Chapter 2

Ginzburg-Landau Phenomenology

The divergence of the correlation length in the vicinity of a second-order phase transition
indicates that the properties of the critical point are insensitive to microscopic details of
the system. This redundancy of information motivates the search for a phenomenological
description of critical phenomena which is capable of describing a wide range of model
systems. In this chapter we introduce and investigate such a phenomenology known as the
Ginzburg-Landau theory. Here we will explore the ‘mean-field’ properties of the equilibrium
theory and perturbatively investigate the influence of spatial fluctuations.

2.1 Ginzburg-Landau Theory

Consider the magnetic properties of a metal, say iron, close to its Curie point. The
microscopic origin of magnetism is quantum mechanical, involving such ingredients as
itinerant electrons, their spin, and the exclusion principle. Clearly a microscopic descrip-
tion is complicated, and material dependent. Such a theory would be necessary if we
are to establish which elements are likely to produce ferromagnetism. However, if we
accept that such behaviour exists, a microscopic theory is not necessarily the most useful
way to describe its disappearance as a result of thermal fluctuations. This is because
the (quantum) statistical mechanics of a collection of interacting electrons is excessively
complicated. However, the degrees of freedom which describe the transition are long-
wavelength collective excitations of spins. We can therefore “coarse-grain” the magnet
to a scale much larger than the lattice spacing, and define a magnetisation vector field
m(x), which represents the average of the elemental spins in the vicinity of a point x. It
is important to emphasise that, while x is treated as a continuous variable, the functions
m do not exhibit any variations at distances of the order of the lattice spacing a, i.e.
their Fourier transforms involve wavevectors with magnitude less than some upper cut-off
Λ ∼ 1/a.
In describing other types of phase transitions, the role of m(x) is played by the appro-
priate order parameter. For this reason it is useful to examine a generalised magnetisation
vector field involving n-component magnetic moments existing in a d-dimensional space,
i.e.

x ≡ (x1 , · · · xd ) ∈ Rd (space), m ≡ (m1 , · · · mn ) ∈ Rn (spin).

Phase Transitions and Collective Phenomena

12 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

Some specific problems covered in this framework include:

n = 1: Liquid-gas transitions; binary mixtures; and uniaxial magnets;

n = 2: Superfluidity; superconductivity; and planar magnets;

n = 3: Classical isotropic magnets.

While most applications occur in three-dimensions, there are also important phenomena
on surfaces (d = 2), and in wires (d = 1). (Relativistic field theory is described by a
similar structure, but in d = 4.)
A general coarse-grained Hamiltonian can be constructed on the basis of appropriate
symmetries:

1. Locality: The Hamiltonian should depend on the local magnetisation and short
range interactions expressed through gradient expansions:
!
βH = dx f [m(x), ∇m, · · ·]

2. Rotational Symmetry: Without magnetic field, the Hamiltonian should be isotropic

in space and therefore invariant under rotations, m %→ Rn m.

βH[m] = βH[Rnm].

3. Translational and Rotational Symmetry in x: This last constraint finally leads

to a Hamiltonian of the form
! "t
βH = dx m2 + um4 + · · ·
2
K L N #
+ (∇m)2 + (∇2 m)2 + m2 (∇m)2 + · · · − h · m (2.1)
2 2 2

Eq. (2.1) is known as the Ginzburg-Landau Hamiltonian. It depends on a set of phe-

nomenological parameters t, u, K, etc. which are non-universal functions of microscopic
interactions, as well as external parameters such as temperature, and pressure.1
1
It is essential to appreciate the latter point, which is usually the source of much confusion. The
probability for a particular configuration of the field is given by the Boltzmann weight exp{−βH[m(x)]}.
This does not imply that all terms in the exponent are proportional to (kB T )−1 . Such dependence
holds only for the true microscopic Hamiltonian. The Ginzburg-Landau Hamiltonian is more accurately
regarded as an effective free energy obtained by integrating over the microscopic degrees of freedom
(coarse-graining), while constraining their average to m(x). It is precisely because of the difficultly of
carrying out such a first principles program that we postulate the form of the resulting effective free energy
on the basis of symmetries alone. The price we pay is that the phenomenological parameters have an
unknown functional dependence on the original microscopic parameters, as well as on external constraints
such as temperature (since we have to account for the entropy of the short distance fluctuations lost in
the coarse-graining procedure).

Phase Transitions and Collective Phenomena

2.2. LANDAU MEAN-FIELD THEORY 13

2.2 Landau Mean-Field Theory

The original problem has been simplified considerably by focusing on the coarse-grained
magnetisation field described by the Ginzburg-Landau Hamiltonian. The various ther-
modynamic functions (and their singular behaviour) can now be obtained from the cor-
responding partition function
!
Z[T, h] = Dm(x) e−βH[m,h] (2.2)

Since
$ the degrees of freedom appearing in the Hamiltonian are functions of x, the symbol
Dm(x) refers to the functional integral. As such, it should be regarded as a limit of
discrete integrals, i.e., for a one-dimensional Hamiltonian,
! ! %
N
Dm(x) z[m(x), ∂m, · · ·] ≡ lim dmi z[mi , (mi+1 − mi )/a, · · ·].
a→0,N →∞
i=1

In general, evaluating the functional integral is not straightforward. However, we can

obtain an estimate of Z by applying a saddle-point or mean-field approximation to the
functional integral

Z[T, h] ≡ e−βF [T,h] , βF [T, h] ( minm [βH [m, h]]

where minm [βH[m, h]] represents the minimum of the function with respect to variations
in m. Such an approach is known as Landau mean-field theory. For K > 0, the
minimum free energy occurs for a uniform vector field m(x) ≡ m̄êh , where êh points
along the direction of the magnetic field, and m̄ is obtained by minimizing the Landau
free energy density
βF t
f (m, h) ≡ = m2 + um4 − hm
V 2
In the vicinity of the critical point m̄ is small, and we are justified in keeping only the
lowest powers in the expansion of f (m, h). The behaviour of f (m, h) depends sensitively
on the sign of t (see Fig. 2.1).

1. For t > 0 the quartic term can be ignored, and the minimum occurs for m̄ ( h/t.
The vanishing of the magnetisation as h → 0 signals paramagnetic behaviour, and
the zero-field susceptibility χ ∼ 1/t diverges as t → 0.

2. For t < 0 a quartic term with a positive value of u is required to ensure stability (i.e.
to keep the magnetisation finite). The function f (m, h) now has degenerate minima
at a non-zero value of m̄. At h = 0 there is a spontaneous breaking of rotational
symmetry in spin space indicating ordered or ferromagnetic behaviour. Switching
on an infinitesimal field h leads to a realignment of the magnetisation along the field
direction and breaks the degeneracy of the ground state.

Phase Transitions and Collective Phenomena

14 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

t>0 t<0
Ψ (m) Ψ (m)
h=0
h>0 h<0

m m
h=0
h>0

Figure 2.1: Schematic diagram of the mean-field Landau free energy.

Thus a saddle-point evaluation of the Ginzburg-Landau Hamiltonian suggests para-

magnetic behaviour for t > 0, and ferromagnetic behaviour for t < 0. Without loss of
generality (i.e. by adjusting the scale of the order parameter), we can identify the param-
eter t with the reduced temperature t = (T − Tc )/Tc . More generally, we can map the
phase diagram of the Ginzburg-Landau Hamiltonian to that of a magnet by setting
t(T, · · ·) = (T − Tc )/Tc + O(T − Tc )2 ,
u(T, · · ·) = u0 + u1 (T − Tc ) + O(T − Tc )2 ,
K(T, · · ·) = K0 + K1 (T − Tc ) + O(T − Tc )2 ,
where u0, K0 are unknown positive constants depending on material properties of the
system. With this identification, let us determine some of the thermodynamical properties
implied by the mean-field analysis.
$ Magnetisation: An explicit expression for the average magnetisation m̄ can be
found from the stationary condition
∂f &&
& = 0 = tm̄ + 4um̄3 − h.
∂m m=m̄
In zero magnetic field we find
'
0 t > 0,
m̄ = (
−t/4u t < 0,

which implies a universal exponent of β = 1/2, while the amplitude is material

dependent.
$ Heat Capacity: For h = 0, the free energy density is given by
'
βF && 0 t > 0,
f (m, h = 0) ≡ & = 2
V h=0 −t /16u t < 0.

Thus, by making use of the identities

∂ ln Z ∂ ∂ ∂
E=− , = −kB T 2 ( −kB Tc ,
∂β ∂β ∂T ∂t

Phase Transitions and Collective Phenomena

2.3. GAUSSIAN AND FUNCTIONAL INTEGRATION 15

the singular contribution to the heat capacity is found to be

'
∂E 0, t > 0,
Csing. = =
∂T kB /8u t < 0.

This implies that the specific heat exponents α+ = α− = 0. In this case we observe
only a discontinuity in the singular part of the specific heat. However, notice that
treating, by higher orders, we can in principle obtain non-zero critical exponents.
$ Susceptibility: The magnetic response is characterised by the (longitudinal) sus-
ceptibility
'
∂ m̄ && −1 ∂h && &
2& t t > 0,
χl ≡ & , χl = & = t + 12um̄ & =
∂h h=0 ∂ m̄ h=0 h=0 −2t t < 0,

which, as a measure of the variance of the magnetisation, must be positive. From

this expression, we can deduce the critical exponents γ+ = γ− = 1. Although the
amplitudes are parameter dependent, their ratio χ+ −
l /χl = 2 is also universal.

$ Equation of State: Finally, on the critical isotherm, t = 0, the magnetisation

behaves as
) *1/3
h
m̄ = ∼ h1/δ
4u

giving the exponent δ = 3.

This completes our survey of the critical properties of the Ginzburg-Landau theory in
the Landau mean-field approximation. To cement these ideas one should attempt to find
the mean-field critical exponents associated with a tricritical point (see, for example,
Question 2 of problem set 2.10). To complement these notes it is also useful to refer to
section 4.2 (p 151-154) of Chaikin and Lubensky on Landau theory.
Landau mean-field theory accommodates only the minimum energy configuration. To
test the validity of this approximation scheme, and to determine spatial correlations it
is necessary to take into account configurations of the field m(x) involving spatial fluc-
tuations. However, before doing so, it is first necessary to acquire some familiarity with
the method of Gaussian functional integration, the basic machinery of statistical (and
quantum) field theory.

2.3 Gaussian and Functional Integration

$ Info: Before defining the Gaussian functional integral, it is useful to recall some results
involving integration over discrete variables. We begin with the Gaussian integral involving a
single (real) variable φ,
! ∞ + , + 2,
φ2 √ Gh
Z1 = dφ exp − + hφ = 2πG exp .
−∞ 2G 2

Phase Transitions and Collective Phenomena

16 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

Now derivatives of Z1 on h generate Gaussian integrals involving powers of φ. Thus, if the

integrand represents the probability distribution of a random variable φ, logarithmic derivatives
can be used to generate moments φ. In particular,
∂ ln Z1
*φ+ ≡ = hG.
∂h
Subjecting ln Z1 to a second derivative obtains (exercise)

∂ 2 ln Z1
= *φ2 + − *φ+2 = G.
∂h2
Note that, in general, the second derivative does not simply yield the second moment. In fact
it obtains an object known as the ‘second cummulant’, the physical significance of which will
become clear later. However, in the present case, it is simple to deduce from the expansion,
*φ+ = hG, and *φ2 + = h2 G2 + G.
Higher moments are more conveniently expressed by the cummulant expansion2

∂ r && - .
*φr +c = & ln ekφ
∂kr k=0

Applied to the first two cummulants, one obtains *φ+c = *φ+ = hG, and *φ2 +c = *φ2 + − *φ+2 = G
(as above), while *φr +c = 0 for r > 2. The average *ekφ + is known as the joint characteristic
function.
Gaussian integrals involving N (real) variables
! N
∞ % + ,
1 T −1
ZN = dφi exp − φ G φ + h · φ , (2.3)
−∞ i=1 2

can be reduced to a product of N one-dimensional integrals by diagonalising the (real symmetric)

matrix G−1 . (Convergence of the Gaussian integration is assured only when the eigenvalues
are positive definite.) Denoting the unitary matrices that diagonalise G by U, the matrix
/ −1 = UG−1 U−1 represents the diagonal matrix of eigenvalues. Making use of the identity
G
(i.e. completing the square)
1 T −1 / −1 χ − 1 hT U−1 GUh
/
φ G φ − h · φ = χT G
2 2
/
where χ = Uφ − GUh, and changing integration variables (since the transformation is unitary,
the corresponding Jacobian is unity) we obtain
! N
∞ % + ,
1 T / −1 1 T −1 /
ZN = dχi exp − χ G χ + h U GUh ,
−∞ i=1 2 2
+ ,
1
= det(2πG)1/2 exp hT Gh . (2.4)
2
2
The moments are related to the cummulants by the identity
0%
*φn + = *φnα +c
P α
1
where P represents the sum over all partitions of the product φn into subsets φnα labelled by α.

Phase Transitions and Collective Phenomena

2.3. GAUSSIAN AND FUNCTIONAL INTEGRATION 17

Regarding ZN as the partition function of a set of N Gaussian distributed random variables,

{φi }, the corresponding cummulant expansion is generated by

∂ ∂ && - .
*φi · · · φj +c = ··· & ln ek·φ
∂ki ∂kj k=0

where the joint characteristic function is equal to

- . + ,
1
ek·φ = exp hT Gk + kT Gk (2.5)
2

Applying this result we find that the first two cummulants are given by
0
*φi +c = Gij hj , *φi φj +c = Gij , (2.6)
j

while, as for the case N = 1, cummulants higher than the second vanish. The latter is a unique
property of Gaussian distributions. Applying Eq. (2.5), we can further deduce the important
result that for any linear combination of Gaussian distributed variables A = a · φ,
2 % /2
*eA + = e$A%c +$A c

Now Gaussian functional integrals are a limiting case of the above. Consider the points i
as the sites of a d-dimensional lattice and let the spacing go to zero. In the continuum limit, the
set {φi } translates to a function φ(x), and the matrix G−1 ij is replaced by an operator kernel
−1 &
or propagator G (x, x ). The natural generalisation of Eq. (2.4) is
! + ! ! ! ,
1 & −1 & &
Dφ(x) exp − dx dx φ(x) G (x, x ) φ(x ) + dx h(x)φ(x)
2
+ ! ! ,
1
∝ (det Ĝ)1/2 exp dx dx& h(x) G(x, x& ) h(x& ) ,
2

where the inverse kernel G(x, x& ) satisfies the equation

!
dx& G−1 (x, x& ) G(x& , x&& ) = δd (x − x&& ) (2.7)

The notation Dφ(x) is used to denote the measure of the functional integral. Although the
constant of proportionality, (2π)N left out is formally divergent in the thermodynamic limit,
it does not affect averages that are obtained from derivatives of such integrals. For Gaussian
distributed functions, Eq. (2.6) then generalises to
!
*φ(x)+c = dx G(x, x& ) h(x& ), *φ(x)φ(x& )+c = G(x, x& )

Later, in dealing with small fluctuations in the Ginzburg-Landau Hamiltonian we will fre-
quently encounter the quadratic form,
! ! !
1 2 3 1
βH[φ] = dx (∇φ)2 + ξ −2 φ2 ≡ dx dx& φ(x& )δd (x − x& )(−∇2 + ξ −2 ) φ(x) (2.8)
2 2

Phase Transitions and Collective Phenomena

18 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

which (integrating by parts) implies an operator kernel

G−1 (x, x& ) = Kδd (x − x& )(−∇2 + ξ −2 )

Substituting into Eq. (2.7) and integrating we obtain (−∇2 +ξ −2 )G(x) = δd (x). The propagator
can thus be identified as nothing but the Green function.
In the present case, translational invariance of the propagator suggests the utility of the
Fourier representation3

0 ! L
iq·x 1
φ(x) = φq e , φq = d dx φ(x) e−iq·x
q
L 0

where q = (q1 , · · · qd ), with the Fourier elements taking values qi = 2πm/L, m integer. In
$L !
this representation, making use of the identity 0 dx e−i(q+q )·x = Ld δq,−q! , the quadratic form
above becomes diagonal in q4

10 2
βH[φ] = (q + ξ −2 )|φq |2
2 q

where, since φ(x) is real, φ−q = φ∗q . The corresponding propagator is given by G(q) = (q2 +
ξ −2 )−1 . Thus in real space, the correlation function is given by
0 !
G(x, x& ) ≡ *φ(x)φ(x& )+c = eiq·(x−x ) G(q).
q

Here we have kept L finite and the modes discrete to emphasize the connection between the
discrete Gaussian integrations ZN and the functional integral. Hereafter, we will focus on the
thermodynamic limit L → ∞.
——————————————–
3
Here the system is supposed to be confined to a square box of dimension d and volume Ld . In the
thermodynamic limit L → ∞, the Fourier series becomes the transform
! ∞ ! ∞
dq
φ(x) = φ(q) eiq·x , φ(q) = dx φ(x) e−iq·x
−∞ (2π)d −∞

Similarly,
! ∞ ! ∞
! dq !
dx ei(q+q )·x = (2π)d δ d (q + q# ), d
e−iq·(x+x ) = δ d (x + x# )
−∞ −∞ (2π)

In the formulae above, the arrangements of (2π)d is not occasional. In defining the Fourier transform,
it is wise to declare a convention and stick to it. The convention chosen here is one in which factors of
(2π)d are attached to the q integration, and the δ-function in q.
4
Similarly, in the thermodynamic limit, the Hamiltonian takes the form
! ∞
dq 1 2
βH[φ] = d 2
(q + ξ −2 )|φ(q)|2 .
−∞ (2π)

Phase Transitions and Collective Phenomena

2.4. SYMMETRY BREAKING: GOLDSTONE MODES 19

2.4 Symmetry Breaking: Goldstone Modes

With these important mathematical preliminaries, we return to the consideration of the
influence of spatial fluctuations on the stability of the mean-field analysis. Even for h = 0,
when βH has full rotational symmetry, the ground state of the Ginzburg-Landau Hamil-
tonian is ordered along some given direction for T < Tc — a direction of ‘magnetisation’ is
specified. One can say that the onset of long-range order is accompanied by the spon-
taneous breaking of the rotational symmetry. The presence of a degenerate manifold of
ground states obtained by a global rotation of the order parameter implies the existence
of low energy excitations corresponding to a slowly varying rotations in the spin space.
Such excitations are characteristic of systems with a broken continuous symmetry and
are known as Goldstone modes. In magnetic systems the Goldstone modes are known
as spin-waves, while in solids, they are the vibrational or phonon modes.
The influence of Goldstone modes can be explored by treating fluctuations within the
framework of the Ginzburg-Landau theory. For a fixed magnitude of the n-component
order parameter or, in the spin model, the magnetic moment m = m̄êh , the transverse
fluctuations can be parametrized in terms of a set of n − 1 angles. One-component, or
Ising spins have only a discrete symmetry and possess no low energy excitations. Two-
component, or XY-spins, where the moment lies in a plane, are defined by a single angle
θ, m = m̄(cos θ, sin θ) (c.f. the complex phase of ‘superfluid’ order parameter). In this
case the Ginzburg-Landau free energy functional takes the form
!
K̄
βH[θ(x)] = βH0 + dx (∇θ)2 (2.9)
2

where K̄ = K m̄2 /2.

Although superficially quadratic, the multi-valued nature of the transverse field θ(x)
makes the evaluation of the partition function problematic. However, at low temperatures,
taking the fluctuations of the fields to be small θ(x) . 2π, the functional integral can be
taken as Gaussian. Following on from our discussion of the Gaussian functional integral,
the operator kernel or propagator can be identified simply as the Laplacian operator.
The latter is diagonalised in Fourier space, and the corresponding degrees of freedom are
associated with spin-wave modes.
Then, employing the results of the previous section, we immediately find the average
phase vanishes *θ(x)+ = 0, and the correlation function takes the form

Cd (x − x& )
G(x, x& ) ≡ *θ(x)θ(x& )+ = − , ∇2 Cd (x) = δ d (x)
K̄

where Cd denotes the Coulomb potential for a δ-function$ charge distribution.

4 Exploiting
the symmetry of the field, and employing Gauss’ law, V dx ∇2 Cd (x) = dS · ∇Cd , we
find that Cd depends only on the radial coordinate x, and

dCd 1 x2−d
= d−1 , Cd (x) = + const., (2.10)
dx x Sd (2 − d)Sd

Phase Transitions and Collective Phenomena

20 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

where Sd = 2π d/2 /(d/2 − 1)! denotes the total d-dimensional solid angle.5 Hence
5 6 2 3 |x|>a 2(|x|2−d − a2−d )
[θ(x) − θ(0)]2 = 2 *θ(0)2 + − *θ(x)θ(0)+ = ,
K̄(2 − d)Sd
where the cut-off, a is of the order of the lattice spacing. Note that the case where d = 2,
the combination |x|2−d /(2 − d) must be interpreted as ln |x|.
The long distance behaviour changes dramatically at d = 2. For d > 2, the phase
fluctuations approach some finite constant as |x| → ∞, while they become asymptotically
large for d ≤ 2. Since the phase is bounded by 2π, it implies that long-range order
(predicted by the mean-field theory) is destroyed. This result becomes more apparent by
examining the effect of phase fluctuations on the two-point correlation function,
5 6
*m(x) · m(0)+ = m̄2 Re ei[θ(x)−θ(0)] .
(Since amplitude fluctuations are neglected, we are in fact looking at the transverse
correlation function.) For Gaussian distributed variables we have already seen that
*exp[αθ]+ = exp[α2 *θ2 +/2]. We thus obtain
+ , + ,
2 1 2 2 (x2−d − a2−d )
*m(x) · m(0)+ = m̄ exp − *[θ(x) − θ(0)] + = m̄ exp − ,
2 K̄(2 − d)Sd
implying a power-law decay of correlations in d = 2, and an exponential decay in d < 2,
'
m̄&2 d > 2,
lim *m(x) · m(0)+ =
|x|→∞ 0 d ≤ 2.
The saddle-point approximation to the order parameter, m̄ was obtained by neglecting
fluctuations. The result above demonstrates that the inclusion of phase fluctuations leads
to a reduction in the degree of order in d = 2, and to its complete destruction in d < 2.
This result typifies a more general result known as the Mermin-Wagner Theorem (N.
D. Mermin and H. Wagner, Phys. Rev. Lett. 17, 1133 (1967)). The theorem states that
there is no spontaneous breaking of a continuous symmetry in systems with short-range
interactions in dimensions d ≤ 2. Corollaries to the theorem include:
• The borderline dimensionality of two, known as the lower critical dimension dl
has to be treated carefully. As we shall show in Chapter 5, there is in fact a phase
transition for the two-dimensional XY-model (or superfluid), although there is no
true long-range order.
5
An important consequence of Eq. (2.10) is the existence of an unphysical ultraviolet divergence
of the theory (i.e. x → 0 ←→ q → ∞) in dimensions d ≥ 2. In the present case, this divergence can
be traced to the limited form of the effective free energy which accommodates short-range fluctuations
of arbitrary magnitude. In principle one can account for the divergence by introducing additional terms
in the free energy which control the short-range behaviour more precisely. Alternatively, and in keeping
with the philosophy that lies behind the Ginzburg-Landau theory, we can introduce a short-length scale
cut-off into the theory, a natural candidate being the “lattice spacing” a of the coarse-grained free energy.
Note, however, that were the free energy a microscopic one — i.e. a free field theory — we would be
forced to make sense of the ultraviolet divergence. Indeed finding a renormalisation scheme to control
ultraviolet aspects of the theory is the subject of high energy quantum field theory. In condensed matter
physics our concern is more naturally with the infrared, long-wavelength divergence of the theory which,
in the present case (2.10), appear in dimensions d ≤ 2.

Phase Transitions and Collective Phenomena

2.5. FLUCTUATIONS, CORRELATIONS & SUSCEPTIBILITIES 21

• There are no Goldstone modes when the broken symmetry is discrete (e.g. for n =
1). In such cases long-range order is possible down to the lower critical dimension
of dl = 1.

2.5 Fluctuations, Correlations & Susceptibilities

Our study of Landau mean-field theory showed that the most probable configuration was
spatially (
uniform with m(x) = m̄ê1 , where ê1 is a unit vector (m̄ is zero for t > 0, and
equal to −t/4u for t < 0). The role of small fluctuations around such a configuration
can be examined by setting
n
0
m(x) = [m̄ + φl (x)] ê1 + φt,α (x)êα
α=2

where φl and φt refer respectively to fluctuations longitudinal and transverse to the

axis of order ê1 . The transverse fluctuations can take place along any of the n−1 directions
perpendicular to ê1 .
After substitution into the Ginzburg-Landau Hamiltonian, a quadratic expansion of
the free energy functional with

(∇m)2 = (∇φl )2 + (∇φt )2 ,

m2 = m̄2 + 2m̄φl + φ2l + φ2t ,
m4 = m̄4 + 4m̄3 φl + 6m̄2 φ2l + 2m̄2 φ2t + O(φ3l , φl φ2t ),

generates the perturbative expansion of the Hamiltonian

) * ! + ,
t 2 4 K 2 t + 12um̄2 2
βH = V m̄ + um̄ + dx (∇φl ) + φl
2 2 2
! + ,
K 2 t + 4um̄2 2
+ dx (∇φt ) + φt + O(φ3l , φl φ2t ). (2.11)
2 2
For spatially uniform fluctuations, one can interpret the prefactors of the quadratic
terms in φ as “masses” or “restoring forces” (c.f. the action of a harmonic oscillator).
These effective masses for the fluctuations can be associated with a length scale defined
by
'
K 2 t t > 0,
≡ t + 12um̄ =
2
ξl −2t t < 0,
K 7
t t > 0,
2
≡ t + 4um̄2 = (2.12)
ξt 0 t < 0.
(The physical significance of the length scales ξl and ξt will soon become apparent.)
Note that there is no distinction between longitudinal and transverse components in the
paramagnet phase (t > 0), while below the transition (t < 0), there is no restoring force for
the transverse fluctuations (a consequence of the massless Goldstone degrees of freedom
discussed previously).

Phase Transitions and Collective Phenomena

22 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

S(q) S (q) S (q)

l t
t>0 t<0 t<0

1/ξ q 1/ξl q q

Figure 2.2: Typical scattering amplitude for t > 0 and t < 0.

To explore spatial fluctuations and correlation functions, it is convenient to switch to

the Fourier representation, wherein the Hamiltonian becomes diagonal (c.f. discussion of
Gaussian functional integration). After the change of variables
! ∞
dq iq·x
φ(x) = d
e φ(q),
−∞ (2π)

the quadratic Hamiltonian becomes separable into longitudinal and transverse modes,
!
dq K 28 2 −2
9 2
8 2 −2
9 2
3
βH[φl , φt ] = q + ξ l |φ l (q)| + q + ξ t |φ t (q)| .
(2π)d 2
Thus, each mode behaves as a Gaussian distributed random variable with zero mean,
while the two-point correlation function assumes the form of a Lorentzian,

*φα (q)φβ (q& )+ = δαβ (2π)d δ d (q + q& ) Gα (q), G−1 2 −2

α (q) = K(q + ξα ) (2.13)

where the indices α, β denote longitudinal and transverse components. In fact, this
equation describing correlations of an order parameter in the vicinity of a critical point
was first proposed by Ornstein and Zernike as a means to explain the phenomenon
of critical opalescence in the light scattering from a fluid in the vicinity of a liquid-gas
transition. To understand the mechanism by which ξ sets the characteristic length scale
of fluctuations let us consider the scattering amplitude.
In the case of the ferromagnetic model, the two-point correlation function of magnetisa-
tion can be observed directly using spin-polarised scattering experiments. The scattering
amplitude is related to the Fourier density of scatterers S(q) ∝ *|m(q)|2 + (see Fig. 2.2).
The Lorentzian form predicted above usually provides an excellent fit to scattering line
shapes away from the critical point. Eq. (2.12) indicates that in the ordered phase lon-
gitudinal scattering still gives a Lorentzian form (on top of a δ-function at q = 0 due to
the spontaneous magnetisation), while transverse scattering always grows as 1/q2 . The
same power law decay is also predicted to hold at the critical point, t = 0. In fact, actual
experimental fits yield a power law of the form

1
S(q, T = Tc ) ∝
|q|2−η

Phase Transitions and Collective Phenomena

2.5. FLUCTUATIONS, CORRELATIONS & SUSCEPTIBILITIES 23

1
χ
Gcl(x) κ Sd(2-d) x d-2 -x/ ξ χ t
e

κ x (d-1)/2 ξ(d-3)/2 χ
l

0 ξ x Tc T

Figure 2.3: Decay of the two-point correlation of magnetisation, and the divergence of
the longitudinal and transverse susceptibility in the vicinity of Tc .

with a small positive value of the universal exponent η.

Turning to real space, we find that the average magnetisation is left unaffected by fluc-
tuations, *φα (x)+ ≡ *mα (x) − m̄α + = 0, while the connected part of the two-point correla-
tion function takes the form Gcαβ (x, x& ) ≡ *(mα (x) − m̄α )(mβ (x& ) − m̄β )+ = *φα (x)φβ (x& )+
where
!
& δαβ & dq eiq·x
*φα (x)φβ (x )+ = − Id (x − x , ξα ), Id (x, ξ) = − . (2.14)
K (2π)d q2 + ξ −2
The detailed profile of this equation6 is left as an exercise, but leads to the asymptotics
(see Fig. 2.5)


 |x|2−d
 Cd (x) = |x| . ξ,
(2 − d)Sd
Id (x, ξ) ( 2−d (2.15)

 ξ exp[−|x|/ξ]
 |x| 2 ξ.
(2 − d)Sd |x/ξ|(d−1)/2

From the form of this equation we can interpret the length scale ξ as the correlation
length.
6
This Fourier transform is discussed in Chaikin and Lubensky p 156. However, some clue to under-
standing the form of the transform can be found from the following: Expressed in terms of the modulus
q and d − 1 angles θd , the d-dimensional integration measure takes the form

dq = q d−1 dq sind−2 θd−1 dθd−1 sind−3 θd−2 dθd−2 · · · dθ1 ,

where 0 < θk < π for k > 1, and 0 < θ1 < 2π. Thus, by showing that
! 1/a d/2
1 q dq
Id (x, ξ) = − Jd/2−1 (q|x|),
d/2
(2π) |x| d/2−1
0 q 2 + ξ −2
one can obtain Eq. (2.15) by asymptotic expansion. A second approach is to present the correlator as
! ∞ !
dq iq·x−t(q2 +ξ−2 )
Id (x, ξ) = − dt d
e ,
0 (2π)
integrate over q, and employ a saddle-point approximation.

Phase Transitions and Collective Phenomena

24 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

Using Eq. (2.12) we see that close to the critical point the longitudinal correlation
length behaves as
'
(K/t)1/2 t > 0,
ξl =
(−K/2t)1/2 t < 0.

−ν±
The singularities
√ can be described by ξ± √ ( ξ0 B± |t| , where ν± = 1/2 and the ratio
B+ /B− = 2 are universal, while ξ0 ∝ K is not. The transverse correlation length is
equivalent to ξl for t > 0, while it is infinite for all t < 0. Eq. (2.15) implies that precisely
at Tc , the correlations decay as 1/|x|d−2 . Again, the decay of the exponent is usually
given by 1/|x|d−2−η .
These results imply a longitudinal susceptibility of the form (see Fig. 2.5)

! Gcl (x)
> ?@ A ! ξl
dx
χl ∝ dx *φl (x)φl (0)+ ∝ ∝ ξl2 ( A± t−1
0 |x|d−2

The universal exponents and amplitude ratios are again recovered from this equation. For
T < Tc there is no upper cut-off length for transverse fluctuations, and the divergence of
the transverse susceptibility can be related to the system size L, as

! Gct (x)
> ?@ A ! L
dx
χt ∝ dx *φl (x)φl (0)+ ∝ ∝ L2 (2.16)
0 |x|d−2

2.6 Comparison of Theory and Experiment

The validity of the mean-field approximation is assessed in the table below by comparing
the results with (approximate) exponents for d = 3 from experiment.

Transition type Material α β γ ν

C ∼ |t|−α *m+ ∼ |t|β χ ∼ |t|−γ ξ ∼ |t|−ν

Ferromag. (n = 3) Fe, Ni −0.1 0.34 1.4 0.7

Superfluid (n = 2) He4 0 0.3 1.3 0.7
Liquid-gas (n = 1) CO2 , Xe 0.11 0.32 1.24 0.63
Superconductors 0 1/2 1 1/2

Mean-field 0 1/2 1 1/2

The discrepancy between the mean-field results and experiment signal the failure of
the saddle-point approximation. Indeed, the results suggest a dependence of the critical
exponents on n (and d). Later we will try to explore ways of going beyond the mean-
field approximation. However, before doing so, the experimental results above leave a

Phase Transitions and Collective Phenomena

2.7. FLUCTUATION CORRECTIONS TO THE SADDLE-POINT 25

dilemma. Why do the critical exponents obtained from measurements of the supercon-
ducting transition agree so well with mean field theory? Indeed, why do they differ from
other transitions which apparently belong to the same universality class? To understand
the answer to these questions, it is necessary to examine more carefully the role of fluc-
tuations on the saddle-point.

2.7 Fluctuation Corrections to the Saddle-Point

We are now in a position to determine the corrections to the saddle-point from fluctuations
at quadratic order. To do so, it is necessary to determine the fluctuation contribution
to the free energy. Applying the matrix (or functional) identity ln det G−1 = −tr ln G to
Eq. (2.11) we obtain the following estimate for the free energy density
!
ln Z t 2 4 1 dq
f =− = m̄ + um̄ + ln[K(q2 + ξl−2 )]
V 2 2 (2π)d
!
n−1 dq
+ d
ln[K(q2 + ξt−2 )].
2 (2π)

Inserting the dependence of the correlation lengths on reduced temperature, the singular
component of the heat capacity is given by
 !

 n dq 1
2  0+ t > 0,
∂ f 2 ! (2π) (Kq + t)2
d 2
Csing. ∝ − 2 = 1 dq 1 (2.17)
∂t 

 +2 t < 0.
8u (2π)d (Kq 2 − 2t)2

The behaviour of the integral correction changes dramatically at d = 4. For d > 4 the
integral diverges at large q and is dominated by the upper cut-off Λ ≈ 1/a, while for
d < 4, the integral is convergent in both limits. It can be made dimensionless by rescaling
q by ξ −1 , and is hence proportional to ξ 4−d . Therefore
'
1 a4−d d > 4,
δC ( 2 (2.18)
K ξ 4−d d < 4.

In dimensions d > 4 fluctuation corrections to the heat capacity add a constant term
to the background on each side of the transition. However, the primary form of the
discontinuity in Csing. is unchanged. For d < 4, the divergence of ξ ∝ t−1/2 at the
transition leads to a correction term that dominates the original discontinuity. Indeed,
the correction term suggests an exponent α = (4 − d)/2. But even this is not reliable — a
treatment of higher order corrections will lead to yet more severe divergences. In fact the
divergence of δC implies that the conclusions drawn from the saddle-point approximation
are simply no longer reliable in dimensions d < 4. One says that Ginzburg-Landau models
which belong to this universality class exhibit an upper critical dimension du of four.
Although we reached this conclusion by examining the heat capacity the same conclusion
would have been reached for any physical quantity such as magnetisation, or susceptibility.

Phase Transitions and Collective Phenomena

26 CHAPTER 2. GINZBURG-LANDAU PHENOMENOLOGY

C d>4 C d<4

Tc T Tc T

Figure 2.4: Sketch of the heat capacity in the vicinity of the critical point.

2.8 Ginzburg Criterion

We have thus established the importance of fluctuations as the probable reason for the
failure of the saddle-point approximation to correctly describe observed exponents. How,
therefore, it is possible to account for materials such as superconductors in which the
exponents agree well with mean-field theory?
Eq. (2.18) suggests that fluctuations become important when the correlation length
begins to diverge. Within the √ saddle-point approximation, the correlation length diverges
as ξ ( ξ0 |t|−1/2 , where ξ0 ≈ K represents the microscopic length scale. The importance
of fluctuations can be assessed by comparing the two terms in Eq. (2.17), the saddle-point
discontinuity ∆Csp ∝ 1/u, and the correction, δC. Since K ∝ ξ02 , and δC ∝ ξ0d t(4−d)/2 ,
fluctuations become important when
) *−d ) *
ξ0 (d−4)/2 ∆Csp 1
t 2 =⇒ |t| . tG ≈
a kB [(ξ0 /a)d (∆C sp /kB )]
2/(4−d)

This inequality is known as the Ginzburg Criterion. Naturally, in d < 4 it is satisfied

sufficiently close to the critical point. However, the resolution of the experiment may
not be good enough to get closer than the Ginzburg reduced temperature tG . If so, the
apparent singularities at reduced temperatures t > tG may show saddle-point behaviour.
In principle, ξ0 can be deduced experimentally from scattering line shapes. It has to
approximately equal the size of the units that undergo ordering at the phase transition.
1/3
For the liquid-gas transition, ξ0 can be estimated by vc , where vc is the critical atomic
volume. In superfluids, ξ0 is approximately equal to the thermal wavelength λ(T ). Taking
∆Csp /kB ∼ 1 per particle, and λ ∼ 2−3Å we obtain tG ∼ 10−1 −10−2 , a value accessible in
experiment. However, for a superconductor, the underlying length scale is the separation
of Cooper pairs which, as a result of Coulomb repulsion, typically gives ξ0 ≈ 103 Å. This
implies tG ∼ 10−16 , a degree of resolution inaccessible by experiment.
The Ginzburg criterion allows us to restore some credibility to the mean-field theory.
As we will shortly see, a theoretical estimate of the critical exponents below the upper
critical dimension is typically challenging endeavour. Yet, for many purposes, a good

Phase Transitions and Collective Phenomena

2.9. SUMMARY 27

Fluctuations Destroy
No ordered Mean-Field Behaviour Saddle-Point
Phase But Not Order Exponents Valid

2 4 d
Lower Critical Upper Critical
Dimension Dimension

qualitative understanding of the thermodynamic properties of the experimentally relevant

regions of the phase diagram can be understood from the mean-field theory alone.

2.9 Summary
A summary of our findings for the Ginzburg-Landau Hamiltonian based on mean-field
theory and Gaussian fluctuations is shown in Fig. 2.8.

Phase Transitions and Collective Phenomena