PRAMANA c Indian Academy of Sciences

° Vol. 64, No. 5

— journal of May 2005
physics pp. 695–707

Fluctuations and large deviations in

non-equilibrium systems
B DERRIDA
Laboratoire de Physique Statistique, Ecole Normale Supérieure, 24 rue Lhomond,
75231 Paris Cedex 05, France
E-mail: derrida@lps.ens.fr

Abstract. For systems in contact with two reservoirs at different densities or with two
thermostats at different temperatures, the large deviation function of the density gives
a possible way of extending the notion of free energy to non-equilibrium systems. This
large deviation function of the density can be calculated explicitly for exclusion models in
one dimension with open boundary conditions. For these models, one can also obtain the
distribution of the current of particles flowing through the system and the results lead to
a simple conjecture for the large deviation function of the current of more general diffusive
systems.

Keywords. Non-equilibrium systems; large deviations; current fluctuations.

PACS Nos 02.50.-r; 05.40.-a; 05.70.Ln; 82.20.-w

1. Introduction

The goal of this talk is to give a short review on results [1–8] obtained recently on
the steady state of non-equilibrium systems such as systems in contact with two
heat baths at different temperatures Ta and Tb .
When Ta = Tb = T , i.e. when the system is in equilibrium, one knows that each
microscopic configuration C of energy E(C) is occupied according to its Boltzmann
weight
· ¸
1 E(C)
P (C) = exp − . (1)
Z T
One can then define the free energy
à · ¸!
X E(C)
F = −T log Z = −T log exp − (2)
T
C

from which macroscopic properties such as the average energy or the specific heat
can be obtained.

695
 B Derrida

When Ta 6= Tb , the system reaches in the long time limit a steady state, but
there is no longer a general expression such as (1) for the steady state probabilities
of microscopic configurations and there is no a priori definition of the free energy.
Moreover new questions arise related to the flow of energy Qt through the system
during time t, i.e. the energy Qt transferred
³ from
´ one reservoir to the other or
1 1
equivalently the entropy production Qt Tb − Ta . One can for example try to
determine the probability distribution P (Qt ) of this energy flow.

2. Large deviation function of the density

One way of extending the notion of free energy to non-equilibrium systems is to

define the large deviation function [9] of the density. If one considers a box of
volume V containing V r particles, the probability Pv (n) of finding n particles in a
subvolume v has the following v dependence for large v,
h ³ n ´i
Pv (n) ∼ exp −va , (3)
v
where a(ρ) is the large deviation function at density ρ. Figure 1 shows the typical
shape of a(ρ) for a homogeneous system (i.e. not at a coexistence between different
phases) with a single minimum at r = 0 where a(r) = 0.
The large deviation function a(ρ) can be defined for equilibrium systems as well
as for non-equilibrium systems. For equilibrium systems, one can show that a(ρ)
is closely related to the free energy: if the volume v is sufficiently large and if the
interactions are short ranged, the large deviation function a(ρ) is given by
f (ρ) − f (r) − (ρ − r)f 0 (r)
a(ρ) = , (4)
kT
where f (ρ) is the free energy per unit volume. This can be seen by noticing that
for v ¿ V and when v 1/d is much larger than the range of the interactions
Zv (n) ZV −v (N − n) d−1
Pv (n) = exp[O(v d )], (5)
ZV (N )

a (ρ)

ρ
r
Figure 1. A typical shape of the large deviation function a(ρ).

696 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems
d−1
where the term exp[O(v d )] represents the interactions between pairs of particles,
one of which is the volume v and the other one in V − v. Then taking the log of
(5) and using the fact that
log Zv (vρ) f (ρ)
lim =− (6)
v→∞ v kT
leads to (4).

3. Large deviation of the current

One can define in a similar way the large deviation function of the current of
particles for a system in contact with two reservoirs at densities ρa and ρb as in
figure 2 (or of the heat current for a system in contact with two thermostats). The
probability of observing during a long time t an average current j takes the form
µ ¶
Qt
Pro = j ∼ e−tF (j) , (7)
t
where F (j) is the large deviation function of the current j. This large deviation
function F (j) has a shape similar to a(ρ), with a minimum at some value j, the
typical value where F (j) = 0.
The knowledge of the large deviation function determines, by expanding around
j, all the cumulants of Qt for large t.

4. The symmetric simple exclusion process (SSEP)

There are only few examples of non-equilibrium steady states for which the large
deviation function a(ρ) of the density or F (j) of the current can be calculated
[2,4–8]. One of the simplest cases for which the calculations can be done is the
symmetric simple exclusion process [10–12] shown in figure 3.

Qt

ρ SYSTEM ρ
a b

Figure 2. System in contact with two reservoirs of particles at densities ρa

and ρb .

γ δ

α 1 1 1 1 β

1 L

Figure 3. The symmetric simple exclusion process.

Pramana – J. Phys., Vol. 64, No. 5, May 2005 697

 B Derrida

The model is defined as a one-dimensional lattice of L sites with open bound-

aries, each site being either empty or occupied by a single particle. During every
infinitesimal time interval dt, each particle has a probability dt of jumping to the
left if the neighboring site on its left is empty, dt of jumping to the right if the
neighboring site on its right is empty. At the two boundaries the dynamics is mod-
ified to mimic the coupling with reservoirs of particles: at the left boundary, during
each time interval dt, a particle is injected on site 1 with probability α dt (if this
site is empty) and a particle is removed from site 1 with probability γ dt (if this
site is occupied). Similarly on site L, particles are injected at rate δ and removed
at rate β.
One can show [4,5,8] that these choices of the rates α, γ, β, δ correspond to the
left boundary being connected to a reservoir at density ρa and the right boundary
to a reservoir at density ρb with ρa and ρb given by
α δ
ρa = ; ρb = . (8)
α+γ β+δ
For the SSEP, it is easy to determine the steady state profile: if τi = 0 or 1 is a
binary variable indicating whether site i is occupied or empty, the time evolution
of the average occupation hτi i is given by

dhτ1 i
= α − (α + γ + 1)hτ1 i + hτ2 i, (9)
dt
dhτi i
= hτi−1 i − 2hτi i + hτi+1 i, for 2 ≤ i ≤ L − 1 (10)
dt
dhτL i
= hτL−1 i − (1 + β + δ)hτL i + δ, (11)
dt

and the steady state density profile (obtained by writing that dhτi i = 0) is [5]
dt
1 1
ρa (L + β+δ − i) + ρb (i − 1 + α+γ)
hτi i = 1 1 . (12)
L+ α+γ + β+δ −1

The average current

ρa − ρb ρa − ρb
j = hτi − τi+1 i = 1 1 ' (13)
L+ α+γ + β+δ − 1 L

is proportional to the gradient of the density (with a coefficient of proportionality

which is simply 1 here) and therefore follows Fick’s law.

5. Current fluctuations in the SSEP

In the long time limit the generating function of the total charge Qt transferred
from the left reservoir to the system during time t grows exponentially with time

698 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems
­ ®
eλQt ∼ eµ(λ)t , (14)
where µ(λ) is related to the large deviation function F (j) by a Legendré transfor-
mation
µ(λ) = max[λj − F (j)]. (15)
j

Because the evolution is Markovian, µ(λ) can be determined as the largest eigen-
value of a certain matrix [2]: if Pt (C) is the probability of finding the system in a
configuration C and M (C, C 0 )dt is the probability of transition from configuration
C 0 to configuration C during time dt, the evolution of Pt (C) is given by
dPt (C) X
= M (C, C 0 )Pt (C 0 ). (16)
dt 0 C

Among the matrix elements M (C, C 0 ), some correspond to exchanges of particles

with the left reservoir and some represent internal moves in the bulk or exchanges
with the right reservoir. One can decompose the matrix M (C, C 0 ) into three matrices
M (C, C 0 ) = M1 (C, C 0 ) + M0 (C, C 0 ) + M−1 (C, C 0 ), (17)
where the index is the number of particles transferred from the left reservoir to the
system during time dt, when the system jumps from the configuration C 0 to the
configuration C. One can then show [2] that µ(λ) is simply the largest eigenvalue
of the matrix
eλ M1 + M0 + e−λ M−1 . (18)
The steady state weights P (C) for the SSEP are known exactly [4,5,13]: they
determine the eigenvector of the matrix eλ M1 + M0 + e−λ M−1 associated to the
eigenvalue µ(λ) when λ = 0.
In [2] a perturbation theory was developed to calculate µ(λ) in powers of λ. The
main outcome of this perturbation theory is that µ(λ), which in principle depends
on L, λ and on the four parameters α, β, γ, δ, takes for large L a simple form
µ ¶
1 1
µ(λ) = R(ω) + O , (19)
L L2
where ω is defined as
ω = (eλ − 1)ρa + (e−λ − 1)ρb − (eλ − 1)(1 − e−λ )ρa ρb . (20)
The perturbation theory gives up to fourth order in ω,
ω2 8ω 3 4ω 4
R(ω) = ω − + − + O(ω 5 ). (21)
3 45 35
The fact that µ(λ) depends only on ρa , ρb and λ through the single parameter
ω is the outcome of the calculation, but so far there is no physical explanation
why it is so. However, ω remains unchanged under a number of symmetries [2]
(left–right, particle–hole, Gallavotti–Cohen [14,15] symmetry) implying that µ(λ)
remains unchanged as it should under these symmetries.
From the knowledge of R(ω) up to fourth order in ω, one can determine the first
four cumulants of the integrated current Qt for arbitrary ρa and ρb :

Pramana – J. Phys., Vol. 64, No. 5, May 2005 699

 B Derrida

• For ρa = 1 and ρb = 0, one finds

µ ¶
hQt i 1 1
= +O , (22)
t L L2
µ ¶
hQ2t ic 1 1
= +O , (23)
t 3L L2
µ ¶
hQ3t ic 1 1
= +O , (24)
t 15L L2
µ ¶
hQ4t ic −1 1
= +O . (25)
t 105L L2
Surprisingly these cumulants are the same as the ones known for a differ-
ent problem of current flow: the case of non-interacting fermions through a
mesoscopic disordered conductor [16–18].
• For ρa = ρb = 21 which corresponds to an equilibrium case with the same
density 1/2 in the two reservoirs, one finds that all odd cumulants vanish as
they should and that
µ ¶
hQ2t ic 1 1
= +O , (26)
t 2L L2
µ ¶
hQ4t ic 1
=O . (27)
t L2

6. A conjecture for the distribution of current

Because µ(λ) depends on the parameters ρa , ρb and λ through the single parameter
ω, if one knows µ(λ) for one single choice of ρa and ρb , then (20),(21) determine
µ(λ) for all other choices of ρa , ρb .
In [2], it was conjectured that for the particular case ρa = ρb = 21 , not only
the fourth cumulant vanishes as in (27), but also all the higher cumulants vanish,
so that the distribution of Qt is Gaussian (to leading order in 1/L). This fully
determines the function R(ω) to be
£ ¡√ √ ¢¤2
R(ω) = log 1 + ω + ω . (28)
One can then check that, with this expression of R(ω), all the higher cumulants of
Qt in the case ρa = 1 and ρb = 0 coincide with those of fermions through mesoscopic
conductors [2,17].

7. The additivity principle

One can formulate another conjecture, the additivity principle [1], based on a sim-
pler physical interpretation, which leads to the same expressions (20),(28) as pre-
dicted by the previous conjecture and can be generalized to other diffusive systems.

700 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems

We have seen that, for a system of length L + L0 in contact with two reservoirs
of particles at densities ρa and ρb , the probability of observing, during a long time
t, an integrated current Qt = jt has the following form

ProL+L0 (j, ρa , ρb ) ∼ e−tFL+L0 (j,ρa ,ρb ) . (29)

The idea of the additivity principle is to relate the large deviation function
FL+L0 (j, ρa , ρb ) of the current to the large deviation functions of subsystems by
writing that for large t,

ProL+L0 (j, ρa , ρb ) ∼ max [ProL (j, ρa , ρ) × ProL0 (j, ρ, ρb )] . (30)

ρ

This means that the probability of transporting a current j over a distance L + L0

between two reservoirs at densities ρa and ρb is the same (up to boundary effects
which give for large L subleading contributions) as the probability of transporting
the same current j over a distance L between two reservoirs at densities ρa and
ρ times the probability of transporting the current j over a distance L0 between
two reservoirs at densities ρ and ρb . One can then argue that choosing the optimal
ρ makes this probability maximum. From (30) one gets the following additivity
property of the large deviation function

FL+L0 (j, ρa , ρb ) = max [FL (j, ρa , ρ) + FL0 (j, ρ, ρb )] . (31)

ρ

Suppose we consider a diffusive system for which we know the two functions D(ρ)
and σ(ρ) defined as follows:

1. For ρa − ρb small,
hQt i ρ a − ρb
lim = D(ρa ) . (32)
t→∞ t L

2. For ρa = ρb ,

hQ2t i σ(ρa )
lim = . (33)
t→∞ t L

If one accepts the additivity property (31) of the large deviation function, one can
cut the system into more and more pieces so that [1] when these pieces have length
L dx, one gets
·Z 1 ¸
1 [jL + D(ρ)ρ0 ]2
FL (j, ρa , ρb ) = max dx , (34)
L ρ(x) 0 2σ(ρ)

where the optimal ρ(x) should satisfy ρ(0) = ρa and ρ(1) = ρb .

Using the fact that for the SSEP D(ρ) = 1 and σ(ρ) = 2ρ(1 − ρ), one can recover
expression (28) from (34) [1]. For more general diffusive systems, one can also
determine all the cumulants of the integrated current for arbitrary ρa and ρb . For
example

Pramana – J. Phys., Vol. 64, No. 5, May 2005 701

 B Derrida

hQ4t ic 1 3(5I4 I12 − 14I1 I2 I3 + 9I23 )

= , (35)
t L I15
where
Z ρa
In = D(ρ)σ(ρ)n−1 dρ. (36)
ρb

Recently, (34) was also derived in the context of fermions in mesoscopic conductors
[18]. It has been checked for weakly interacting lattice gases [19]. It was also
shown [20] that there are restrictions on σ(ρ) and D(ρ) for (34) to be valid and
that, as FL (j, ρa , ρb ) defined in (29) should be a convex function of j, for some
σ(ρ) and D(ρ) one has to replace the expression (34) of FL (j, ρa , ρb ) by its convex
envelope.

8. Free energy functional

If one divides a system of linear size L into n boxes of linear size l (in dimension d,
one has n = Ld /ld such boxes), one can try to determine the probability of finding
a certain density profile {ρ1 , ρ2 , . . . , ρn }, i.e. the probability of seeing ld ρ1 particles
in the first box, ld ρ2 particles in the second box,. . . , ld ρn in the nth box. For large
L one expects the following L dependence of this probability

Pro(ρ1 , . . . , ρn ) ∼ exp[−Ld F(ρ1 , ρ2 , . . . , ρn )], (37)

where F is a large deviation function which generalizes a(ρ) introduced in (3). If

one defines a reduced coordinate ~x by

~r = L~x (38)

and if one takes the limit l → ∞ with l ¿ L so that the number of boxes becomes
infinite, one can define a functional F(ρ(~x)) for an arbitrary density profile ρ(~x),

Pro(ρ(~x)) ∼ exp[−Ld F(ρ(~x))]. (39)

1 2

Figure 4. One specifies the density ρi in each box i.

702 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems

At equilibrium the functional F can be expressed in terms of the free energy f (ρ)
per unit volume (6) at density ρ: if one considers V r particles in a volume V = Ld ,
one can show that for systems with short range pair interactions
· µ d ¶¸
Zv (vρ1 ) . . . Zv (vρn ) L
Pro(ρ1 , . . . , ρn ) = exp O , (40)
ZV (V r) l

where v = ld . Comparing with (37), in the limit L → ∞, l → ∞, keeping n fixed

gives
n
1 1X
F(ρ1 , ρ2 , . . . , ρn ) = [f (ρi ) − f (r)]. (41)
kT n i=1

In the limit of an infinite number of boxes, this becomes

Z
1
F(ρ(~x)) = d~x [f (ρ(~x)) − f (r)]. (42)
kT
Thus the large deviation functional F is fully determined by the knowledge of the
free energy f (ρ) per unit volume. From (42), we see that this functional is a local
functional of ρ(~x). It is also convex (as the free energy is a convex function of ρ). By
expanding (42) around the most likely profile ρ(~x) = r, one obtains from (42) that
the fluctuations of the density profile are Gaussian with a variance related to the
second derivative of the free energy, i.e. related to the compressibility as predicted
by Smoluchowski and Einstein. Lastly one knows (by the Landau argument) that,
with short range interactions, there is no phase transition in one dimension.
For non-equilibrium systems, the large deviation functional of the density has
been calculated so far in very few cases [4–8,19,21]. For the asymmetric simple
exclusion process (ASEP) with open boundaries, which is an extension of the SSEP
discussed in §4, where the hopping rates in the bulk become asymmetric (a rate
1 to jump to the right and a rate q to jump to the left), the exact expression of
F(ρ(~x)) obtained in [6,7] shows that, depending on q, ρa , ρb , the functional may
be non-local, non-convex. It was also shown that for the ASEP, the fluctuations
of the density are non-Gaussian [3,6,7] and are not related to the expansion of the
functional around the most likely profile. It is also well-known that non-equilibrium
systems may exhibit phase transitions in one dimensions [13,22–25].
For the SSEP (in one dimension), the functional F(ρ(x)) is given by the following
exact expression:

• at equilibrium, i.e. for ρa = ρb = r,

Z 1
F(ρ(x)) = B(ρ(x), r)dx, (43)
0

where
1−ρ ρ
B(ρ, r) = (1 − ρ) log + ρ log . (44)
1−r r

Pramana – J. Phys., Vol. 64, No. 5, May 2005 703

 B Derrida

This can be derived easily. When ρa = ρb = r, the steady state is a Bernoulli

measure where all the sites are occupied independently with probability r.
Therefore if one divides a chain of length L into L/l intervals of length l, one
has
L/l
Y l!
Pro(ρ1 , . . . , ρn ) = rlρi (1 − r)l(1−ρi ) (45)
i
[lρi ]![l(1 − ρi )]!

and using Stirling’s formula one gets (43), (44).

• In the non-equilibrium case, i.e. for ρa 6= ρb , it was shown in [4,5,21] that
Z 1 · ¸
F 0 (x)
F(ρ(x)) = dx B(ρ(x), F (x)) + log , (46)
0 ρb − ρa

where the function F (x) is the monotone solution of the differential equation
F (1 − F )F 00
ρ(x) = F + (47)
F 02

satisfying the boundary conditions F (0) = ρa and F (1) = ρb .

This expression shows that F is a non-local functional of the density profile
ρ(x) as F (x) depends (in a non-linear way) on the profile ρ(y) at all points y.
For example if the difference ρa − ρb is small, one can expand F and obtain
an expression where the non-local character of the functional is clearly visible
Z 1
(ρa − ρb )2
F= dx B(ρ(x), ρ(x)) +
0 [ρa (1 − ρa )]2
Z 1 Z 1
× dx dy x(1 − y)(ρ(x) − ρ(x))(ρ(y)− ρ(y)) + O(ρa − ρb )3 .
0 x

Here ρ(x) is the most likely profile given by

ρ(x) = (1 − x)ρa + xρb . (48)

9. Derivation of the functional for the SSEP

The expressions (46), (47) have been derived by two different approaches:
• The matrix approach: For exclusion processes in one dimension with open bound-
ary conditions, the weights of all microscopic configurations are known exactly
[13,26,27]. In [13] it was shown that the probability of a microscopic configu-
ration {τ1 , τ2 , . . . , τL } can be written as the matrix element of a product of L
matrices
hW |X1 X2 . . . XL |V i
Pro(τ1 , τ2 , . . . , τL ) = , (49)
hW |(D + E)L |V i

704 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems

where the matrix Xi depends on the occupation τi of site i,

Xi = τi D + (1 − τi )E (50)

and the matrices D and E satisfy the following algebraic rules:

DE − ED = D + E,
hW |(αE − γD) = hW |,
(βD − δE)|V i = |V i. (51)

Let us introduce the following left and right eigenvectors of the operators ρa E −
(1 − ρa )D and (1 − ρb )D − ρb E,

hρa , a| [ρa E − (1 − ρa )D] = ahρa , a|,

[(1 − ρb )D − ρb E] |ρb , bi = b|ρb , bi. (52)

It is easy to see, using the definition (8), that the vectors hW | and |V i are given
by

hW | = hρa , (α + γ)−1 |,
|V i = |ρb , (β + δ)−1 i. (53)

It is then possible to show, using (51) that DE − ED = D + E and the definition

of the eigenvectors (52) that (for ρb < ρa )
I
hρa , a|Y1 Y2 |ρb , bi dρ (ρa − ρb )a+b
=
hρa , a|ρb , bi 2iπ (ρa − ρ)a+b (ρ − ρb )
ρb <|ρ|<ρa

hρa , a|Y1 |ρ, bi hρ, 1 − b|Y2 |ρb , bi

× , (54)
hρa , a|ρ, bi hρ, 1 − b|ρb , bi

where Y1 and Y2 are arbitrary polynomials of matrices D and E. (To prove (54)
0
it is sufficient to establish it when both Y1 and Y2 are of the form E n Dn as
any polynomial can be reduced to a sum of such terms by the relation DE −
ED = D + E. One can also, and this is easier, prove (54) for Y1 of the form
0 00 000
[ρa E − (1 − ρa )D]n [D + E]n and Y2 of the form [D + E]n [(1 − ρb )D − ρb E]n
and show using DE − ED = D + E that any polynomial Y1 or Y2 can be reduced
to a finite sum of such terms.)
To calculate the large deviation functional for a given profile ρ(x), one
needs to sum the weights of all microscopic configurations consistent with
this profile. One way of doing it [4,5] consists in breaking the system into
subsystems and in using (54) to relate the density functional of a large sys-
tem to the functionals of its subsystems (when the subsystems are them-
selves large enough, one can evaluate the integral in (54) by a saddle point
method).

Pramana – J. Phys., Vol. 64, No. 5, May 2005 705

 B Derrida

• The macroscopic fluctuation theory: Bertini et al [21,28,29] have developed a

different and more general approach to calculate this large deviation functional.
Their idea is that to observe a certain density profile ρ(x) at time t, one has
to find out how this fluctuation is produced. For large L, one has to find the
optimal path ρ(x, s) for −∞ < s < t in the space of profiles which goes from the
typical profile ρ(x) to the desired profile ρ(x). In other words,
( Z t Z 1 )
dρ 2
[j + D(ρ) dx ]
Pro(ρ(x)) ∼ max exp −L ds dx , (55)
ρ(x,s) −∞ 0 2σ(ρ)

where the current j(x, s) is related to the density profile ρ(x, s) by

dρ(x, s) dj(x, s)
=− (56)
ds dx

and the optimal path ρ(x, s) satisfies

ρ(x, −∞) = ρ(x),

ρ(x, t) = ρ(x).

So far it has not been possible to find the explicit expression of the functional F
for general D(ρ) and σ(ρ). For the SSEP, however, this approach gives (46), (47).
It also leads to the same expression of F as found by the matrix approach [8] in
the weak asymmetric ASEP.

10. Conclusion

The results on the distribution of current flowing through a system in its non-
equilibrium steady state are so far rather limited. It would be nice to calculate
µ(λ) exactly for a SSEP of finite length L and to justify that way the conjecture
(28). It would also be interesting to establish under what conditions on σ(ρ) and
D(ρ), the additivity principle [1] is valid.
Extending the results on the current fluctuations to the case of a strong asym-
metry, the case of several species of particles or situations where more than one
quantity is conserved (numbers of particles, energy, momentum..) would also be
very interesting.
Lastly it would be nice to obtain an explicit expression of the functional F(ρ(x))
for general σ(ρ) and D(ρ) or for models with several species of particles [25].

Acknowledgements

The author would like to thank T Bodineau, B Douçot, C Enaud, J L Lebowitz,

P E Roche and E R Speer. Most results reported in this talk were obtained in
a series of works done in collaboration with them. The author would also like to
thank E Brunet, M R Evans, V Hakim, D Mukamel and V Pasquier for useful
discussions.

706 Pramana – J. Phys., Vol. 64, No. 5, May 2005

 Fluctuations and large deviations in non-equilibrium systems

References

[1] T Bodineau and B Derrida, Phys. Rev. Lett. 92, 180601 (2004)
[2] B Derrida, B Douçot and P E Roche, J. Stat. Phys. 115, 717–748 (2004)
[3] B Derrida, C Enaud and J L Lebowitz, J. Stat. Phys. 115, 365–382 (2004)
[4] B Derrida, J L Lebowitz and E R Speer, Phys. Rev. Lett. 87, 150601 (2001)
[5] B Derrida, J L Lebowitz and E R Speer, J. Stat. Phys. 107, 599–634 (2002)
[6] B Derrida, J L Lebowitz and E R Speer, Phys. Rev. Lett. 89, 030601 (2002)
[7] B Derrida, J L Lebowitz and E R Speer, J. Stat. Phys. 110, 775–810 (2003)
[8] C Enaud and B Derrida, J. Stat. Phys. 114, 537–562 (2004)
[9] C Kipnis, S Olla and S R S Varadhan, Commun. Pure Appl. Math. 42, 115–137
(1989)
[10] H Spohn, Large scale dynamics of interacting particles (Springer-Verlag, Berlin, 1991)
[11] T M Liggett, Stochastic interacting systems: contact, voter and exclusion processes,
324, (Springer, 1999)
[12] D Chowdhury, L Santen and A Schadschneider, Phys. Rep. 329, 199–329 (2000)
[13] B Derrida, M R Evans, V Hakim and V Pasquier, J. Phys. A26, 1493–1517
(1993)
[14] G Gallavotti and E D G Cohen, J. Stat. Phys. 80, 931–970 (1995)
[15] J L Lebowitz and H Spohn, J. Stat. Phys. 95, 333–366 (1999)
[16] Y M Blanter and M Buttiker, Phys. Rep. 336, 1–166 (2000)
[17] H Lee, L S Levitov and A Yu Yakovets, Phys. Rev. B51, 4079–4083 (1995)
[18] A N Jordan, E V Sukhorukov and S Pilgram, J. Math. Phys. 45, 4386–4417 (2004),
cond-mat/0401650
[19] F van Wijland and Z Rácz, J. Stat. Phys. 118, 27–54 (2005), cond-mat/0404358
[20] L Bertini, A De Sole, D Gabrielli, G Jona-Lasinio and C Landim, Phys. Rev. Lett.
94, 030601 (2005), cond-mat/0407161
[21] L Bertini, A De Sole, D Gabrielli, G Jona-Lasinio and C Landim, J. Stat. Phys. 107,
635–675 (2002)
[22] J Krug, Phys. Rev. Lett. 67, 1882–1885 (1991)
[23] M R Evans, Braz. J. Phys. 30, 42–57 (2000)
[24] Y Kafri, E Levine, D Mukamel, G M Schütz and J Torok, Phys. Rev. Lett. 89, 035702
(2002)
[25] M Clincy, B Derrida and M R Evans, Phys. Rev. E67, 066115 (2003)
[26] B Derrida, E Domany and D Mukamel, J. Stat. Phys. 69, 667–687 (1992)
[27] G M Schütz and E Domany, J. Stat. Phys. 72, 277–296 (1993)
[28] L Bertini, A De Sole, D Gabrielli, G Jona-Lasinio and C Landim, Phys. Rev. Lett.
87, 040601 (2001)
[29] L Bertini, A De Sole, D Gabrielli, G Jona-Lasinio and C Landim, J. Stat. Phys. 116,
831–841 (2004), cond-mat/0310072

Pramana – J. Phys., Vol. 64, No. 5, May 2005 707