You are on page 1of 14

PHYSICAL REVIEW B

VOLUME 35, NUMBER 16

Numerical

simulation

JUNE 1987

of dynamics in the XY model

Richard Loft and Thomas A. Deorand


Physics Department,

of Colorado, Boulder, Colorado 80309


(Received 15 October 1986)
University

We study the dynamics of the classical two-dimensional XY model via numerical simulation using
both microcanonical and Langevin equations of motion. We investigate three specific phenomena in
this regard: the annealing of vortex defects during a quench from high to low temperature, the
Brownian motion of an isolated vortex due to spin-wave interactions, and the temperature-dependent
behavior of real-time spin-spin correlation functions. We compare the results of our simulations with
simple theoretical models, whenever applicable.

I.

INTRODUCTION

This paper investigates the real-time behavior of the


classical two-dimensional XY model at finite temperature.
The model consists of a two-dimensional rectangular lathich interact through a potice of variables P; E [ tr, 7r] w
tential
V

= g g cos(P; P;+) .

Besides being interesting in its own right as a model for


films or for smectic-C liquid crystals, the
liquid-helium
XY model serves as a paradigm for more complicated systems in condensed matter and particle physics. This is
particularly the case with respect to the topological defects
of the XY model, the vortices, and the role they play in
the critical behavior of the model.
Our goal is to explore the real-time behavior of the
model via numerical simulation, via the Langevin and
While the dynamics of
molecular dynamics techniques.
Monte Carlo (MC) updating have been shown to coincide
with the results of first-order Langevin dynamics through
an appropriate choice of scale ~Mc with units of sec per
MC sweep, the relationship of Monte Carlo "time" to real
time remains implicit. ' Thus the direct connection of the
Langevin approach to physical time, through a dissipative
drag coef5cient, makes it preferable for our purposes.
There is extensive literature discussing dynamical phenomena in condensed-matter physics (for a good overview
of the theory, see Ref. 2), but most of the numerical studies involve systems with discrete variables (for a summary
of the numerical work, see Ref. 3). The updating algorithms used are typically Glauber, Kawasaki, or Creutz
dynamics, which are effectively cellular automatic rules,
for which the connection to physical time is somewhat obscure. In the case of the XY model, however, the dynamiThus it is possible to use
cal variables are continuous.
differential equations for which the time has an obvious
We choose to simulate the real-time
physical meaning.
behavior of the system using both deterministic (microcanonical) and stochastic (Langevin) diff'erential equations.
This calculational method is most similar to the molecular
dynamics techniques of Schneider and Stoll, which they
have employed successfully on a number of condensed35

matter systems.
Recently, Langevin techniques have become popular in
particle physics for numerical simulation of lattice gauge
These studies focus only on computing static
theory.
quantities; the Langevin time is an extra, artificial fifth dimension through which the system evolves. Our goals are
We are studying classical field theory
quite different.
rather than quantum field theory, and there is a direct relationship between our Langevin time and clock time.
Furthermore, since the equations of motion are classical,
Langevin techniques may be applied straightforwardly.
This is a great advantage over the more complicated and
which
less physical technique of stochastic quantization,
encounters convergence problems due to the introduction
of a complex action. Importantly, there are interesting
matter systems for which such classical
condensed
Langevin dynamics are expected to be applicable. In this
paper we have studied three different problems: the evolution of vorticity through a quench, the dynamics of an
isolated vortex, and the measurement of real-time correlation functions.
The annealing of defects and growth of domains in
quenched condensed matter systems have been extensively
studied in, for example, the Ising and Potts' models.
We have measured the metastable number of vortices as a
function of time for a quenched lattice of planar spins
Unlike discrete state rnodobeying Langevin dynamics.
els, which possess defects (domain walls) only in the lowtemperature magnetically ordered phase, the XY model
possesses defects (vortices) in the high-temperature
phase.
This permits us to track the annealing process through the
phase transition. We have studied therefore, three basic
types of quenches: above T, to above Tabove T, to T
and above T, to below T, .
For the first type of experiment the system was
quenched from T=5.0 to 1.2 (the transition occurs at
about T, =0.89). In this instance we have observed that
initially the number of vortices declines according to an
exponential law. This proceeds until a critical density of
defects is reached, at which point the rate crosses over to
t '. We explain these observations in
a power law
terms of a simple kinetic model. At very high temperatures, the system is saturated with vortex-antivortex pairs.
Since the pairs behave as "particles" that decay at a rate
8528

1987

The American Physical Society

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

35

proportional to the number of pairs, an exponential annihilation results. Eventually the system begins to be
dominated by isolated vortices. The slowest step in the
annealing kinetics is the Brownian diffusion of vortices
into antivortices. We show that this process is governed
by the simple kinetic equation p, =
Kp, where p, is the
vortex density. This equation leads naturally to t ' behavior.
The second and third experiment, in which the model
was quenched to T, or below, showed interesting hysteresis behavior during the phase transition, and t ' behavior otherwise. Apart from the hysteresis effects, which
we have shown to correspond to states of energy near the
critical energy, these studies are in accord with the kinetic
model previously described.
In order to study the physical properties of an isolated
vortex, we have created a cylindrically symmetric "dish"
with an initial net vorticity: thus the system will always
We have
support one vortex, even at low temperatures.
verified that the motion of the vortex is truly Brownian
and that (v ), a vortex mass m, , and viscosity time constant ~, may be measured. Finally we discuss at least one
experimental realization of this situation, that is, in thin
films of smectic liquid crystals. '
In the last section of the work, we investigate the static
and the dynamical correlation functions for the planar
spin model. The outline of the paper is as follows. In
Sec. II we describe the details of the computation. In Sec.
III we study quenches. Section IV is a description of the
dynamics of a single vortex, and in Sec. V we discuss
of correlation functions. Our conclusions
measurements
and future plans are found in Sec. VI.

"

8529

(T(t).T(s) ) =2II kT5(t s)

(2.5)

We define the time constant I to be b /m. If L; is the angular momentum of the rotor at site i, we can now write
the equation of motion as

L; = I L;
g

gP sin(P, P;+)+T (t) .

We introduce the dimensionless


t =yt,

time t,

=+g iI,

(2.7)

and define the dimensionless


less temperature T:

coupling

T =kT/g,

(2.9}

and the scaled noise g(t), such that

(q(t)g(s)) =2I T5(t

s)

(2. 10)

Then (2.6) may be written as

rt=0
P;+ I P;+ g sin(P; P;+)+

(2. 11)

We take (2. 10) and (2. 11) to be the semiphenomenological


Langevin equations which describe the dynamics of the
planar spin system.
The long time behavior of Eq. (2. 11) is given by a
Boltzmann distribution:
the probability that the spins at
sites i lie between P; and P; + d P; is

&(y)/dy;= /de;exp
I

II.

=icos(y; y; )
1

DERIVATION OF LANGEVIN EQUATION


OF MOTION FOR THE PLANAR SPIN MODEL

(2. 12)

Let us imagine, as a simplest case, a rigid dumbbell rotor with moment of inertia I =mr 0, where ro is the rotor
radius. The Langevin force arises from collisions with
particles of an imaginary fluid in which the rotor is immersed. This fluid can represent degrees of freedom for
the rotor which have been integrated out or are otherwise
neglected. If T(t) is the instantaneous torque due to all of
these random collisions, we may write the torque-torque
correlation function

(2. 1)

Since the rotor is constrained to rotate only in a twodimensional plane, we may rewrite (2. 1) as

(T(t) T(s)) =2ro(si. ng(t)sinO(s)F(t)F(s)) .

and dimension-

(2.8)

(T(t) T(s)) =2([ro(t)XF(t)].[ro(s)XF(s)]) .

(2.6)

(2.2)

Most Langevin simulations of which we are aware are devoted entirely to computing the static properties of statistical systems. We on the other hand are interested in real
time properties of the system as given by Eq. (2. 11). Now
for smectic-C liquid crystals, which appear to be modeled
well by the two-dimensional
classical XY model, the moment of inertia of the molecules is small and the drag
term dominates the equation of motion. The films are observed to obey dissipative dynamics. For such systems we
must consider the first-order Langevin equation,

bP;+g

g sin(P; P;+)+g=0,

(2.13)

in our studies, where b is the viscosity: of course (2.13) is


just the large-viscosity limit of Eq. (2. 11). Now it is more
convenient to scale time according to

For a Markovian force this is approximately

(T(t) T(s)) =2ro(sin 8)(F(t)F(s)) .

All values of 0 are equally probable so (sin 8) = ,', and


(F(t)F(s) ) is given by the fluctuation-dissipation theorem

(F(t)F(s) ) =2I mkT5(t s),

(2. 14}

(2.3)

(2.4)

"medium.
where b is the viscosity of the surrounding
Thus the torque-torque noise correlation function is

"

This yields the dimensionless

P; sin(P;

gP

equation

P;+)+q=0-

and the fluctuation-dissipation

(g(t)g(s)) =2T5(t s) .

(2. 15)

theorem

(2. 16)

RICHARD LOFT AND THOMAS A. DeGRAND

8530

We use the standard Eulerian method to integrate the


Langevin equations. ' For example to integrate Eq. (2. 11)
we introduce the scaled angular momentum L;(t), defined
by

L;(F) =

L, (t)

(2. 17)

yI

where y is given by Eq. (2.7). For a time step E, the


difference equations are

6;(t+@)=P; (t)+eL; (t)

(2. 18)

and

L;(t+e)=L;(t)

eg

El

L;(t
)

sin[/;(t)

P;+(t)]+jw;(t),

(2. 19)

g=(24T

e)'

crocanonical updating was performed using a standard


fourth-order Runge-Kutta algorithm with a time step
@=0.02. The update time for one spin using either the
Langevin or microcanonical equation of motion was about
0.4 msec.
Much of this paper is concerned with measurements of
vortices. We use the Chester-Tobochnik'
definition of
vorticity, and we now remind the reader of it. Vortices
live on a simple one by one square or plaquette. To determine the vorticity of a square, we compute the successive
differences in angles around the square and define those

differences as P;
n;+0, where 0 is restricted to
P;+~. 2m
the range ( rr, n
The. vorticity of a square is equal to
)
the sum of the n s around the square, and may be either
positive or negative. For periodic boundary conditions
the number of vortices equals the number of antivortices.
For an extensive study of equilibrium configurations of
vortices, see Ref. 14.

III.

where g is defined to be
1

35

(2.20)

The discrete time analog of the fluctuation-dissipation


theorem for il;(t ) = gw; (t ) is
(2.21)
choosing w;(t ) to be a uniform random number over the
0.5 to 0.5. Since such random numbers are
range
easier to generate than Gaussian random numbers, we obtain a considerable improvement in algorithm speed. We
have confirmed that this choice of noise still reproduces
for long times the equilibrium solution of the FokkerPlanck equation, Eq. (2.12). In choosing w;(t) to be uniformly distributed rather than Gaussian we are following
the work of Creutz, who showed that for the Ising model
spin dynamics were sufficiently complicated that the system would still come into equilibrium
when very low
quality random numbers, or even none at all, were used.
Further, we have integrated the Brownian motion of a
classical harmonic oscillator using both distributions of
random numbers, and found the results to be consistent
with each other.
A11 the simulations reported here were carried out on a
VAX-11-780. Our lattices (except for the studies of Sec.
IV) were either 39 )& 40 using skew-periodic boundary
conditions or 40)&40 using periodic boundary conditions.
We took @=0.005. For this value of e, we are not sensitive to the O(e) terms by which the equilibrium distribution of P's, first noticed by Batrouni et al. , differ from
(2. 12).
Some of our simulations were carried out using microcanonical updating. In this case, we take a configuration
of spins prepared by Langevin undating, set both the
damping term I P and the noise to zero, and integrate the
partial differential equation. The temperature of the system may be determined by monitoring the mean kinetic
energy E~ of the system, which in equilibrium must have
the equipartition value ,'kT per degree of freedom. Mi-

QUKNCHES

We begin our investigation of real-time behavior in the


XY model by studying the approach of the system to equilibrium. We prepare a system in equilibrium at high temperature T; and then at a time t=0 abruptly decrease the
temperature [magnitude of the noise term in Eq. (2. 11)] to
some final temperature TF. We hold the damping parameter I constant during the quench. The system cools and
eventually comes into equilibrium at TF. For this first
study we consider the three cases TF & TTF=Tand

TF)T

'

There is an extensive literature on the dynamics of ordering in far-from-equilibrium


This work is
systems.
somewhat different from the simulations we have performed. The models studied are typically discrete spin
models, i.e. , Z~ clock models or Q-state Potts models,
and the quenches are from infinite temperature to zero
temperature (or to a final temperature at least within the
lowest temperature magnetically ordered phase of the system). The approach to equilibrium is characterized by the
coalescence of magnetic domains: the average domain
size is pararnetrized as

R'(t)-t"

(3. 1)

The updating method is normally Glauber dynamics.


The XY model is sufficiently different from the discrete
models that we have chosen to parametrize its approach
to equilibrium differently. The XY model has no domains
and hence the characterization of the approach to equilibrium using Eq. (3.1) cannot be made in the usual fashion.
(We will comment on a simple relationship at the end of
this section. ) Instead, we choose to track the density of
vortices and antivortices with time during a quench, and
to describe the approach to equilibrium of the XY model
via a simple Boltzmann equation for the vortices.
We begin first with a quench from T=5.0 to T=0.8
(much less than T, ). In Fig. 1 we show a set of typical
snapshots of the distribution of vortices (circles) and antivortices (crosses) at various times into the quench.
Qualitatively speaking, the t= 0 equilibrium configuration
and the configurations of the first few time units after the

35

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

quench appear to be dominated by clumps of vortices and


antivortices. Many of these annihilate without moving.
Later on, isolated vortices and antivortices are left behind;
they slowly drift toward one another and annihilate. Oc-

Q +
0
+
Q . +
+
0 . 0
+ + . 0
0

00+

0 0

+
0 .

+
+

0 + .

+
+

. 0

+
G

. +

.
Q

0+

0+

. + . + 0
+ . . 0
0 + . 0

0.
Q

0
+ . 0

+
+ 0 . 0 +
+

0
+

0 +

. .
.

. .

0.+ . .
0. 0+0
. .
0+.

0 +

0+

+ p

0 0
0

0
+

+.
Q

+ 0 +
+ .

. .

0
+

+
. 0

.
.

. +
. . 0

+ 0
+ p

+
,

0.

0
0

p +

+
+

. .

.
,

0 . 0

+
Q

. +
0

+00+

Q,

+
+

0 +

0..
+

+
Q

. . 0
. + . +

0+

+0.
. +0
0.+0.
. . . 0
0
.

0
+

0
Q

0 .
. .
. . Q
+ . .

+
+

. 0
+

+ 0

. 0

0 . 0

0+ 0+

. .

. +

+
+ + p
+

Q.
+

. .

. .

+ +

0 +
0 + . 0

. 0

0 +

0 +

. .
.

+
+

Q.+

. .

0
+

. +

+
0

+ 0
+

+.

+
+

+ . +
0 + 0

+
p

. +

0 +

. 0

. +

+
p +

+ + 0

.
. .
. 0

p.+

0
Q

. +

0 +

0
0 +

0 . 0 +

+.0

.
. .

0 + . 0
+ . . Q +
0

. +

0 . 0

+
0
+ p +
Q

0
0

0 +

0 +

0. 0+

+
+ p

+ +
00+
. +

+ 0 + 0

0 0
0

p +
+ Q

P +
+

0+.
0.+ 0
.
+

+ Q

+ 0

.
. .
. . Q
+
. +

.
+ +

thermal fluctuations produce vortex-antivortex


of course, the annihilation rate
equilibrium,
production rate.
write down a simple kinetic equation for the

00+
+

0 0

0
+

+
Q

+
+

0 +
+

0
+

. .

. .
+,
0 0

0
+

+. 0.

+ +

0+0

0 +

0 .

0
'I

. +

0 +

. 0

+ 0 +

0 +

. . + + 0 0
Q. . . . +
. . 0 .
. 0 0
+ .
+

p + 0
+
+

0
Q

+
Q

0 +

. 0

0
Q

casionally
pairs; at
equals the
We can

8531

Q
Q

0
+

0 .
+

+0. 0

0 +

(b)

Q,

. +

. +

0
+

0
p

0
+

0 + 0

0
+

. .

(c)

0
+

0
0

+ )1

(e)

FICs. 1. Snapshots of a quench of the planar spin model from T=5.0 to 0. 5. o, vortices
1 vorticity) and
+ antivortices ( 1
vorticity). In equilibrium at T=5.0, (a) shows large patches of vortex-antivortex pairs covering the system. Pairs rapidly annihilate
immediately after the quench (b). (c) (f) show defects (underlined) attracting each other at time t=5.5, and ultimately annihilating in
the final frame at time t=8.0. At very late times, only isolated defects remain.

(+

RICHARD LOFT AND THOMAS A. DeGRAND

8532

35

late stage of vortex-antivortex annihilation.


If we assume
a uniform distribution of vortices in position space, a
short-range annihilation cross section, and assume that
annihilation
all other scattering,
then the
dominates
Boltzmann equation for the excess vortex or antivortex
density is

af, (p, t) =
d 'p, dAo-( r), )uif, (p, t)f, (p t) ,-

o. is the vortex-antivortex

where
tion,

Vv

~ rel

(3.2)

annihilation

cross sec-

Vv

and 0, is the scattering angle. Defining the number density of vortices (equal to the number of antivortices) as

f, (t) =p, (t)F(p, t),

In

FIG. 2. Log-log plot of excess vorticity vs time for the average of 20 quenches through T, : T=5.0 initially and is suddenly
reduced to T=0.8 at t=0. The relaxation is dominated at later
times by t ' behavior which is explained by a simple kinetic
model of defect annihilation.
The kinklike deviation near
t=12.0 is probably the result of critical slowing down.

with

I' p, td

p=l

for all t, (3.2) is

' =
(ou)p,',

at

(3.3)

where

(o.u)

f d'p , drtF(pt)-F(pt)oui. (3.

4)

of o. v average over the vortex and antivortex distributions.


Since vortex-antivortex annihilation into spin waves is
exothemic we expect the quantity (ou) to be a constant
independent of time. Then the solution to (3.3) is
is the average

1
N(t)= (o.
u )t +1/N(0)

We plot in Figs. 2 4, the excess vortex density versus


time for three quenches. We display in Table I fits to the
data over various ranges of t. We parametrize the vortex
density as

N(t)=

(a

+bt)r

(t)

time become very long and the response of the vortices is


affected. Similar results are obtained in ordinary Monte
Carlo simulations when the temperature is stepped quickly through a phase transition.
The quench from T=5.0 to T=0.9 Fig. 3 shows behavior similar to the T=0.8 quench. Finally, the quench
from T=5.0 to T=1.2, Fig. 4, the log-log plot on the left
shows 1/t behavior for long times. At small t, however,
the excess vortex density decays exponentially in time, according to Ap, (t) =exp(
k, t) as shown in the righthand
semi-log plot of Fig. 4. We find from this data that
A, , =0. 280+0.007. A simple
model accounts for this behavior. At early times in the quench the vortices and an-

(3.5)

and determine a, 6, and y by a three-parameter nonlinear


least-squares fit.
Figure 2 shows a quench from T=5.0 to T=0.8. The
behavior of the excess vorticity shows several distinct regimes. For lnt & 1 there is a relatively slow change in the
vorticity which will be discussed in greater detail below.
Then the vorticity falls like 1/t. Between lnt=2. 5 and 3.0
the vorticity shows irregular behavior, and then falls
smoothly again like 1/t until the signal disappears in the
noise.
We can quantify the irregular region by computing the
temperature as a function of the energy density. In Fig. 3
we have also plotted the energy density as a function of
the time. The irregular region begins at T=0.99 and ends
at T=0.91. This is the region of the phase transition.
The departure from 1/t behavior is probably due to critical slowing down: the correlation length and correlation

In

(t)

FIG. 3. Same as Fig. 2 with quench T=5.0 to 0.9. The results are qualitatively the same. Considerably more kinking is
observed, which is likely due to the system's proximity to the
phase transition.

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

35

tivortices are present in clumps, and individual vortices do


not have to move in order to annihilate.
Most of the
vortex-antivortex pairs exists because of the presence of
excited links between them. These links may be thought
of as "particles" which are created and annihilated with
constant probabilities per unit time. This model makes
two predictions. First, link decay should cease to be important as the density of links approaches its equilibrium
value. Thus we expect to see the excess vorticity in later
times dominated by isolated vortices. Examining our
snapshots of a quenched system Figs. 1(a) through 1(e) reveal just this sort of behavior. Second, the excess potential energy per rotor, Ae, should also be decaying exponentially with the same time constant as the vortex density, since each link (vortex-antivortex pair) should have a
constant energy E~, defined by EI Ae/Ap, . The time

8533

TABLE I. Kinetic equation parameters y and (crv) for various quenches.

Region

0.8
0.8
0.9
0.9
1.2

B
B
B

5.91+0.16
5.95+0.97
4. 88+0. 36
6.01+ 1.00
11.61+6. 83

1.038+0.027
0.993+0. 130
1.001+0.003
1.122+0. 110
1.083+0. 102

of EI is shown in Fig. 5 for Tf =1.2 and


For Tf 1.2, E~ is constant through the
0.8,
7. 50+0.09. For Tf
quench: we find also that EI
however, E~ is not constant. This is the result of the

dependence

Tf 0.8.

dN'erence in the equilibrium


vortex density at the two
temperatures.
During the quench to low temperature, the
mean vortex separation R (t) becomes much greater than
the separation found in the quench to high temperature.
The increase in the observed value of EI can be accounted
for through the inclusion of the "Coulombic" potential
energy of isolated vortices separated by a mean distance
R (r). Finally the inverse of A,
is the lifetime of a
vortex-antivortex
pair. We find the measured value of
~~ = is 3.57+0.089.
We conclude this section with an attempt to construct
an analog of Eq. (3. 1). In a discrete model at low temperature, the out-of-equilibrium
state is characterized by the
presence of domain wall defects, which separate regions of
difterent ground states from one another. The growth of
the domains during the quench is parametrized in terms
of a length scale, R (t), proportional to the square root of
the domain area. In the equilibrium state there are no
domain walls, and R (t) is infinite. In the XY model,
while there are no domains, the obvious analog are the
vortices: the low-temperature ground state has essentially
no vortex pairs, while the out-of-equilibrium
configuration
is characterized by the presence of vortex-antivortex pairs.

r~,

a 3

)jN

)( n

(a)

T =0. 8

IO

l5

(b)

FIG. 4. Quench

T=5.0 to 1.2.

Two markedly different


regions are observed. (a) is a log-log plot of excess vortex density
as a function of time and (b) is a semi-log plot of the same data
for early times. For times less than 9.0 exponential decay of vortex pairs is the rule, while for late times the power-law t ' dominates. The same early signature can be seen in Figs. 2 and 3 but
is much more pronounced in this plot.
from

IO

l5

20

FIG. 5. Semi-log plot of the energy per vortex pair vs time for
the quench from T=5.0 to 1.2. For T= 1.2 the ratio is constant,
supporting the idea that the exponential decay is associated with
the decay of vortex pairs. The difference in the two curves can
be explained by the contribution of the Coulombic potential to
the Tf =0.8 case.

RICHARD LOFT AND THOMAS A. DeCrRAND

8534

The relevant length scale for the system is the mean


vortex-antivortex
In terms of the
separation distance.
metastable vortex density p, it is

(t)=+1/p,

From the observed power-law dependence of the vortex


density, we find that R (t) obeys the scaling law

(t)-t'

35

of a single vortex, as well as obscuring the source of


the interactions causing the Brownian motion. We have
circumvented this problem by studying a configuration
with a net global winding number N =1. Such a system
supports a single, isolated vortex at low temperatures for
all times, and thus is ideally suited for the measurements
we desire to make. Taking our lattice to be an approximately circular grid of points (shown in Fig. 6), we choose
the initial conditions at grid point i =(x,y):
ing

y
2~
arctan ,
y

this is the same power law obtained for


Interestingly,
quenched Ising systems and small-N Zz models, but it is
very different from the scaling law for large-N Z~ models
quenched into the magnetically ordered phase.

P;(0)=

'

IV. DYNAMICS OF AN INDIVIDUAL VORTEX


In this section we investigate the dynamical behavior of
a vortex in the XY model, since our quench studies support a physical picture in which the isolated vortices are
free to move around. A simple model for this process is
to suppose that an isolated vortex undergoes Brownian
motion as a result of its interactions with the surrounding
"sea" of spin waves. Assuming that the vortex has a
mass m, and experiences a drag b then its motion can
be described by the phenomenological Langevin equation

m, r+b, r'=g(t),

'

y
arctan ,
y
X

the noise rt(t) is a Markovian force due to the


chaotic spin-wave background, with correlation

(g(t)q(t') ) =2kTb, 6(t

t')

(4.7)

&0

(4. 8)

which starts the vortex at the center of the grid and establishes N=1. We make use of three different equations
of motion to describe the dynamics of an individual rotor.
First- the second-order microcanonical equation
~ ~

(4.9)

Second, we study the second-order


rived in Sec. II):

Langevin equation (de-

~I'(4 0+~)
a

(4. 1)

where

)0

(4. 10)

where q is given by Eq. (2. 10). Third we use a first-order


Langevin equation of the form
~

(4.2)

BV

(4 0+p)+n

(4. 1 1)

We define the Brownian relaxation time

w,

to be

to describe dynamics of interest in liquid-crystal

mt'

physics.

(4.3)

b,

Of course, the solution to Eq. (4. 1) is found in terms of


the moment

(r ). For

(r') =2(u') r,t

))r,

we have

(4.4)

Now, since the vortex is a collection of modes of the


system, it need not have the canonical number of degrees
of freedom for a particle. We suppose that the vortex acThen in
tually acts as an effective number of particles n
thermal equilibrium, the energy of the vortex (which is all
kinetic energy if we neglect edge effects) is

'.

'm, (U2) =n *kT .

(4.&)

We define the mass per degree of freedom


and thus

(U')

= 2kT
m~,

=m, /n*
(4.6)

),

and the slope of (r ) versus t in


By measuring (u
numerical experiments, the values of the mass m, and
A maBrownian relaxation time ~, may be determined.
jor difticulty with this sort of measurement is that one
finds unbound vortices in the XY model only at high temperature, where production of vortex-antivortex pairs is
important. Pair production greatly complicates the track-

FICx. 6. The orientation of the rotors in our circular grid with


a radius of 15 during a numerical simulation. The + marks the
final location of the vortex (vorticity 1). The temperature is

0.6.

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

35

A. Microcanonical case

In integrating the equations of motion for the XY model, we must specify the initial momenta of each rotor, as
well as the position. To speed up convergence to equilibrium at a temperature T, we randomly assign momenta
such that the total initial energy of the system correT. The simulation begins by
sponds to a temperature
monitoring the mean kinetic energy of the spin until the
equipartition value equivalent to of ,kT is attained. By
tracking the vortex in its motion about the grid over many
realizations, we may thus study (r )(t) and (u ) . Figure 7 shows a plot of (r )(t) versus time measured over
40 realizations on a grid of radius 20 lattice units. As expected from Eq. (4.6), we notice that the plot is indeed
linear. As a check, we plotted (U ) as a function of time,
to ensure that in fact that ( v ) had relaxed to its equilibrium value, denoted (v )
Now if indeed the mass m
of the vortex is a true mass, it should not be a function of
the temperature, at least in the spin-wave phase. As Fig.
8 illustrates, the measured mass shows, to within experimental error, no temperature dependence. For the largest
lattice studied, a grid of radius 20 with 1264 points, we
find
the mass per particle degree of freedom
is
m =2. 673+0. 120 with a 7 per degree of freedom of
0.310, averaged over the region T=0.25 and 0.7. Similarly the Brownian relaxation time w, shows no discernible
temperature dependence in this region of the spin-wave
phase. We observe ~, =0. 514+0.032 with a g per degree
of freedom of 1.610, on a grid of radius 20. We remind

the reader of the units of the above dimensionless quantities: mass scale is set by the moment of inertia density
and the quantity &I/g is the natural unit of time.
The important point to emphasize here is that this stochastic motion of the vortex in the microcanonical simulation can only arise as a result of vortex-spin-wave coupling. Thus the characteristic time ~, is an indirect measure of the coupling strength between spin waves and vortices. Although, to our knowledge, no analytical technique for computing this coupling strength exists, our results suggest that the very interesting physics of such interactions is accessible numerically.
Finally, we discuss the dependence of our results on the
lattice size r. The lower plot of Fig. 9 shows the mean
value for the mass for lattices of radii 12, 15, 17, and 20.
The upper plot of Fig. 9 shows the analogous plot for w, .
Neither of our results show much sensitivity to the lattice
size. Probably this is because we have taken great care to
measure these properties while the vortex was near the
center of the grid, where edge e6'ects are small.

I/a,

B.

I.O

0.5

0
=

0.5

O.

0.2

0.3

0.4

0.5

0.6

O. 7

0.8

0.9

.0

FIG. 7. Experimental

I.O

O. 6

0
0

Second-order Langevin equation

The natural extension of the microcanonical simulations described above is to add fluctuating and dissipating
terms to the equation of motion, resulting in Eq. (2. 11), a

T'

8535

verification of vortex Brownian motion


is obtained from this plot of
vs t. (r') is averaged over 40
random walks of the vortex, which started from the center of the
circular grid. The slope of the line is the diffusion constant of
the vortex.

(r')

FIG. 8. Isolated vortex mass per particle degree of freedom


m, units of I/a (lower), and the Brownian time ~, (upper) vs T.
The system is a circular grid of radius 20 lattice units, with a net
vorticity of
1 throughout,
obeying deterministic equations of
motion. In the spin-wave phase, from T=0.25 to 0.70, no temperature dependence is discernible.

RICHARD LOFT AND THOMAS A. DeGRAND

8536

second-order

Our purpose in studythe dependence


of the phenomenological parameters m, and ~, on the dissipative time constant I &. We expect that the mass of the
vortex should be a parameter independent of drag. To
verify this, we conducted 40 realizations on a grid with a
radius of 15 units at various values of the time constant.
In all cases the time step used was 0.005. Figure 10
shows the scaled vortex mass obtained when I ~ is varied
over the range 0.0 to 1.2: within the quoted experimental
errors the measured mass is independent of the drag.
Now consider the Brownian time ~, . We define the total vortex dissipative time constant I
by
ing this type

equation.

Langevin

35

d&e, )

of dynamics is to understand

(4. 14)

for these mean velocities we obtain

Using the expressions


the result

d(e, )

I,

d &eq)

Ig

(4. 15)
m

We make the reasonable assumption that d ( e, ) /d ( e~) is


a constant and defines an efFective number of rotors n
which makes up the vortex. This gives us the result

(4. 12)
where I"o is the time constant measured using the deterdeministic equations of motion. To understand how
for
loss
rate
a
mean
the
look
at
I
we
energy
pends on &,
single rotor:

I,

'( ')

(4. 16)
Our numerical studies confirm this simple linear prediction, as shown in Fig. 11. Determination of the slope permits us to compute the value of n from these results: We
find that n =3.29+0.33. Since rn is the mass of the vortex per rotor degree of freedom, we may obtain an estimate of the total mass of the vortex, mby equating n *
and n:

(4. 13)

as compared to the energy loss rate for a vortex

m, =n, rn

=9.04+0. 81

(4. 17)

The chemical potential of an isolated vortex p is related to


its mass and spin-wave propagation velocity c by
(4. 18)
p=m, c

The chemical potential of an isolated vortex is expected


theoretically (See Ref. 15) to be

'
p =2~( y+ ,log2)

= 10. 17,

(4. 19)

where y is the Euler-Mascheroni constant. Our result for


is remarkably close to this prediction. This result supports the numerically measured value of the chemical potential of vortex pairs in Ref. 4. There Chester and Tobochnik found p~,=9.4+0.3, a value also somewhat lower
than the expected value from the decoupled vortex model

of
I

IO

20

sr.

C. First-order Langevin equation

The first order Langevin equation of motion for a rotor,


Eq. (2.13), is an interesting limiting case of the second or-

IO

20

FIG. 9. Dependence of m =m/n (lower) and r, on circular


grid radius r. No trends due to edge effects are observed in the
data for grids with radius of 12, 15, 17, and 20 lattice units. The
smallest "dish" studied contained 448 sites (r=12), the largest
1264

(r=20).

0. 5

I.O

FICi. 10. Effect on the measured vortex mass of introducing


and dissipation into the second-order equations of

Auctuations

motion. As the drag time constant I ~ is increased the mass


remains constant. The temperature is 0.5 throughout this measurement.

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

35

8537

D, from the plot of (r ) versus t is D=0. 181+0.014,


with a g per degree of freedom of 1.20 at T of 0.05. This
compares favorably to the D predicted from (4.23). Using
the values

of

and n obtained previously


we find
Finally (U ) is observed to be very
large, and to increase in size as the time step of integration is reduced. This is expected, since the finite time step
acts as a cutoff for the divergent velocity of the defect.
m

D=0. 161+0.031.

V. SPIN CORRELATION FUNCTIONS


Finally we turn to the measurement of spin-spin correHere it is more convenient to work
lation functions.
of a spin vector s(r) with components
in terms
(cosP(r), sing(r)). We measure correlation functions first
at equal time

I (r) =Re(s(r). s(0) )

(5. 1)

and their Fourier transform

I (k)=Re(s(k). s( k) )
and then the real-time structure factor

0. 5

(5.2)

I.O

I (r, t)=Re(s(r, t) s(0, 0))


FICy. 11. Dependence of the vortex dissipative time constant,
on I ~, the individual rotor time constant, in second-order
Langevin dynamics. The slope gives a measure of the effective
number of rotors composing a vortex.

I,

der dynamics. Condensed matter systems, such as smectic liquid crystals, are known to obey such equations of
motion.
Since the dynamical equation of an individual rotor
does not contain a second order "inertial" term, and there
seems to be no mechanism to generate such a term in the
corresponding vortex Langevin equation, we consider the
first-order equation for the defect:

b, r=g,

(4.20)

(5.3)

and its Fourier transform

I (k, t)=Re(s(k, t). s( k, 0)) .

(5.4)

A. Static correlation functions

Our discussion will be brief, since this topic has already


been covered by, e.g. , Tobochnik and Chester. ' Langevin
techniques have no real advantages over conventional
Monte Carlo for the measurement of static properties.
that for
The Kosterlitz- Thouless
predicts
theory
T & T, I (r) falls like a power

(r)-r

(5.5)

g= T/2~ deep in the spin-wave phase, and g= at


T =T, . Villain' has shown that for T &0. 9T, (in the

when

where g forms the correlation

( r)(t)i)(t') ) = 2n *kTb5(t t')

(4.21)

'

dilute-monopole
ventions),

approximation

and converted to our con-

identical to Eq. (4.2). It can be shown that the solution to


Eq. (4.20) for the moment (r ) is

(5.6)

Ip
7
(r &=4

(4.22)

where we have scaled r and t by r = r/a, t = (g /b~ )t, and


used the relationships b~ ~ and b, =ml, . We get
the scaled diffusion constant D from the slope of the ( r )
versus r plot. Using Eq. (4.16), D is predicted to be

II

(4.23)

nm

Finally, referring back to (4.20), it is easy to convince


one's self that ( U ) is expected to be infinite in first-order
Langevin dynamics.
We have numerically studied this system on a grid with
radius of 15.0 lattice spacings. The measured value of the

and the Kosterlitz

renormalization

for t

small

=(T, T)/T,
'
c ~/t,
r) =

group predicts'

(5.7)

where c is a (nonuniversal)

(r)=r

'

that

constant.

For T ~ T
(5.8)

where g is the correlation length.


We extract i) by measuring I (r) along the principle
axes from r=1 to r=10 on a 40/40 lattice and fitting to
(5.3). Figure 12 shows the results of that procedure. The
temperature dependence of g is shown in Fig. 13. The
curves labeled V and K are Eqs. (5.6) and (5.7), respectively. We performed a very careful measurement of q for

DeGRAND

AND THOM
RD LOFT A

8538

35

d o 90, and then fit the


etween T=
tern perature betw
.
transition temp
em erate
data to Eq'

0 9]0
were strong y corre I a t e d Our best
as
e
and 9= 0481 where
ey may be best consi'dered
consistent
upper booundS. These results aree reasonably co
The constant c was found to be
h ex
expectations.
0 24+0.03 b t is ot expecteted to be unjvers al. Above T
q iSs very close to
~

B.

ructure factors
Dynamic struc
0. 2

to study two different real-tim e d y nam-

0.3

will concentra
t
19, and so we wi
ics case.

'

e uation dythe simulation n for wave equa


onsisted of a
E h

to microcanoni
ted the corre a io
lues o t, an
'
r many runns to acis p
d
tion I (k, t), re
r p eating t ss
statistics.
l tion functionn r(k, o),
We first m easure d h
'ust the Fourier transform o f the sta ic
I (r). The resu
esultss for T= . a
orrelation f
T=1.0 are showwn' in ig.
tter, w ic
ion functio natT=
'
nature o cr
tho r fort evarariation of
We are unaware o any d
switched

T = 0.5

0.5

Tc

&.

'

functionn ex onent q vs T. The


relation
a i
13. Static corre
5.6) and (5.7), resp
es labeled V an
respectively.
c an
co
ra
tempeerature and tthee constant
(For the latter,r thee transition
'ust

the exponent t v aare the best


beloww T, . )

ucture factor witith


it temperature. e. A ssimple
the dynamic struc
spin- wave analyl sis predict
p

k(0)) )
k, 0) ) = cos(cot) ( s(k, t). s(
sin(clfr)

Bs(,

(5.9)

lO

T= 0.8

T=0. 5

l5

I .0

0.5

0
l. 5

I.O
k

= ( s ik
orrelation functIon I ( k, 0)=

FI
temperature s.. T

n
I (r) vs r for a variety of
rrelation funnction

esr

=1.0 (solid

circles). Th e l

critical scaling.
Ing associated witith cri

0).s( k, o)

e clearly shows

for

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

35

where co = k .
We can fit our data to
~

r(k, r) =cos(~r +y)r(k)


of
co versus k in Fig. 15 for a temperature
We see no deviation from spin-wave behavior,
Eq. (5.9), in our data, which range from T=0.5 to
T=1.0. This is a rather cursory study of the secondorder undamped equation of motion. We are interested
primarily in dissipative dynamics. Another study of wave
dynamics may be found in Ref. 19.
The liquid-crystal system described in Sec. IV, as well
as most theoretical models, evolves according to a firstorder dissipative equation of motion similar to Eq. (2. 13).
In the limit of small k and long times we expect the continuum limit of I (k, t) to have the form
and we plot

T=0.5.

I (k, t) exp(

Dk r) .

(5. 10)

On the lattice the discretized version of Eq. (5. 10) is

I (k, t) =exp

2Dt

cosk)
gP (1

(5. 1 1)

Above T, the position space correlation function decays


exponentially with distance. The finite correlation length
where
of the system g results in a mass gap m,
' =m, and I
(k, t) takes the form on the lattice of

I (k, t) =exp

Dt 2

gP(1

cosk)+m-

(5. 12)

the properties
since
the diffusion constant is expected to be accessible in light

We are primarily

interested in measuring

of D in an XY model obeying diffusive dynamics,

8539

scattering experiments proposed by Heinekamp and Pelcovits in liquid-crystal


systems near the smectic- A to
smectic-C phase transition.
We studied the mass gap as
a function of temperature simply as a consistency check of
our program, since the mass gap has been extensively
studied both theoretically, and in numerical simulations. '
Our measurements were made on a 40)&40 lattice with
periodic boundary conditions using a time step of 0.005.
During a typical run, the lattice Fourier transform was
performed every 0.05 time units up to a maximum temporal separation of 0.5, for k modes ranging from (1, 1) to
(5, 5). The system was then allowed to evolve unobserved
for 2.0 time units. This cycle was repeated 80 to 100
times.
Figure 16 shows the comparison between our results for
the mass gap and the results reported in Ref. 21, for temperatures between 1.0 and 2.0. We obtain reasonable
agreement with the values obtained by others. The measurement of the diffusion constant reveals some very interesting structure.
Figure 17 shows the plot of the
diffusion constant D, scaled in units of ga /b, versus T.
At low temperatures
we obtain the naively expected
"bare" result D=1.0. However at temperatures near the
phase transition, between 0.9 and 1.05 the value of D
drops precipitously to approximately 0.5. This value of
the diffusion constant appears constant between T=1.1
and 2.0. The temperature dependence of D can be exand
plained by assuming that b remains unrenormalized,
that the renormalized coupling constant g (T) determines
the value of D through

D(T)= g(T)

(5. 13)

Thus Fig. 17 provides a direct plot of the temperature


dependence of the coupling constant. Both Vi11ain, ' and
have made predictions regarding
Kosterlitz and Nelson
the temperature
dependence of the coupling constant,
relevant to this data. Villain predicts the low-temperature

I, O

l. p

0.5

l. 5

0.5
p

FIG. 15. Dispersion

I.Q

1.5

2. 0

0.5

relation for the dynamic structure factor


T=0.5. As expected, the system obeys
co vs k at a temperature
the wave dispersion relation co =ck.

FIG. 16. Spin wave mass, or mass gap for the XY model,
m, , vs temperature derived from the dynamic correlation function Eq. (5. 12). The b, symbols denote results obtained by Chester and Tobochnik; )& indicates results reported by Ref. 21.

35

RICHARD LOFT AND THOMAS A. DeGRAND

8540

behavior to be

g(T)
g

'

(5. 14)

I.O

We find this prediction in good agreement with data up to


a dimensionless temperature T=0.8. In Ref. 22 the limiting value of the coupling constant as T, is approached
from below is predicted to be

g(T, )

2T,

77

0. 5

For T, =0.9 this yields a value for D(T, ) of 0.573. Although somewhat lower than the observed value, the
agreement is good considering the uncertainties and rapidly changing values of the data. Finally we expect at temperature higher than those considered here that D will approach the high-temperature Villain expression

g(T)

(5. 16)

4 lnT

It would be interesting to experimentally observe the


change in D strictly as the result of this behavior of the
XY model. In the liquid crystal systems discussed in Ref.
22, this behavior would have to be deconvoluted from the
superimposed liquid-crystal phase transition, in which the
tilt angle of the molecules, and hence the XY coupling
constant drops to zero. Although beyond the scope of
this discussion, this does not appear to be an insurmountable task experimentally.
VI. CONCLUSIONS
The purpose of this work was to explore the real-time
behavior of a classical field theory at finite temperature.
We used Langevin and microcanonical methods to analyze several situations involving the two dimensional XY
bemodel, including both equilibrium and nonequilibrium
havior. We found that our methods were stable and reliable, and were able to generate interesting results in all
the cases which we studied.
Although we have not devoted much effort to it in this
paper, we believe that it is possible to use these methods
to study dynamical critical exponents. (In this regard, see
also Ref. 20.) Here the main problem is statistics: near
Tcritical slowing down occurs, the correlation time becomes long, and error analysis becomes complicated.
One particular example to which we hope to return is a
study of the real-time dynamics of vortex-antivortex annihilation into spin waves. The idea is to track isolated
vortices during the late stages of a quench (or to run a
simulation as in Sec. IV with appropriate initial conditions), watch as a vortex and antivortex come together,
and focus on the production of spin waves during the an-

P. Meakin, H. Metiu, R. G. Petchek, and D. J. Scalapino, J.


Chem. Phys. 79, 1948 (1983).
2S. W. Lovesey, Condensed Matter Physics: Dynamic
tions (Benjamin/Cummings,
Reading, Mass. , 1986).

Correla-

0.5

1.0

1.5
2.0
T
FIG. 17. Diffusion constant for spin waves under first-order
D decreases
Langevin dynamics as a function of temperature.
slowly in the spin-wave phase from the low-temperature limiting
value of 1.0. Near the T=1.0 the diffusion constant drops suddenly to about 0.5.

nihilation. As the annihilation of topological objects is a


nonperturbative
process (in terms of the spin-spin coupling constant) the experiments could be quite interesting.
There are obvious analogies with the annihilation of magnetic monopoles in the early Universe or of skyrmions
and antiskyrmions, in elementary particle physics.
Another obvious extension of this work is to models in
The three dimensional
a higher number of dimensions.
XY model supports topological excitations which are line
defects. As far as we know, there has been very little
work done on the evolution of structures involving line
defects during a quench. The subject has interesting applications to astrophysics, where the physical realization
of the quench is an early Universe phase transition which
leads to the production of cosmic strings.
Interesting
quench dynamics have also been observed in liquid crystals which may have an analog in the three dimensional
XY model or a variation thereof.
Finally, there is the extremely interesting problem of
It
real-time behavior of quantum mechanical systems.
may be that some variation on our techniques may be
applicable to these systems. At present, however, we have
no idea even how to begin to discuss this problem.
ACKNOWLEDGMENTS

We would like to thank Paul Beale, John Bjorkman,


Philip Ensign, Noel Clark, Chris Muzny, and David Srolovitz for useful and enjoyable conversations.
This work
was supported by the U. S. Department of Energy.

3K. Binder, Applications of the Monte Carlo Methods in Statisti


cal Physics, (Springer-Verlag, Berlin, 1984).
4M. Cretz, Comput. Phys. Commun. 33, 361 (1985).
~T. Schneider and E. Stoll, Phys. Rev. B 17, 1302 (1978).

35

NUMERICAL SIMULATION OF DYNAMICS IN THE XY MODEL

T. Schneider and E. Stoll, Phys. Rev. B IS, 6468 (1978).


~G. G. Batrouni, G. R. Katz, A. S. Kronfeld, G. P. Lepage, B.
Svetitsky, and K. G. Wilson, Phys. Rev. D 32, 2736 (1985).
8G. Parisi and Y. Wu, Sci. Sin. 24, 483 (1981).
P. S. Sahni, G. Dee, J. D. Gunton, M. Phani, J. L. Lebowitz,
and M. H. Kalos, Phys. Rev. B 24, 410 (1981).
P. S. Sahni, D. J. Srolovitz, G. S. Grest, M. Anderson, and S.
A. Safran, Phys. Rev. B 28, 2705 (1983).
D. Toussaint and F. Wilczek, J. Chem. Phys. 78, 2642 (1983).
' N. A. Clark and R. Pindak, Phys. Rev. Lett. 45, 1193 (1980).
' H. Risken, The Fokker-P/anck Equation (Springer-Verlag, Berlin, 1984), pp. 60-62.
J. Tobochnik and G. V. Chester, Phys. Rev. B 20, 3671 (1979).

'

J. M.

8541

Kosterlitz and D. J. Thouless, J. Phys. C 6, 1181 (1973).


C. Y. Young, R. Pindak, N. A. Clark, and R. B. Meyer, Phys.
Rev. Lett. 40, 773 (1978).
' J. Villain, J. Phys. (Paris) 36, 581 (1973).
8J. M. Kosterlitz, J. Phys. C 7, 1046 (1974).
' C. Kawabata, M. Takeuchi, and A. R. Bishop, J. Magn. Magn.
Mater. 54-57, 871 (1986).
S. W. Heinekamp and R. A. Pelcovits (unpublished).
'A. Patkos and N. D. Hari Dass, Phys. Lett. 119B, 391 (1982).
J. M. Kosterlitz and D. R. Nelson, Phys. Rev. Lett. 39, 1201
(1977)
C. T. Vachaspati and A. Vilenkin, Phys. Rev. D 30, 2036
(1984).
~

You might also like