Home Search Collections Journals About Contact us My IOPscience

The law of action and reaction for the effective force in a non-equilibrium colloidal system

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2006 J. Phys.: Condens. Matter 18 2825

(http://iopscience.iop.org/0953-8984/18/10/008)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 132.236.27.111
This content was downloaded on 28/05/2014 at 10:11

Please note that terms and conditions apply.

INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 18 (2006) 2825–2836 doi:10.1088/0953-8984/18/10/008

The law of action and reaction for the effective force in

a non-equilibrium colloidal system
Kumiko Hayashi1 and Shin-ichi Sasa
Department of Pure and Applied Sciences, University of Tokyo, Komaba, Tokyo 153-8902, Japan

E-mail: hayashi@jiro.c.u-tokyo.ac.jp

Received 25 October 2005, in final form 27 January 2006

Published 20 February 2006
Online at stacks.iop.org/JPhysCM/18/2825

Abstract
We study a non-equilibrium Langevin many-body system containing two ‘test’
particles and many ‘background’ particles. The test particles are spatially
confined by a harmonic potential, and the background particles are driven by an
external driving force. Using numerical simulations of the model, we formulate
an effective description of the two test particles in a non-equilibrium steady
state. In particular, we investigate several different definitions of the effective
force acting between the test particles. We find that the law of action and
reaction does not hold for the total mechanical force exerted by the background
particles, but that it does hold for the thermodynamic force defined operationally
on the basis of an idea used to extend the first law of thermodynamics to non-
equilibrium steady states.

1. Introduction

Colloids are microscopic particles suspended in liquid. Generally, they are of the order of
micrometres in size, and thus they are much larger than molecules but still small enough to
exhibit Brownian motion. Colloids are convenient to study experimentally because, unlike
molecules, they can be easily observed with ordinary microscopes. In recent years, as the
technology used in their manipulation has developed, experimental studies of colloids have
advanced greatly [1–6].
In colloidal many-body systems, melting, freezing, glass transitions and non-equilibrium
statistical mechanics have been studied. Among the studies of non-equilibrium steady states
(NESSs), that of Dzubeilla et al investigated a model that describes two fixed colloidal particles
(test particles) in an environment containing many driven particles (background particles).
They reported that in a NESS, the effective interaction forces they defined between the two
fixed particles violate the law of action and reaction [7]. This result implies that an effective
potential seems not to be constructed in NESS.
1 Author to whom any correspondence should be addressed.

0953-8984/06/102825+12$30.00 © 2006 IOP Publishing Ltd Printed in the UK 2825

2826 K Hayashi and S-I Sasa

Figure 1. Schematic representation of the model we study. Note that the background particles are
not driven in the equilibrium case.

The effective interaction force defined in [7] represents the two-body effect of the total
force exerted by the background particles, which is extracted by subtracting from it the total
force in the case when there are only colloidal particles. Although their definition seems
plausible because of its simplicity, we expect that there is room to define a different type
of effective force in NESS. With this motivation, we attempt to provide another idea for an
effective interaction force between the two fixed particles.
In this paper we reconsider effective forces in NESSs by studying a system similar to that
used in [7]. First, we review the concept of effective forces in equilibrium from three points
of view: statistical mechanics, mechanics and thermodynamics. Next, we investigate effective
forces for non-equilibrium systems by considering the results of numerical experiments. In
particular, we demonstrate that in the NESS we consider the law of action and reaction holds
for the effective force defined according to a conjectured thermodynamic relation applied to
NESSs.

2. Model

The model we study describes a system in which two ‘test’ particles are trapped side by side
in the centre of the system by a harmonic potential, and N ’background’ particles are driven by
an external force (see figure 1). The test particles are unaffected by the driving force, and the
background particles are unaffected by the trapping potential. Hereafter, we refer to the test
particles as particle 1 and particle 2.
For the sake of simplicity, we consider the idealized situation of a two-dimensional system
of length L x and L y in the x and the y directions, with periodic boundary conditions in both
directions. Then, letting R
1 and R
2 be the positions of particle 1 and particle 2, we consider
particle 2 fixed for simplicity, with particle 1 trapped by the harmonic potential
k
Vt ( R
1 ) = ( R
1 − R
c )2 , (1)
2
where R
c corresponds to the centre of the potential and k is a spring constant chosen to be
sufficiently large that particle 1 can be regarded as almost fixed. It is important to note that
although it may seem simpler to fix particle 1 the dynamical degrees of freedom of particle
1 must be taken into account when we consider the distribution function of the effective
description in which the degrees of freedom of the background particles are integrated. (If
particle 1 is also fixed, no variable remains in the distribution function of the effective
description (see the next section).) As a further simplification, it is assumed that there is no
direct interaction between particles 1 and 2 and that there is no direct interaction among the
background particles. Then, we stipulate that the interaction between particle 1 or particle 2
Effective forces in non-equilibrium 2827

and a background particle is represented by a short-ranged potential function Vint ( R
i − r
k ),
where r
k is the position of the k th background particle (k = 1, . . . , N ), of the form

 2 
 r

Vint (
r ) = V0 exp −2  . (2)
σ
With the system as described above, the motion of particle 1 is described by the Langevin
equation
N
∂ Vint ( R
1 − r
k )
γ R
˙1 = −k( R
1 − R
c ) − + ξ
(t), (3)
k=1 ∂ R
1

where γ is a friction constant and ξ
(t) = (ξx (t), ξ y (t)) represents Gaussian white noise that
satisfies
 
ξα (t)ξα (t  ) = 2γ T δ(t − t  )δα,α . (4)
Here, the Boltzmann constant is set to unity, and T (=1/β ) is the temperature of the
environment. The motion of the k th background particle is described by
2
∂ Vint( R
i − r
k )
γ r
˙k = f
− + ξ
k (t), (5)
i=1
∂ r
k

where f
= ( f, 0) is an external driving force and ξ
k (t) = (ξk,x (t), ξk,y (t)) is Gaussian white
noise that satisfies
 
ξk,α (t)ξk ,α (t  ) = 2γ T δk,k δ(t − t  )δα,α . (6)
Below, all the quantities are converted into dimensionless forms by normalizing σ , γ and
T to unity. We set the parameters used for our numerical experiments as L x = 11, L y = 6,
N = 200, k = 50, V0 = 7, 0
f
0.3, R
c = (3, 3) and R
2 = (4.5, 3). Also, all the
numerical results in this paper are obtained using a finite difference method of the Langevin
equations with a time step 2.5 × 10−4 .

3. Equilibrium case

3.1. Review of the effective description

Before analysing non-equilibrium cases, we review the effective description of particles 1 and
2 in equilibrium, in which the stationary distribution function of the model is the canonical
distribution
 N
exp{−β[Vt( R
1 ) + 2i=1 k=

1 Vint ( Ri − r
k )]}



pc ( R1 , {
rk }; R2 ) = , (7)
Z
where Z is a normalization constant. In the equilibrium case, employing the well-established
framework of statistical mechanics, we can derive the effective Hamiltonian of particles 1 and
2 by integrating (7) over the degrees of freedom of the background particles.
For the canonical distribution pc ( R
1 , {

rk }; R
2 ) given by (7) and the interaction given


by (2), there exists a function R1 − R2 , which we write Veff ( R
1 − R
2 ), satisfying the equation


 N




exp{−β[Veff ( R1 − R2 ) + Vt ( R1 )]} = dr
k pc ( R
1 , {

rk }; R
2 ). (8)
k=1

It is natural to refer to the function Veff ( R
1 − R
2 ) as the effective interaction potential. The fact
that this effective interaction potential depends only on the relative displacement R
1 − R
2 is
2828 K Hayashi and S-I Sasa

due to the spatial symmetry of the system. Then, as the left-hand side of (8) depends only on
R
1 and R
2 , it is interpreted as the steady-state distribution for the effective description:
Peff ( R
1 ; R
2 ) = exp{−β[Vt ( R
1 ) + Veff ( R
1 − R
2 )]}. (9)
Clearly the effective potential Veff ( R
1 − R
2 ) obtained in the above manner is properly defined,
as this Peff ( R
1 ; R
2 ) is in canonical form.
With the above argument establishing the proper definition of Veff ( R
1 − R
2 ), the statistical
mechanical force between particles 1 and 2 is defined naturally by
∂ Veff ( R
1 − R
2 )
F
istat ≡ − , (10)
∂ R
i
with i = 1, 2. Because Veff ( R
1 − R
2 ) is a function of the relative displacement R
1 − R
2 , we
immediately find the relation
F
1stat = − F
2stat , (11)
which represents the law of action and reaction for statistical mechanical forces in equilibrium.
The statistical mechanical force defined above is closely related to the mechanical force exerted
by all the background particles:
N
∂ Vint ( R
i − r
k )
F
imec ≡ . (12)
k=1
∂ r
k
Indeed, using (8) and (10), we can prove the relation

s
s
F
istat = F
imec , (13)
f =0 f =0

where  sf =0 represents the statistical average in the steady state with f = 0 (see section 3.1
for the proof of (13)). From (11) and (13), we obtain the law of action and reaction for F
imec :

s
s
F
1mec = − F
2mec . (14)
f =0 f =0

The statistical mechanical force in equilibrium can also be understood within a

thermodynamic framework by considering the case in which an operation is performed on the
system. We now explain this understanding. As an example, we consider the simple operation
of shifting the position of particle 2 as R
2 → R
2 +  R
2 instantaneously at time t = 0 (see
figure 2(a) depicting the case f = 0). After the application of this operation, the system is left
until a new equilibrium state is realized. Then, the work done by the external agent carrying
out this operation, W op1 , satisfies the first law of thermodynamics:
 op1

op1 
2 N
W = Q 1  f =0 + Q b  f =0 + Vt ( R
1 ) Vint ( R
i − r
k )
op1 op1
op1
+ . (15)
f =0
i=1 k=1 f =0
op1
Here  f represents the statistical average with respect to this operation. The quantities Q 1
and Q b are the energies transferred to the heat bath from the degrees of freedom of particle 1
and the background particles. We refer to each of these transferred quantities of energy as heat.
According to [8], the quantities

J1 ≡ (γ R
˙1 − ξ
) ◦ R
˙1 , (16)

N
Jb ≡ (γ r
˙k − ξ
k ) ◦ r
˙k (17)
k=1
Effective forces in non-equilibrium 2829

(a) (b)

Figure 2. (a) A simple operation shifting the position of particle 2 as R
2 → R
2 +  R
2
instantaneously at time t = 0. Note that the background particles are not driven in the equilibrium
case. (b) A simple operation in which the roles of particles 1 and 2 in (a) are exchanged.

represent the heat transferred per unit time, and therefore Q 1 and Q b can be defined as
 ∞
Q1 ≡ dt J1 (t), (18)
0 ∞
Qb ≡ dt Jb (t). (19)
0

The symbol ◦ in (16) and (17) indicates that the product here is the stochastic Stieltjes integral
in the Stratonovich sense [8].
Now, using the work W op1 , we define the thermodynamic force by
 

W op1 W op1
F2thermo
≡ lim − ,− . (20)
| R
2 |→0 ( R
2 )x ( R
2 ) y
Then, integrating the Langevin equations (3) and (5), and using the expression

R
˙2 =  R
2 δ(t) (21)

to represent the operation, we can prove the relation

s
F
2thermo = F
2mec , (22)
f =0

(see section 3.2 for the proof of (22)).

In the same manner that we used to define the thermodynamic force F
2thermo , we can also
define F
1thermo by exchanging the roles of particles 1 and 2 (see figure 2(b) depicting the case
f = 0). That is, particle 1 is fixed, while particle 2 is trapped by the harmonic potential.
Because this exchange is equivalent to the spatial reflection with respect to R
2 /2, we find

F
1thermo = − F
2thermo . (23)

Then, from (14), (22) and (23), we derive

s
F
1thermo = F
1mec . (24)
f =0

From the above results, we conclude that in the equilibrium case, because  F
istat sf =0 =
 F
imec sf =0 = F
ithermo , we can regard any one of these quantities as the effective interaction
force between particles 1 and 2.
2830 K Hayashi and S-I Sasa

3.2. Proofs of (13) and (22)

Proof of (13). The differentiation of (8) with respect to R
2 yields

∂ Veff ( R
1 − R
2 )
exp{−β[Veff ( R
1 − R
2 ) + Vt ( R
1 )]}
∂ R
2
 
N N
∂ Vint ( R
2 − r
k )
= dr
k pc ( R
1 , r
k ; R
2 ). (25)
k=1 k=1 ∂ R
2
Integrating this with respect to R
1 , we derive
 s  s
∂ Veff ( R
1 − R
2 ) N
∂ Vint ( R
2 − r
k )
= (26)
∂ R
2 f =0 k=1 ∂ R
2 f =0
 s
 ∂ Vint ( R
2 − r
k )
N
=− . (27)
k=1
∂ r
k
f =0

Using the definitions (10) and (12), we obtain (13).

Proof of (22). From (3) and (5) with f = 0, we obtain the energy balance equations:
∂ Vt ( R
1 )
˙  ∂ Vint ( R
1 − r
k ) ˙
N
(γ R
˙1 − ξ
) ◦ R
˙1 = − ◦ R1 − ◦ R
1 (28)
∂ R
1 k=1 ∂ R
1

2
∂ Vint ( R
i − r
k ) ˙
(γ r
˙k − ξ
k (t)) ◦ r
˙k = − ◦ r
k , (29)
i=1
∂ r
k
where k = 1, . . . , N . The addition of (28) and (29) with k = 1, . . . , N leads to

2 
N 
N
∂ Vint ( R
2 − r
k )
J1 + Jb = − V̇int ( R
i − r
k ) − V̇t ( R
1 ) + ·  R
2 δ(t), (30)
i=1 k=1 k=1 ∂ R
2
where we have used (21). Here, we perform the time integration and consider its statistical
average. Noting that the third term on the right-hand side of (30) becomes − F
2mec sf =0 · R
2,
we obtain
 op1

s 
2 
N
op1
− F
2 ·  R
2 = Q 1  f =0 + Q b  f =0 + Vint ( R
i − r
k ) + Vt ( R
1 )
op1 op1
mec
.
f =0 f =0
i=1 k=1 f =0
(31)
Comparing this with the definition of F2thermo given in (15), we obtain

s
F
2thermo = F
2mec . (32)
f =0

4. Non-equilibrium case

In this section, which consists of three subsections, we study the non-equilibrium case. We
first note that the canonical distribution cannot be employed in the derivation of an effective
description for a NESS, in contrast to the equilibrium case. In the case of NESSs, the
relationships between F
istat , F
imec and F
ithermo are not yet understood. In this section, we
investigate these three forces as defined above in the non-equilibrium, f = 0 case. We hope
that this will allow us to obtain a proper effective description of particles 1 and 2 in a NESS. We
Effective forces in non-equilibrium 2831

2 f
mec s
>
7

-<F
6

5
, 1 f
mec s
>
4
<F

3
0 0.1 0.2 0.3
f

Figure 3. F1mec sf (open symbols) and −F2mec sf (solid symbols) are plotted as functions of f .

first present the numerically computed results determining F
imec and F
ithermo as defined in (12)
and (20) for a NESS. Then, we derive F
istat from the expression of the steady-state distribution
and compare this F
istat with F
imec and F
ithermo .
Throughout this section, we consider only the case of forces in the x direction and denote
the x component of F
imec , F
ithermo and F
istat as Fimec , Fithermo and Fistat .

4.1. Mechanical forces exerted by the background particles

In figure 3, F1mec sf and −F2mec sf are plotted as functions of f . In the case f = 0, it is
seen that F1mec sf =0 = −F2mec sf =0 , which is consistent with (14). However, it is seen that
for f = 0, these quantities are not equal, and indeed the discrepancy between them increases
monotonically as a function of f . Hence, the law of action and reaction does not hold with
respect to F1mec sf and F2mec sf in the non-equilibrium case, unlike in the equilibrium case.
This asymmetry is interpreted as follows. Particle 2 is partially shielded from the effect of the
background particles by particle 1 in the case f > 0 (see figure 1).

4.2. Thermodynamic definition

Let us now reconsider the first law of thermodynamics represented by (15), which provides
the definition of F
ithermo (i = 1, 2). In contrast to equilibrium cases, (15) cannot be employed
directly for NESSs because Jb , defined by (17), takes a non-zero value even when there is no
operation applied to the system, and hence Q b becomes infinite. (Recall that the background
particles are driven by the external driving force, f .)
In order to extend thermodynamic considerations to NESSs, we introduce the idea of the
net heat generated by an operation, following the framework of steady-state thermodynamics
proposed by Oono and Paniconi [9]. In this framework, the heat necessary to maintain a
steady state (the so-called housekeeping heat) is regarded as being independent of the change
undergone by the system caused by the operation, and the excess heat is defined by subtracting
the housekeeping heat from the total heat dissipated to the heat bath. They conjecture that
thermodynamics can be extended to NESSs by using this excess heat.
op1
Here, as one example, in figure 4  Jb (t) f is plotted as a function of time for the case
that  R
2 = (R2 , 0) = (0.25, 0) and f = 0.3. This graph shows that  Jb (t) f reaches a
op1

steady-state value, Jb (∞), a sufficiently long time after the application of the operation. Then,
2832 K Hayashi and S-I Sasa

40

35

30

f
op
<Jb>
25

20

15

10
0 0.05 0.1 0.15 0.2 0.25
t

op1 op2
Figure 4. Jb (t) f (open symbols) and Jb (t) f (solid symbols) as functions of time t in the
op1 op2
non-equilibrium case with f = 0.3. Here, R2 = 0.25 and R1 = 0.25. Jb (t) f and Jb (t) f
are computed in the situations described in figures 2(a) and (b), respectively.

employing the idea of the excess heat, it is reasonable to define

 ∞
Q̃ b ≡ dt (Jb (t) − Jb (∞)). (33)
0

By replacing Q b in (15) with Q̃ b , we hypothesize the following, an extended form of the first
law of thermodynamics:
 op1

op1
op1 
2  N
+ Vt ( R
1 ) Vint ( R
i − r
k )
op1
W = Q 1  f + Q̃ b
op1
+ . (34)
f f
i=1 k=1 f
op1
Using this W , we define the thermodynamic force in the present NESS by (20).
 N
op1
In figures 5 and 6, we plot Vt  f and  2i=1 k=
op1
1 Vint ( Ri − r
k ) f as functions of
R2 in the case f = 0.3. It is seen that these quantities depend linearly on R2 in the range
plotted. This implies that R2 = 0.25 is sufficiently small to be used in (20) to numerically
compute the force. Using this value, we obtain  Q̃ b  f = 0.61, Vt ( R
1 ) f = −0.11,
op1 op1
2  N
 i=1 k=1 Vint ( R
i − r
k ) f = 0.79. Also, we find that Q 1  f is sufficiently small to be
op1 op1

ignored (within the numerical accuracy of our simulations). Using these values, we estimate
F2thermo = −5.2. (35)
Next, by exchanging the roles of particles 1 and 2 (see figure 2(b)), we can define F1thermo
in the same way:

op2
op2
+ Vt ( R
2 )
op2
−F1thermo R1 = Q 2  f + Q̃ b
f f
 op2

2 N
+ Vint ( R
i − r
k ) + O(R12 ). (36)
i=1 k=1 f
 N
In figures 5 and 6, we plot Vt  f
op2
and  2i=1 k=
op2
1 Vint ( Ri − r
k ) f as functions of
op2
R1 for the case of f = 0.3. Also, in figure 4, Jb (t) f is plotted as a function of time
= 0.64, Vt ( R
1 ) f = −0.05,
op2 op2
for the case of R1 = 0.25. Then, obtaining  Q̃ b  f
 N
 2i=1 k=
op2
1 Vint ( Ri − r
k ) f
op2
= 0.79, and ignoring the small contribution from Q 2  f ,
we estimate
F2thermo = 5.5. (37)
Effective forces in non-equilibrium 2833

f
op
<∆Vt >
-0.05

-0.1

0 0.05 0.1 0.15 0.2 0.25

∆R

op1 op2
Figure 5. Vt  f as a function of R2 , and Vt  f as a function of R1 . The open symbols
and the solid symbols represent the quantities obtained in the situations described in figures 2(a)
and (b), respectively.

1
f
op

0.75
<Σi Σk∆Vint(Ri-rk)>

0.5

0.25

0
0 0.05 0.1 0.15 0.2 0.25
∆R
 N  N
Figure 6.  2i=1 k=
op1
as a function of R2 , and  2i=1 k=

1 Vint ( Ri − r
k ) f 1 Vint ( Ri −
op2
r
k ) f as a function of R1 . The open symbols and the solid symbols represent the quantities
obtained in the situations described in figures 2(a) and (b), respectively.

The numerical results given in (35) and (37) are consistent with the law of action and reaction
with respect to F1thermo and F2thermo . While it is desirable to obtain more precise results (the
op1
uncertainty on the values of Fithermo is due mainly to the difficulty in determining  Q̃ b  f and
op2
 Q̃ b  f (see figure 4)), it is clear from the present results that even if the law of action and
reaction is violated for Fithermo in the NESS we consider, the extent of this violation is much
less than that in the case of Fimec (recall that we found F1mec sf = 7.0 while F2mec sf = −3.6
(see figure 3)).

4.3. Statistical mechanical definition

In this subsection, we investigate the interaction force between particles 1 and 2 in a NESS
from a statistical mechanical point of view, employing the steady-state distribution function,
ps ( R
1 , {
rk }; R
2 ). It is known that ps ( R
1 , {
rk }; R
2 ) is given by

ps ( R
1 , {
rk }; R
2 ) = lim pc ( R
1 , {
rk }; R
2 ) exp (−β

τ ), (38)
τ →∞
2834 K Hayashi and S-I Sasa

with
 τ 
N

τ = dt f
· r
˙ k , (39)
0 k=1

where A denotes the statistical average of A over noise histories under the condition that
the initial condition ( R
1 (0), {
rk (0)}) is fixed as the argument ( R
1 , {
rk }) of the distribution
function ps ( R
1 , {

rk }; R
2 ). The expression (38) indicates that the deviation of ps ( R
1 , {
rk }; R
2 )
from the canonical distribution, pc ( R
1 , {
rk }; R
2 ), is represented by the entropy production
β
τ . This form of the steady-state distribution is similar to that obtained by Zubarev and
McLennan [10, 11]. (Regarding the derivation for stochastic systems, see section 4.2 of [12].)
Now, referring to the equilibrium case, we define an effective potential Veff ( R
1 ; R
2 ) by
 N
exp[−β(Veff ( R
1 ; R
2 ) + Vt ( R
1 ))] = lim dr
k pc ( R
1 , {

rk }; R
2 ) exp(−β
τ ). (40)
τ →∞
k=1
With this effective potential, we define the statistical mechanical force as
∂ Veff ( R
1 ; R
2 )
F
2stat ≡ − . (41)
∂ R
2
Then, differentiating (40) with respect to R
2 , assuming that the relation
 
∂ ∂
exp(−β
τ ) = −β
τ exp(−β
τ ) (42)
∂ R
2 ∂ R
2
holds, we derive

s
s 
N
∂
F
2stat = F
2mec − lim d R
1 dr
k ps ( R
1 , {

rk }; R
2 )
τ . (43)
f f τ →∞
k=1 ∂ R
2
Note that (42) is valid to linear order in f
because the both sides of (42) become
∂
−β
τ + O(| f
|2 ). (44)
∂ R
2
We expect that (42) might be approximately valid even for large f
, due to many-body effects.
In the argument below we assume (42).
On the other hand, using the energy balance equation obtained from the Langevin
equations (3) and (5), we can rewrite (34) as
 op1  s 
 τ  τ 
  N N
F2thermo R2 = F2mec f R2 − lim f
·  r
˙ k r
˙ k  .
s
dt − dt (45)
τ →∞ 0 0
k=1 f k=1 f
op1
Then, recalling that f
is the statistical average with respect to the instantaneous shift
R
2 → R
2 +  R
2 at t = 0, we can write
 op1  s
τ 
N τ N
dt ˙
r
k − dt ˙
r
k
0 k=1 0 k=1
f f
  τ 
N
∂ N
= d R
1 dr
k ps ( R
1 , {

rk }; R
2 ) dt r
˙ k  R
2 + O(| R
2 |2 ). (46)
k=1 ∂ R
2 0 k=1
Substituting this result into (45) and comparing the obtained expression with (43), we find
 s
F2thermo = F2stat f . (47)
Effective forces in non-equilibrium 2835

Thus, the equality F1thermo = −F2thermo observed numerically in the last subsection suggests
that F
1stat defined by
∂ Veff ( R
2 ; R
1 )
F
1stat ≡ − (48)
∂ R
1
satisfies the relation
F
1stat = − F
2stat . (49)
We further assume that
Veff ( R
1 ; R
2 ) = Veff ( R
2 ; R
1 ). (50)
Then, (49) and (50) imply that there exists a function Ṽeff ( R
1 − R
2 ) such that
Ṽeff ( R
1 − R
2 ) = Veff ( R
1 ; R
2 )
= Veff ( R
2 ; R
1 ). (51)
It is for future work to develop a theoretical argument for the suggestion (49).

5. Discussion

In this paper we have studied three types of force, Fistat , Fimec and Fithermo , in a NESS. In
equilibrium, these forces satisfy the relations Fistat sf =0 = Fimec sf =0 = Fithermo . From the
results of our numerical simulations, we found that in the NESS that we study F1thermo =
−F2thermo but F1mec sf = −F2mec sf , and hence the law of action and reaction holds for forces
of the former kind but not the latter. We also demonstrated that Fithermo defined with respect to a
form of the first law of thermodynamics hypothesized to hold for our NESS is equal to Fistat sf
derived from the expression for the steady-state distribution under the assumption represented
by (42).
The law of action and reaction holds for any force given by the derivative of a potential
function. Thus, the fact that this law holds with respect to the thermodynamic force defined
by (34) implies that in the NESS we considered, there may exist a potential function
associated with this force. It is important to clarify the physical conditions under which the
thermodynamic force obtained using the method employed in this paper is indeed a potential
force.
In equilibrium, a potential function is related to the thermodynamic free energy. For
this reason, it is natural to conjecture that even in a NESS, the potential force measured
experimentally is related to thermodynamics. Such a conjecture led us to introduce the
thermodynamic force based on the idea of Oono and Paniconi [9]. Here we remark that
the framework proposed by Oono and Paniconi is embodied in the non-equilibrium Langevin
model [13]. In that work, Hatano and Sasa derived an identity which yields an extended second
law relating the Shannon entropy to the excess heat they defined. Recently, the validity of this
identity was confirmed experimentally for a bead system [14]. In the analysis employed in the
present paper, we conjecture that Q̃ b defined in (33) has a certain connection with the excess
heat defined in [13]. However, we do not yet understand this relation, though apparently the
quantity Q̃ b employed here takes a simpler form than that proposed by Hatano and Sasa.
As another important subject related to the present work, we now consider the problem of
force decomposition. As seen in section 4.1 (see figure 3), the law of action and reaction does
not hold for the total mechanical force acting on the test particle,  F
imec sf . If the potential force
in a NESS can be defined properly on the basis of thermodynamic considerations, it might be
possible to decompose  F
imec sf into its potential and non-potential parts. Furthermore, because
2836 K Hayashi and S-I Sasa

F
imec fluctuates in time, we would like to decompose F
imec into a fluctuating dissipative force,
a fluctuating interaction potential force and the remaining part.
With regard to the decomposition of fluctuating forces, recently a significant result has
been obtained for a model of a Brownian particle that is driven by an external driving force and
subject to a spatially periodic potential. As this particle is driven in one direction, it experiences
repeated ‘collisions’ with the periodic potential barriers. Studying a finite time average of the
force exerted by the periodic potential under non-equilibrium conditions, we have found that
a simple orthogonality condition provides a proper force decomposition of the time averaged
force into a dissipative force and a non-dissipative part [15]. This result has led us to construct a
mathematical technique for re-expressing a Langevin equation in a form in which the response
function appears explicitly [16]. Using this technique, an interesting equality that connects
energy dissipation with the violation of a fluctuation-response relation has been proved [17].
Considering these developments, the problem of decomposing  F
imec sf can be regarded as a
natural extension of the study presented in [15], involving the elimination of the dynamical
degrees of freedom.
To summarize the most important point of this work, we have found that the law of action
and reaction with respect to thermodynamic forces has a deep connection with fundamental
aspects of non-equilibrium statistical mechanics. We believe that the numerical findings
reported in this paper will stimulate further theoretical and experimental studies.

Acknowledgments

This work was supported by grants from JSPS Research Fellowships for Young Scientists
(grant no. 1711222) and the Ministry of Education, Science, Sports and Culture of Japan (grant
no. 16540337).

References

[1] Leunissen M E, Christove C G, Hynnien A P, Royall C P, Cambell A I, Imhof A, Dijkstra M, Roij R V and
Blaaderen A V 2005 Nature 437 235
[2] Brunner M, Bechinger C, Streep W, Lobaskin V and von Grünberg H H 2002 Europhys. Lett. 58 926
[3] Brunner M, Bechinger C, Herz U and von Grünberg H H 2003 Europhys. Lett. 63 791
[4] Anderson V A and Lekkerkerker H N W 2002 Nature 416 811
[5] Pham K N, Puertas A M, Bergenholtz J, Egelhaaf S U, Moussaı̈d A, Pusey P N, Schofield A B, Cates M E,
Fuchs M and Poon W C K 2002 Science 296 104
[6] Harada T and Yoshikawa K 2004 Phys. Rev. E 69 031113
[7] Dzubiella J, Löwen H and Likos C N 2003 Phys. Rev. Lett. 91 248301
[8] Sekimoto K 1997 J. Phys. Soc. Japan 66 1234
[9] Oono Y and Paniconi M 1998 Prog. Theor. Phys. Suppl. 130 29
[10] Zubarev D N 1974 Nonequilibrium Statistical Thermodynamics (New York: Consultants Bureau)
[11] McLennan J A 1989 Introduction to Nonequilibrium Statistical Mechanics (Englewood Cliffs, NJ: Prentice-Hall)
[12] Hayashi K and Sasa S 2005 Preprint cond-mat/0507719
[13] Hatano T and Sasa S 2001 Phys. Rev. Lett. 86 3463
[14] Trepangnier E H, Jarzynski C, Ritort F, Crooks G E, Bustamante C J and Liphardt J 2004 Proc. Natl Acad. Sci.
USA 101 15038
[15] Hayashi K and Sasa S 2005 Phys. Rev. E 71 020102(R)
[16] Harada T, Hayashi K and Sasa S 2005 J. Phys. A: Math. Gen. 38 3799
[17] Harada T and Sasa S 2005 Phys. Rev. Lett. 95 130602