The Kosterlitz-Thouless Transition

Henrik Jeldtoft Jensen

Department of Mathamtics
Imperial College

Keywords: Generalised rigidity, Topological defects, Two Dimensional

XY-model, Superconductivity, Two dimensional Melting, Renormalization

1 Introduction

Matter in the universe is organised in a hierarchical structure. At the bottom we have

elementary particles: quarks, gluons, electrons etc. At present we don’t know what these
particles are made off. But we believe we know that quarks, gluons etc. make up protons,
neutrons, etc. These then go together to make atoms, that are the building blocks of
molecules. Atoms and molecules gives us gas, liquids and solids from which we get stars
and planets which are grouped together in galaxies, that then form clusters and eventually
we arrive at the entire universe. Or from atoms and molecules we get macro molecules like
proteins and DNA, that are the building blocks of organelles, which together form the cells.
From cells we get organs, that put together form organisms: animals and plants of a great
variety of species. The totality of individuals and species constitute the entire ecology.

One branch of science is concerned with the breaking-up of systems into smaller and
smaller parts. The behaviour and properties are studied at each respective level. Statistical
Mechanics is concerned with the opposite quest. Namely, from the interactions between the
components, say atoms, at one given level the aim is to understand the collective coherent
behaviour which emerges as many atoms are but together and the next level if formed.
Often the microscopic details of the properties of the individual building blocks are not so
crucial. Rather it happens that the collective behaviour is controlled by general properties
of the interaction between the building “atoms”.

In these lectures we shall discuss a particular case, where it is possible to follow in

detail, how components at one level go together and form certain collective coherent struc-
tures: topological defects or topological charges. These charges can be Coulomb charges
in two dimensions, dislocations in two dimensional crystals, vortices in two dimensional
superconductors and more. The interaction between the topological charges depends in all
cases logarithmically on the spatial separation and this leads to some very general collective
behaviour, most spectacular it causes a certain type of phase transition: the Kosterlitz-
Thouless transition [1].

1
2 The Two Dimensional XY-Model

We will use the 2d XY-model as our reference model. The model consists of planar rotors of
unit length arranged on a two dimensional square lattice. The Hamilatonian of the system
is given by X X
H = −J Si · Sj = −J cos(θi − θj ). (1)
hi,ji hi,ji

Here hi, ji denotes summation over all nearest neighbour sites in the lattice, and θi denotes
the angle of the rotor on site i with respect to some (arbitrary) polar direction in the two
dimensional vector space containing the rotors.

If we assume that the direction of the rotors varies smoothly from site to site, we can
approximate cos(θi − θj ) by the first two terms 1 − 21 (θi − θj )2 in the Taylor expansion of
cos. The sum over the nearest neighbours corresponds to the discrete Laplace operator,
which we can express in terms of partial derivatives through θi − θj = ∂x θ for two site i
and j which differs by one lattice spacing in the x-direction. This leads to the continuum
Hamiltonian
JZ
H = E0 + dr(∇θ)2 . (2)
2
Here E0 = 2JN is the energy of the completely aligned ground state of N rotors.

The thermodynamics of the system is obtained from the partition function

Z
JZ
Z = e−βE0 dr(∇θ)2 },
D[θ] exp{−β (3)
2
a functional integral over all possible configurations of the director field θ(r). The integral
over θ(r) can be divided into a sum over the local minima θvor of H[θ] plus fluctuations θsw
around the minima
−βE0
XZ 1Z Z
δ2H
Z=e D[θsw ] exp{−β(H[θvor ] + dr1 dr2 θsw (r1 ) θsw (r2 ))}. (4)
θvor 2 δθ(r1 )δθ(r2 )
The field configurations corresponding to local minima of H are solutions to the extremal
condition
δH
= 0 ⇒ ∇2 θ(r) = 0. (5)
δθ(r)
There are two types of solutions to this equation. The first consists of the ground state
θ(r) = constant. The second type of solutions consist of vortices in the director field (see
Fig. 1) and are obtained by imposing the following set of boundary conditions on the cir-
culation integral of θ(r):

1) For all closed curves encircling the position r0 of the centre of the vortex
I
∇θ(r) · dl = 2πn. (6)

2) For all paths that don’t encircle the vortex position r0

I
∇θ(r) · dl = 0. (7)

2
 Condition 1) imposes a singularity in the director field. Note the circulation integral
must be equal to an integer times 2π since we circle a closed path and therefore θ(r) has to
point in the same direction after traversing the path as it did when we started.

We can estimate the energy of a vortex in the following way. The problem is spherical
symmetric, hence the vortex field θvor must be of the form θ(r) = θ(r). The dependence on
r can be found from Eq. 6. We calculate the circulation integral along a circle of radius r
centred at the position r0 of the vortex
I
2πn = ∇θ(r) · dl = 2πr|∇θ|. (8)

We solve and obtain |∇θ(r)| = n/r. Substitute this result into the Hamiltonian Eq. 2
JZ
Evor − E0 = dr[∇θ(r)]2 (9)
2
Jn2 Z 2π Z L 1
= rdr 2 (10)
2 0 a r
L
= πn2 J ln( ). (11)
a
The circulation condition Eq. 6 creates a distortion in the phase field θ(r) that persists
infinitely far from the centre of the vortex. |∇θ| decays only as 1/r leading to a logarithmic
divergence of the energy. Hence we need to take into account that the integral over r in
Eq. 10 is cut-off for large r-values by the finite system size L and for small r-values by the
lattice spacing a. We recall that our continuum Hamiltonian is an approximation to the
lattice Hamiltonian in Eq. 1. A vortex with the factor n in Eq. 6 larger than one is called
multiple charged. We notice that the energy of the vortex is quadratic in the charge. In an
macroscopically large system even the energy of a single charge vortex will be large.

Consider now a pairH of single charged vortex and an anti-vortex. When we encircle
the vortex we pick up dl · ∇θ = 2π and when we encircle the anti-vortex we pick up
dl · ∇θ = −2π. Hence, if we choose a path large enough to enclose both vortices we pick
H

up a circulation of the phase equal to 2π + (−2π) = 0. I.e. the distortion of the phase field
θ(r) from the vortex–anti-vortex pair is able to cancel out at distances from the centre of
the two vortices large compared to the separation R between the vortex and the anti-vortex,
see Fig. 2. This explains why the energy of the vortex pair is of the form
E2vor (R) = 2Ec + E1 ln(R/a). (12)
Where Ec is the energy of the vortex cores and E1 is proportional to J. In detail, the phase
field θ2vor (r) of a vortex located at r = (−a, 0) and an anti-vortex located at r = (a, 0) is
given by
2ay
 
θ2vor (r) = arctg 2 . (13)
a − r2

3 Lack of Ordering in Two Dimensions

In order to highlight the peculiarity of two dimensions we consider the d-dimensional XY-
model. We imagine a d-dimensional cubic lattice. Each lattice site contains a planar rotor

3
or a phase. In the continuum limit the Hamiltonian is still given by Eq. 2 except the
integral over r is now a d-dimensional integral and therefore the factor J is replaced by
Ja2−d . The average size of the projection of the rotors along the x-direction in S space, i.e.
the magnetisation, is

hSx i = hcos θ(r)i (14)

= hcos θ(0)i. (15)

We neglect the singular vortex contributions (which is perfectly safe at low temperature)
and Fourier transform the phase field
dkZ
θ(r) = θ̂(k)e−ik·r (16)
(2π)d
Z
dk
θ(0) = θ̂(k) (17)
(2π)d
Z
2
Z
dk 2 ˆ ˆ
dr(∇θ) = k θ(k)θ(−k). (18)
(2π)d
These eqs. are substituted into the expression

D[θ] cos(θ(0))e−βH
R Z 
−βH+iθ(0)
hSx i = = Re D[θ] cos(θ(0))e /Z . (19)
D[θ]e−βH
R

After some algebra one obtains the following expression

Z π/a !
T
hSx i = exp − Sd dkk d−3 . (20)
2Ja2−d π/L

The behaviour of hSx i is controlled by the integral

Z π/a
I(L) = dkk d−3 . (21)
π/L

The behaviour of I(L) strongly depends on the dimension d. For d < 2 we have I(L) ∼
L2−d → ∞ as → ∞. Hence, hSx i = 0 in the limit of large systems for dimensions less than
2. For d > 2 we have that  d−2
1 π
I(L) → A = (22)
d−2 a
and therefore
Sd
hSx i = exp(− AT ) > 0. (23)
2Ja2−d
Finally for d = 2 the integral I(L) is logarithmically divergent I(L) = ln(L/a) which is
sufficient to force hSx i to zero for any non-zero temperature.

We conclude that there is no ordered low temperature phase for d ≤ 2. For d < 2 this
means that there is no phase transition. The situation is different for d = 2. Although
for any non-zero temperature the rotors are unable to order along a common direction the
vortices are able to induce a phase transition. Though a transition of a special type. One
for which there is no local order parameter, there is no magnetization that goes to zero at

4
a critical temperature. This is the Kosterlitz-Thouless transition which we shall return to
in a little while. First we want to investigate the correlations in the XY-model for different
dimensions. We will again neglect vortex excitations and repeat the calculation we did for
hSx i. This time we calculate the correlation function hS(r)S(0)i and obtain
hS(r)S(0)i = hcos(θ(r)θ(0)i (24)
= Rehexp(i(θ(r) − θ(0))i (25)
= exp[g(r)]. (26)
Where the integral
2−ddk 1 − e−ik·r
Z
g(r) = T ja (27)
(2π)d k2
behaves asymptotically for |r| → ∞ in the following way
  d−2
Sd π


 d−2 L
for d > 2
g(r) ' 1
ln(r/L) for d = 2 (28)

 2π

r/2 for d = 1
From this it follows that the long distance behaviour of the correlation function is
e−const.T



  −η
for d > 2
hS(r)S(0)i ' r (29)
 L
for d = 2
 T

exp(− 2Ja r) for d = 1.
We notice that for d > 2 the correlations survive, hS(r)S(0)i decays to a non-zero constant
as r → ∞. This indicates long range order, a certain degree of alignment of the rotors, i.e.
the model posses an ordered phase at low temperature. In d = 1 the correlation function
decays to zero exponentially over a correlation length ξ = 2T a/T that diverges in the limit
of T → 0.

The situation is very different in two dimensions. Here the correlation function depends
algebraically on r with an exponent η = T /2πJ that continuously changes with temperature.
Algebraic decay of the correlation function is what we expect when the temperature is tuned
to the critical temperature of a continuous phase transition. In the 2d XY-model we find
critical algebraic correlations for all temperatures for which our calculation is valid. In the
calculation we have neglected vortices, so we expect our results to break down when the
temperature becomes high enough to excite vortex pairs.

We conclude, that although in two dimensions there is no long range order with a non-
zero value of hSx i for any temperature above zero, the correlations of the two dimensional
model are algebraic. This is the usual case, say in Ising systems, precisely at the critical
temperature where the correlations are algebraic and the order parameter is still zero, though
it will become non-zero if the temperature is lowered and infinitesimal amount.

4 Vortex Unbinding

We mentioned at the end of the previous section that we expect vortices to become important
as the temperature is increased. To see this we estimate the free energy of a single vortex.

5
The Helmholtz free energy is given by the difference between the energy and the entropy
multiplied by the temperature F = E − T S. The energy is given by Eq. 11. We estimate
the entropy from the number of places where we can position the vortex centre, namely on
each of the L2 plaquette of the square lattice, i.e., S = kB ln(L2 /a2 ). Accordingly the free
energy is given by
F = E0 + (πJ − 2kB T ) ln(L/a). (30)
For T < πJ/2kB the free energy will diverge to plus infinity as L → ∞. At temperatures
T > πJ/2kB the system can lower its free energy by producing vortices: F → −∞ as
L → ∞. This simple heuristic argument points to the fact that the logarithmic dependence
on system size of the energy of the vortex combines with the logarithmic dependence of the
entropy to produce the subtleties of the vortex unbinding transition. Assume a different
dependence of the energy on systems size and one will either have thermal activation of
vortices at all temperatures (in case Evor → const. < ∞) or vortices will not be activated at
any temperature (in case Evor ∼ (L/a)b with b > 0). It is the logarithmic size dependence
of the 2d vortex energy that allows the outcome of the competition between the entropy
and the energy to change qualitatively at a certain finite temperature TKT .

In reality it is not single vortices of the same sign that proliferates at a certain temper-
ature. What happens is that the larger vortex pairs which are bound together for temper-
atures below TKT unbind at TKT . This is a collective effect. The vortex pairs induced as
one approaches TKT disturbs the phase field so much that the effective value of the vortex
binding term E1 in the vortex pair free energy 1 is driven to zero for large vortex separations.
In the next section we shall see in detail how this happens, but preliminary insight can be
obtained from the following

Exercise: Use the expression in Eq. 12 for the energy of a vortex pair to calculate, as
function of temperature, the average separation hRi of a vortex pair.

4.1 The Spin Wave Stiffness

The effect of the thermally activated vortex pairs is describe by the temperature dependent
spin wave stiffness ρRs . This is an example of what Philip W Anderson calls a generalised
rigidity[2]. The spin wave stiffness describes how much free energy it costs to apply a twist,
or gradient, to the rotors (also called spins):

θ(r) = θ0 (r) + vex · r, (31)

here θ0 (r) is allowed to vary according to the canonical ensemble. The increase in the free
energy is given by
1
F (vex ) − F (0) = V ρR 2
s vex . (32)
2
An number of comments concerning the notation are illuminating. The notation vex for the
gradient applied to the phase field θ(r) has its origin in the fact that the same physics, as
we describes here, applies to superfluid films and superconducting films. In these cases the
1
That is Eq. 12 generalised to non-zero temperature

6
field θ(r) is the phase of the complex order parameter, the wave function of the super-fluid.
Being the phase of a quantum mechanical wave function the gradient of θ(r) is related to
a probability current and thereby to the velocity field of the super-fluid. The notation ρRs
is meant to remind one that this phase rigidity, is determined by the density of superfluid
in the case of a superfluid or a superconductor. The superscript R in ρR s indicates that
thermal excitations renormalise the quantity. It follows immediately from the Hamiltonian
in Eq. 2 that at zero temperature ρR s = J = ρs . The spin wave stiffness is similar to the
shear constant of a material. The shear constant determines how the (free) energy increase
of a shear deformation. As temperature is increased the shear constant decreases and drops
abruptly to zero when the solid melts into a liquid.

To obtain ρR
s one calculates the left hand side of Eq. 32. Details can be found in the
wonderful book by Chaikin and Lubensky[3]. The phase field is split into two parts
θ0 (r) = θs (r) + θv (r), (33)
where the first term describes smooth spin waves and the second term contains the singular
vortex contribution. The free energy is obtained from F = kb T ln Z and the partition
function in Eq. 3 by introducing Fourier transforms of the phase field. After quite a bit of
algebra one arrives at the following simple expression
1 ρ2s hn̂(k)n̂(−k)i)
ρR
s = ρs − lim . (34)
2 T k→0 k2
which expresses the renormalized stiffness in terms of the correlation function of the Fourier
transform of the vortex density function
X
n(r) = nα δ(r − rα ), (35)
α

for a collection of vortices of charge nα (see Eq. 6) with centres located at positions rα . The
thermodynamic average is performed over the canonical ensemble with no twist imposed,
hence the subscript 0. Eq. 34 can be used to determine how the spin wave stiffness behave
at large distances as a function of temperature. We will discuss how in the next section.

Exercise: A very enlightening and stimulating activity for a quiet afternoon is to go thor-
ough the details leading to Eq. 34. The simplest way to do this is to study Chaikin and
Lubensky’s book [3], but it is also strongly recommendable to dig out the original papers
by Kosterlitz and Thouless [1, 7].

4.2 The KT transition

Let us first summarise the phenomenology of the Kosterlitz-Thouless transition. As the

temperature is increased more and more vortex pairs are thermally activated. This makes
ρR
s decrease, see Eq. 34. This corresponds to a decrease in the increment of the free energy
induced by a certain twist vex . We can understand this from the fact that the phase field
θ(r) becomes more and more distorted as the temperature is increased, hence the extra
perturbation caused by vex becomes relatively less important. Quantitatively one finds
−
(
ρRs (TKT )[1 + const.(TKT − T )
1/2
] for T < TKT
ρR
s = (36)
0 for T > TKT .

7
Here, TKT is the Kosterlitz-Thouless temperature at which vortex pairs unbind. The value
of TKT differ from one system to another. In the 2d XY-model TKT /J ' 0.893 ± 0.002 [4].
The remarkable thing is, as we shall see below, that the ratio
−
ρR
s (TKT )/TKT = 2/π (37)
+
is universal for all systems that undergoes a KT-transition. Since ρR s (TKT ) = 0 Eq. 37 is
referred to as the universal jump. The correlation length ξ(T ) behaves in a very unusual way
as one approaches TKT from above. We are used to a relatively slow algebraic divergence of
the correlation length as the critical temperature is approached. For the KT-transition the
divergence is, however, much faster
!
const.
ξ(T ) ∼ exp for T > TKT . (38)
(T − TKT )1/2

Can we in a simple way understand this exponential divergence. Yes, we can. The phase
field is significantly distorted by unbound vortices, since these vortices are not screened
by a nearby anti-vortex. I.e. the phases θ(r) can remain correlated over distances shorter
√
than the typical distance hDi = 1/ nub between unbound vortices of density nub [5]. Or in
other words, we expect the correlation length ξ ∼ D. The vortices are thermally induced
and therefor their density is expected to depend on the temperature through a Boltzmann
factor exp(−Evor /T ).2 This argument can indicate the cause of the exponential dependence
of ξ. But it is no more than an indication since the exponential dependence in Eq. 38 is
significantly different from a simple Boltmann factor. This difference is due to corrective
renormalization effects.

Continuous phase transitions are accompanied by divergences in thermodynamic quan-

tities caused by the divergence of the correlation length as the critical temperature Tc is
approached. The singular part of the free energy density f can be estimated as the amount
of thermal energy Tc within a correlated volume ξ d of f ∼ Tc /ξ d . The specific heat cV is
given by the second derivative of the free energy cV = −T ∂ 2 f /∂T 2 ∼ ∂ 2 ξ −d ∂T 2 . For the
KT-transition the exponential divergence of ξ(T ) in Eq. 38 is so rapid and occur over such
a narrow temperature range that the divergence in cV cannot be resolved in simulations or
in experiment. The measured cV is sketched in the Fig. 3 and is smooth through TKT with
a broad peak above TKT induced by the entropy released by the unbounding of the vortices
(see [3]).

4.3 The Green’s function for the vortex-vortex interaction

The behaviour describe in the previous subsection is obtained by a real space renormaliza-
tion procedure first devised by Kosterlitz [7]. A detailed and readable presentation of this
calculation can be found in Chaikin and Lubensky’s book [3]. Here we only briefly mention
the main ingredients of this calculation and leave it as an
2
The situation described here is exactly what happens in the one dimensional so-called φ4 model. This
model supports thermally activated solitons. The correlation length is set by the inverse of the soliton
density and diverges exponentially as the temperature goes to zero [6].

8
Exercise: To go through in details the calculations described in Kosterlitz’s 1974 paper
[7], the 1978 paper by Josè, Kadanoff, Kirkpatrick, and Nelson [8] and spelled out in more
detail in eg. Chaikin and Lubensky’s book [3].

The aim of the calculation is to study the behaviour of ρR

s in Eq. 34. We want to extract
R
the behaviour of ρs in the limit where the applied twist in Eq. 32 varies very slowly. One
can think of this as probing the rigidity between two rotors very far apart. According to Eq.
34 the crucial quantity we need to analyse is the vortex density-density correlation function.
This is done by rewriting the Hamiltonian in a way that allows a more direct interpretation
of how the vortices interact.

The phase field is separated into a spin wave and a vortex part as in Eq. 33. This makes
the Hamiltonian in Eq. 2 split into to terms

H = Hsw + Hvor (39)

where the vortex part can be expressed in terms of the Fourier transform of the vortex
density function given in Eq. 35

1 KZ n̂(k)n̂(−k)
Hvor = dk . (40)
T 2 k2
Here the temperature dependent “coupling constant” K = ρs /T = J/T is introduced in ac-
cordance with standard stat. mech. tradition. It turns out to be advantageous to transform
the Hamiltonian in Eq. 40 into real space. One obtains
1 K Z Z
H = (2π)2 dr1 dr2 n(r1 )G(r1 − r2 )n(r2 ). (41)
T 2
Where the Green’s function mediating the interaction between two vortices is obtained from
Z
dk eik·r
G(r) = (42)
(2π)2 k 2
Z π/a
1 Z 2π
= dk cos[kr cos(φ)]. (43)
π/L k 0
The integral needs to be regularized by the introduction of upper and lower cut-offs. The
integral over φ introduces the zero order Bessel function

1 − 12 z 2 + · · · for |z|  1
(
J0 (z) ' q
2 (44)
πz
(cos(z − π/4) + · · ·) for |z|  1

The substitution u = rk leads to

πr
1 Z a J0 (u)
G(r) = du . (45)
2π πr
L
u

From the asymptotic behaviour listed in Eq. 44 we note that

1 − J0 (u)
∼u (46)
u

9
can be integrated in the limit u → 0 and that
J0 (u) 1
∼ 3/2 (47)
u u
can be integrated in the limit u → ∞. This suggest that the integral over u should be split
up in the following way
πr
( )
1 Z a 1 1 − Jo (u)
G(r) = du − (48)
2π πr L
u u
πr
1 − J0 (u) Z a 1 − J0 (u)
  Z 1
1 L
= [ln − πr − (49)
2π a L
u 1 u
1 L r
   
= [ln − ln + const. (50)
2π a a
We arrive at a logarithmic dependence of the vortex-vortex interaction as anticipated by
the simple argument leading to Eq. 12. Next step is to substitute this expression for G(r)
into Eq. 41. In the limit L → ∞ charge neutrality α nα = 0 is needed to obtain a finite
P

energy. We also need to include a core energy Ec associated with each vortex. When we
put this together and return to the original lattice notation we obtain
!
1 |r1 − r2 | Ec X
n(r)2 .
X
Hvor = −πK n(r1 ) ln n(r2 ) + (51)
T r1 6=r2 a T r

The leading contributions to the thermodynamic averages come from configurations for
which
n(r) ∈ {−1, 0, 1}. (52)
Multiple charges leads to much higher energies. The tick is now to use the Hamiltonian in
Eq. 51 to perform the average of hn̂(k)n̂(−k)i0 over the canonical ensemble. The result of
this calculation is substituted into Eq. 34 for the renormalized spin wave stiffness and leads
to
1 1 Z L
dr r 3−2πK
 
= + 2π 3 y 2 . (53)
KR K a a a
Where KR = ρR s /T and y = exp(−βEc ) is called the vortex fugacity. Eq. 53 is derived under
the assumption of low temperature, i.e. Eq. 52, or low density of vortices. Those people
who did the first Exercise about the average size hRi of a vortex pair will recognize that the
exponent 3 − 2πK may lead to interesting behaviour. At low temperature K = ρs /T is large
and the integral in Eq. 53 will converge as L → ∞. The renormalized spin wave stiffness
is accordingly corrected by a small amount of order y 2 . When the temperature is increased
K will become larger than 2/π and the large r limit will diverge. The renormalization
procedure is able to extract the behaviour of KR even when the integral in Eq. 53 diverges.

Our next task is to use Eq. 53 to establish a set of renormalization group equations for
KR and y from which we can extract the long distance behaviour.

4.4 The RG equations for the KT transition

The standard procedure of the renormalization group is to integrate over short wave length
variations. One can think of this as if one is looking at how the system behaves at ever

10
larger length scales. In the Wilson RG scheme one works directly with the Hamiltonian
in Fourier space and “filter” out the modes corresponding to large k-vectors or short wave
lengths. In the Kadanoff-Migdal real space approach one decimates the lattice by elimi-
nating successively every other spin. In Kosterlitz’s approach one study how the spin wave
stiffness changes as one gradually increases the lower cut-off a → edl a. Increasing a does
correspond to eliminating the short distance fluctuations. For instance, since a is the min-
imum separation of two vortices, increasing a excludes configurations of small separation.
One way to see the effect of increasing a is to break the integral in Eq. 53 up [8] in the
following way
Z L Z ∞ Z aedl Z ∞
7→ = + . (54)
a a a aedl

The integral in Eq. 53 is broken up in this way and a new renormalized stiffness3 K̃is
introduced according to
Z aedl  3−2πK
1 1 dr r
= + 2π 3 y 2 . (55)
K̃ K a a a
This leads to the equation
Z ∞  3−2πK
1 1 dr r
= + 2π 3 . (56)
KR K̃ aedl a a
Next step is to write this equation in a form that is invariant with respect to the change
a 7→ aedl . This is done by introducing a renormalized fugacity

ỹ = ye2−πK . (57)

We are interested in the changes of K̃ and ỹ induced by an infinitesimal change in a, i.e. for
small dl. Expanding in dl one arrives at the Kosterlitz renormalization equations (correct
to order y 4 and y 3 respectively)

dK̃ −1
= 2π 3 ỹ 2 (58)
dl
dỹ
= (2 − π K̃)ỹ. (59)
dl
This set of equations are integrated to obtain the trajectories as function of length scale l
in the phase plane (K̃ −1 , y). The trajectories are sketched in the figure on the next page.

3
This time renormalized due to change of the length scale, not as before when we introduced ρR
s where
the renormalization was relative to the zero temperature value ρs .

11
Phase Plane Trajectories for the Kosterlitz
RG Flow
• Note: y = 0 is a line of fixed points

• Note: y(l) is a decreasing function of l for K̃ −1 (l) < π/2 and increasing with l for
K̃ −1 (l) > π/2.

• Note: The straight line indicates the set of initial conditions for different temperatures.
Increasing the temperature corresponds to moving up to the right along the line.

• Note: Most interesting is what happens in the limit l → ∞. For initial conditions
below the separatrix liml→∞ (K̃ −1 , ỹ) = (K̃ −1 (∞), 0) where K̃ −1 (∞) < ∞. For initial
conditions above the separatrix liml→∞ (K̃ −1 , ỹ) = (∞, ∞)
−
• Note: The separatrix corresponds to T = TKT , and liml→∞ K̃ −1 (TKT ) = π/2. Whereas
+
liml→∞ K̃ −1 (TKT ) = 0, this is the universal jump mentioned in Eq. 37. We see
from Eq. 29 that the universal jump determines the correlation function exponent at
T = TKT to be η(TKT ) = 1/4.

12
5 The Vortex Unbinding Transition in Other Systems

We have a number of times alluded to the fact that not only the XY-model exhibit the
Kosterlitz-Thouless vortex unbinding transition. Any two dimensional system that sup-
ports thermally induced “charges” or topological defects that interact logarithmically will
undergo this transition. The U (1) symmetry of the phase field θ(r) of the XY-model is
also present in the Ginzburg-Landau free energy of superfluids and of superconductors. The
topological excitations in the case of a superfluid consist of vortices in the flow of the su-
perfluid. Vortices like the those observed when one empty a bath tub. For thin superfluid
helium film experiments find that the destruction of the superfluid phase with increasing
temperature occurs according to the scenario of the KT-transition.

The situation is slightly more complicated in superconductors. Because the superfluid

in this case is charged (the superconducting pairs of electrons) screening effects play a
role. However, for thin superconducting films of thickness δ the effective screening length is
given by λef f = λ2 /δ which can easily become a macroscopic length. In this case the loss of
superconductivity is caused by the unbinding of vortex pairs according to the KT-transition.
The broken pairs can move freely when they respond to an applied electric current. As they
move they cause phase slips in the superconducting order parameter. These phase slips
induce a voltage drop according to the Josephson relation. The superconductor is now
unable to support an electric current without a voltage drop, i.e. it is not a superconductor
any longer.

Dislocations in two dimensional crystals interact through the strain field. Two edge
dislocations of opposite sign correspond to an extra row of atoms inserted along the line
connecting the location of the two dislocation cores. The extra line of atoms produces strain
and leads to an increase in the energy which is logarithmic in the separation between the
two dislocations. Thus the situation is very similar to the one encountered in the XY-model.
When the dislocations unbind free dislocations are produced. A shear applied to the sys-
tem can now be accommodated by the mobile dislocations without an increase in the (free)
energy. I.e., the shear constant has dropped to zero and the system is melted. The 2d
melting theory of Kosterlitz-Thouless-Halperin-Nelson-Young predicts that melting occur
in two stages. At the first stages dislocations unbind and make the shear constant drop to
zero. The dislocations are topological defects, their effect on the order of the lattice are,
however, not very dramatic. Before the unbinding of dislocations the translational and the
orientational order of the lattice are both described by correlation functions that depends al-
gebraically on distance. When the dislocations unbind the translational correlation function
becomes exponential but the orientational correlations remain algebraic. At a somewhat
higher temperature topological defects called disclinations unbind with the effect that the
orientational order becomes exponential. Details can be found in Chaikin and Lubensky [3].

There are many other cases where the logarithmic vortex interaction and the KT-
transition plays a role. For instance, the shape of surfaces in three dimensions may undergo
a transition from smooth to rough. This transition can be described in the limit of a discrete
surface constructed by columns of different height4 as a KT-transition.
4
This is called the solid-on-solid model

13
6 Conclusion

We have describe a specific example where one can follow in great detail the step from one
level of structure to the next in the hierarchical order of matter. Topological defects arise as
coherent structures of the “atoms” of one level and can be considered as (composite) particles
at the next level. Their interaction can be derived from the behaviour of the constituent
“atoms”. Many different systems may support composite particles that interact in the same
way. In the case we have discussed the specific physical properties and the subtleties of the
Kosterlitz-Thouless transition are caused by the logarithmic interact between the topological
defects. The one most important fact in determining the KT-transition is the fact that
both energy and entropy depends logarithmically on length scale for the two dimensional
topological charges.

References
[1] J.M. Kosterlitz and D.J. Thouless, Metastability and Phase Transitions in Two-
Dimensional Systems, J. Phys. C 6, 1181 (1973).

[2] P.W. Anderson, Basic Notions of Condensed Matter Physics, The Benjamin/Cummings
Publishing Company, 1984.

[3] P.M. Chaikin and T.C. Lubensky, Principles of condensed matter physics, Cambridge
University Press, 1995.

[4] P. Olsson and P. Minnhagen, On the helicity modulus, the critical temperature and
Monte Carlo simulations for the two-dimensional XY-model. Phys. Scr. 43, 203 (1992).

[5] H. J. Jensen and H. Weber,A Phenomenological Study of Vortices in a Two Dimensional

XY-Model in a Magnetic Field. Phys. Rev. B. 45, 10468(1992).

[6] H. J. Jensen and H. C. Fogedby, Phonon-Kink Interference in the φ4 Model. Physica

Scripta, 31, 210 (1985).

[7] J.M. Kosterlitz, J. Phys. C 7, 1046 (1974).

[8] J. Josè,, L.P. Kadanoff, S. Kirkpatrick, and D.R. Nelson, Phys. Rev. B 16, 1217 (1977)
[with an erra. in Phys. Rev. B 17, 1477 (1978)].

14