PHYSICAL REVIEW E 80, 061138 共2009兲

Nonequilibrium phase transition in an exactly solvable driven Ising model with friction

Alfred Hucht
Fakultät für Physik und CeNIDE, Universität Duisburg–Essen, D-47048 Duisburg, Germany
共Received 3 September 2009; published 30 December 2009兲
A driven Ising model with friction due to magnetic correlations was proposed by Kadau et al. 关Phys. Rev.
Lett. 101, 137205 共2008兲兴. The nonequilibrium phase transition present in this system is investigated in detail
using analytical methods as well as Monte Carlo simulations. In the limit of high driving velocities v the model
shows mean-field behavior due to dimensional reduction and can be solved exactly for various geometries. The
simulations are performed with three different single spin-flip rates: the common Metropolis and Glauber rates
as well as a multiplicative rate. Due to the nonequilibrium nature of the model all rates lead to different critical
temperatures at v ⬎ 0, while the exact solution matches the multiplicative rate. Finally, the crossover from Ising
to mean-field behavior as function of velocity and system size is analyzed in one and two dimensions.

DOI: 10.1103/PhysRevE.80.061138 PACS number共s兲: 05.50.⫹q, 68.35.Rh, 04.20.Jb

I. INTRODUCTION II. MODEL

Magnetic contributions to friction due to spin correlations Let us start with the simplest case denoted 1d in Fig. 1
have attracted increasing interest in recent years. One inter- and consider two Ising chains with spin variables ␴ = ⫾ 1,
esting aspect is the energy dissipation due to spin waves in nearest-neighbor coupling K = ␤J 共␤ = 1 / kBT and we set kB
magnetic force microscopy, where magnetic structures are = 1兲 and L储 sites each, interacting with boundary coupling
investigated by moving a magnetic tip over a surface 关1–3兴. Kb = ␤Jb and moving along each other with relative velocity
On the other hand, magnetic friction is also present in bulk v. In the Monte Carlo simulation the upper system is moved
magnetic systems which are in close proximity. In this con- v times by one lattice constant a0 with respect to the lower
text, Kadau et al. 关4兴 proposed a simple model for magnetic system during each random sequential Monte Carlo sweep
friction mediated solely by spin degrees of freedom. In this 共MCS兲. As one MCS corresponds to a typical spin-relaxation
time t0 = O共10−8 s兲 关12兴 and a0 = O共10−10 m兲, the velocity v
model an Ising spin system is moved over a second spin
is given in natural units a0 / t0 = O共1 cm/ s兲 共we will set a0
system with constant velocity v along a boundary. This per-
= t0 = 1 in the following兲.
manent perturbation drives the system to a steady state far
To simplify the implementation, instead of moving the
away from equilibrium, leading to a permanent energy flow upper part of the lattice with respect to the lower part we
from the boundary to the heat bath. reorder the couplings at the boundary with time. This proce-
This problem can be analyzed for several different geom- dure is analogous to the Lees-Edwards or moving boundary
etries in one, two, and three dimensions, as shown in Fig. 1: condition in molecular-dynamics simulations of fluids 关13兴
besides the original problem of two half-infinite two- and leads to a system as shown in Fig. 2. Assuming periodic
dimensional systems moving along the one-dimensional boundary conditions 共PBCs兲 ␴k,l ⬅ ␴k,l mod L储 in the parallel
boundary, denoted by 2db in the following, we will consider direction, the time-dependent Hamiltonian reads
the homogeneous cases 1d and 2d where all spins are at the
boundary, as well as the experimentally relevant three-
dimensional case 共3db兲. Additionally, we will extend the
1d 2d
analysis to sheared systems in two 关5–7兴 and three 关8兴 di-
mensions, denoted by 1 + 1d and 2 + 1d. These systems are
experimentally accessible within the framework of shear
flow in binary liquid mixtures 共for a review, see 关9兴兲, though
with conserved order parameter, while we deal with a non-
2db 3db
conserved order parameter.
This model has some similarities to the driven lattice gas
共DLG兲 proposed by Katz et al. 关10兴 共see 关11兴 for a review兲,
where a system is driven out of equilibrium by an applied
field which favors the motion of particles in one direction. 1+1d 2+1d
We will discuss these similarities throughout this work.
The paper is organized as follows: in the first part we will
introduce the model and geometries and present, in the sec-
ond part, an exact solution of the model in the limit of high FIG. 1. 共Color online兲 Overview of the geometries considered in
driving velocities v → ⬁, which will be checked numerically this work. The gray regions are the magnetic systems, while the
in the last part using Monte Carlo 共MC兲 simulations. There green 共dark兲 regions are the moving boundaries. The arrows indi-
we will also investigate the case of finite velocities v. cate the motion of the subsystems.

1539-3755/2009/80共6兲/061138共13兲 061138-1 ©2009 The American Physical Society

ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

k=1 k=1

k=0 k=0
l−2 l−1 l l+1 l+2
⇓
FIG. 2. 共Color online兲 Sketch of geometry 1d after ⌬ = 2 moves. k=1
Spin ␴0,l interacts with spin ␴1,l+2 with coupling Jb 关green 共gray兲
lines兴, while all other couplings are J 共black lines兲.
k=0
l−2 l−1 l l+1 l+2
1 L储 L储
␤H共t兲 = − K 兺 兺 ␴k,l␴k,l+1 − Kb 兺 ␴0,l␴1,l+⌬共t兲 , 共1兲 FIG. 4. 共Color online兲 Mapping of the 1d driven system, shown
k=0 l=1 l=1 for ⌬ = 2, on two disconnected 1D systems with fluctuating fields.
with the time-dependent displacement
can be understood as follows: in the limit v → ⬁ the interac-
⌬共t兲 = vt. 共2兲 tion Kb␴0,l␴1,l+⌬共t兲 across the driven boundary becomes un-
The second geometry considered in this work is the 2db correlated, as, in the Monte Carlo simulations, at large v the
case shown in Fig. 3, which already was investigated by spin ␴1,l+⌬共t兲 is different in every trial step and can, for sim-
Kadau et al. 关4兴. Here we have a square lattice with plicity, be a randomly chosen spin ␴1,rnd from row 1. Note
L储 ⫻ L⬜ sites and periodic boundary conditions in both that this simplification was checked within the simulations
directions, i.e., ␴k,l ⬅ ␴k mod L⬜,l mod L储. Note that especially and indeed gave the same results, enabling us to perform
simulations at v = ⬁. Thus the boundary coupling can be re-
␴L⬜,l ⬅ ␴0,l. The Hamiltonian of this system becomes
placed by the action of a fluctuating boundary field ␮, e.g.,
L⬜ L储
␤H共t兲 = − 兺 兺 K␴k,l␴k,l+1 + K⬜,k␴k,l␴k+1,l+⌬k共t兲 , 共3兲 ␴0,l␴1,l+⌬共t兲 → ␴0,l␴1,rnd → ␴0,l␮0,l , 共5兲
k=1 l=1
with stochastic variables ␮kl = ⫾ 1共k = 0 , 1兲 under the con-
with ⌬k共t兲 ⬅ 0 and K⬜,k = K for all rows except row k = 0, straint 具␮kl典 = 具␴kl典 = mb, where mb denotes the magnetization
where the couplings to row k = 1 are shifted with constant at the driven boundary. Here we used the translation symme-
velocity ⌬0共t兲 ⬅ ⌬共t兲 = vt. The coupling K⬜,0 ⬅ Kb across the try 具␴kl典 = mk and the reflection symmetry at the boundary,
boundary is allowed to be different from K. For v = 0 and mk = m1−k. In Fig. 4 this mapping of the driven system onto a
Jb = J = 1 this system simplifies to the two–dimensional 共2D兲 system with fluctuating boundary fields is illustrated for the
Ising model in equilibrium, which was solved exactly by 1d case. The next step will be to map the fluctuating fields
关14兴 and shows a continuous phase transition at onto static fields by integrating out the degrees of freedom
2 ␮kl.
Tc,eq = = 2.269 185 3 . . . . 共4兲
ln共1 + 冑2兲
A. Ising model in a fluctuating field
Note that both systems are translationally invariant in 储 di- Consider a general Ising model with arbitrary couplings
rection under the transformation l → l + 1 and obey reflection
Kij in a static external field hext and additional fluctuating
symmetry at the boundary under k → 1 − k. i
fields of strength ki 共note the factor ␤ in all field quantities兲

III. EXACT SOLUTION AT HIGH VELOCITIES ␤H␮ = − 兺 Kij␴i␴ j − 兺 共hext

i + k i␮ i兲 ␴ i , 共6兲
i⬍j i
In Ref. 关4兴 it was shown that for high velocities v Ⰷ 1 the
properties of the 2db system become independent of v. This where the ␮i = ⫾ 1 are stochastic variables at site i with given
average

k=2 具 ␮ i典 = m i . 共7兲
As this condition is given a priori, averages containing ␮i
k=1 can be calculated using the trace formula

k=0
Tr␮ f共␮i兲 = 兺
␮i=⫾1
f共␮i兲pi共␮i兲 共8兲

k = −1
with the probability distribution pi共␮i兲 = 共1 + ␮imi兲 / 2, as then

l−2 l−1 l l+1 l+2 具␮i典 = Tr␮ ␮i = 兺

␮i=⫾1
␮ i p i共 ␮ i兲 = m i
FIG. 3. 共Color online兲 Sketch of geometry 2db after ⌬ = 2
moves. as assumed. With the decomposition

061138-2
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

␤ H ␮ = ␤ H 0 − 兺 k i␮ i␴ i , 共9兲 As 1 = ⳵mb,eq / ⳵mb 兩mb=0 at criticality, we obtain a very

i useful connection between the reduced zero-field
boundary susceptibility of the equilibrium model
the degrees of freedom ␮ in the partition function Z can be 共0兲
␹b,eq 共K兲 = ⳵mb,eq / ⳵hb 兩hb=0 and the critical temperature Tc of
traced out,
the driven system by expanding Eq. 共15兲 to first order around
Z = Tr␴␮ e−␤H␮ = Tr␴ e−␤H0 Tr␮ 兿i ek ␮ ␴ i i i mb = 0, namely,
共0兲
␹b,eq 共Kc兲tanh Kb,c = 1. 共16兲
= Tr␴ e −␤H0
兿i ␮兺
=⫾1
e ki␮i␴i
p i共 ␮ i兲 In the following we will apply these results to the one- and
i
two-dimensional models introduced in Sec. II.
= 兿 cosh ki Tr␴ e−␤H0 兿 关1 + ␴imi tanh ki兴, 共10兲
i i
C. 1d case
The effective Hamiltonian of the system 1d in a fluctuat-
where we used the fact that ␴i = ⫾ 1.
ing field reads
On the other hand, the Hamiltonian of the equilibrium
Ising model without fluctuating fields in a static field hi can L储
be written as ␤H = − 兺 K␴l␴l+1 + 共hext + Kb␮l兲␴l . 共17兲
l=1
␤Heq = − 兺 Kij␴i␴ j − 兺 hi␴i = ␤H0 − 兺 bi␴i , 共11兲
i⬍j i i
Applying the self-consistence condition Eq. 共15兲 to the well-
known expression for the equilibrium magnetization of the
with H0 from Eq. 共9兲 if we let bi = hi − hext
i . The partition one-dimensional 共1D兲 Ising model 共cf. 关15兴兲
function of this model clearly fulfills
sinh h
Zeq = Tr␴ e−␤Heq meq共K,h兲 = , 共18兲
冑e −4K
+ sinh2 h
= Tr␴ e−␤H0 兿 ebi␴i we obtain the zero-field magnetization of the 1d driven sys-
i
tem in the ordered phase for velocity v → ⬁,
= 兿 cosh bi Tr␴ e−␤H0 兿 关1 + ␴i tanh bi兴.
i i
共12兲
m共K,Kb兲 = 冑 cosh 2Kb − coth 2K
cosh 2Kb − 1
, 共19兲
Comparing Eqs. 共10兲 and 共12兲, we conclude that under the
condition with critical temperature fulfilling

tanh bi = mi tanh ki , 共13兲 e2Kc tanh Kb,c = 1, 共20兲

共0兲
the partition function Z can be expressed in terms of Zeq, as ␹eq 共K兲 = e2K in this case. Interestingly, Eq. 共19兲 is equal to

冏兿 冏
the spontaneous surface magnetization of the 2D equilibrium
Z=
cosh ki
Zeq . 共14兲 Ising model 关关16兴, Chap. VI, Eq. 5.20兴 if we identify K and
i cosh bi Kb with the couplings 储 and ⬜ to the surface and conse-
Eq.共13兲
quently has the identical critical temperature Tc. For the spe-
To summarize, the coupling with strength ki to fluctuating cial case K = Kb this gives the well-known value from Eq.
fields ␮i = ⫾ 1 with given average 具␮i典 = mi can be written as 共4兲. However, we regard this equality as coincidence without
coupling to static effective fields bi with strength given by deeper meaning, as Eq. 共19兲 is solution of a simple quadratic
Eq. 共13兲. In the next section we will use this mapping to equation with small integer coefficients when written in the
exactly solve the driven Ising model for high velocities natural variables. Nevertheless, we checked this identity in
v → ⬁. the 2d case and found that we do not get the surface magne-
tization of the 3D system by the same procedure, as the
B. Application to the driven Ising model critical temperature is Tc ⬇ 4.058 关Eq. 共60兲兴 instead of the
correct value Tc = 4.511 424共53兲 关17,18兴.
The general condition 关Eq. 共13兲兴 for the effective
To calculate other quantities we use the transfer-matrix
static fields bi simplifies for the systems considered in this
共TM兲 formulation: the TM of the 1D equilibrium Ising model
work: as all boundary spins are equivalent, mi = mb, with
reads 共cf. 关52兴兲
coupling ki = Kb, leading to a uniform effective field
hb = artanh共mb tanh Kb兲 at the boundary, as we assume no
additional external fields, hexti = 0. Inserting this into the equi-
librium expression for the boundary magnetization
Teq = 冉 eK+h e−K
e−K eK−h
冊 共21兲

mb,eq共K , hb兲 = ⳵ ln Zeq / ⳵hb, we end with the self-consistence and the partition function of a periodic system with L储 spins
condition can be expressed as
mb,eq关K,artanh共mb tanh Kb兲兴 = mb 共15兲 Zeq = Tr Teq
L储
. 共22兲
for the nonequilibrium order parameter mb. Using Eq. 共14兲 and the conditions Eq. 共13兲 we can write

061138-3
ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

Z = Tr TL储 共23兲
0.5
with the TM 共we set hext = 0 from now on兲 c T 

T= 冏 cosh Kb
cosh h
Teq 冏 tanh h=m tanh Kb
, 共24兲 0.0

which can be written as

0.5
T = cosh Kb 冉 eK共1 + sin ␺兲
e−K cos ␺
e−K cos ␺
eK共1 − sin ␺兲
冊 共25兲 e T 

1.0
using
0.0 0.5 1.0 1.5
sin ␺ = m tanh Kb . 共26兲 T
Tc
The angle ␺ decreases from ␺ = ␲ / 2 at T = 0 to ␺ = 0 at T
FIG. 5. 共Color online兲 Internal energy e储共T兲 关Eq. 共35兲兴 and spe-
ⱖ Tc. The eigenvalues ␭␮ of T fulfill
cific heat c储共T兲 关Eq. 共36兲兴 of the 1d driven system at v → ⬁. The
T兩t␮典 = ␭␮兩t␮典 共27兲 points are MC results for L储 = 211, and the dashed lines are results
for the one-dimensional Ising model in equilibrium.
and are given by

␭0,1 = 再 eK⫾Kb
cosh Kb共eK ⫾ e−K兲 T ⱖ Tc ,
T ⱕ Tc
冎 共28兲
Sec. IV we will argue that in finite systems the spin fluctua-
tions are not only mediated by the spins ␴i but also by the
self-consistent field m which fluctuates at finite L储, an effect
where ␭0 denotes the larger eigenvalue dominant in the ther- which vanishes in the exact solution, as L储 → ⬁.
modynamic limit. Note that in this limit the analog to the From the nearest-neighbor correlation function we can
free-energy density calculate the internal energy e储 = −J具␴l␴l+1典 in 储 direction

冦 冧
1 Je−2K−Kb
f=− ln ␭0 = − 共J + Jb兲 共29兲 − 1 T ⱕ Tc
␤ e储 = sinh 2K sinh Kb 共35兲
of the driven system is simply a constant in the ordered − J tanh K T ⱖ Tc
phase T ⱕ Tc 关35兴. Nevertheless, we can calculate physical
quantities within this TM notation using expectation values as well as the specific heat c储 = ⳵e储 / ⳵T in 储 direction 共Fig. 5兲

冦 冧
as the whole information of the half-infinite system is con- 2K2
tained in the normalized eigenvectors 共coth Kb − 1兲 T ⬍ Tc
sinh2 K
兩t0典 = 冉 冊 cos ␾
sin ␾
, 兩t1典 = 冉 − sin ␾
cos ␾
冊 , 共30兲
c储 =
K2
cosh2 K
T ⬎ Tc .
共36兲

with cos 2␾ = m. Using the normalized TM T̂ = T / ␭0 and the On the other hand, the internal energy in ⬜ direction is sim-
Pauli matrix M = diag共1 , −1兲, the magnetization 关Eq. 共19兲兴 ply given by
can be expressed as
e ⬜ = − J bm 2 共37兲
m = 具t0兩M兩t0典, 共31兲
as the related spins are uncorrelated.
while the correlation function in 储 direction becomes Now we turn to dynamical properties of this system under
g储共n兲 = 具␴l␴l+n典 − 具␴l典具␴l+n典 a concrete MC Glauber dynamics 共see Sec. IV B for details兲
and calculate the spin-flip acceptance rate A = 具pflip典 and the
= 具t0兩MT̂nM兩t0典 − 具t0兩M兩t0典2 energy dissipation rate P = ⳵E / ⳵t: let 具␨ᐉ␨␨r典 denote the prob-
ability of picking a spin ␴ with direction ␨ = ↑ , ↓ and left and
0 具t0兩M兩t1典 ,
= ␭n1␭−n 2
共32兲
right neighbors ␴ᐉ,r with direction ␨ᐉ,r. These probabilities
as Tn = 兺␮␭␮n 兩t␮典具t␮兩. We get the result can be calculated using the matrices P↑ = diag共1 , 0兲 and

g储共n兲 = 再 共1 − m2兲e−2nKb T ⱕ Tc
tanhn K T ⱖ Tc ,
冎 共33兲
P↓ = diag共0 , 1兲, e.g., 具↑ ↑ ↓典 = 具t0兩P↑T̂P↑T̂P↓兩t0典. As the third
coupling partner ␮ of spin ␴, with direction ␨␮, is uncorre-
lated at infinite velocity, the probability of a particular spin
configuration becomes

再 冎
leading to the inverse correlation length
␭0 2Kb T ⱕ Tc 具␨ᐉ␨␨r典具␨␮典 = 具t0兩P␨ᐉT̂P␨T̂P␨r兩t0典具t0兩P␨␮兩t0典. 共38兲
␰−1
储 = ln = 共34兲
␭1 ln coth K T ⱖ Tc .
The spin-flip probability of a given configuration is pflip共⌬E兲,
Note that ␰储 does not diverge at the critical point, a feature with ⌬E = ⌬E1 + ⌬E2 = 2J␴共␴ᐉ + ␴r兲 + 2Jb␴␮, and A becomes
which would lead to a correlation length exponent ␯ = 0. In the sum over all 24 possible cases

061138-4
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

A= 兺 pflip共⌬E兲具␨ᐉ␨␨r典具␨␮典,
␨ ,␨,␨ ,␨ =↑,↓
共39兲 0.6 AT 


















ᐉ r ␮
PT 















which can be written as
AMC T 

AT , PT 
0.4  PMC T 







A= 兺
␨ᐉ,␨,␨r=↑,↓
ⴱ
pflip 共⌬E1兲具␨ᐉ␨␨r典 兺
␨␮=↑,↓
ⴱ
pflip 共⌬E2兲具␨␮典

0.2


 
= 兺 X␨共e−2Kb具␨典 + 具¯␨典兲 共40兲







␨=↑,↓






0.0









 
ⴱ ⴱ ⴱ
for the multiplicative rate pflip 共⌬E兲 = pflip 共⌬E1兲pflip 共⌬E2兲 in-
troduced in Sec. IV B 关Eq. 共69兲兴 using the abbreviation 0 1 2 3
T
Tc
X␨ = 兺 ⴱ
pflip 共⌬E1兲具␨ᐉ␨␨r典
␨ᐉ,␨r=↑,↓ FIG. 6. 共Color online兲 Spin-flip probability A 关Eq. 共43兲兴 and
energy dissipation rate P 关Eq. 共47兲兴 versus reduced temperature
¯ 典 + 具¯␨␨␨
= e−4K具␨␨␨典 + 2e−2K具␨␨␨ ¯ 典. 共41兲 T / Tc for the 1d system at v → ⬁, together with MC data for
L储 = 211.
Note that the two terms in Eq. 共40兲 are equal and the accep-
tance rate is independent of spin ␨ because m is stationary.

冦 冧
The resulting acceptance rate becomes 2e−8K coth 2K
T ⱕ Tc
tanh K sinh 2K

冦 冧
cosh共K + Kb兲 − sinh共K − Kb兲 P= 共47兲
T ⱕ Tc 2e−3K tanh K
A = 4e2共K+Kb兲 sinh K cosh2 K sinh Kb 共42兲 T ⱖ Tc .
cosh K
e−Kb cosh Kb共1 − tanh K兲2 T ⱖ Tc ,
These results are shown in Fig. 6, together with data from
which simplifies for J = Jb to MC simulations. Note that these results are only valid for the
ⴱ
from Eq. 共69兲.

冦 冧
multiplicative rate pflip
e−4K coth 2K
T ⱕ Tc Finally we list the critical exponents for the 1d driven
sinh 2K system at v → ⬁ to be
A= 共43兲
e−3K
T ⱖ Tc . 1
cosh K ␤= , ␥ = 1, ␣ = 0. 共48兲
2
The calculation of the energy dissipation rate P per spin is
The behavior of this system at finite velocities v will be
very similar to the acceptance rate A 关Eq. 共40兲兴 and gives
discussed in Sec. IV.
P = − 2Jb 兺
␨=↑,↓
X␨共e−2Kb具␨典 − 具¯␨典兲. 共44兲
D. 2db case
Furthermore, P / A can be calculated for arbitrary dimensions The 2db case can be solved exactly using the expression
and geometries, as it is solely a property of the fluctuating for the equilibrium surface magnetization mb,eq共z , y b兲 of the
field. We find 2D Ising model in a static surface field hb obtained by Mc-
Coy and Wu 关关16兴, Chap. VI, Eq. 5.1兴, with z = tanh K and
P 兺␨=↑,↓X␨共e−2Kb具␨典 − 具¯␨典兲 y b = tanh hb. The integral representation given in their work
= − 2Jb can be further evaluated and written in closed form, the re-
A 兺 X 共e−2Kb具␨典 + 具¯␨典兲
␨=↑,↓ ␨ sults are given in Appendix A 关Eq. 共A2兲兴. If we again use Eq.
e−2Kb具␨典 − 具¯␨典 共15兲 and set y b = mbzb, with zb = tanh Kb, we can calculate the
= − Jb 兺 nonequilibrium boundary magnetization mb共z , zb兲 numeri-
␨=↑,↓ e−2Kb具␨典 + 具¯␨典 cally as solution of the self-consistence condition
2Jb共m2 + 1兲tanh Kb mb,eq共z,mbzb兲 = mb , 共49兲
= . 共45兲
1 − m2 tanh2 Kb
which is shown for J = 1 and several values of Jb in Fig. 7.
For the magnetization 关Eq. 共19兲兴 of the 1d system this gives The critical temperature Tc of the system can be evaluated
from the reduced zero-field boundary susceptibility ␹b,eq共z兲

冦 冧
2Jbe−4K 关Eq. 共A3兲兴 to give
P T ⱕ Tc
= tanh Kb 共46兲 Tc = 2.661 472 565 575 2. . . 共50兲
A
2Jb tanh Kb T ⱖ Tc , 共0兲
for the case Jb = J = 1 using ␹b,eq 共zc兲zb,c = 1
关Eq. 共16兲兴.
which, multiplied with A from Eq. 共71兲 and for J = Jb, be- As the critical temperature Tc 关Eq. 共50兲兴 is larger than the
comes equilibrium critical temperature Tc,eq = 2.269 18. . ., the driven

061138-5
ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

1.0 k=1

Jb  1.25 k=0
Jb 1
mb T 

Jb  0.75
0.5 k = −1
Jb  0.5
Jb  0.25
Jb  0.05 ⇓
Jb 0
k=1
0.0
1.0 1.5 2.0 2.5
T k=0

FIG. 7. 共Color online兲 Boundary magnetization mb共T兲 关Eq. 共49兲兴

of the 2db system for J = 1 and several values of Jb. For Jb = 0 the mb k = −1
reduces to the surface magnetization of the 2D equilibrium Ising
model 关Eq. 共A5兲兴. l−2 l−1 l l+1 l+2

FIG. 8. 共Color online兲 Mapping of the 1 + 1d sheared system,

boundary induces a surface phase transition where only the shown for ⌬ = 2, on L⬜ disconnected 1D systems with fluctuating
driven surface has long range order above Tc,eq. The velocity fields.
dependence of this transition and the resulting phase diagram
is discussed in more detail in Sec. IV.
1
Ci = 共1 − m2i + 共1 + m2i 兲cosh 2ki兲. 共53兲
2
E. 1 + 1d sheared case
Equating Eq. 共52兲 with Eq. 共12兲 we conclude that static fields
If the motion of the lattice described by Eq. 共3兲 is not
bi can replace the fluctuating fields ␮ij, with average mi, if
restricted to one row but applied to the whole system, we get
a system with uniform shear. Then all ⌬k共t兲 ⬅ ⌬共t兲 = vt are 2mi sinh 2ki
equal, and we assume K⬜,k ⬅ K⬜ to get tanh bi = . 共54兲
1− m2i + 共1 + m2i 兲cosh 2ki
L⬜ L储
The sheared system is translationally invariant in both direc-
␤H共t兲 = − 兺 兺 K储␴k,l␴k,l+1 + K⬜␴k,l␴k+1,l+⌬共t兲 . 共51兲 tions, leading to homogeneous values mi = m, ki = K⬜, and
k=1 l=1
bi = h. Inserting Eq. 共54兲 into Eq. 共19兲 we get the order pa-
Note that this system is translationally invariant in both di- rameter of the sheared 1 + 1d system
rections, a fact that drastically simplifies the analysis of the 冑1 − 2e4K + 2e2K 冑e4K − 1 + tanh2 K⬜
储 储 储

critical behavior. m共K储,K⬜兲 = ,

Now we will investigate this system in the limit v → ⬁. tanh K⬜
Then each spin ␴kl interacts, as depicted in Fig. 8, with its 共55兲
neighbors ␴k⫾1,l⫾⌬共t兲 via fluctuating fields, while the interac-
tion to the parallel neighbors ␴k,l⫾1 remains unchanged. Thus with critical temperature fulfilling
the system decomposes into L⬜ identical 1D Ising models 2e2K储,c tanh K⬜,c = 1, 共56兲
which again can be solved exactly: the coupling to two fluc-
tuating fields ␮i,1 and ␮i,2 with equal strength ki on each site which gives 冑3 + 冑17兲 = 3.465 91. . . for
Tc = 1 / ln共 21
can be traced similar to Eq. 共10兲 to give J储 = J⬜ = 1. A generalization of Eq. 共52兲 from two to f fluctu-
ating fields per spin is straightforward and leads to the gen-
2 eral criticality condition
Z = Tr␴ e −␤H0
Tr␮ 兿i 兿
j=1
ek ␮ ␴ i ij i
共0兲
␹eq 共Kc兲f tanh Kb,c = 1. 共57兲
2
Although this geometry can be solved exactly at v = ⬁ we
= Tr␴ e−␤H0 兿 兿 兺 ek ␮ ␴ pi共␮ij兲 i ij i
expect the phase transition to be strongly anisotropic 共see,
i j=1 ␮ =⫾1ij e.g., 关19兴兲 with two different correlation length exponents
␯储 ⬎ ␯⬜. In fact we found such behavior, with strong evi-
= Tr␴ e −␤H0
兿i 关cosh ki + mi␴i sinh ki兴2 dence for the exponents ␯储 = 3 / 2 and ␯⬜ = 1 / 2, details on this

冋 册
will be published elsewhere 关20兴.
mi
= 兿 Ci Tr␴ e−␤H0 兿 1 + ␴i sinh 2ki , 共52兲 F. Other geometries
i i Ci
For two more cases we can derive highly accurate esti-
with mates for the critical temperature Tc of the driven system

061138-6
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

when v → ⬁, namely, the 2d Ising double layer 关20兴 with J

1 L储

Hamiltonian Eb,储 = − 兺 兺 ␴k,l␴k,l+1 ,

2L储 k=0 l=1
共62b兲
1 L储 L储
␤H共t兲 = − 兺 兺 兺 关K␴klm共␴k,l,m+1 + ␴k,l+1,m兲 L储
Jb
兺 ␴0,l␴1,l+⌬共t兲 ,
k=0 l=1 m=1
Eb = − 共62c兲
+ Kb␴0lm␴1,l+⌬共t兲,m兴 共58兲 L储 l=1

and the experimentally relevant 2 + 1d sheared case as well as the corresponding bulk quantities. From these
L⬜ L储 L储
time-dependent quantities we calculate the averages of the
magnetization, reduced susceptibility, Binder cumulant, in-
␤H共t兲 = − 兺 兺 兺 关K ␴klm共␴k,l,m+1 + ␴k,l+1,m兲
储
ternal energy, and specific heat at the boundary,
k=1 l=1 m=1

+ K⬜␴klm␴k+1,l+⌬共t兲,m兴, 共59兲 mb,abs = 具兩M b兩典, 共63a兲

both on simple cubic lattices: with Eq. 共57兲 we can express

␹b,abs = 2L储共具M 2b典 − 具兩M b兩典2兲, 共63b兲
Tc using the high-temperature series expansion for the re-
共0兲
duced zero-field susceptibility ␹eq 共K兲 of the 2D Ising model,
which was calculated to higher than 2000th order recently 具M 4b典
Ub = 1 − , 共63c兲
using a highly efficient polynomial time algorithm 关21兴. 3具M 2b典2
Using this extremely accurate result we find, for J = Jb = 1, the
critical temperatures
eb = 具Eb典, 共63d兲
Tc = 4.058 782 423 137 980 000 987 775 040 680. . .
共60兲 cb = L储␤2共具E2b典 − 具Eb典2兲. 共63e兲
for the two 2d layers and Note that we have absorbed the factor ␤−1 into ␹. Near criti-
cality these quantities show power-law behavior and fulfill
Tc = 5.264 750 414 514 743 550 598 017 203 424. . .
共61兲 mb,abs共␶兲 ⬀ 共− ␶兲␤ , 共64a兲

for the 2 + 1d sheared system with f = 2 analogous to the

␹b,abs共␶兲 ⬀ 兩␶兩−␥ , 共64b兲
1 + 1d sheared system. Note that due to the high accuracy of
the series these values can be calculated to approximately
500 and 700 digits, respectively. c b共 ␶ 兲 ⬀ 兩 ␶ 兩 −␣ , 共64c兲
Just for reference we also give the critical temperatures
for two more cases: the experimentally relevant 3db case with reduced temperature ␶ = T / Tc − 1 and critical exponents
shown in Fig. 1 as well as the quite theoretical 3d case of ␤, ␥, and ␣. But before we present the results, we have to
two three-dimensional systems in direct contact along the take a closer look at the used spin-flip rates.
fourth dimension. In the 3db case we find Tc = 4.8共1兲
using the eighth-order high-temperature series from B. Integrable algorithm
Ref. 关22兴 共Table IV兲, while in the 3d case we obtain While equilibrium properties are most efficiently investi-
Tc = 5.983 835共1兲 using the 32th order series from 关23兴. gated in Monte Carlo simulations using cluster algorithms,
All these higher-dimensional geometries are expected to nonequilibrium systems have to be treated with random se-
show strongly anisotropic behavior with two different corre- quential single spin-flip dynamics such as the nonconserved
lation length exponents, the reader is referred to Ref. 关20兴. Glauber dynamics 关24兴 or the conserved Kawasaki dynamics
关25兴. The driven system is permanently under an external
IV. MONTE CARLO SIMULATIONS
perturbation which drives it out of equilibrium, while the
A. Method internal degrees of freedom are coupled to a heat bath in
thermal equilibrium. From this coupling the spin-flip prob-
We now describe the algorithms used to investigate the
ability pflip共⌬E兲 of a given energy change ⌬E fulfills the
driven system: for finite velocities v we shift the boundary
detailed balance condition
couplings by increasing ⌬共t兲 from Eq. 共2兲 after every N / v
random sequential single spin-flip attempts, where N denotes pflip共⌬E兲
the total number of spins. Using 105 – 106 MCS per tempera- = e−␤⌬E , 共65兲
ture, we measured the following boundary properties: the pflip共− ⌬E兲
boundary magnetization per spin and the energy per bond just like in the equilibrium case 共for details, see 关20兴兲.
parallel to and across the boundary of a given configuration The most common rates fulfilling Eq. 共65兲 are the Me-
1 L储 tropolis rate 关26兴 and the Glauber rate 关24兴,
1
Mb = 兺 兺 ␴k,l ,
2L储 k=0 l=1
共62a兲 M
pflip 共⌬E兲 = min共1,e−␤⌬E兲, 共66a兲

061138-7
ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

r 1





Metropolis
 Glauber

j’
i
j 
 multiplicative

p flip E
FIG. 9. 共Color online兲 Interactions of surface spin ␴i in the 1d 

case.


G
pflip 共⌬E兲 =
1
. 共66b兲 


1 + e␤⌬E 0

6
4
2 0 2 4 6
Using these rates in simulations of, e.g., the 1d driven system
关Eq. 共1兲兴, it turns out that for all v ⬎ 0 the critical temperature E
Tc共v兲 depends on the used rate 共see also Fig. 15兲: we find, for FIG. 10. 共Color online兲 Spin-flip probabilities of the Metropolis
v → ⬁ and Jb = J = 1, the values TM c = 1.910共2兲 and Tc
G
rate 关Eq. 共66a兲兴 共dashed black line, circles兲, the Glauber rate
= 2.031共2兲 for the Metropolis and Glauber rate, respectively, 关Eq. 共66b兲兴 共dotted blue line, squares兲, and the multiplicative rate
while the exact solution 关Eq. 共20兲兴 of the model presented in 关Eq. 共69兲兴 共red line, diamonds兲 for the 1d system at criticality
Sec. III gives Tc = 2.269. . .. Note that a similar dependency 共J = Jb = 1兲.
was recently found in the DLG by Kwak et al. 关27兴.
How can these discrepancies be understood? And can we larger than one. For our example 关Eq. 共67兲兴 we find
construct a rate that matches the analytical treatment, i.e., has ⌬E1,min = −2zJ and ⌬E2,min = −2Jb to fulfill Eq. 共68兲. This rate
the same Tc? This is indeed possible: consider a microscopic reproduces the calculated critical temperatures in all consid-
change, i.e., a spin flip, of spin ␴i at the boundary 共see Fig. 9兲 ered geometries, e.g., Tⴱc = 2.269共1兲 for the 1d case at
with energy difference v → ⬁.
⌬E = 2J␴i 兺 具j典⬘ ␴ j + 2Jb␴i␴r ,
z The resulting spin-flip rates for the 1d case at criticality
are shown in Fig. 10. Clearly, the multiplicative algorithm
⌬E2 关Eq. 共69兲兴 has a smaller overall acceptance rate than
⌬E1 共67兲 Eqs. 共66兲 and is thus slightly less efficient: a finite-size scal-
where the sum runs over the z neighbors of ␴i in the same ing analysis of the acceptance rate A = 具pflip典 at criticality in
subsystem 共z = 2 in the 1d case兲, while ␴r is from the other the 1d case yields AM G
c = 0.476共2兲, Ac = 0.366共2兲 and
ⴱ
side of the moving boundary. The idea of the exact solution Ac = 0.242共2兲 for the three algorithms, rendering this method
presented in the last section was to treat spin ␴r as a fluctu- roughly two times slower than the Metropolis algorithm. In
ating variable ␮i at site i with appropriate statistics. By con- fact, Ac = 3冑2 – 4 = 0.242 64. . . can be calculated exactly from
trast, correlations of different strength are introduced be- Eq. 共43兲.
tween the two subsystems by the rates 关Eq. 共65兲兴 because the Note that the Metropolis and Glauber rates can be consid-
influence of spin ␴r depends on the actual state of the z spins ered as many particle rates, as pflip depends on the many
␴ j. This can be seen most easily in the case of the Metropolis particle state of all coupling partners, while the multiplica-
rate 共Jb = J兲: if, e.g., ␴i = −␴ j then ⌬E1 = −2zJ and pflip
M
= 1 in- tive rate corresponds to a product of two particle contribu-
dependent of ␴r 共note that ⌬E2 = ⫾ 2J兲, while in the parallel tions. We believe that the dynamics generated by the multi-
case 共␴i = ␴ j兲⌬E1 = 2zJ and pflip M
strongly depends on ␴r 共see plicative rate is generally simpler than the one generated by
Fig. 10兲. Metropolis or Glauber rates, making an exact solution more
Fortunately, these rate-induced correlations can be com- feasible. Whether this differentiation only holds for the non-
pletely eliminated by requiring that the flipping probability is conserved Glauber dynamics or also for the conserved Ka-
multiplicative, wasaki dynamics is subject of future work.
In the next two sections we will investigate finite-size
pflip共⌬E1 + ⌬E2兲 = pflip共⌬E1兲pflip共⌬E2兲. 共68兲 effects in the 1d case as well as the crossover behavior at
finite velocities v in the 1d as well as in the 2db case. We first
Clearly this condition is not satisfied for the rates in Eq. 共65兲, turn to the 1d case.
M
e.g., pflip 共−2zJ + 2J兲 = 1, while pflip M
共−2zJ兲pflip
M
共2J兲 = e−2K
共again we assume Jb = J兲.
Instead, for simulations of driven systems we propose the C. 1d case
rate The exact solution presented in Sec. III was derived in the
ⴱ
pflip 共⌬E兲 =e −␤/2共⌬E−⌬Emin兲
, 共69兲 thermodynamic limit L储 → ⬁ as we assumed a constant and
nonfluctuating order parameter m in the self-consistence con-
which is uniquely defined by the detailed balance condition dition 关Eq. 共15兲兴. This led to the result that the correlation
关Eq. 共65兲兴 and the multiplicity condition 关Eq. 共68兲兴 关36兴. The length ␰储 关Eq. 共34兲兴 remains finite at criticality. However, in a
constant ⌬Emin is the minimum possible value of ⌬E at given finite system the assumption m = const is not fulfilled and
ⴱ
geometry; this assures that pflip 共⌬E兲 is maximal but never finite-size effects occur, leading to a nontrivial dependency

061138-8
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

1.0




























































































































































































































 






































10.0





















L  24





















































 








 
L  26















  1





 




 































  2




 L  28




















5.0 





1
Νeff




mT 








 


L  210  







 





 
0.5   








 



 











 


L  212







 


  

 




 





 


 


 





   



 





 0
 




 L  214





 










 2.0

∆Τ v 1
2









 


 






 

  






  

 



  
L  216  








 



 

 







  









 



  

 


  
  

  

 



Exact





















 
 1.0 10
2 100 102 104
0.0 




























































































































































































 L
v
0.0 0.5 1.0 1.5 



0.5  v 4 
T
Tc  v  16 





 


  


  

  
   
  

 v  64  
FIG. 11. 共Color online兲 Magnetization mabs共T兲 关Eq. 共64a兲兴 of the  v  256
1d system at v = ⬁ for several system sizes L储 from Monte Carlo 0.2  v  1024
simulations, together with the exact solution Eq. 共19兲.
10
2 10
1 100 101 102 103 104
of the physical quantities on system size. The fluctuating L
v
order parameter gives rise to additional correlations between
spins at large distances not included in the exact solution. As FIG. 13. 共Color online兲 Velocity dependent crossover behavior
the driven system shows mean-field behavior, we can use the in the 1d case. Shown is the rescaled width of the critical region
standard finite-size scaling theory for mean-field systems: ␦␶v1/2 as function of the crossover scaling variable L储 / v for several
near criticality the correlation length parallel to the boundary velocities v and several system sizes L储 = 24 , . . . , 216 共see text兲. The
fulfills ␰储共␶兲 ⬀ 兩␶兩−␯储 with critical exponent ␯储 = 2 / db, where db inset shows the corresponding crossover of the effective correlation
denotes the boundary dimension. We have db = 1 in both the length exponent ␯eff−1
from ␯eff
−1
= 1 / 2 共MF, dotted line兲 to ␯eff −1
=0
1d and the 2db case, leading to ␯储 = 2 in these cases. 共Ising noncritical, dashed line兲.
To illustrate these finite-size effects in the 1d case, in Fig.
11 we show the magnetization mabs共T兲 关Eq. 共64a兲兴 as func- the magnetization mabs共T兲 and specific heat c共T兲 关Eq. 共64c兲兴,
tion of temperature for v = ⬁ and several system sizes L储. The verifying the other two exponents ␤ = 1 / 2 and ␣ = 0.
exact solution 关Eq. 共19兲兴 is only approached in the limit In summary, the 1d and the 2db systems with boundary
L储 → ⬁. dimension db = 1 have the standard mean-field exponents and
The finite-size scaling behavior is demonstrated exem- fulfill the exponent relations
plarily for the susceptibility ␹abs共T兲 关Eq. 共64b兲兴, which is
shown in a finite–size scaling plot in Fig. 12: after rescala- 2 − ␣ = 2 ␤ + ␥ = d b␯ 储 . 共70兲
tion of the MC data in the usual way we indeed find the We now turn to finite velocities v: then the 1d system
expected mean-field exponents ␥ = 1 and ␯储 = 2, furthermore always shows a crossover from mean field to Ising behavior
the data falls onto the universal finite-size scaling function with increasing system size L储. Only in the limit v → ⬁ the
calculated in Ref. 关28兴. The same analysis was performed for system undergoes a phase transition at finite temperatures. To
investigate this velocity-dependent crossover, we measured
the width ␦␶ of the critical region by analyzing the Binder
L  212


 
cumulant 关Eq. 共63c兲兴. Using least-squares fits of the simula-
1  c2 L
1
2


 

 
1 






 



 








 
13 




 




L 2
 











 





 14









 tion data to the simple approximation,
L 2
 

 
 





冦 冧
  

 
 
  
 





 

 


  









 
15 








L 2 1
关1 + tanh共˜␶/␦␶兲兴 ˜␶ ⱕ 0


  


 




 

 

 

 16

L 2
 

 

 
 

3

 


 



 

 




 

Ub共T兲 ⬇ 共71兲



  

 

















 









 

 
FSSF

Γ
Ν






 










 


 1 1

˜␶ ⬎ 0,




 




 








 
 


 

 

ΧT  L



 







3 1 + ˜␶/␦␶
 
 










 
















 








 






 









































with ˜␶ = T / T̃c − 1 and fit parameters T̃c and ␦␶, for several
0 velocities v and system sizes L储 we determined ␦␶ and plot-

2
1 0 1 2 ted them in Fig. 13. We find that the crossover scaling vari-
T
Tc
1 L1
Ν

able is L储 / v in this case, while the y axis has to be rescaled as
␯储
␦␶v1/2 to get the correct limit L1/ 储 ␦␶ = const with ␯储 = 2 in the
FIG. 12. 共Color online兲 Finite-size scaling plot of the reduced limit v → ⬁. At finite v the width ␦␶ stops shrinking at
susceptibility ␹abs共T兲 关Eq. 共64b兲兴 of the 1d system for v = ⬁ and L⫻储 ⬇ 9 v , where L 储
⫻
denotes the crossover system size, and
system sizes L储 = 212 , . . . , 216, together with the exact mean-field only goes to zero for v → ⬁, indicating a sharp phase transi-
finite-size scaling function 共black line兲 from Ref. 关28兴. The correc- tion in this limit. The inset shows the effective exponent ␯eff
tion factor c2 = 2.7. obtained from the logarithmic derivative,

061138-9
ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

0.10 TABLE I. Velocity dependent critical temperatures Tc共v兲 for the

0.68 exp
L
Ξ ,c  2db case, estimated using the multiplicative rate, 关Eq. 共69兲兴 with
  MC data 400.000 MC sweeps per temperature as well as using the Metropo-
Tc L 
Tc

lis rate 关Eq. 共66a兲兴 with 50.000 MC sweeps per temperature.

0.05

v Tⴱc 共v兲 c 共v兲

TM

0.00     1/16 2.301共2兲
2 3
2 4
2 5
2 6
2 7
28 1/4 2.33共1兲
L 1 2.41共1兲 2.30共2兲
4 2.52共1兲 2.37共2兲
FIG. 14. 共Color online兲 Influence of the system size L⬜ on the
critical point in the 2db case at v = ⬁ and L储 = 256. The effective 16 2.61共1兲 2.42共2兲
critical temperature Tc共L⬜兲 shifts to higher values if L⬜ ⱗ 10␰⬜,c 64 2.644共3兲 2.44共2兲
共see text兲. 256 2.654共2兲 2.44共2兲
1024 2.659共2兲 2.45共2兲
⳵ ln ␦␶ ⬁ 2.661共1兲 2.45共2兲
␯eff
−1
=− , 共72兲
⳵ ln L储
whose value changes from ␯eff−1
= 1 / 2 共MF兲 to ␯eff
−1
= 0 共Ising兲 temperature were used. The results are given in Table I and
with growing system size. In the next section we will see that are compiled into a phase diagram of the 2db case shown in
this behavior changes substantially in the 2db case. Fig. 15. An important aspect of this phase diagram is the
possibility of a velocity driven nonequilibrium phase transi-
tion at fixed temperature 共double arrow兲: while the system is
D. 2db case paramagnetic at v = 0 and up to vc共T兲 共thick blue line兲, the
In the 2db case the moving boundary is coupled to a two- boundary shows long range order above that velocity. It
dimensional Ising model, which undergoes a phase transition would be interesting to see this transition in experiments,
at Tc,eq 关Eq. 共4兲兴, independent of the velocity v. In addition, which could be performed in the corresponding geometry
the moving boundary shows a boundary phase transition at 3db 共see Fig. 1兲, e.g., using two close rotating magnets
temperature Tc共v兲, which grows with increasing v and even- slightly above the Curie temperature. The magnets should be
tually approaches the value given in Eq. 共50兲 for v → ⬁. As isolating to avoid eddy currents 关4兴.
Tc共v兲 ⬎ Tc,eq for all v ⬎ 0 we expect a boundary phase tran- In the 2db case the crossover scaling variable can be de-
sition with paramagnetic bulk. Then the correlation length ␰⬜ termined from the Tc共v兲 dependency discussed above. The
perpendicular to the boundary is finite at criticality and has correlation length ␰eq at the critical point of the driven sys-
the Ising value
Tc,eq TcM   Tc  
␰⬜,c共v兲 = ␰eq关Tc共v兲兴, 共73兲 

256 

with 关16兴

再 冎
64 

4K − 2 ln coth K T ⬍ Tc,eq bulk surface
␰eq
−1
共T兲 = 共74兲
ln coth K − 2K T ⬎ Tc,eq .
Velocity v

order order
For that reason, in the finite-size scaling analysis it is suffi- 16 

cient for given v to simulate systems with varying length L储 velocity driven
while holding the height L⬜ fixed at a value L⬜ Ⰷ ␰⬜,c共v兲. To phase transition
illustrate this behavior, in Fig. 14 we show the effect of dif- no
4 

ferent values of L⬜ for v = ⬁ and L储 = 256. Only below order
L⬜ ⬇ 32 the system feels the finite width L⬜, resulting in a 1 

shift of the effective critical temperature Tc共L⬜兲 to higher 0



2.1 2.2 2.3 2.4 2.5 2.6 2.7
values. The strength of the shift is proportional to the corre-
lation function in ⬜ direction, 具␴0,l␴L⬜,l典 ⬀ exp共−L⬜ / ␰⬜,c兲. Temperature T
The curves collapse for L⬜ ⬎ 32 showing that a ratio FIG. 15. 共Color online兲 Phase diagram of the 2db case. Below
L⬜ / ␰⬜,c ⬇ 10 is sufficient as ␰⬜,c共⬁兲 = 3.663 23. . . in this Tc,eq the two-dimensional bulk is ordered, while surface order is
case. observed even above Tc,eq up to the velocity dependent phase
We performed MC simulations and determined the critical boundary Tc共v兲. The position of this boundary depends on the al-
temperatures for different velocities v by performing a finite- gorithm, the blue line holds for the multiplicative rate 关Eq. 共69兲兴,
size scaling analysis of the boundary susceptibility ␹abs,b共T兲 while the thin red dotted line holds for the Metropolis rate 关Eq.
and the boundary cumulant Ub共T兲. For the multiplicative al- 共66a兲兴. At fixed temperatures between Tc,eq and Tc共v兲 a velocity
gorithm, Eq. 共69兲, we used 400.000 MC steps per tempera- driven phase transition is possible. The points are results from MC
ture, while for the Metropolis algorithm 50.000 MC steps per simulations.

061138-10
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

102 driving velocities v → ⬁ the model was solved exactly by



integrating out the nonequilibrium degrees of freedom. The
1 




resulting exact self-consistence equation was analyzed for
 
  various geometries, leading in many cases to precise values

 

1
Νeff

101 of the critical temperature Tc of the nonequilibrium phase



 transition. In the limit v → ⬁ the system always shows mean-
  
 
 
 





1




field behavior due to dimensional reduction independent of
∆Τ ΞeqTc v

2

geometry. In the simplest one-dimensional case denoted 1d a
100  v  1
1024 
 10
1 100 101 102 103 complete analysis of both equilibrium as well as nonequilib-
 v  1
256  rium quantities has been presented. These exact results are

  L
Ξeq Tc v 
 v  1
64 another example of mean-field critical behavior in an exactly

v  1
16  

 solvable driven system, just as in the case of the DLG in a
 v  1
4 

10
1  v 1 

 certain limit 关30兴.
 v 4 



 v  16 

The analytic results were reproduced using a multiplica-


 v  64 


 tive Monte Carlo rate originally introduced in 关30兴, which
 v  256 


 eliminates correlations due to many particle dynamics intro-
 v  1024
10
2 v
duced by the common Metropolis and Glauber rates. We
claim that this algorithm is generally favorable to the Me-
10
1 100 101 102 103 tropolis and Glauber rates if an analytical treatment is con-
L
ΞeqTc v  sidered.
The finite-size effects naturally emerging in the simula-
FIG. 16. 共Color online兲 Velocity dependent crossover behavior tions were analyzed using finite-size scaling techniques, a
in the 2db case. Shown is the rescaled width of the critical region perfect agreement with exactly known universal finite-size
␦␶␰eq关Tc共v兲兴 as function of the crossover scaling variable scaling functions 关28兴 were found.
L储 / ␰eq关Tc共v兲兴 for several velocities v and different system sizes We analyzed the critical behavior at finite velocities and
L储 = 24 , . . . , 210 共see text兲. The inset shows the corresponding cross- studied the crossover behavior from low to high velocities:
over of the effective correlation length exponent ␯eff from ␯eff = 1 we found that the 1d system only has a phase transition in
共Ising, dashed line兲 to ␯eff = 2 共MF, dotted line兲. the thermodynamic limit for v = ⬁, while systems with finite
v always become noncritical at the crossover system size
tem, Tc共v兲, plays a key role: The system is Ising-like as long L⫻储 ⬇ 9 v . On the contrary, the two-dimensional case 2d b al-

as correlations span the whole system in both directions 储 ready has an Ising type phase transition at v = 0, which
and ⬜, i.e., as long as the system size L储 is of the order of the changes to mean-field behavior for all finite v ⬎ 0 in the
bulk correlation length ␰eq at the critical point Tc共v兲 of the thermodynamic limit, at a crossover length L⫻ 储 ⬇ 6 ␰ eq关T c共 v 兲兴.

driven system, leading to the crossover scaling variable In this sense, the velocity v is a relevant perturbation, always
L储 / ␰eq关Tc共v兲兴. Again, the rescaling of the y axis can be de- driving the system to a nonequilibrium state.
termined by requiring that a data collapse is obtained in the The 1d system changes from mean field to noncritical
limit v → 0, leading to the expression ␦␶␰eq关Tc共v兲兴, as ␰eq Ising universality, while the 2db case changes from Ising to
cancels in this case and we get the required condition mean-field type with growing system size L储. This somewhat
L储␦␶ = const, as ␰eq ⬀ ␶−␯eq in this limit and ␯eq = 1. puzzling fact can be understood in terms of the critical width
␯
The resulting crossover scaling plot is shown in Fig. 16. ␦␶ of the transition as follows: as in general ␦␶ ⬀ L−1/ 储 at
For all finite v ⬎ 0 the critical behavior changes from Ising to criticality, in the two-dimensional Ising case ␦␶ ⬀ L储 , while
−1

mean field at the crossover system size L⫻ 储 ⬇ 6 ␰ eq关T c共 v 兲兴:

in the mean-field case with one-dimensional boundary
below this value ␦␶ shrinks according to ␦␶ ⬀ L−1 储 共Ising, ␦␶ ⬀ L−1/2
储 . Third, ␦␶ ⬀ L0储 = const in the 1d case at finite v. In
dashed line兲, while above this value ␦␶ ⬀ L−1/2 holds 共MF, the crossover the actual critical width ␦␶ is always governed
储
dotted line兲. As the shift exponent ␪ at small velocities de- by the largest contribution and so at sufficiently large system
fined by size L储 the contribution with smallest ␯−1 dominates and de-
termines the critical behavior. As consequence in both cases
Tc共v兲 − Tc共0兲 ⬀ v␪ 共75兲 the effective inverse correlation length exponent ␯eff −1
changes
⫻ −␪ from a larger value at small L储 to a smaller value at large L储
is close to 1/2 we have, for small v, L储 ⬀ v ⬇ v . The −1/2
共1 / 2 → 0 in the 1d case, 1 → 1 / 2 in the 2db case兲.
shift exponent ␪ = 1 / 2 has also been found in a field theoret-
Comparing the results to the driven lattice gas 共DLG兲
ical calculation of the 2 + 1d system 关29兴.
关10兴, we note that the DLG also shows a continuous nonequi-
librium phase transition from an ordered to a disordered state
V. SUMMARY at a critical temperature which grows with growing driving
field. However, in the DLG the particle number is conserved,
In this work we investigated a recently proposed driven while we deal with a nonconserved magnetization.
Ising model with friction due to magnetic correlations. The Finally some remarks on strongly anisotropic critical be-
nonequilibrium phase transition present in this system was havior: the sheared system denoted 1 + 1d shows strongly
investigated in detail using analytical methods and Monte anisotropic behavior at criticality and v → ⬁, with strong evi-
Carlo simulations. In the far from equilibrium limit of high dence for the correlation length exponents ␯储 = 3 / 2 and ␯⬜

061138-11
ALFRED HUCHT PHYSICAL REVIEW E 80, 061138 共2009兲

= 1 / 2, details on this will be published elsewhere 关20兴. Re-

markably, this is a rare case of an exactly solvable nonequi-
共0兲
␹b,eq 共z兲 = 冉 冊
1
z2
−1

冋 册
librium system with strongly anisotropic critical behavior.
K共16w2兲 E共16w2兲 1
⫻ 共1 + 2w − 8w2兲 − − ,
ACKNOWLEDGMENTS 4␲w 4␲w 4
Special thanks go to Dietrich E. Wolf for very valuable 共A3兲
discussions, criticism, and comments within the framework
of the Sonderforschungsbereich 616, “Energy Dissipation at with the abbreviations
Surfaces.” Thanks also to Sebastian Angst, Lothar Brendel,
and Felix Schmidt for helpful discussions and to Sven
z共1 − z2兲
Lübeck for critical reading of the paper. w= , 共A4a兲
共1 + z2兲2
APPENDIX A: SURFACE MAGNETIZATION OF THE 2D
ISING MODEL
1 − 2z − z2
The equilibrium surface magnetization mb,eq of the 2D a= , 共A4b兲
1 + z2
Ising model in a static surface field hb obtained by McCoy
and Wu 共关16兴, Chapter VI, Eq. 5.1兲 as well as the reduced
zero-field boundary susceptibility 1 + 2z − z2

冏 冏
b= , 共A4c兲
共0兲 ⳵ mb,eq 1 + z2
␹b,eq = 共A1兲
⳵ hb hb→0

can be written in closed form not present in the literature yet 2z

共0兲
关31兴. ␹b,eq is sometimes denoted ␹11, and a high-temperature c= , 共A4d兲
1 + z2
series expansion was derived up to tenth order in Ref. 关32兴
and up to 23th order in Ref. 关33兴. As the expressions for
anisotropic couplings K储 and K⬜ become way too compli-
cated, we only give the results for the isotropic Ising model
with K储 = K⬜ = K here: using the definitions z = tanh K,
Y= 冉 az
c2y 2
+1 冊冉 by 2
c 2z
−1 冊 −1
, 共A4e兲

y = tanh hb we find

冋
and the complete elliptic integrals 关37兴 of the first, second,
b2 共a + 兲
y2 2
z−1 − z b2 z and third kind, K共m兲, E共m兲, and ⌸共n , m兲. Note that the vari-
mb,eq共z,y兲 = K共16w2兲 +
z
− y
2␲ 4␲w 1 − by2 able w is also used in high-temperature series analysis of the

册
y z
bulk zero-field susceptibility 关21兴. For hb = 0 the surface mag-

冉 冊
c 2z

共1 − bzy 兲2 netization Eq. 共A2兲 reduces to the well-known expression

2
Y 1/2 − Y −1/2 1
⫻⌸ 2 ,16w2 + − ,
by 2共z−1 − z兲
冑
1− 4
c 2z
cosh 2K − coth 2K
共A2兲 mb,eq共K兲 = . 共A5兲
cosh 2K − 1

关1兴 C. Fusco, D. E. Wolf, and U. Nowak, Phys. Rev. B 77, 174426 关9兴 A. Onuki, J. Phys.: Condens. Matter 9, 6119 共1997兲.
共2008兲. 关10兴 S. Katz, J. L. Lebowitz, and H. Spohn, Phys. Rev. B 28, 1655
关2兴 M. P. Magiera, L. Brendel, D. E. Wolf, and U. Nowak, EPL 共1983兲.
87, 26002 共2009兲. 关11兴 B. Schmittmann and R. K. P. Zia, in Phase Transitions and
关3兴 M. P. Magiera, D. E. Wolf, L. Brendel, and U. Nowak, IEEE Critical Phenomena, edited by C. Domb and J. L. Lebowitz
Trans. Magn. 45, 3938 共2009兲. 共Academic Press, London, 1995兲, Vol. 17.
关4兴 D. Kadau, A. Hucht, and D. E. Wolf, Phys. Rev. Lett. 101, 关12兴 M. B. Stearns, in Magnetic Properties of Metals, Landolt-
137205 共2008兲. Börnstein, New Series, Vol. III/19a, edited by H. P. J. Wijn
关5兴 C. K. Chan and L. Lin, Europhys. Lett. 11, 13 共1990兲. 共Springer, Berlin, 1986兲.
关6兴 F. Corberi, G. Gonnella, and A. Lamura, Phys. Rev. Lett. 81, 关13兴 M. P. Allen and D. J. Tildesley, Computer Simulation of Liq-
3852 共1998兲. uids 共Clarendon Press, Oxford, 1987兲.
关7兴 E. N. M. Cirillo, G. Gonnella, and G. P. Saracco, Phys. Rev. E 关14兴 L. Onsager, Phys. Rev. 65, 117 共1944兲.
72, 026139 共2005兲. 关15兴 R. J. Baxter, Exactly Solved Models in Statistical Mechanics
关8兴 T. Imaeda and K. Kawasaki, Prog. Theor. Phys. 73, 559 共Academic Press, London, 1982兲.
共1985兲. 关16兴 B. M. McCoy and T. T. Wu, The Two-Dimensional Ising Model

061138-12
NONEQUILIBRIUM PHASE TRANSITION IN AN EXACTLY… PHYSICAL REVIEW E 80, 061138 共2009兲

共Harvard University Press, Cambridge, 1973兲. 关27兴 W. Kwak, D. P. Landau, and B. Schmittmann, Phys. Rev. E
关17兴 A. M. Ferrenberg and D. P. Landau, Phys. Rev. B 44, 5081 69, 066134 共2004兲.
共1991兲. 关28兴 D. Grüneberg and A. Hucht, Phys. Rev. E 69, 036104 共2004兲.
关18兴 N. Ito, K. Hukushima, K. Ogawa, and Y. Ozeki, J. Phys. Soc. 关29兴 G. Gonnella and M. Pellicoro, J. Phys. A 33, 7043 共2000兲.
Jpn. 69, 1931 共2000兲. 关30兴 H. van Beijeren and L. S. Schulman, Phys. Rev. Lett. 53, 806
关19兴 A. Hucht, J. Phys. A 35, L481 共2002兲.
共1984兲.
关20兴 S. Angst, A. Hucht, and D. E. Wolf 共unpublished兲.
关31兴 B. M. McCoy 共private communication兲.
关21兴 S. Boukraa, A. J. Guttmann, S. Hassani, I. Jensen, J.-M. Mail-
lard, B. Nickel, and N. Zenine, J. Phys A: Math. Theor. 41, 关32兴 K. Binder and P. C. Hohenberg, Phys. Rev. B 6, 3461 共1972兲.
455202 共2008兲, URL http://www.ms.unimelb.edu.au/~iwan/ 关33兴 I. G. Enting and A. J. Guttmann, J. Phys. A 13, 1043 共1980兲.
ising/Ising_ser.html 关34兴 M. E. J. Newman and G. T. Barkema, Monte Carlo Methods in
关22兴 K. Binder and P. C. Hohenberg, Phys. Rev. B 9, 2194 共1974兲. Statistical Mechanics 共Clarendon Press, Oxford, 1999兲.
关23兴 H. Arisue and T. Fujiwara, Nucl. Phys. B 共Proc. Suppl.兲 119, 关35兴 TM calculations for stripes of width L⬜ ⬎ 1 show that this is
855 共2003兲; http://www.rccp.tsukuba.ac.jp/kenkyukai/asia- only the case for L⬜ = 1.
pacific/program/transparency/arisue/[email protected]. 关36兴 Note that Eq. 共69兲 is mentioned in the literature 关30,34兴 with-
关24兴 R. J. Glauber, J. Math. Phys. 4, 294 共1963兲. out stressing the multiplicative property 关Eq. 共68兲兴.
关25兴 K. Kawasaki, Phys. Rev. 145, 224 共1966兲. 关37兴 We use the definition of elliptic functions without square, e.g.,
关26兴 N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H.
K共m兲 = 兰␲0 /2共1 − m sin2 ␪兲−1/2d␪.
Teller, and E. Teller, J. Chem. Phys. 21, 1087 共1953兲.

061138-13