F (x, t) = Most experimental evidence on dynamic
F0 + DF (x, t) around F0 yields critical phenomena comes from methods
such as inelastic scattering of neutrons or
∂
∂t
{
DF ( x, t ) = − G0 ( r + 3 u F02 ) DF ( x, t ) light, NMR or ESR spectroscopy, and
ultrasonic attenuation. Although the over-
R2 2 ⎫ all agreement between theory and experi-
− ∇ [ DF ( x, t )]⎬ (4-96) ment is satisfactory, only a small fraction
d ⎭
of the various theoretical predictions have
Assuming that DF (x, t) is produced by been thoroughly tested so far. For reviews,
a field H (x) = Hq exp (i q · x) which is see Hohenberg and Halperin (1977) and
switched off at time t = 0, we obtain (see Enz (1979).
also Sec. 4.2.2) Finally, we also draw attention to nonlin-
(4-97) ear critical relaxation. Consider, for exam-
DFq (t ) ple, an ordering alloy which is held well
= exp [ − w ( q ) t ], w ( q ) = G0 / c ( q )
DFq ( 0 ) below Tc in the ordered phase and assume
that it is suddenly heated to T = Tc . The or-
where Eq. (4-34) has been used. Since c (q) der parameter is then expected to decay to-
diverges for q = 0, T Æ Tc , the characteris- wards zero with time in a singular fashion,
tic frequency w (q = 0) ∫ t –1 vanishes as F (t) ~ t – b /n z (Fisher and Racz, 1976). Sim-
w (q = 0) ~ c T–1 ~ x –g /n = x – (2 – h ), and by ilarly, if the alloy is quenched from a tem-
comparison with Eq. (4-94) the result of perature T > Tc to T = Tc , the superstructure
the “conventional theory” is obtained: peak appears and grows in a singular fash-
z=2–h (4-98) ion with the time t after the quench,
S (q, t) ~ t g /zn (Sadiq and Binder, 1984). Re-
Although Eq. (4-98) suggests that there is a lated predictions also exist for the nonlin-
scaling relationship linking the dynamic to ear relaxation of other quantities (ordering
the static exponents, this is not true in gen- energy, electrical resistivity) (Binder and
eral if effects due to critical fluctuations are Stauffer, 1976 a; Sadiq and Binder, 1984).
taken into account. In fact, for noncon- A very important topic is the dynamics
served systems, z slightly exceeds 2 – h and of first-order phase transitions, which will
extensive Monte Carlo studies were needed not be discussed here because it is dis-
 4.2 Phenomenological Concepts 279

cussed elsewhere in this book (see Chap- rameter space of the corresponding multi-
ters by Binder and Fratzl (2001) and by component system show up like a rounding
Wagner et al. (2001)). of the first-order transition.
In solids, however, diffusion of impurity
atoms is often negligibly small and such a
4.2.5 Effects of Surfaces and of Quenched
disorder due to frozen-in, immobile impur-
Disorder on Phase Transitions:
ities is called “quenched disorder” (Brout,
A Brief Overview
1959). Other examples of quenched disor-
It a system contains some impurity at- der in solids are due to vacancies, in addi-
oms which are mobile, we refer to “an- tion to extended defects such as disloca-
nealed disorder” (Brout, 1959). From the tions, grain boundaries, and external sur-
point of view of statistical thermodynam- faces. Quenched disorder has a drastic effect
ics, such mobile impurities act like addi- on phase transitions, as will be seen below.
tional components constituting the system Since the disorder in a system is usually
under consideration. For a second-order assumed to be random (irregular arrange-
transition, small fractions of such addi- ment of frozen-in impurities, for example),
tional components have rather minor ef- we wish to take an average of this random
fects: the transition point may be slightly disorder. We denote this averaging over the
shifted relative to that of the pure material, disorder variables, which we formally de-
and as the concentration of these impurities note here as the set {xa }, by the symbol
is normally strictly conserved, in principle […]av in order to distinguish it from the
the so-called “Fisher renormalization” thermal averaging ·…ÒT implied by statisti-
(Fisher, 1968) of critical exponents is ex- cal mechanics. Thus the average free en-
pected, as discussed in Sec. 4.2.2. For ergy which needs to be calculated in the
small impurity contents, the region around presence of quenched disorder is
the transition point where this happens is
extremely narrow, and hence this effect is F = − kB T [ln Z {xa}]av (4-100 )
not important.
The effect of such impurities on first-or-
[ ( {F }
= − kB T ln tr exp( − Ᏼ {xa ,Fi }/ kB T )
i
) ]av
der transitions is usually more important, whereas in the case of annealed disorder it
e.g., for a pure one-component system, the would be the partition function rather than
melting transition occurs at one well-de- the free energy which is averaged over the
fined melting temperature, whereas in the disorder:
presence of mobile impurities this transi-
Fann = − kB T ln[ Z {xa}]av (4-101)
tion splits up into two points, correspond-
ing to the “solidus line” and “liquidus line”
in the phase diagram of a two-component
( {F } i
)
= − kB T ln tr [exp( − Ᏼ {xa ,Fi }/ kB T )]av

system. This splitting for small impurity = − kB T ln ( tr exp( − Ᏼ {xa ,Fi }/ kB T ))

{F , x }i a
contents is simply proportional to the con-
centration of the impurities of a given type. Eq. (4-101) again shows that annealed dis-
In practice, where a material may contain order just means that we enlarge the space
various impurities of different chemical na- of microscopic variables which need to be
tures and different concentrations which included in the statistical mechanics treat-
often are not known, these splittings of the ment, whereas the nature of quenched aver-
transition line in a high-dimensional pa- aging (Eq. (4-100)) is different.
280 4 Statistical Theories of Phase Transitions

Experiments can be carried out with just impurity concentration, respectively) is

one probe crystal and no averaging over smaller. Since no ideal long-range order is
many samples as is implied by Eq. (4-100) established at the critical point, this corre-
needs to be performed. This is because we sponds to a rounding of the phase transi-
can think about the parts of a macroscopic tion.
sample as subsystems over which such an In the case n = 1 and d = 3 dimensions,
average as written in Eq. (4-100) is actu- very weak random fields do not destroy
ally performed. This “self-averaging” uniform long-range order (Imbrie, 1984;
property of macroscopically large systems Nattermann, 1998), although long-range
means that the average […]av over the set order is destroyed if HR exceeds a certain
{xa } could also be omitted: it need only be threshold value HRc (T ) which vanishes as
kept for mathematical convenience. T Æ Tc . But the random field drastically
We first discuss the physical effects of changes the nature of both static and dy-
randomly quenched impurities (point-like namic phenomena (Villain, 1985; Grin-
defects!) on phase transitions, assuming stein, 1985). In d = 2 dimensions, arbitrary
the concentrations of these defects to be weak random fields destroy even Ising-
very small. Various cases need to be distin- like (n = 1) order, producing domains of
guished, depending on the nature of the size x D with ln x D ~ (J/HR )2 (see Binder,
local coupling between the defect and the 1983 a).
local order parameter. This coupling may We now turn briefly to the physical real-
be linear (“random field”), bilinear (“ran- ization of “random fields”. For symmetry
dom bond”) or quadratic (“random aniso- reasons, no such defects are expected for
tropy”). As an example, let us consider ferromagnets, but they may be induced in-
magnetic phase transitions described by the directly by a uniform field acting on an
n-vector model of magnetism, with an ex- antiferromagnet (Fishman and Aharony,
change interaction depending on the dis- 1979; Belanger, 1998). Since antiferro-
tance between lattice points ai and aj : magnets in a field can be “translated” into
models for order – disorder phenomena in
Ᏼ = − ∑ J ( ai − a j ) Si ⋅ S j − ∑ Hi Siz (4-102) alloys (the magnetization of the antiferro-
i≠ j i
magnet “translates” into the concentration
In Eq. (4-102) we have assumed that im- of the alloy, see Sec. 4.3.1) or (in d = 2 di-
purities produce a random field Hi which mensions) in adsorbed layers (the magnet-
has the properties ization then “translates” into the coverage
[Hi ]av = 0 , [Hi2 ]av = HR2 (4-103) of the layer), random-field effects are im-
portant for ordering alloys or order – disor-
and Si is an n-component unit vector in the der transitions in monolayers, also. A nice
direction of the local order parameter. Now experimental example was provided by
we can show that for n ≥ 2 arbitrary small Wiechert and Arlt (1993) who showed that
amplitudes of the random field HR (which the transition of CO monolayers on graph-
physically is equivalent to arbitrary small ite near 5 K to a ferrielectric phase is
impurity concentration) can destroy uni- rounded by small dilution with N2 mole-
form long-range order (Imry and Ma, cules, in accord with the theory of random-
1975): the system is broken into an irregu- field effects. Furthermore, random fields
lar arrangement of domains. The mean act on many structural phase transitions
size of the domains is larger as HR (or the owing to impurity effects: ions at low-sym-
 4.2 Phenomenological Concepts 281

metry, off-center positions may produce for a review). Of course, the impurities will
both a random electric field and a random always produce some shift of the critical
strain field. Also, symmetry generally per- temperature, which decreases as p in Eq.
mits random fields to act on tensorial order (4-106) decreases. When p becomes
parameters in diluted molecular crystals smaller than a critical threshold value pc ,
(Harris and Meyer, 1985). For diluted mo- only finite clusters of spins are still mutu-
lecular crystals (such as N2 diluted with Ar, ally connected by nonzero exchange bonds
or KCN diluted with KBr, etc.) it is still un- J > 0, and long-range order is no longer
clear whether the resulting “orientational possible. The phase transition at T = 0 pro-
glass” phase is due to these random fields duced by variation of p in Eq. (4-106) can
or to random bonds (for reviews, see Loidl hence be interpreted purely geometrically
(1989); Höchli et al. (1990); Binder and in terms of the connectivity of finite clus-
Reger (1992); Binder (1998)). ters or an infinite “percolating” network of
The “random bond” Hamiltonian differs spins (Stauffer and Aharony, 1992). This
from Eq. (4-102) by the introduction of dis- transition is again described by a com-
order into the exchange terms rather than pletely different set of exponents.
–
the “Zeeman energy”-type term, For the case where | J | O D J in Eq.
(4-105), a new type of ordering occurs,
Ᏼ = − ∑ Jij Si ⋅ S j − H ∑ Siz (4-104) which is not possible in systems without
i≠ j i
quenched disorder; a transition occurs to a
Whereas in Eq. (4-102) we have assumed state without conventional ferro- or antifer-
the pairwise energy to be translationally in- romagnetic long-range order but to a “spin
variant (the exchange energy J depends glass phase” where the spins are frozen-in
only on the distance ai – aj between the in a random direction (Binder and Young,
spins but not on their lattice vectors ai and 1986; Young, 1998). The order parameter
aj separately), we now assume Ji j to be a of the Ising spin glass was introduced by
random variable, e.g., distributed accord- Edwards and Anderson (1975) as
ing to a Gaussian distribution
– qEA = [·Si Ò2T ] av (4-107)
P (Ji j ) µ exp [– (Ji j – J )2/2 (DJ)2 ] (4-105)
The nature of the phase transition in spin
or according to a two-delta function distri- glasses and the properties of the ordered
bution phase have been the subject of intense re-
P (Ji j ) = p d (Ji j – J) + (1 – p) d (Ji j ) (4-106) search (Binder and Young, 1986). This
great interest in spin glasses can be under-
–
If J o DJ (Eq. (4-105)) or if 1 – p O 1 (Eq. stood because many different systems show
(4-106)), the ferromagnetic order occurring spin glass behavior: transition metals with
–
for Eq. (4-104) if J > 0 or J > 0 is only a small content x of magnetic impurities
weakly disturbed, both for d = 3 and for such as Au1– x Fex and Cu1– x Mnx , diluted
d = 2 dimensions. Following an argument insulating magnets such as Eux Sr1– x S, var-
presented by Harris (1974), we can see that ious amorphous alloys, and also mixed
the critical behavior of the system remains dielectric materials such as mixtures of
unaltered in the presence of such impurities RbH 2PO4 and NH 4 H 2PO4 , where the spin
provided that the specific heat exponent represents an electric rather than magnetic
a < 0, whereas a new type of critical behav- dipole moment. A related random ordering
ior occurs for a ≥ 0 (see Grinstein (1985) of quadrupole moments rather than dipole
282 4 Statistical Theories of Phase Transitions

moments if found in “quadrupolar glasses” glasses) the critical dynamics again have a
(also called “orientational glasses”) such very different character, and dynamics
as K(CN)1– x Brx and (N2)x Ar1– x , (Loidl, characteristic of thermally activated pro-
1989; Höchli et al., 1990; Binder and cesses with a broad spectrum of relaxation
Reger, 1992; Binder, 1998). However, it times are often observed.
is still debated (Franz and Parisi, 1998) It is also interesting to discuss the effect
whether or not concepts appropriate for the of quenched impurities on first-order tran-
description of spin glasses are also useful sitions (Imry and Wortis, 1979). It is found
for the structural glass transition of under- that typically “precursor phenomena” near
cooled liquids (Jäckle, 1986). the first-order transition are induced, and
Finally we mention systems with ran- the latent heat associated with the transi-
dom anisotropic axes which can be mod- tion in the pure systems can be signifi-
eled by the following Hamiltonian (Harris cantly reduced. It is also possible that such
et al., 1973): impurity effects may completely remove
(4-108) the latent heat discontinuity and lead to a
Ᏼ = − ∑ J ( ai − a j ) Si ⋅ S j − ∑ ( ei ⋅ S j )2 rounding of the phase transition.
i≠ j i
Such rounding effects on phase transi-
where the ei are vectors whose components tions also occur when extended defects
are independent random variables with a such as grain boundaries and surfaces are
Gaussian distribution. This model is also considered. These disrupt long-range order
believed to exhibit destruction of long- because the system is then approximately
range order due to break-up in domains homogeneous and ideal (i.e., defect-free)
similar to the random field systems. It is only over a finite region in space. While
also suggested that spin glass phases may the description of the rounding and shifting
occur in these systems. Again Eq. (4-108) of phase transitions due to finite size ef-
is expected to be relevant, not only for dis- fects has been elaborated theoretically in
ordered magnetic materials but also for detail (Privman, 1990), only a few cases
dielectrics where the spin represents elec- exist where the theory has been tested ex-
tric dipole moments, or displacement vec- perimentally, such as the normal fluid –
tors of atoms at structural phase transitions, superfluid transitions of He4 confined to
etc. pores, or the melting transition of oxygen
Very small fractions of quenched impur- monolayers adsorbed on grafoil where the
ities which do not yet have an appreciable substrate is homogeneous over linear di-
effect on the static critical properties of a mension L of the order of 100 Å (see Marx,
second-order phase transition can already 1989). While at critical points, a rounding
affect the critical dynamics drastically. An and shifting of the transition normally set
example of this behavior is the occurrence in when the linear dimension L and the cor-
of impurity-induced “central peaks” for relation length x are comparable (Fisher,
structural phase transitions in the scattering 1971; Binder, 1987b), at first-order transi-
function S (q, w ) at frequency w = 0 in addi- tions the temperature region DT over
tion to the (damped) soft phonon peaks which the transition is rounded and shifted
(Halperin and Varma, 1976). In the above is inversely proportional to the volume of
systems where the impurities disrupt the the system (Imry, 1980; Challa et al., 1986).
conventional ordering more drastically An important effect of extended defects
(such as random field systems and spin such as grain boundaries and surfaces is
 4.3 Computational Methods Dealing with Statistical Mechanics 283

that they often induce precursor effects to Transitions occur where the basic degree of
phase transitions, e.g., the “wetting transi- freedom is the (thermally activated) diffu-
tion” (Dietrich, 1988; Sullivan and Telo da sion process of atoms between various lat-
Gama, 1985) where a fluid layer condens- tice sites. This happens for unmixing alloys
ing at a surface is a precursor phenomenon such as Al – Zn or for ordering alloys such
to gas – fluid condensation in the three- as b-CuZn or the Cu –Au system. The mod-
dimensional bulk volume. Similarly “sur- eling of such systems will be discussed in
face melting” and “grain-boundary melt- Sec. 4.3.1. Many structural transitions are
ing” phenomena can be interpreted as the of a very different nature: we encounter pe-
intrusion of a precursor fluid layer at a sur- riodic lattice distortions where atomic dis-
face (or grain boundary) of a crystal (Di placements are comparable to those of lat-
Tolla et al., 1996). Similarly, at surfaces of tice vibrations. Short wavelength distor-
ordering alloys such as Cu3Au, the effect tions may give rise to “antiferrodistortive”
of the “missing neighbors” may destabilize and “antiferroelectric” ordering, as exem-
the ordering to the extent that “surface-in- plified by the perovskites SrTiO3 , PbZrO3 ,
duced disordering” occurs when the system etc. Long wavelength distortions corre-
approaches the transition temperature of sponding to acoustic phonons give rise to
the bulk (Lipowsky, 1984). Finally, we “ferroelastic” ordering. Sec. 4.3.2 will deal
draw attention to the fact that the local crit- with the mean field treatment of such phase
ical behavior at surfaces differs signifi- transitions where the order parameter is a
cantly from the critical behavior in the bulk phonon normal coordinate, and Sec. 4.3.3
(Binder, 1983 b). Such effects are outside is devoted to numerical methods going be-
the scope of this chapter. yond mean-field theory.

4.3 Computational Methods 4.3.1 Models for Order – Disorder

Dealing with the Statistical Phenomena in Alloys
Mechanics of Phase Transitions Since the diffusive motion of atoms in
and Phase Diagrams substitutional alloys (Fig. 4-12) is so much
slower than other degrees of freedom (such
While the phenomenological theories in as lattice vibrations), we may describe the
Sec. 4.2 yield a qualitative insight into configurational statistics of a substitutional
phase transitions, the quantitative descrip- binary alloy, to which we restrict attention
tion of real materials needs a more detailed for simplicity, by local occupation vari-
analysis. In this section, we complement ables {ci }. If lattice site i is taken by an
the phenomenological theory by a more atom of species B, ci = 1, if it is taken by an
microscopic approach. In the first step, the atom of species A, ci = 0.
essential degrees of freedom for a particu- Neglecting the coupling between these
lar transition are identified and an appro- variables and other degrees of freedom, the
priate model is constructed. In the second Hamiltonian is then:
step, the statistical mechanics of the model
are treated by suitable approximate or nu- Ᏼ = Ᏼ 0 + ∑ [ci c j v BB ( xi − x j )
merical methods. i≠ j
Fig. 4-12 illustrates the modeling of or- + 2 ci (1 − c j ) v AB ( xi − x j ) ( 4-109)
der – disorder phase transitions in solids. + (1 − ci ) (1 − c j ) v AA ( xi − x j )] + …
284 4 Statistical Theories of Phase Transitions

Figure 4-12. Degree of free-

dom essential for the descrip-
tion of various order – disor-
der transitions in solids. For
further explanations, see text.

where vAA , vAB , and vBB are interactions where Dm is the chemical potential differ-
between pairs of AA, AB, and BB atoms. ence between the two species.
In fact, terms involving three- and four- The same mapping applies for the lat-
body interactions may also occur, but are tice-gas model of fluids, which at the same
not considered here. Also, effects due to time can be considered as a model of ad-
vacancies may easily be included but are sorbed layers on crystalline surface sub-
neglected here. strates (in d = 2 dimensions) (see Binder
As is well known, Eq. (4-109) can be re- and Landau (1989) for a review) or as a
duced to the Ising model, Eq. (4-56), by the model of interstitial alloys, such as hydro-
transformation Si = 1 – 2 ci = ± 1, apart from gen or light atoms such as C, N and O in
a constant term which is of no interest to us metals (Alefeld, 1969; Wagner and Horner,
here. The “exchange interaction” Jij 1974; Alefeld and Völkl, 1978). If we
between spins i and j and “magnetic field” interpret B in Eqs. (4-109) – (4-111) as an
H in Eq. (4-56) are related to the interac- occupied site and A as a vacant site, then
tion parameters of Eq. (4-109) by usually vAB = vAA= 0, such that
(4-112)
J (xi – xj) ∫ Jij = [2 vAB (xi – xj ) (4-110) Ᏼ = Ᏼ 0 + ∑ ci c j v ( xi − x j ) + e ∑ ci
– vAA (xi – xj ) – vBB (xi – xj )]/4 i≠ j i

e being the binding energy which a particle

1⎛ ⎞
H= ⎜ ∑ [v AA ( xi − x j ) − v BB ( xi − x j )]⎟ feels at lattice site i (for adsorbates at sur-
2 ⎝ j (≠ i) ⎠ faces, this is a binding to the substrate, and
− Dm ( 4-111) a similar enthalpy term is expected when
 4.3 Computational Methods Dealing with Statistical Mechanics 285

an interstitial is dissolved in a metal, transformation {Si , H} Æ {– Si , – H} . This

whereas for modeling of an ordinary gas – implies a particle–hole symmetry of the
liquid transition we would put e = 0). The phase diagram of the lattice gas model, or a
relationships analogous to Eqs. (4-110) and symmetry against interchange of A and B
(4-111) are then: in the phase diagram of the binary alloy
(4-113)
model. Even in very ideal cases, where
[
H = − (e + m )/ 2 + ∑ v ( xi − x j )/ 4
j (≠ i)
] both partners A and B of a binary alloy
crystallize in the same structure and have
Jij = – v (xi – xj)/4 (4-114) similar lattice spacings, the phase diagram
is in reality not symmetric around the line
where m is the chemical potential and ·ci Ò = 1/2 (see the Cu – Au phase diagram
v (xi – xj) the pairwise interaction of parti- reproduced in Fig. 4-13 as an experimental
cles at lattice sites xi and xj. example). There are various conceivable
Of course, it is straigthforward to gener- reasons for this asymmetry of real phase
alize this approach to binary (A – B) alloys diagrams: (i) the assumption of strictly
including as a third option that a lattice site pairwise potentials (Eq. (4-109)) fails, and
is vacant (V) or to ternary alloys; then terms such as ci cj ck , ci cj ck cl need to be in-
more complicated models result, e.g., the cluded in the effective Hamiltonian; and
Potts (1952) model, Eq. (4-20), or the (ii) the potentials vAA , vAB , and vBB in Eq.
model of Blume et al. (1971). Another gen- (4-109) are not strictly constant but depend
eralization occurs if one species (B) of a
binary alloy is magnetic, e.g., Fe in Fe –Al
alloys (see Dünweg and Binder (1987)).
In the latter case, the Hamiltonian is (the
true magnetic spin variable is now denoted
as s i )

Ᏼ = Ᏼ0 ( 4-115)
+ ∑ {ci c j [v BB ( xi − x j ) − Jm ( xi − x j ) s i ⋅ s j ]
i≠ j
+ 2 ci (1 − c j ) v AB ( xi − x j )
+ (1 − ci ) (1 − c j ) v AA ( xi − x j )} + …
where Jm (xi – xj) denotes the strength of Figure 4-13. Partial phase diagram of copper – gold
the magnetic interaction, which we have alloys in the temperature – concentration plane indi-
assumed to be of the same type as used in cating the existence regions of the three ordered
the Heisenberg model (Eq. (4-102)) with phases Cu3Au, CuAu, and CuAu3 (cf. Fig. 4-8 a and
n = 3 and Hi = 0. c). These phases are separated from each other (and
from the disordered phase occurring at higher tem-
We will not go further into the classifica- peratures) by two-phase coexistence regions. The
tion of the various models here, but note boundaries of these regions are indicated by full and
one general property of both models in broken lines. Region II is a long-period modulated
Eqs. (4-109) and (4-112) which becomes version of the simple CuAu structure occurring in re-
evident when mapping the Ising Hamilton- gion I. Note that for strictly pairwise constant inter-
actions (of arbitrary range!) in a model such as Eq.
ian Eq. (4-56): for H = 0, this Hamiltonian (4-109) the phase diagram should have mirror sym-
is invariant against a change of sign of all metry around the line cAu = 50%. From Hansen
spins. For H ≠ 0, it is invariant against the (1958).
286 4 Statistical Theories of Phase Transitions

on the average concentration ·cÒ of the al-

loy, since the effective lattice parameter a
and the Fermi wavenumber kF and other
physical characteristics entering the effec-
tive potential of the alloy also depend
on concentration ·cÒ. In fact, both reasons
probably contribute to some extent in real
alloys – the first-principles derivation of
model Hamiltonians such as Eq. (4-109)
from the electronic structure theory for sol-
ids is a challenging problem (Bieber and
Gautier 1984 a, b; Zunger, 1994). Since the
effective interaction typically is oscillatory
in sign (Fig. 4-14) according to the Friedel
form
v (| x |) = A cos (2 | kF || x | + j)/| x | 3 (4-116)
where A and j are constants, small changes
in | kF | and/or the distances lead to rela-
tively large changes in v (| x |) for distant
neighbors. Whereas the interaction Eq.
(4-116) results from the scattering of con-
duction electrons at concentration inhomo-
geneities, in other cases such as H in metals
the effective interactions are due to the
elastic distortions produced by these inter-
stitials in their environment. The resulting
elastic interaction is also of long range Figure 4-14. (a) Interaction potential V lmn = – 2 J (x)
(Wagner and Horner, 1974). (note that x = (l, m, n) –a2 , a being the lattice spacing) vs.
-----------
Since the effective interactions that distance | x | (measured as ÷l 2 + m2 + n2 ) for NiCr0.11
should be used in Eqs. (4-109) to (4-115) at T = 560 °C, as deduced from (b) short-range order
often cannot be predicted theoretically in a parameters a (x) measured by diffuse neutron scatter-
ing. The circles show the results of the high-temper-
reliable way, it may be desirable to extract ature approximation, Eq. (4-121), and crosses the re-
them from suitable experimental data. For sult of the inverse Monte Carlo method. Full and
metallic alloys, such suitable experimental broken curves represent a potential function of the
data are the Cowley (1950) short-range or- Friedel form, Eq. (4-116), with different choices of
der parameters a (x) ∫ a (xi – xj) which de- the Fermi wavevector kF for ·100Ò and ·110Ò direc-
tions, amplitude A and phase F being treated as fit-
scribe the normalized concentration corre- ting parameters. From Schweika and Hauboldt (1986).
lation function, c ∫ ·ciÒ = (1 – M)/2, M ∫ ·SiÒ
〈 ci c j 〉 − c 2 〈 Si S j 〉 − M 2 resentation (Eq. (4-117) for a ferromagnet
a ( xi − x j ) ≡ =
c (1 − c ) 1 − M2 is equivalent to the phenomenological Eq.
(4-117) (4-43)). In the disordered phase of an Ising
which correspond to the normalized spin spin system, it is straightforward to obtain
correlation functions in the Ising spin rep- the wavevector-dependent response func-
 4.3 Computational Methods Dealing with Statistical Mechanics 287

tion c (q) describing the response to a (cf. Sec. 4.2.2); a more accurate procedure
wavevector-dependent field (cf. Eqs. (4-32), to deduce J (x) from experimental data on
(4-34), and (4-40)), if we apply the molec- a (x) is the “inverse Monte Carlo method”
ular field approximation (see the next sub- (Gerold and Kern, 1986, 1987; Schweika,
section). The result is (Brout, 1965): 1989), but this method also relies on the as-
(4-118)
sumption that an Ising model description as
1− M 2
S ( q ) = kB T c ( q ) = written in Eq. (4-109) is appropriate. On
1 − [ J ( q )/ kB T ] (1 − M 2 )
˜
the level of the molecular field approxima-
where J̃ (q) is the Fourier transform of the tion (for alloys this is usually referred to as
“exchange interaction” J (x), the Bragg – Williams (1934) approxima-
(4-119) tion) or the Bethe (1935) approximation, it
˜J ( q ) = ∑ J ( xi − x j ) exp [ i q ⋅ ( xi − x j )] is possible to avoid models of the type of
j (≠ i) Eq. (4-109) and include the configurational
degrees of freedom in an electronic struc-
In Eq. (4-118), we have once again invoked
ture calculation (Kittler and Falicov, 1978,
that there is a “fluctuation relation”
1979). Although such an approach sounds
(= static limit of the so-called “fluctuation-
very attractive in principle, the results are
dissipation theorem”) relating c (q) to the
not so encouraging in practice, as shown in
structure factor S (q) which is just the
Fig. 4-15. The results of this method for
Fourier transform of the correlations ap-
Cu3Au are compared with Monte Carlo data
pearing in Eq. (4-117):
(4-120) of a nearest neighbor Ising model (Binder,
S (q) = ∑ exp [ i q ⋅ ( ri − rj )] ( 〈 Si S j 〉 − M 2 ) 1980), with the cluster-variation treatment
j (≠ i) of the same model (Golosov et al., 1973)
and with experimental data (Keating and
Combining now Eqs. (4-118) and (4-120)
Warren, 1951; Moss, 1964; Orr, 1960).
we see that the reciprocal of this diffuse
The conclusion of this subsection is that
scattering intensity S (q) in q-space is sim-
the development of microscopic models for
ply related to the Fourier transformation of
the description of order – disorder phenom-
the interactions as
ena in alloys is still an active area of re-
4 c (1 − c ) 4 J˜ ( q ) search, and is a complicated matter, be-
=1− c (1 − c ) (4-121) cause the validity of the models can only be
S (q) kB T
judged by comparing results drawn from
For the model defined in Eqs. (4-109) to the models with experimental data. How-
(4-111), this expression even is exact in ever, these results also depend on the ap-
an expansion of 1/S (q) in a power series in proximation involved in the statistical-me-
1/T to leading order in 1/T, and is one of chanical treatment of the models (e.g., the
the standard tools for inferring information full and broken curves in Fig. 4-15 refer to
on interactions in alloys from diffuse scat- the same nearest-neighbor Ising model,
tering data (Clapp and Moss, 1966, 1968; whereas the dash-dotted curve refers to a
Moss and Clapp, 1968; Krivoglaz, 1969; different model). In the next subsections,
Schweika, 1994). An example is shown we consider various sophistications of the
in Fig. 4-14. Close to the order – disorder statistical mechanics as applied to various
phase transition, we expect corrections to models for phase-transition phenomena.
the molecular field expression Eq. (4-121) More details about all these problems can
due to the effects of statistical fluctuations be found in the Chapter by Inden (2001).
288 4 Statistical Theories of Phase Transitions

of lattice gas models is discussed), for mag-

netic transitions (De Jongh and Miedema,
1974), etc. Usually only the source of ex-
perimental data used to extract the interac-
tion parameters is different: e.g., for ferro-
magnets such as EuS, rather than using dif-
fuse magnetic neutron scattering in analogy
with Eq. (4-118) it is more convenient to
extract J̃ (q) from the measurements of spin
wave dispersion curves found from inelas-
tic neutron scattering (Bohn et al., 1980).

4.3.2 Molecular Field Theory

and its Generalization
(Cluster Variation Method, etc.)
The molecular field approximation
(MFA) is the simplest theory for the de-
scription of phase transition in materials;
despite its shortcomings, it still finds wide-
spread application and has been described
in great detail in various textbooks (Brout,
1965; Smart, 1966). Therefore we do not
treat the MFA in full detail here, but rather
Figure 4-15. (a) Order parameters for the A3B struc- indicate only the spirit of the approach.
ture (Fig. 4-8 c) on the fcc lattice: Long-range order
parameter Y (LRO) and short-range order parameter
We start with the Ising ferromagnet, Eq.
– a1 for the nearest-neighbor distance (SRO) vs. tem- (4-56). The exact Helmholtz energy can be
perature, according to the Monte Carlo method found formally from the minimum of the
(Binder, 1980), the cluster variation (CV) method in functional (Morita, 1972)
the tetrahedron approximation (Golosov et al., 1973),
and the Kittler – Falicov (1978, 1979) theory. Data for Ᏺ = ∑ Ᏼ Ising ({Si }) P ({Si })
Cu3Au after Keating and Warren (1951) (LRO) and {Si = ± 1}
after Moss (1964) (SRO). (b) Ordering energy DU
(normalized to zero at Tc in the disordered state) vs.
+ kB T ∑ P ({Si }) ln ( P{Si }) (4-122)
{Si = ± 1}
temperature for the fcc A3B alloy (top) and the fcc
AB alloy (bottom). Theoretical curves are from the where the sums extend over all configura-
same sources as in (a); experimental data were taken tions of the spins in the system and P ({Si})
from Orr (1960) and Orr et al. (1960). is the probability that a configuration {Si}
occurs. Eq. (4-122) thus corresponds to the
thermodynamic relation F = U – TS where
We have concentrated in this subsection the entropy S is written in its statistical
on alloys, but a similar discussion could interpretation. Minimizing Eq. (4-122) for-
have also been presented for order – disor- mally with respect to P yields the canonical
der phenomena in adsorbed monolayers distribution
(see for example, Binder and Landau (1981,
Peq ({Si}) ~ exp [– Ᏼ Ising ({Si})/kB T]
1989), where the modeling of systems such
as H adsorbed on Pd (100) surfaces in terms as desired.
 4.3 Computational Methods Dealing with Statistical Mechanics 289

The MFA can now be defined by factor- These probabilities can be expressed in
izing the probability P ({Si}) of a spin con- terms of the multi-spin correlation func-
figuration of the whole lattice into a prod- tions gnc (i) ∫ ·Si Sj 1 … Sjn Ò, where the set of
uct of single-site probabilities pi which vectors xj 1 – xi , …, xjn – xi defines the n-
can take two values: p+ = (1 + M )/2 is the point cluster of type c located at lattice site
probability that the spin at site i is up i. The Helmholtz energy functional to be
and p– = (1 – M )/2 is the probability that it minimized in this cluster variation method
is down, p+ – p– = M is the magnetization. (Kikuchi, 1951; Sanchez and De Fontaine,
Now the expression Jij Si Sj pi pj (cf. Eq. 1980, 1982; Finel, 1994) is a more compli-
(4-56)) summed over the possible values cated approximation of Eq. (4-122) than
p+ and p– simply yields Jij M 2, and hence Eq. (4-123). If the largest cluster consid-
Eq. (4-122) reduces to, using Eq. (4-119) ered exceeds the interaction range, the en-
ergy term in F = U – TS is treated exactly;
1 MFA 1 ˜
Ᏺ = J (q = 0) M 2 − H M (4-123) unlike Eq. (4-123), the entropy is approxi-
N 2 mated. We find
⎡1 + M ⎛ 1 + M ⎞
+ kB T ⎢ ln
⎣ 2 ⎝ 2 ⎠ 1
Ᏺ= ∑ ∑ Jij g2, rj (i ) + kB T ∑ ∑ g nc
1 − M ⎛1 − M⎞⎤ 2 i j i n, c
+ ln
2 ⎝ 2 ⎠ ⎥⎦ 2n
× ∑ pn, c ( k , i ) (4-125)
Minimizing Ᏺ MFA
with respect to M now k =1

yields the elementary self-consistency where the coefficients gnc are combinato-
equation: rial factors depending on the lattice geome-
1 ˜ try and the clusters included in the approx-
M = tanh [ J (q = 0) M + H ] (4-124) imation (Kikuchi, 1951), e.g., in the tetra-
kB T
hedron approximation for the f.c.c. lattice,
As is well known, Eq. (4-124) implies a the sum over c in Eq. (4-125) includes the
second-order transition at Tc ∫ J̃ (q = 0)/kB (nearest-neighbor) tetrahedron, the near-
with the same exponents as in the Landau est-neighbor triangle, the nearest-neighbor
theory. pair, and the single site.
Clearly, in factorizing P ({Si}) into a Assuming the ordered structure to be
product of single-site probabilities and known, the symmetry operations of the as-
solving only an effective single-site prob- sociated group can be applied to reduce
lem, we have disregarded correlation in the the number of variational parameters in
probabilities of different sites. A system- Eq. (4-125). In the MFA, there is a single
atic improvement is obtained if we approx- non-linear self-consistent equation (Eq.
imate the probability of configurations not (4-124)) or a set of equations involving the
just by single-point probabilities but by us- order-parameter components if a problem
ing “cluster probabilities”. We consider more complicated than the Ising ferromag-
probabilities pnc (k, i) that a configuration k net is considered. In the CV method, a
of the n spins in a cluster of geometric con- much larger set of coupled non-linear
figuration c occurs (c may be a nearest- equations involving the short-range order
neighbor pair, or a triangle, tetrahedron, etc.). parameters gnc (i) is obtained when we min-
Note k = 1, … , 2n for Ising spins whereas imize Eq. (4-125). Therefore, whereas the
k = 1, … , qn for the q-state Potts model. simple MFA is still manageable for a wide
290 4 Statistical Theories of Phase Transitions

variety of systems (Brout, 1965; Smart, priori, it is often convenient to introduce

1966), the CV method is essentially re- Fourier transformations
stricted to Ising-type problems relevant for
1
phase transitions in metallic alloys (De X (q) = ∑ x i exp [ − i q ⋅ x i ] (4-130)
Fontaine, 1979; Finel, 1994). We discuss N i

the merits of the various approaches in the which yield:

next section.
N
At this point, we return to the formula- DᏲ MFA = ∑ f˜ ′′ ( q ) | X ( q )|2
tion of the MFA for the case of an arbitrary 2 q
type of ordering, rather than the simple N
+ f0′′′ ∑ X ( q ) X ( q ′ ) X ( q ′′ )
transition from paramagnetic to ferromag- 3! q, q ′, q ′′
netic considered in the above treatment × d ( q + q ′ + q ′′ − G )
(Eqs. (4-123) and (4-124)) of the Ising N IV
model (Eq. (4-56)). Hence Eq. (4-123) can + f0 ∑ X ( q ) X ( q ′ ) X ( q ′′ ) X ( q ′′′ )
4! q, q ′, q ′′, q ′′′
obviously be generalized as follows (see
also De Fontaine, 1975): × d ( q + q ′ + q ′′ + q ′′′ − G ) (4-131)

Ᏺ MFA = − 2 ∑ J ( xi − x j ) x i x j (4-126) where G is a reciprocal lattice vector and

i≠ j f̃ ¢¢(q) can be written as
+ kB T ∑ [ci ln ci + (1 − ci ) ln (1 − ci )] f̃ ¢¢(q) = – 4 J˜ (q) + kB T/[c (1 – c)] (4-132)
i

where xi = ci – c = (M – Si)/2, and constant Comparing Eqs. (4-121) and (4-132), we

terms have been omitted. Expanding Eq. realize that the inverse structure factor (or
(4-126) in terms of xi yields, again omitting inverse “susceptibility”, c – 1 (q)) is simply
a constant term, proportional to f̃ ¢¢(q), as it should be, since
1 X (q) and H (q) (apart from constants) are
DᏲ MFA = ∑ f0′′( xi − x j ) x i x j (4-127) canonically conjugate thermodynamic var-
2 i, j
iables. Of special importance now are the
1 1 IV points qc where J˜ (q) has its maximum and
+ f0′′′∑ x i3 + f0 ∑ x i4
3! i 4! i correspondingly f˜ ¢¢(q) has a minimum, be-
with the coefficients cause for these wavevectors a sign change
of f˜ ¢¢(q) in Eq. (4-131) occurs first (at the
f0¢¢(x ≠ 0) = – 4 J (xi – xj) highest temperature). Hence a stability
f0¢¢(x = 0) = kBT /[c (1 – c)] (4-128) limit of the disordered phase is predicted as

and kBTc = 4 J˜ (qc) c (1 – c) (4-133)

f0¢¢¢ = – kBT (2 c – 1)/[c (1 – c)]2 (4-129) If at this temperature a second-order transi-

tion occurs, it should actually be described
f0IV = 2
2 kBT [3 c – 3 c + 1)]/[c (1 – c)] 3
by a “concentration wave” X (qc) as an or-
If we group the x i s properly into the sub- der parameter (see also the discussion
lattices reflecting the (known or assumed) following Eq. (4-31)). In many cases the
state, Eq. (4-127) essentially yields the maxima of J˜ (q) occur at special symme-
Landau expansion in terms of the order-pa- try points of the Brillouin zone, and sym-
rameter components, as discussed in Sec. metry considerations in reciprocal space
4.2.1. Rather than defining sublattices a are useful for discussing the resulting or-
 4.3 Computational Methods Dealing with Statistical Mechanics 291

dered structures (De Fontaine, 1975, 1979; of an order-parameter component for the
Khachaturyan, 1973, 1983). transition: in mean-field theory, the asso-
This formulation of order – disorder tran- ciate eigenfrequency vanishes at a temper-
sitions in alloys is analogous to the treat- ature Tc (“soft phonon”). If this happens for
ment of structural transitions, for which a a phonon with wavevector k0 at the Bril-
model Hamiltonian similar to Eq. (4-127) louin zone edge, we have an antiferro-
can be written: rather than concentration electric order, provided that the phonon is
deviations xi we now have displacement polar, i.e., it produces a local dipole mo-
vectors ul (x) associated with a lattice vec- ment. For non-polar phonons, such as for
tor x for the l th atomic species in the unit the transition in SrTiO3 at Tc = 106 K where
cell. Just as it is useful to relate xi to the k0 = p (–12 , –12 , –12 )/a (the soft phonon there
Fourier transform X (q) of the concentra- physically corresponds to an antiphase ro-
tion deviation, it is useful to relate ul (x) to tation of neighboring TiO6 octahedra, as
the phonon normal coordinate Qk, l , de- indicated schematically in Fig. 4-12), the
fined as transition leads to “antiferrodistortive” or-
(4-134) der. Long-wavelength distortions corre-
1
ul ( x ) = ∑ exp( i k ⋅ x ) el ( k , l ) Qk, l sponding to optical phonons give rise
N Ml k, l
to ferroelectric ordering (an example is
where Ml is the mass of the atom of type l Pb5Ge3O11, where k0 = 0 and Tc = 450 K
at site Rli in the i th unit cell, el (k, x) is a (Gebhardt and Krey, 1979)). There are also
phonon polarization vector, l labels the paraelectric – ferroelectric transitions of
phonon branch and k its wavevector (Fig. first order, e.g., the cubic – tetragonal tran-
4-16). In this case, ·Qk0 , l0 ÒT plays the role sition of BaTiO3, such that no soft mode
occurs.
Just as the “macroscopic” ferroelectric
ordering can be associated with the normal
coordinate Qk0 = 0, l of the associate optical
phonon as a microscopic order parameter
characterizing the displacements on the
atomic scale, the “macroscopic” ferroelas-
tic orderings, where in the phenomenologi-
cal theory, Eq. (4-27), a component of the
strain tensor emn is used as an order pa-
rameter, can be related to acoustical pho-
nons. Examples are the martensitic mate-
rial In-25 at.% Tl where the combination
c11 – c12 of the elastic constants nearly sof-
tens at Tc = 195 K, and LaP5O14 where c55
Figure 4-16. (a) Schematic temperature variation of softens at Tc = 400 K and the structure
order parameter F = ·Q k0 , l 0 Ò T and square of the changes from orthorhombic to monoclinic
“soft-mode” frequency w 2 (k0 , l 0 ) at a displacive (Gebhardt and Krey, 1979). Of course, this
structural transition. (b) Schematic phonon spectrum
correspondence between the microscopic
of the solid at T > Tc . Note that either optical or
acoustic phonons may go soft, and for optical pho- description of displacements in crystals in
nons a softening may often occur at the boundary of terms of phonons and the phenomenologi-
the first Brillouin zone rather than at its center. cal macroscopic description in terms of po-
292 4 Statistical Theories of Phase Transitions

larization fields, strain fields, etc., is a component structural transition as follows

well-known feature of condensed-matter (Bruce and Cowley, 1981):
theory (Kittel, 1967).
ᏴF 4 = ∑ ⎡⎢ r Fi2 + uFi4 ⎤⎥
1 1
As is well known, the Qk, l are defined
i ⎣ 2 4 ⎦
such that the Hamiltonian of the crystal in
1
the quasi-harmonic approximation is diag- + ∑ C (Fi − F j )2 (4-136)
onalized (Born and Huang, 1954): 〈 i, j 〉 2

where r, u and C are phenomenological co-

1
Ᏼ = U0 + ∑ [ ∂ 2U /( ∂xil )a ( ∂x lj′ )b ] efficients. The last term corresponds to
2 i, j , l , l ′, a , b the harmonic interaction between displace-
× ula ( xi ) ulb′ (x j ) ments in neighboring lattice cells, while
1 the anharmonicity has been restricted to the
= U0 + ∑ w 2 ( k , l ) | Qk, l |2 (4-135) “single-site Hamiltonian” [r Fi2/2 + u Fi4/4].
2 k, l
Note that Eq. (4-136) is fully equivalent to
Eq. (4-135) corresponds precisely to Eqs. the “Hamiltonian” DᏲ MFA in Eq. (4-127) if
(4-127) and (4-131) if only the “harmonic f0¢¢¢= 0 is chosen, since (Fi – Fj)2 = Fi2 + Fj2
terms” (i.e., quadratic terms in the expan- – 2 Fi Fj, and the coefficient C is thus
sion with respect to the xi s) are retained. equivalent to f0¢¢(xi – xj) where xi – xj is a
The vanishing of a soft mode, w2 (k0 , l0) ~ nearest-neighbor distance, a. On the other
(T – Tc ), is again equivalent to the vanish- hand, the “F 4-model” in Eq. (4-136) can
ing of a coefficient r ~ (T – Tc ) of a quad- be thought of as a lattice analog of the
ratic term in a Landau expansion. Of Helmholtz energy functional Eq. (4-10),
course, as in the Landau theory, higher-or- putting (Fi – Fj) ≈ a · —Fi (see Milchev
der “anharmonic” terms in Q are crucial for et al. (1986) for a discussion).
the description of the ordered phase in An interesting distinction concerns this
terms of a stable order parameter ·Qk0 , l0 Ò effective single-site Hamiltonian felt by the
for T < Tc . atoms undergoing the distortion. We have
These anharmonic terms thoroughly assumed that the ordered structure is
modify the picture of the transition as ob- doubly degenerate; the atoms below Tc can
tained from MFA, as they lead to a cou- sit in the right or the left minimum of a
pling of the soft mode Qk0, l0 with other double-well potential. If the single-site po-
noncritical modes. This coupling among tential above Tc is essentially of the same
modes gives rise to a damping of the soft type, and only the distribution of the atoms
mode. In fact, under certain circumstances over the minima is more or less random,
even an overdamped soft mode and the ap- the transition is called “order – disorder
pearance of a central peak are expected type”. This occurs, for example, for hydro-
(Gebhardt and Krey, 1979; Bruce and gen-bonded ferroelectrics and is analogous
Cowley, 1981). to the sublattice ordering described above
A treatment of the anharmonic higher- for alloys. On the other hand, if the single-
order terms of a Landau-like expansion site potential itself changes above Tc to a
in reciprocal space is cumbersome, as Eq. single-well form, the transition is called
(4-131) demonstrates. Denoting the ampli- “displacive”. Whereas it was often thought
tude of the displacement vector produced that displacive structural transitions exhibit
by the soft mode Qk0 , l0 in the unit cell i as well-defined soft phonons right up to their
Fi , we may formulate a model for a one- transition temperature Tc, it has now be-
 4.3 Computational Methods Dealing with Statistical Mechanics 293

come clear that all these structural transi- different structures can be handled by per-
tions with a one-component order parame- forming this calculation for both phases
ter acquire characteristics of order – disor- and identifying the temperature Tc where
der transitions close to Tc, as expected from the free energy branches of the two phases
the “universality principle”, and therefore cross. Since the quasi-harmonic theory is a
the distinction between the character of a calculation of the mean-field type, as
structural transition as being “order – disor- pointed out above, first order transitions
der” or “displacive” is not a sharp one also show up via stability limits of the
(Bruce and Cowley, 1981). phases, where the soft modes vanish; thus
It should also be noted that the softening we are not locating Tc but rather tempera-
of w (k0, l0) near Tc does not mean that dis- tures T0 or T1 (cf. Fig. 4-6 b), which are
placements ul (x) become very large. In often not very far from the actual transi-
fact, the mean square displacement of an tion temperature. This quasi-harmonic ap-
atom at a structural transition is only ex- proach to structural phase transitions has
pected to have an energy-like singularity been tried for many materials. Typical ex-
(Meißner and Binder, 1975) amples include RbCaF3 (Boyer and Hardy,
1981) and the systems CaF2 and SrF2
·u2l ÒT – ·u2l ÒTc ~ (T /Tc – 1)1 – a (4-137)
(Boyer, 1980, 1981 a, b), which show phase
where a is the specific heat exponent. This transitions to a superionic conducting state.
is important because ·u2l ÒT is easily de-
duced experimentally from the Mössbauer
effect, from the Debye – Waller factor de- 4.3.3 Computer Simulation Techniques
scribing the temperature variation of Bragg In a computer simulation, we consider a
peaks in X-ray or neutron scattering, etc. finite system (e.g., a cubic box of size L3
We end this subsection with a comment with periodic boundary conditions to avoid
on the theory of first-order structural tran- surface effects) and obtain information on
sitions. The common approach is to restrict the thermodynamic properties, correlation
the analysis entirely to the framework of functions, etc. of the system (as specified
the quasi-harmonic approximation, in by its model Hamiltonian) which is exact,
which the Helmholtz energy at volume V apart from statistical errors. However, this
and temperature T is written as approach is restricted to classical statistical
mechanics (including quasi-classical mod-
F (T , V ) = U − T S
els such as Ising or Potts models), although
1
= U0 ( V ) + ∑
w V ( k , l ) (4-138) remarkable progress on application to
2 k, l quantum problems has been made (Kalos,
+ kB T ∑ ln [1 − exp( −
w V ( k , l )/ kB T )] 1985; Kalos and Schmidt, 1984; De Raedt
k, l and Lagendijk, 1985; Suzuki, 1992; Ceper-
Therefore, if the effective potentials speci- ley, 1995). The principal approaches of this
fying the dynamical matrix ∂2 U/[(∂xil )a type are the molecular dynamics (MD)
∂xjl )b ] in Eq. (4-135) are known, the pho- technique and the Monte Carlo (MC) tech-
non frequencies wV (k, l) for a given vol- nique (for reviews of MD, see Ciccotti
ume and the free energy F (T, V) are ob- et al., 1987; Hoover, 1987; Hockney and
tained. Of course in this approach, knowl- Eastwood, 1988; Binder and Ciccotti,
edge of the structure of the material is 1996; of MC, see Binder, 1979, 1984 a;
assumed. First-order transitions between Mouritsen, 1984; Binder and Heermann,
294 4 Statistical Theories of Phase Transitions

1988; Binder and Ciccotti 1996). In the principle, this problem is well understood
MD method, we numerically integrate New- (Fisher 1971; Challa et al., 1986; Binder,
ton’s equation of motion which follows 1987 b; Privman, 1990). In practice, this
from the chosen Hamiltonian, assuming er- makes it difficult to distinguish between
godic behavior, and the quantities of inter- second-order and weakly first-order transi-
est are obtained as time averages from the tions. For example, extensive MD work
simulation. This method requires that the was necessary to obtain evidence that the
relevant physical time scales involved in melting transition of two-dimensional sol-
the problem are not too different from each ids with pure Lennard – Jones interaction is
other. The MD approach has often been ap- first order (Abraham, 1983, 1984; Bakker
plied successfully to liquid – solid transi- et al., 1984), and that the suggested two
tions. It would not be suitable to study or- continuous transitions involving the hexa-
der – disorder phenomena in solid alloys, tic phase do not occur in these systems
since the time step in the MD method for a (Nelson and Halperin, 1979). However,
solid must be much less than a phonon fre- this conclusion has been called into ques-
quency, and this time scale is orders of tion by recent simulations for hard-disk
magnitude smaller than the time between fluids (Jaster, 1998) providing evidence for
diffusive hops of atoms to neighboring va- continuous two-dimensional melting.
cant sites, which is the process relevant to Another difficulty is that the periodic
equilibration of configurational degrees of boundary condition (for a chosen shape of
freedom (Fig. 4-12). the box) prefers certain structures of a solid
For problems of the latter type, the MC and suppresses others which do not “fit”:
method clearly is to be preferred. In the this is particularly cumbersome for incom-
MC method, random numbers are used to mensurate modulated structures (Selke, 1988,
construct a random walk through the con- 1989; 1992) and for off-lattice systems,
figuration space of the model system. Us- such as studies of the fluid – solid transition
ing the Hamiltonian, transition probabilities or phase transitions between different lat-
between configurations are adjusted such tice symmetries. For example, particles in-
that configurations are visited according to teracting with a screened Coulomb potential
their proper statistical weight (Binder, (this is a model for colloidal suspensions or
1979, 1984 a; Binder and Ciccotti, 1996). colloidal crystals (see Alexander et al.,
Again averages are obtained as “pseudo- 1984)) may exhibit a fluid phase in addi-
time averages” along the trajectory of the tion to an f.c.c. and a b.c.c. crystal, and the
system in phase space, the only difference determination of a complete phase diagram
from the MD method being that the tra- is correspondingly difficult (Kremer et al.,
jectory is now stochastic rather than deter- 1986, 1987; Robbins et al., 1988). The tra-
ministic. Both methods have been exten- ditional approach to dealing with such prob-
sively reviewed (see the references quoted lems is to repeat the calculation for differ-
above); therefore, we do not give any de- ent box shapes and compare the free ener-
tails here, but only briefly mention the dif- gies of the different phases. An interesting
ficulties encountered when phase transi- alternative method has been proposed by
tions are studied, and discuss a few typical Parrinello and Rahman (1980) and Parri-
examples of their application. nello et al. (1983), who generalized the MD
One principal difficulty is the finite-size method by including the linear dimensions
rounding and shifting of the transition. In of the box as separate dynamic variables.
 4.3 Computational Methods Dealing with Statistical Mechanics 295

Another severe problem is the occur-

rence of metastability and hysteresis; the
system may become trapped in a meta-
stable state, the lifetime of which is longer
than the observation time of the simulation.
The distinction of such long-lived meta-
stable states from true equilibrium states is
difficult and may require computation of
the Helmholtz energies of the phases in
question.
These problems must be considered
when first-order transitions are studied by
computer simulation so that suitable box
sizes and observation times are chosen and
the initial state is prepared accordingly.
Then, employing sufficient effort in com-
puting time, and performing a careful anal-
ysis of all the possible pitfalls mentioned
above, very reliable and useful results can
be obtained which are superior in most
cases to any of the other methods that have
been discussed so far (see Fig. 4-15 for a
comparison in the case of f.c.c. alloys).
Misjudgements of the problems mentioned
above have also led to erroneous con- Figure 4-17. Temperature dependence of the long-
clusions: in the f.c.c. Ising antiferromagnet range order parameter Y of the Cu3Au structure and
of the nearest- and next-nearest-neighbor short-range
with nearest interaction J, the triple point order parameter a 1 and a 2 . Curves are Monte Carlo
between the AB and A3B structures results of Binder (1986). Points show experimental
(Fig. 4-8 a, c) was suggested to occur at data of Cowley (1950), Schwartz and Cohen (1965),
T = 0 (Binder, 1980) while later a nonzero Moss (1964), Bardham and Cohen (1976), and Keat-
ut low temperature was found, kBT /| J | ~ 1.0 ing and Warren (1951). From Binder (1986).
(Gahn, 1986; Diep et al., 1986; Kämmerer
et al., 1996). Unlike in the other methods,
such problems can always be clarified by sive (Jnn) and next-nearest-neighbor attrac-
substantially increasing the computational tive interaction Jnnn (Eq. (4-110)), for vari-
effort and carrying out more detailed anal- ous choices of their ratio R = Jnnn/Jnn. The
yses of the simulation “data”. comparison with experimental data shows
A significant advantage of such a “com- that a reasonable description of Cu3Au is
puter experiment” is that interaction pa- obtained for R = – 0.2, for T < Tc , whereas
rameters of a model can be varied system- better experimental data for T > Tc are
atically. As an example, Fig. 4-17 shows needed before a more refined fit of effec-
the temperature dependence of the long- tive interaction parameters can be per-
range order parameters and some short- formed.
range order parameters of an f.c.c model of As an example showing that the MC
an A3B alloy with nearest-neighbor repul- method can deal with very complex phase
296 4 Statistical Theories of Phase Transitions

diagrams, we consider the model in Eq.

(4-115) for alloys with one magnetic com-
ponent. However, in order to facilitate
comparison with the MFA and CV meth-
ods, the magnetic interaction is chosen to be
of Ising rather than Heisenberg type (Dün-
weg and Binder, 1987). Fig. 4-18 shows
the resulting phase diagrams for the choice
of interaction parameters Jnnn/Jnn = 0.5 and
Jm /| Jnn | = 0.7. In this case, the MFA pre-
dicts a wrong phase diagram “topology”
(e.g., a direct transition between paramag-
netic D03 and A2 phases never occurs,
there is always a B2 phase in between), and
it also grossly overestimates the transition
temperatures. In contrast, the CV method
yields the phase diagram “topology” cor-
rectly, and it overestimates the transition
temperature by only a few per cent. Al-
though the CV method does not always
perform so well (see Fig. 4-15, and Binder
(1980) and Diep et al. (1986) for a discus-
sion in the case of f.c.c. alloys), it is always
much superior to the simple MFA, which in
many cases fails dramatically.
One great advantage of the MC method
over the CV method is that it can also be
applied straightforwardly to models with
continuous degrees of freedom, for which
the CV method would be cumbersome to
work out. As examples, Figs. 4-19 and
Figure 4-18. Phase diagram of the body-centered
4-20 show the phase diagrams of the model
cubic binary alloy model with nearest and next near-
in Eqs. (4-93) and (4-136), namely the an- est neighbor crystallographic interactions Jnn and
isotropic classical Heisenberg antiferro- Jnnn (both being “antiferromagnetic” if Ising-model
magnet in a uniform field H|| along the easy terminology is used) and a ferromagnetic nearest-
axis and the F 4 model on the square lat- neighbor interaction Jm between one species, for
tice. Note that in the latter case the result Jnnn /Jnn = 0.5, Jm / | Jnn | = 0.7; c denotes the concentra-
tion of the magnetic species. (a) Result of the MFA
of the MFA would be off-scale in this (Bragg – Williams approximation); (b) result of the
figure (e.g., Kc– 1 = 4 in the Ising limit). Fi- CV method in the tetrahedron approximation, and
nally, Fig. 4-21 shows the phase diagram of part (c) the MC result. The A2 phase is the crystallo-
a model for EuxSr1 – xS (Binder et al., graphically disordered phase, the orderings of the B2
1979): and D03 phases are shown in Fig. 4-3. Two-phase co-
existence regions are shaded. The magnetic ordering
of the phases is indicated as para(-magnetic) and
1
Ᏼ=− ∑ Jij xi x j Si ⋅ S j (4-139) ferro(-magnetic), respectively. From Dünweg and
2 i≠ j Binder (1987).
 4.3 Computational Methods Dealing with Statistical Mechanics 297

Figure 4-19. Phase diagram of a uniaxial classical Figure 4-20. Critical line Kc–1 of the F 4 model
Heisenberg antiferromagnet on the simple cubic lat- (Eq. (4-136)) on the square lattice, shown in the
tice, as a function of temperature T and field H|| ap- space of couplings K –1, L–1 defined as K = – r C/u,
plied in the direction of the easy axis. The anisotropy L = r (r + 4 C)/(4 u). Note that for L–1 = 0, and K finite,
parameter D in the Hamiltonian Eq. (4-93) is chosen the square Ising model results, the transition temper-
as D = 0.2. Both Monte Carlo results (crosses, circles, ature of which is exactly known (Onsager, 1944; ar-
full curves) and the Landau theory fitted to the phase row). (쐌) Monte Carlo simulation of Milchev et al.
diagram in the region off the true bicritical point is (1986); (+) MD results of Schneider and Stoll
shown (dash-dotted straight line). The Landau theory (1976); (왌, ¥) from a real space renormalization
would overestimate the location of the bicritical tem- group calculation of Burkhardt and Kinzel (1979).
perature Tb (Tb* > Tb ), and fails to yield the singular The disordered phase occurs above the critical line.
umbilicus shape of the phase diagram lines Tc^, Tc|| From Milchev et al. (1986).
near the bicritical point. The broken straight lines de-
note the appropriate choices of “scaling axes” for a
crossover scaling analysis near Tb . The triangles nearest neighbors only, further-neighbor
show another phase diagram, namely when the field interactions being negligibly small, with
is oriented in the perpendicular direction to the easy
axis. The nature of the phases is denoted as AF (anti-
Jnnn/ Jnn = – 1/2. Thus Fig. 4-21 exhibits a
ferromagnetic), SF (spin-flop) and P (paramagnetic). remarkable agreement between computer
See Fig. 4-9 a for a schematic explanation of this simulation and experiment (Maletta and
phase diagram. From Landau and Binder (1978). Felsch, 1979); there are no adjustable pa-
rameters involved whatsoever. Note that
where Si are unit vectors in the direction of analytical methods for such problems with
the Eu magnetic moment, and from spin quenched disorder would be notoriously
wave measurements (Bohn et al., 1980) it difficult to apply.
is known that EuS essentially exhibits To conclude this subsection, we state
superexchange between nearest and next- that computer simulation techniques are
298 4 Statistical Theories of Phase Transitions

4.4 Concepts About Metastability

Metastable phases are very common in
nature, and, for many practical purposes,
not at all distinct from stable phases (re-
member that diamond is only a metastable
modification of graphite!). Also, approxi-
mate theories of first-order phase transi-
tions easily yield Helmholtz energy
branches that do not correspond to the ther-
mal equilibrium states of minimum
Helmholtz energy, and hence are com-
Figure 4-21. Ferromagnetic critical temperature of monly interpreted as metastable or unstable
the model Eq. (4-139) of Eu x Sr1– x S vs. concentration states (cf. Figs. 4-6 and 4-22). From the an-
x of magnetic atoms. Circles denote experimental alog of Eq. (4-12) for H (0,
data due to Maletta and Felsch (1979), triangles the (4-140)
MC results for the diluted classical Heisenberg fcc 1 ⎛ ∂F ⎞ H
⎜ ⎟ = r F + uF 3 − =0
kB T V ⎝ ∂F ⎠ T
ferromagnet with nearest (Jnn ) and next-nearest
(Jnnn ) exchange, Jnnn = – –12 Jnn . From Binder et al.
kB T
(1979).
the limit of metastability where c T =
(∂F /∂H )T diverges as follows:
well suited for studying the phase diagrams 1 1
r + 3 uFs2 = =0
of various systems, magnetic systems, me- kB T c T
tallic alloys, structural transitions, ad-
Fs = − r / 3 u = F0 / 3
sorbed layers at surfaces, etc. provided that
a suitable model Hamiltonian is known. 2r −r
Hc = − (4-141)
Apart from the phase diagram, detailed in- 3 3u
formation on long- and short-range order is
also accessible. Also, for certain problems, Since in the metastable states
the kinetics of phase transitions can be
cT = (3 kB Tu)– 1 (F 2 – F s2)– 1
studied, in particular for alloys where the
Monte Carlo process can directly model we see that cT Æ • as F Æ Fs; in the
the atomic jump processes on the crystal (T, F ) plane, the spinodal curve F = Fs (T )
lattice (Binder, 1979, 1984 a; Kehr et al., plays the part of a line of critical points.
1989). Of course, there is also interest in Similar behavior also occurs in many
clarifying the phase diagrams of materials other theories: for example, the van der
in their fluid states. Computer simulations Waals equation of state describing gas – liq-
can now obtain accurate gas – liquid phase uid condensation exhibits an analogous
diagrams (Wilding, 1997), liquid binary loop of one-phase states in the two-phase
mixtures (Wilding et al., 1998) and surfac- coexistence region.
tants (Schmid, 1999). Unfortunately, although the description
of metastability in the framework of the
MFA seems straightforward, this is not so
if more accurate methods of statistical ther-
modynamics are used. A heuristic compu-
 4.4 Concepts About Metastability 299

tational approach, discussed in Sec. 4.3.2,

is the CV method where we systematically
improve upon the MFA by taking more and
more short-range correlations into account,
the larger the cluster is chosen. Computing
the equation of state for a nearest-neighbor
Ising ferromagnet in this way (Kikuchi,
1967), it is found that the stable branch is
nicely convergent, whereas the metastable
loop becomes flatter and flatter the larger
the cluster, i.e., the critical field Hc (Eq. Figure 4-22. Order parameter F vs. conjugate field
(4-141)) converges towards zero. An exact H according to the phenomenological Landau theory
calculation of the equation of state yields for system at a temperature T less than the critical
the magnetization jump from F0 to – F0 temperature Tc of a second-order phase transition
as H changes from 0+ to 0– in Fig. 4-22, but (schematic). At H = 0, a first-order transition from F0
to – F0 occurs (thick straight line). The metastable
does not yield any metastable state! This is branches (dash-dotted) end at the “limit of meta-
not surprising, however, because statistical stability” or “spinodal point” (Fs , – Hs ), respectively,
mechanics is constructed to yield informa- and are characterized by a positive-order parameter
tion only on thermal equilibrium states. susceptibility c T > 0, whereas for the unstable branch
The partition function is dominated by the (broken curve) c T < 0.
system configurations in the vicinity of the
minimum of the Helmholtz energy. In a
magnet with H < 0, states with negative magnetization in equilibrium is negative,
magnetization have lower Helmholtz en- but in the course of the nonequilibrium re-
ergy than those with positive magnetiza- laxation process from the initial state at F 0
tion. Hence the latter do not result from the a metastable state may occur with Fms > 0,
partition function in the thermodynamic although H < 0, which exists only over a
limit. Similar conclusions emerge from a finite lifetime tms . In order that Fms can
rigorous treatment of the gas – fluid transi- clearly be identified in this dynamic pro-
tion in systems with long-range interac- cess where the time-dependent order pa-
tions (Lebowitz and Penrose, 1966). rameter F (t) relaxes from F (t = 0) = F 0 to
Two concepts emerged for the descrip- the negative equilibrium value F (t Æ •),
tion of metastable states (for a more de- it is necessary that tms is much larger than
tailed discussion see Binder (1987 a)). One any intrinsic relaxation time. Then the time
concept (Binder, 1973) suggests the defini- tms for the decay of the metastable state far
tion of metastability be based on the con- exceeds the time needed for the system to
sideration of kinetics: starting with the sys- relax from the initial state towards the met-
tem in a state of stable equilibrium, the sys- astable state, and, in the “nonequilibrium
tem is brought out of equilibrium by a sud- relaxation function” F (t) the metastable
den change of external parameters (temper- state shows up as a long-lived flat part
ature, pressure, fields, etc.). For example, where F (t) ≈ Fms. The second concept
in the Ising-type ferromagnet of Fig. 4-22, tries to define a metastable state in the
at times t < 0 we may assume we have a framework of equilibrium statistical me-
state at H = 0+, F = F 0 > 0: at t = 0, the field chanics by constraining the phase space so
is switched to a negative value: then the as to forbid the two-phase configurations
300 4 Statistical Theories of Phase Transitions

that otherwise dominate the partition func- F = Fs (T ) acts like a line of critical points,
tion. Langer (1974) suggested that phase as discussed above, we require (cf. Eq.
separation into two phases with order pa- (4-58))
rameters – F 0 and + F 0 coexisting at the
·[F (x) – F]2 ÒT, L O [F – Fs (T )]2 (4-142)
first-order transition at H = 0 in Fig. 4-1
and Fig. 4-22 is suppressed if we consider i.e., the mean-square fluctuations in a vol-
a system (a) at fixed F between – F 0 and ume Ld must be much smaller than the
+ F 0 but (b) constrain the system and di- squared distance of the order parameter
vide it into cells of size Ld, and (c) require from its value at criticality, which in this
that the order parameter is not only glo- case is the value at the spinodal. Now, the
bally fixed at F but also inside each cell. maximum permissible value for L is the
If L is small enough, namely a O L O xcoex correlation length x in the metastable state,
(this is the same condition as noted for which becomes (4-143)
the construction of a continuum model
from the microscopic Hamiltonian, see x = {R/[6 d u Fs (T )]1/2} [F – Fs (T )]– 1/2
Sec. 4.2.2, and Eqs. (4-9) and (4-57)), Eq. (4-143) exhibits the critical singularity
phase separation inside a cell cannot occur, of x at Fs (T ), corresponding to that of cT
and hence a coarse-grained Helmholtz en- noted above. With a little algebra we find
ergy density fcg (F ) of states with uniform from Eqs. (4-142) and (4-143) (Binder,
order parameter F is obtained, which has 1984 c)
exactly the double-well shape of the Lan-
dau theory (Fig. 4-6 a). Since fcg (F ) then 1 O R2 x d − 2 [F − Fs ( T )]2
necessarily depends on this coarse-grained ~ R d [Fs ( T )]( 2 − d )/ 2 [F − Fs ( T )]( 6 − d )/ 2
length scale L, it is clear that the metastable
~ R d ( Hc − H )( 6 − d )/ 4 (4-144)
states of fcg (F ) are not precisely those
observed in experiments, unlike the dy- This condition can only be fulfilled if the
namic definition. While fcg (F ) defines a interaction range is very large. Even then
“limit of metastability” or “spinodal curve” for d = 3 this inequality is not fulfilled very
Fs where (∂2 fcg (F )/∂F 2 )T = c T– 1 vanishes close to the spinodal curve.
and changes sign, this limit of metastability In order to elucidate the physical signifi-
also depends on L and hence is not related cance of this condition further and to dis-
to any physical limit of metastability. In the cuss the problem of the lifetime of meta-
context of spinodal decomposition of mix- stable states, we briefly discuss the mean
tures, the problem of the extent to which a field theory of nucleation phenomena. In
spinodal curve is relevant also arises and is practice, for solid materials, heterogeneous
discussed in detail in the chapter by Binder nucleation (nucleation at surfaces, grain
and Fratzl (2001). boundaries, dislocations, etc.) will domi-
In some systems where a mean-field de- nate; and this may restrict the existence of
scription is appropriate, the lifetime of metastable states much more than expected
metastable states can be extremely long, from Eq. (4-144). However, we consider
and then the mean-field concept of a limit here only the idealized case that homoge-
of metastability may be very useful. This neous nucleation (formation of “droplets”
fact can again be understood in terms of of the stable phase due to statistical fluctu-
the Ginzburg criterion concept as outlined ations) is the mechanism by which the met-
in Sec. 4.2.2. Since the spinodal curve astable state decays. Since there is no
 4.4 Concepts About Metastability 301

mechanism by which this process can be

suppressed, metastable states cannot exist
in a region where strong homogeneous nu-
cleation occurs. Therefore we estimate the
intrinsic and ultimate limit of metastability
by the condition that the Helmholtz energy
barrier against homogeneous nucleation
decreases to the order of the thermal en-
ergy kBT (Binder and Stauffer, 1976 b).
As is well known, the nucleation barrier
arises from competition between the favor-
able) volume energy of a droplet of the new
phase and the (unfavorable) surface Helm-
holtz energy between the droplet and the
surrounding metastable background phase.
In order to calculate this Helmholtz energy
barrier in the framework of the Landau the-
ory, we can still use Eqs. (4-10) and (4-33).
Putting H (x) = 0 in Eq. (4-33) and consid-
ering the situation that F (z Æ •) = F 0 ,
F (z Æ – •) = – F 0 and solving for the con-
centration profile F (z) (Cahn and Hilliard,
Figure 4-23. Order-parameter profile F (z) across
1958), we can obtain the interface Helm-
an interface between two coexisting phases ± Fcoex ,
holtz energy associated with a planar (infi- the interface being oriented perpendicular to the z-di-
nitely extended) interface perpendicular to rection (a) and the radial order-parameter profile for
the z-direction (Fig. 4-23 a). In the “classi- a marginally stable droplet in a metastable state
cal theory of nucleation” (Zettlemoyer, which is close to the coexistence curve (b) or close to
the spinodal curve (c). In (a) and (b) the intrinsic
1969), this Helmholtz energy is then used
“thickness” of the interface is of the order of the cor-
to estimate the Helmholtz energy barrier. relation length at coexistence x coex , whereas in (c) it
However, this “classical theory of nuclea- is of the same order as the critical radius R*. From
tion” is expected to be reliable only for Binder (1984 c).
metastable states near the coexistence
curve, where the radius R* of a critical permitted, and a boundary condition
droplet (corresponding to a droplet Helm- F (r Æ •) = F ms is imposed (Fig. 4-
holtz energy exactly at the Helmholtz en- 23 b, c). Whereas for Fms near Fcoex = F 0
ergy barrier DF*) is much larger than the this treatment agrees with the “classical
width of the interfacial profile (which then theory of nucleation”, it differs signifi-
is also of the same order as the correlation cantly from it for F near Fs (T ): then the
length xcoex at the coexistence curve, cf. critical droplet radius R* is of the same or-
Figs. 4-23 a, b). Therefore it cannot be used der as the (nearly divergent!) correlation
close to the limit of metastability. Cahn and length x (Eq. (4-143)), and the profile is
Hilliard (1959) have extended the Landau extremely flat, F (r) reaches in the droplet
theory to this problem, solving Eq. (4-33) center only a value slightly below Fs rather
for a spherical geometry, where only a than the other branch of the coexistence
radial variation of F (r) with radius r is curve. Calculating the Helmholtz energy
302 4 Statistical Theories of Phase Transitions

barrier DF*, we obtain for T near Tc

(Binder, 1984 c; Klein and Unger, 1983)
( 4 − d )/ 2 ( 6 − d )/ 2
DF* ⎛ T⎞ ⎛ F − Fs ⎞
~ Rd ⎜1 − ⎟ ⎜ ⎟
kB Tc ⎝ Tc ⎠ ⎝ Fcoex ⎠
(4-145)
whereas near the coexistence curve the re-
sult is
( 4 − d )/ 2 − ( d − 1)
DF* ⎛ T⎞ ⎛ Fcoex − F ⎞
~ Rd ⎜1 − ⎟ ⎜ ⎟
kB Tc ⎝ Tc ⎠ ⎝ Fcoex ⎠
(4-146)
where all prefactors of order unity are omit-
ted. In a system with a large range R of in-
teraction, the nucleation barrier is very high
in the mean-field critical region, in which
R d (1 – T /Tc)(4 – d )/2 o 1 (cf. Eq. (4-62)); this
factor, which controls the Ginzburg criter-
ion, also controls the scale of the nuclea-
tion barrier as a prefactor (see Fig. 4-24).
Figure 4-24. Schematic plots of the Helmholtz en-
In this region the condition for the limit of ergy barrier for (a) the mean field critical region, i.e.,
metastability, DF*/kBT ≈ 1, is located very R d (1 – T/Tc ) (4 – d )/2 o 1, and (b) the non-mean field
close to the mean-field spinodal. Then the critical region i.e., R d (1 – T/Tc ) (4 – d )/2 O 1. Note that
description of nucleation phenomena in owing to large prefactors to the nucleation rate, the
terms of the diffuse droplets described by constant of order unity where the gradual transition
from nucleation to spinodal decomposition occurs is
Fig. 4-23 c near the spinodal curve is mean- about 101 rather than 100. From Binder (1984 c).
ingful (“spinodal nucleation”). On the
other hand, for a system with short-range
interactions where R (measured in units
of the lattice spacing in Eqs. (4-145) and new phase) are also extremely rare, and the
(4-146)) is unity, the Helmholtz energy Helmholtz energy cost to form them should
barrier becomes of order unity long before be very high, as implied by Eqs. (4-145)
the spinodal curve is reached. The singu- and (4-146).
larity at the spinodal then completely lacks These concepts have been tested by
any physical significance, as the meta- Monte Carlo simulations on simple-cubic
stable state decays to the stable phase long Ising models with various choices of the
before the spinodal is reached. range of the interaction (Fig. 4-25, Heer-
It is instructive to compare the condition mann et al., 1982). For the case of nearest-
DF*/kBT O 1 with Eq. (4-144): this shows neighbor interaction (each spin interacts
that these conditions are essentially iden- with q = 6 neighbors) it is seen that cT– 1 dif-
tical! This is not surprising: the MFA is fers considerably from the MFA result, and
essentially correct as long as effects of the range of fields over which metastable
statistical fluctuations are very small: the states can be observed is not very large. With
“heterophase fluctuations” (droplets of the an increase in the number of neighbors q,
 4.5 Discussion 303

have a strong influence on certain physical

properties. The approach of statistical me-
chanics tries to provide general concepts
for dealing with such phenomena: classifi-
cation methods are developed which also
try to clarify which aspects of a phase
transition are specific for a particular mate-
rial and which are general (“universal”). At
the same time, theoretical descriptions are
available both on a phenomenological
level, where thermodynamic potentials are
expanded in terms of suitable order param-
eters and the expansion coefficients are un-
determined, and can only be adjusted to ex-
perimental data, and on a microscopic
level, where we start from a model Hamil-
tonian which is treated either by molecular
field approximations or variants thereof or
by computer simulation techniques.
This chapter has not given full details of
all these approaches, but rather tried to
Figure 4-25. Inverse susceptibility of Ising ferro-
magnets plotted against h = – H/kB T at T/TcMFA = 4/9 give a discussion which shows what these
for various ranges of the exchange interaction; each methods can achieve, and to give the reader
spin interacts with equal strength with q neighbors. a guide to more detailed literature on the
The full curve is the MFA, the broken curve a fit to a subject. An attempt has been made to sum-
droplet model description. From Heermann et al.
marize the main ideas and concepts in the
(1982).
field and to describe the general facts that
have been established, while actual materi-
als and their phase transitions are men-
this range increases, the value of cT– 1 at tioned as illustrative examples only, with
the field where nucleation becomes appre- no attempt at completenesss being made.
ciable falls also, and the cT– 1 vs. H curve Although the statistical thermodynamics of
quickly converges towards the MFA pre- phase transitions has provided much physi-
diction, except in the immediate neighbor- cal insight and the merits and limitations of
hood of the limit of metastability Hc. the various theoretical approaches are now
It is expected that the ideas sketched well understood generally, the detailed
here will carry over to more realistic sys- understanding of many materials is still ru-
tems. dimentary in many cases; often there are
not enough experimental data on the order-
ing phenomenon in question, or the data
4.5 Discussion are not precise enough; a microscopic
model Hamiltonian describing the interac-
There are many different kinds of phase tion relevant for the considered ordering is
transitions in materials, and the fact that a often not explicitly known, or is so com-
material can exist in several phases may plicated that a detailed theory based on
304 4 Statistical Theories of Phase Transitions

such a model Hamiltonian does not exist. Baxter, R. J. (1982), Exactly Solved Models in Statis-
tical Mechanics. London: Academic.
Many of the more sophisticated methods Bean, C. P., Rodbell, D. S. (1962), Phys. Rev. 126,
(CV method, computer simulation, etc.) 104.
are restricted in practice to relatively sim- Belanger, D. P. (1998), in: Spin Glasses and Random
Fields: Young, A. P. (Ed.). Singapore: World Sci-
ple models. Hence, there still remains a lot entific, p. 251.
to be done to improve our understanding of Berlin, T. H., Kac, M. (1952), Phys. Rev. 86, 821.
phase transitions. Bethe, H. (1935), Proc. Roy. Soc. A 150, 552.
Bieber, A. Gautier, F. (1984 a), J. Phys. Soc. Japan
53, 2061.
Bieber, A., Gautier, F. (1984 b), Z. Phys. B 57, 335.
4.6 References Binder, K. (1973), Phys. Rev. B 8, 3423.
Binder, K. (Ed.) (1979), Monte Carlo Methods in
Abraham, F. F. (1974), Homogeneous Nucleation Statistical Physics. Berlin: Springer.
Theory. New York: Academic. Binder, K. (1980), Phys. Rev. Lett. 45, 811.
Abraham, F. F. (1983), Phys. Rev. Lett. 50, 978. Binder, K. (1983 a), Z. Physik B 50, 343.
Abraham, F. F. (1984), Phys. Rev. B 29, 2606. Binder, K. (1983 b), in: Phase Transitions and Criti-
Aharony, A. (1973), Phys. Rev. B 8, 1973. cal Phenomena, Vol. 8: Domb, C., Green, M. S.
Aharony, A. (1976), in: Phase Transitions and Criti- (Eds.). London: Academic Press, p. 1.
cal Phenomena, Vol. 6: Domb, C., Green, M. S. Binder, K. (1984 a), Applications of the Monte Carlo
(Eds.). London: Academic, p. 358. Method in Statistical Physics. Berlin: Springer
Alefeld, G. (1969), Phys. Stat. Sol. 32, 67. Verlag.
Alefeld, G., Völkl, J. (Eds.) (1978), Hydrogen in Binder, K. (1984 b), in: Condensed Matter Research
Metals I, II. Berlin: Springer Verlag (Topics in Using Neutrons: Lovesey, S. W., Scherm, R.
Appl. Physics, Vols. 28, 29). (Eds.). New York: Plenum Press, p. 1.
Alexander, S. (1975), Phys. Lett. 54 A, 353. Binder, K. (1984 c), Phys. Rev. A 29, 341.
Alexander, S., Chaikin, P. M., Grant, P., Morales, G. Binder, K. (1986), in: Atomic Transport and Defects
J., Pincus, P., Hone, D. (1984), J. Chem. Phys. 80, in Metals by Neutron Scattering: Janot, C., Petry,
5776. T., Richter, D., Springer, T. (Eds.). Berlin: Sprin-
Als-Nielsen, J. (1976), in: Phase Transitions and ger, p. 12.
Critical Phenomena, Vol. 5 A: Domb, C., Green, Binder, K. (1987 a), Rep. Progr. Phys. 50, 783.
M. S. (Eds.). New York: Academic, p. 88. Binder, K. (1987 b), Ferroelectrics 73, 43.
Als-Nielsen, J., Laursen, I. (1980), in: Ordering of Binder, K. (1989), in: Alloy Phase Stability: Gonis,
Strongly Fluctuating Condensed Matter Systems: A., Stocks, L. M. (Eds.). Dordrecht: Kluwer Acad.
Riste, T. (Ed.). New York: Plenum Press, p. 39. Publ., p. 232.
Amit, D. J. (1984), Field Theory, the Renormaliza- Binder, K. (1994), Adv. Polymer Sci. 112, 181.
tion Group and Critical Phenomena. Singapore: Binder, K. (1998), in: Spin Glasses and Random
World Scientific. Fields: Young, A. P. (Ed.). Singapore: World Sci-
Angell, C. A., Goldstein, M. (Eds.) (1986), Dynamic entific, p. 99.
Aspects of Structure Change in Liquids and Binder, K., Fratzl, P. (2001), in: Phase Transforma-
Glasses. New York: N.Y. Acad. of Sciences. tions in Materials: Kostorz, G. (Ed.). Weinheim:
Ausloos, M., Elliott, R. J. (1983), Magnetic Phase Wiley-VCH, Chap. 6.
Transitions. Berlin: Springer Verlag. Binder, K., Heermann, D. W. (1988), Monte Carlo
Bak, P., Domany, E. (1979), Phys. Rev. B 20, 2818. Simulation in Statistical Physics: An Introduction.
Bak, P., Mukamel, D. (1979), Phys. Rev. B 19, 1604. Berlin: Springer.
Bak, P., Mukamel, D., Villain, J., Wentowska, K. Binder, K. Landau, D. P. (1981), Surface Sci. 108, 503.
(1979), Phys. Rev. B 19, 1610. Binder, K., Landau, D. P. (1989), in: Molecule – Sur-
Bakker, A. F., Bruin, C., Hilhorst, H. J. (1984), Phys. face Interactions: Lawley, K. P. (Ed.). New York:
Rev. Let. 52, 449. Wiley & Sons, p. 91.
Bardham, P., Cohen, J. B. (1976), Acta Cryst. A 32, Binder, K., Stauffer, D. (1976 a), Z. Phys. B 24,
597. 407.
Bates, S. F., Rosedale, J. H., Fredrickson, G. H., Binder, K., Stauffer, D. (1976 b), Adv. Phys. 25, 343.
Glinka, C. J. (1988), Phys. Rev. Lett. 61, 229. Binder, K., Young, A. P. (1986), Rev. Mod. Phys. 58,
Baus, M. (1987), J. Stat. Phys. 48, 1129. 801.
Baus, M., Rull, L. F., Ryckaert, J. P. (Eds.) (1995), Binder, K., Meißner, G., Mais, H. (1976), Phys. Rev.
Observation, Prediction and Simulation of Phase B 15, 267.
Transitions in Complex Fluids. Dordrecht: Kluwer Binder, K., Kinzel, W., Stauffer, D. (1979), Z. Phys.
Acad. Publ. B 36, 162.
Baxter, R. J. (1973), J. Phys. C 6, L445. Binder, K. Reger, J. D. (1992), Adv. Phys. 41, 547.
 4.6 References 305

Binder, K., Ciccotti, G. (Eds.) (1996), Monte Carlo Cusack, N. E. (1987), The Physics of Structurally
and Molecular Dynamics of Condensed Matter. Disordered Matter: An Introduction. Bristol:
Bologna: Società Italiana di Fisica. Adam Hilger.
Blinc, R., Levanyuk, A. P. (1986), Incommensurate De Fontaine, D. (1975), Acta Met. 23, 553.
Phases in Dielectrics, Vols. 1, 2. Amsterdam: De Fontaine, D. (1979), in: Solid State Physics, Vol.
North-Holland. 34: Ehrenreich, H., Seitz, F., Turnbull, D. (Eds.).
Blinc, R., Zeks, B. (1974), Soft Modes in Ferro- London: Academic, p. 73.
electrics and Antiferroelectrics. Amsterdam: De Fontaine, D., Kulik, J. (1985), Acta Metall. 33, 145.
North-Holland. De Gennes, P. G. (1974), The Physics of Liquid Crys-
Bloch, D., Mauri, R. (1973), Phys. Rev. B 7, 4883. tals. Oxford: Clarendon Press.
Blöte, H. W., Swendsen, R. H. (1979), Phys. Rev. De Gennes, P. G. (1979), Scaling Concepts in Poly-
Lett. 43, 799. mer Physics. Ithaca: Cornell University Press.
Blume, M. (1966), Phys. Rev. 141, 517. De Jongh, J. L., Miedema, A. R. (1974), Adv. Phys.
Blume, M., Emery, V., Griffiths, R. B (1971), Phys. 23, 1.
Rev. A 4, 1971. Delaey, L. (2001), in: Phase Transformations in Ma-
Bohn, H. G., Zinn, W., Dorner, B., Kollmar, A. terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
(1980), Phys. Rev. B 22, 5447. Chap. 9.
Born, M., Huang, K. (1954), Dynamical Theory of Den Nijs, M. P. M. (1979), J. Phys. A 12, 1857.
Crystal Lattices. Oxford: Oxford Univ. Press. De Raedt, H., Lagendijk, A. (1985), Phys. Repts.
Boyer, L. L. (1980), Phys. Rev. Lett. 45, 1858. 127, 235.
Boyer, L. L. (1981 a), Phys. Rev. Lett. 56, 1172. De Raedt, B., Binder, K., Michel, K. H. (1981), J.
Boyer, L. L. (1981 b), Ferroelectrics 35, 83. Chem. Phys. 75, 2977.
Boyer, L. L., Hardy, J. R. (1981), Phys. Rev. B 24, Diep, H. T., Ghazali, A., Berge, B., Lallemand, P.
2577. (1986), Europhys. Lett. 2, 603.
Bragg, W. L., Williams, E. J (1934), Proc. Roy. Soc. Dietrich, S. (1988), in: Phase Transitions and Criti-
A 145, 699. cal Phenomena, Vol. 12: Domb, C., Lebowitz, J. L.
Brazovskii, S. A. (1975), Soviet Phys. JETP 41, 85. (Eds.). New York: Academic, p. 1.
Bretz, M. (1977), Phys. Rev. Lett. 38, 501. Dillon, J. F., Chen, E. Y., Guggenheim, H. J. (1978),
Broddin, D., Van Tendeloo, G., Van Landuyt, J., Phys. Rev. B 18, 377.
Amelinckx, S., Portier, R., Guymont, M., Loiseau, Di Tolla, D. F., Tosatti, E., Ercolessi, F. (1996), in:
A. (1986), Phil. Mag. A 54, 395. Monte Carlo and Molecular Dynamics of Con-
Brout, R. (1959), Phys. Rev. 115, 824. densed Matter Systems: Binder, K., Ciccotti, G.
Brout, R. (1965), Phase Transitions. New York: Ben- (Eds.). Bologna: Società Italiana di Fisica, p. 345.
jamin Press. Domany, E., Schick, M., Walker, J. S., Griffiths, R.
Bruce, A. D., Aharony, A. (1975), Phys. Rev. B 11, B. (1978), Phys. Rev. B 18, 2209.
478. Domany, E., Shnidman, Y., Mukamel, D. (1982), J.
Bruce, A. D., Cowley, R. A. (1981), Structural Phase Phys. C 15, L495.
Transitions. London: Taylor and Francis. Domb, C., Green, M. S. (Eds.) (1972 – 1976), Phase
Burkhardt, T. W., Kinzel, W. (1979), Phys. Rev. B 20, Transitions and Critical Phenomena, Vols. 1 – 6.
4730. New York: Academic.
Burkhardt, T. W., van Leeuwen, J. M. J. (Eds.) (1982), Domb, C, Green, M. S. (Eds.) (1983 – 1997), Phase
Real-Space Renormalization. Berlin: Springer. Transitions and Critical Phenomena, Vols. 6 – 17.
Buzaré, J. Y., Fayet, J. C., Berlinger, W., Müller, K. New York: Academic.
A. (1979), Phys. Rev. Lett. 42, 465. Dünweg, B., Binder, K. (1987), Phys. Rev. B 36, 6935.
Cahn, J. W., Hilliard, J. E. (1958), J. Chem. Phys. 28, Edwards, S. F., Anderson, P. W. (1975), J. Phys. F 5,
258. 965.
Cahn, J. W., Hilliard, J. E. (1959), J. Chem. Phys. 31, Enz, C. P. (Ed.) (1979), Dynamic Critical Phenom-
688. ena and Related Topics. Proceedings. Berlin:
Ceperley, D. M. (1995), Rev. Mod. Phys. 67, 279. Springer Verlag (Springer Lecture Notes in Phys-
Challa, M. S. S., Binder, K., Landau, D. P. (1986), ics, Vol. 104).
Phys. Rev. B 34, 1841. Ferrell, R. A., Menyhard, N., Schmidt, H., Schwabl,
Chandrasekhar, S. (1992), Liquid Crystals. Cam- F., Szepfalusy, P. (1967), Phys. Rev. Lett. 18, 891.
bridge: Cambridge University Press. Finel, A. (1994), in: Static and Dynamics of Alloy
Ciccotti, G., Frenkel, D., McDonald, I. R. (Eds.) Phase Transformations: Turchi, P. E. A., Gonis, A.
(1987), Simulations of Liquids and Solids. Amster- (Eds.). New York: Plenum Press, p. 495.
dam: North-Holland. Fisher, M. E. (1968), Phys. Rev. 176, 257.
Clapp, P. C., Moss, S. C. (1966), Phys. Rev. 142, 418. Fisher, M. E. (1971), in: Critical Phenomena: Green
Clapp, P. C., Moss, S. C. (1968), Phys. Rev. 171, 754. M. S. (Ed.). London: Academic, p. 1.
Cowley, J. M. (1950), Phys. Rev. 77, 669. Fisher, M. E. (1974), Rev. Mod. Phys. 46, 587.
Cowley, J. M. (1976), Phys. Rev. B 13, 4877. Fisher, M. E. (1975 a), AIP Conf. Proc. 24, 273.
306 4 Statistical Theories of Phase Transitions

Fisher, M. E. (1975 b), Phys. Rev. Lett. 34, 1634. Halperin, B. I., Hohenberg, P. C. (1967), Phys. Rev.
Fisher, M. E., Langer, J. S. (1968), Phys. Rev. Lett. Lett. 19, 700.
20, 665. Halperin, B. I., Varma, C. (1976), Phys. Rev. B14, 4030.
Fisher, M. E., Nelson, D. R. (1974), Phys. Rev. Lett. Halperin, B. I., Hohenberg, P. C., Ma, S. K. (1974),
32, 1350. Phys. Rev. B 10, 139.
Fisher, M. E., Racz, Z. (1976), Phys. Rev. B 13, 5039. Halperin, B. I., Hohenberg, P. C., Siggia, E. (1976),
Fishman, S., Aharony, A. (1979), J. Phys. C 12, L 729. Phys. Rev. B 13, 4119.
Flory, P. J. (1953), Principles of Polymer Chemistry. Hansen, M. (1958), Constitution of Binary Alloys.
Ithaca: Cornell University Press. New York: McGraw Hill.
Folk, R., Iro, H., Schwabl, F. (1976), Z. Phys. B 25, 69. Harris, A. B. (1974), J. Phys. C 7, 1671.
Franz, S., Parisi, G. (1998), Physica A 261, 317. Harris, A. B., Meyer, H. (1985), Can. J. Phys. 63, 3.
Frazer, B. C., Shirane, G., Cox, D. E. (1965), Phys. Harris, R., Plischke, M., Zuckermann, M. J. (1973),
Rev. 140 A, 1448. Phys. Rev. Lett. 31, 160.
Fredrickson, G. H., Binder, K. (1989), J. Chem. Heermann, D. W., Klein, W., Stauffer, D. (1982),
Phys. 91, 7265. Phys. Rev. Lett. 49, 1262.
Fredrickson, G. H., Helfand, E. (1987), J. Chem. Herrmann, H. J., Janke, W., Karsch, F. (Eds.) (1992),
Phys. 87, 697. Dynamic of First Order Phase Transitions. Singa-
Friedmann, L. F., Tunstall, D. P., (Eds.) (1978), The pore: World Scientific.
Metal – Non Metal Transition in Disordered Hockney, R. W., Eastwood, J. W. (1988), Computer
Systems. Edinburgh: SUSSP. Simulation Using Particles. Bristol: Adam Hilger.
Fujimoto, M. (1997), The Physics of Structural Höchli, U. T., Knorr, K., Loidl, A. (1990), Adv. Phys.
Phase Transitions. Berlin: Springer. 39, 405.
Gahn, U. (1986), J. Phys. Chem. Solids 47, 1153. Hohenberg, P. C., Halperin, B. I. (1977), Rev. Mod.
Gebhardt, W., Krey, U. (1979), Phasenübergänge Phys. 49, 435.
und kritische Phänomene – Eine Einführung. Hoover, W. G. (1987), Molecular Dynamics. Berlin:
Braunschweig: Vieweg. Springer.
Gehring, G. A. (1977), J. Phys. C 10, 531. Hornreich, R. M., Luban, M., Shtrikman, S. (1975),
Gehring, G. A., Gehring, K. A. (1975), Rep. Progr. Phys. Rev. Lett. 35, 1678.
Phys. 38, 1. Hulliger, F., Siegrist, T. (1979), Z. Phys. B 35, 81.
Gerold, V., Kern, J. (1986), in: Atomic Transport and Imbrie, J. Z. (1984), Phys. Rev. Lett. 53, 1747.
Defects in Metals by Neutron Scattering: Janot, C., Imry, Y. (1980), Phys. Rev. B 21, 2042.
Petry, T., Richter, D., Springer, T. (Eds.). Berlin: Imry, Y., Ma, S. K. (1975), Phys. Rev. Lett. 35, 1399.
Springer Verlag, p. 17. Imry, Y., Wortis, M. (1979), Phys. Rev. B 19, 3580.
Gerold, V., Kern, J. (1987), Acta Met. 35, 393. Inden, G. (2001), in: Phase Transformations in Mate-
Ginzburg, V. L. (1960), Sov. Phys.-Solid State 2, rials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
1824. Chap. 8.
Giordano, N., Wolf, W. P. (1975), Phys. Rev. Lett. 35, Jäckle, J. (1986), Rep. Progr. Phys. 49, 171.
799. Jaster, A. (1998), Europhys. Lett. 42, 277.
Golosov, N. S., Popov, L. E., Rudan, L. Y. (1973), Jona, F., Shirane, G. (1962), Ferroelectric Crystals.
J. Phys. Chem. Solids 34, 1149 – 1159. New York: Pergamon Press.
Gonis, A., Stocks, L. M. (Eds.) (1989), Alloy Phase Joyce, J. S. (1972), in: Phase Transitions and Criti-
Stability. Dordrecht: Kluwer Acad. Publ. cal Phenomena, Vol. 2: Domb, C., Green, M. S.
Grazhdankina, N. P. (1969), Sov. Phys.-Usp. 11, 727; (Eds.). New York: Academic, Chapter X.
Usp. Fiz. Nauk. 96, 291. Kämmerer, S., Dünweg, B., Binder, K., D’Onorio De
Grinstein, G. (1985), in: Fundamental Problems in Meo, M. (1996), Phys. Rev. B 53, 2345.
Statistical Mechanics, VI: Cohen, E. G. D. (Ed.). Kalos, M. H. (Ed.) (1985), Monte Carlo Methods in
Amsterdam, North-Holland, p. 147. Quantum Problems. Dordrecht: Kluwer Acad. Publ.
Guerard, D., Herold, A. (1975), Carbon 13, 337. Kalos, M. H., Schmidt, K. E. (1984), in: Applications
Gunton, J. D., Droz, M. (1983), Introduction to the of the Monte Carlo Method in Statistical Physics:
Theory of Metastable and Unstable States. Berlin: Binder, K. (Ed.). Berlin: Springer Verlag, p. 125.
Springer. Kaplan, T. A. (1959), Phys. Rev. 116, 888.
Gunton, J. D., San Miguel, M., Sahni, P. S. (1983), Keating, D. T., Warren, B. E. (1951), J. Appl. Phys.
in: Phase Transitions and Critical Phenomena, 22, 286.
Vol. 8: Domb, C., Lebowitz, J. L. (Eds.). London: Kehr, K. W., Binder, K., Reulein, S. M. (1989), Phys.
Academic, p. 267. Rev. B 39, 4891.
Guymont, M. (1978), Phys. Rev. B 18, 5385. Khachaturyan, A. G. (1962), Fiz. Metallov i Metallo-
Guymont, M. (1981), Phys. Rev. B 24, 2647. vedenie 13, 493.
Haasen, P., Gerold, V., Wagner, R., Ashby, M. F. Khachaturyan, A. G. (1963), Fiz. tverd. Tela 5, 26;
(Eds.) (1984), Decomposition of Alloys: The Early ibid 5, 750.
Stages. New York: Pergamon Press. Khachaturyan, A. G. (1973), Phys. stat. sol. (b) 60, 9.
 4.6 References 307

Khachaturyan, A. G. (1983), Theory of Structural Michel, K. H., Naudts, J. (1977), J. Chem. Phys. 67,
Transformation in Solids. New York: J. Wiley & 547.
Sons. Michel, K. H., Naudts, J. (1978), J. Chem. Phys. 68,
Kikuchi, R. (1951), Phys. Rev. 81, 998. 216.
Kikuchi, R. (1967), J. Chem. Phys. 47, 1664. Milchev, A., Heermann, D. W., Binder, K. (1986),
Kittel, C. (1967), Quantum Theory of Solids. New J. Stat. Phys. 44, 749.
York: J. Wiley & Sons. Moncton, D. E., Axe, J. D., Di Salvo, F. J. (1977),
Kittler, R. C., Falicov, L. M. (1978), J. Phys. B18, 2506. Phys. Rev. B 16, 801.
Kittler, R. C., Falicov, L. M. (1979), Phys. Rev. B 19, Morita, T. (1972), J. Math. Phys. 13, 115.
291. Moss, S. C. (1964), J. Appl. Phys. 35, 3547.
Kleemann, W., Schäfer, F. J., Tannhäuser, D. S. Moss, S. C., Clapp, P. C. (1968), Phys. Rev. 171, 764.
(1980), J. Magn. Magnet. Mater. 15 – 18, 415. Mott, N. F. (1974), Metal-Insulator Transitions. Lon-
Klein, W., Unger, C. (1983), Phys. Rev. B 28, 445. don: Taylor and Francis.
Koch, S. W. (1984), Dynamics of First-Ordered Mouritsen, O. G. (1984), Computer Studies of Phase
Phase Transitions in Equilibrium and Nonequilib- Transitions and Critical Phenomena. Berlin:
rium. Berlin: Springer. Springer.
Kötzler, J. (1984), Z. Phys. B 55, 119. Müller, K. A., Berlinger, W. (1975), Phys. Rev. Lett.
Kötzler, J., Raffius, G., Loidl, A., Zeyen, C. M. E. 22, 1547.
(1979). Z. Phys. B 35, 125. Müller-Krumbhaar, H. (1977), in: Crystal Growth
Kosterlitz, J. M., Thouless, D. J. (1973), J. Phys. C 6, and Materials: Kaldis, E. (Ed.). Amsterdam:
1181. North-Holland, p. 79.
Kremer, K., Robbins, M. O., Grest, G. S. (1986), Mukamel, D., Krinsky, S. (1976), Phys. Rev. B 13,
Phys. Rev. Lett. 57, 2694. 5065.
Kremer, K., Grest, G. S., Robbins, O. (1987), J. Mukamel, D., Fisher, M. E., Domany, E. (1976 a),
Phys. A 20, L181. Phys. Rev. Lett. 37, 565.
Krinsky, S., Mukamel, D. (1977), Phys. Rev. B 16, Mukamel, D., Krinsky, S., Bak, P. (1976 b), AIP
2313. Conf. Proc. 29, 474.
Krivoglaz, M. A. (1969), Theory of X-Ray and Ther- Nattermann, T. (1998), in: Spin Classes and Random
mal Neutron Scattering by Real Crystals. New Fields: Young, A. P. (Ed.). Singapore: World Sci-
York: Plenum Press. entific, p. 277.
Landau, D. P., Binder, K. (1978), Phys. Rev. B17, 2328. Nelson, D. R., Halperin, B. I. (1979), Phys. Rev. B19,
Landau, D. P., Binder, K. (1985), Phys. Rev. B31, 5946. 2457.
Landau, L. D., Lifshitz, E. M. (1958), Statistical Nienhuis, B., Riedel, E. K., Schick, M. (1980), J.
Physics. Oxford: Pergamon Press. Phys. A 13, L189.
Langer, J. S. (1974). Physica 73, 61. Nishiyama, Z. (1979), Martensitic Transformation.
Larkin, A. I., Khmelnitskii, D. E. (1969), Zh. Eksp. New York: Academic.
Fiz. 56, 2087; Sov. Phys. JETP 29, 1123. Onsager, L. (1944), Phys. Rev. 65, 117.
Lebowitz, J. L., Penrose, O. (1966), J. Math. Phys. 7, 98. Orr, R. L. (1960), Acta Met. 8, 489.
LeGuillou, J. C., Zinn-Justin, J. (1980), Phys. Rev. Orr, R. L., Luciat-Labry, J., Hultgren, R. (1960), Acta
B 21, 3976. Met. 8, 431.
Leibler, L. (1980), Macromolecules 13, 1602. Parrinello, M., Rahman, A. (1980), Phys. Rev. Lett.
Lifshitz, E. M. (1942), J. Physique 6, 61. 45, 1196.
Lipowsky, R. (1984), J. Appl. Phys. 55, 213. Parrinello, M., Rahman, A., Vashishta, P. (1983),
Loidl, A. (1989), Ann. Rev. Phys. Chem. 40, 29. Phys. Rev. Lett. 50, 1073.
Loiseau, A., Van Tendeloo, G., Portier, R., Ducas- Patashinskii, A. Z., Pokrovskii, V. I. (1979), Fluctua-
telle, F. (1985), J. Physique 46, 595. tion Theory of Phase Transitions. Oxford: Per-
Lynden-Bell, R. M., Michel, K. H. (1994), Rev. Mod. gamon Press.
Phys. 66, 721. Pershan, P. S. (1988), Structure of Liquid Crystal
Ma, S.-K. (1976), Modern Theory of Phase Transi- Phases. Singapore: World Scientific.
tions. Reading: Benjamin Press. Piercy, P., Pfnür, H. (1987), Phys. Rev. Lett. 59, 1124.
Maletta, H., Felsch, W. (1979), Phys. Rev. B 20, 1079. Pimenta, M. A., Echegut, P., Luspin, Y., Hauset, G.,
Mandelbrot, B. B. (1982), The Fractal Geometry of Gervais, F., Abélard, P. (1989), Phys. Rev. B 39,
Nature. San Francisco: Freeman. 3361.
Marx, R. (1989), Phys. Rev. B 40, 2585. Potts, R. B. (1952), Proc. Camb. Phil. Soc. 48, 106.
McCoy, B. M., Wu, T. T. (1973), The Two-Dimen- Privman, V. (Ed.) (1990), Finite Size Scaling and Nu-
sional Ising Model. Cambridge, Mass: Harvard merical Simulation. Singapore: World Scientific.
University Press. Privman, V., Hohenberg, P. C., Aharony, A. (1991),
McGuire, T. R., Gambino, R. J., Pickart, S. J., Alpe- in: Phase Transitions and Critical Phenomena,
rin, H. A. (1969), J. Appl. Phys. 40, 1009. Vol. 14: Domb, C., Lebowitz, J. L. (Eds.). London:
Meißner, G., Binder, K. (1975), Phys. Rev. B 12, 3948. Academic Press, Chapter 1.
308 4 Statistical Theories of Phase Transitions

Pynn, T., Skjeltorp, A. (Eds.) (1983), Multicritical Suzuki, M. (Ed.) (1992), Quantum Monte Carlo
Phenomena. New York: Plenum Press. Methods in Condensed Matter Physics. Singapore:
Riste, T., Sherrington, D. (Eds.) (1989), Phase Tran- World Scientific.
sitions in Soft Condensed Matter. New York: Ple- Swann, P. S., Duff, W. R., Fisher, R. M. (1972),
num Press. Metallurg. Trans. 3, 409.
Robbins, M. O., Kremer, K., Grest, G. S. (1988), J. Swendsen, R. H., Krinsky, S. (1979), Phys. Rev. Lett.
Chem. Phys. 88, 3286. 43, 177.
Rohrer, H. (1975), Phys. Rev. Lett. 34, 1638. Tanisaki, S. (1961), Phys. Soc. Japan 16, 579.
Rohrer, H., Gerber, Ch. (1977), Phys. Rev. Lett. 38, 909. Tolédano, J. C. (1981), Ferroelectrics 35, 31.
Roth, W. L. (1958), Phys. Rev. 110, 1333. Tolédano, P., Pascoli, G. (1980), Ferroelectrics 25,
Sadiq, A., Binder, K. (1984), J. Stat. Phys. 35, 517. 427.
Salanon, B., Fabre, F., Lapoujoulade, J., Selke, W. Tolédano, J. C., Tolédano, P. (1987), The Landau
(1988), Phys. Rev. B 38, 7385. Theory of Phase Transitions. Singapore: World
Salje, E. (1990), Phase Transitions in Ferroelastic Scientific.
and Co-elastic Crystals. Cambridge: Cambridge Tsakalakos, T. (1984), Phase Transformations in Sol-
Univ. Press. ids. Amsterdam: North-Holland.
Sanchez, J. M., De Fontaine, D. (1980), Phys. Rev. Turchi, P. E. A., Gonis, A. (Eds.) (1994), Statics and
B 21, 216. Dynamics of Alloy Phase Transformations. New
Sanchez, J. M., De Fontaine, D. (1982), Phys. Rev. York: Plenum Press.
B 25, 1759. van Hove, L. (1954), Phys. Rev. 93, 1374.
Schick, M. (1981), Progr. Surf. Sci. 11, 245. Villain, J. (1959), J. Phys. Chem. Solids 11, 303.
Schmid, F. (1999), in: Computational Methods in Villain, J. (1985), in: Scaling Phenomena in Disor-
Colloid and Interface Science: Borowko, M. (Ed.). dered Systems: Pynn, R., Skjeltorp, A. (Eds.). New
New York: M. Dekker, Chapter 10. York: Plenum Press, p. 423.
Schmidt, V. H. (1978), in: High Pressure and Low Wagner, H., Horner, H. (1974), Adv. Phys. 23, 587.
Temperature Physics: Chu, C. W., Wollam, J. A. Wagner, R., Kampmann, R., Vorhees, P. W. (2001),
(Eds.). New York: Plenum Press, p. 237. in: Phase Transformations in Materials: Kostorz,
Schneider, T., Stoll, E. (1976), Phys. Rev. B 13, 1216. G. (Ed.). Weinheim: Wiley-VCH, Chap. 5.
Schwartz, L. H., Cohen, J. B. (1965), J. Appl. Phys. Wansleben, S., Landau, D. P. (1987), J. Appl. Phys.
36, 598. 61, 3968.
Schweika, W. (1989), in: Alloy Phase Stability: Stocks, Weeks, J. D. (1980), in: Ordering in Strongly Fluctu-
G. M., Gonis, A. (Eds.). Dordrecht: Kluwer, p. 137. ating Condensed Matter Systems: Riste, T. (Ed.).
Schweika, W. (1994), in: Statics and Dynamics of New York: Plenum Press, p. 293.
Alloy Phase Transformations: Turchi, P. E. A., Wiechert, H., Arlt, S. A. (1993), Phys. Rev. Lett. 71,
Gonis, A. (Eds.). New York: Plenum Press, p. 103. 2090.
Schweika, W., Hauboldt, H.-G. (1986), in: Atomic Wilding, N. B. (1997), J. Phys.: Condens. Matter 9,
Transport and Defects in Metals by Neutron Scat- 585.
tering: Janot, C., Petry, T., Richter, D., Springer, T. Wilding, N. B., Schmid, F., Nielaba, P. (1998), Phys.
(Eds.). Berlin: Springer Verlag, p. 22. Rev. E 58, 2201.
Scott, J. F. (1987), in: Phase Transitions and Critical Wilson, K. G., Kogut, J. (1974), Phys. Repts. 12 C, 75.
Phenomena: Domb, C., Lebowitz, J. L. (Eds.). Wu, F. Y. (1982), Rev. Mod. Phys. 54, 235.
London: Academic Press, Chapter 2. Yelon, W. B., Cox, D. E., Kortmann, P. J. (1974),
Selke, W. (1988), Phys. Repts. 170, 213. Phys. Rev. B 9, 4843.
Selke, W. (1989), in: Alloy Phase Stability: Gonis, A., Yeomans, J. (1992), Statistical Mechanics of Phase
Stocks, L. M. (Eds.). Dordrecht: Kluwer Akad. Publ. Transitions. Oxford: Clarendon Press.
Selke, W. (1992), in: Phase Transitions and Critical Young, A. F. (Ed.) (1998), Spin Glasses and Random
Phenomena, Vol. 15: Domb, C., Lebowitz, J. L. Fields. Singapore: World Scientific.
(Eds.). London: Academic Press, p. 1. Zallen, T. (1983), The Physics of Amorphous Solids.
Semenovskaya, S. V. (1974), Phys. Stat. Sol. (b) 64, New York: J. Wiley and Sons.
291. Zarzycki, J. (Ed.) (1991), Glasses and Amorphous
Smart, J. S. (1966), Effective Field Theories of Mag- Materials (Materials Science and Technology,
netism. New York: W. B. Saunders. Vol. 9). Weinheim: VCH.
Stanley, H. E. (1971), Introduction to Phase Transi- Zettlemoyer, A. C. (Ed.) (1969), Nucleation. New
tions and Critical Phenomena. Oxford: Oxford York: Marcel Dekker.
University Press. Zunger, A. (1994), in: Statics and Dynamics of Alloy
Stauffer, D., Aharony, A. (1992), An Introduction to Phase Transformations: Turchi, P. E. A., Gonis, A.
Percolation Theory. London: Taylor and Francis. (Eds.). New York: Plenum Press, p. 361.
Sullivan, D. E., Telo da Gama, M. M. (1985), in:
Fluid Interfacial Phenomena: Croxton, C. A.
(Ed.). New York: J. Wiley and Sons, p. 45.