Decomposition of force fluctuations far from equilibrium
Extracted source text ยท arXiv cond-mat/0409537
\documentclass[twocolumn,showpacs]{revtex4}
%\documentclass[preprint]{revtex4}
\usepackage[xdvi]{graphicx}%
\usepackage{dcolumn}
\usepackage{amsmath}
\makeatletter
\def\btt#1{\texttt{\@backslashchar#1}}%
\DeclareRobustCommand\bblash{\btt{\@backslashchar}}%
\makeatother
\newcommand{\bra}{\left\langle}
\newcommand{\ket}{\right\rangle}
\newcommand{\brap}{\left.}
\newcommand{\ketp}{\right.}
\newcommand{\ketc}{\right\rangle_{\rm c}}
\newcommand{\kets}{\right\rangle_{\rm s}}
\newcommand{\pder}[2]{\frac{\partial #1}{\partial #2}}
\newcommand{\der}[2]{\frac{d #1}{d #2}}
\newcommand{\Xif}{\Xi_{\rm f}}
\newcommand{\vs}{v_{\rm s}}
\newcommand{\ps}{p_{\rm s}}
\newcommand{\fp}{f^{\rm p}}
\newcommand{\Qp}{Q^{\rm p}}
\newcommand{\Wp}{W^{\rm p}}
\newcommand{\mud}{\mu_{\rm d}}
\newcommand{\taum}{\tau_{\rm m}}
\newcommand{\tauM}{\tau_{\rm M}}
\newcommand{\cD}{{\cal D}}
\newcommand{\cA}{{\cal A}}
\newcommand{\e}{{\rm e}}
\newcommand{\ep}{\epsilon}
%
\begin{document}
%
\title{Decomposition of force fluctuations far from equilibrium}
\author{Kumiko Hayashi and Shin-ichi Sasa}
\affiliation
%\address
{Department of Pure and Applied Sciences,
University of Tokyo, Komaba, Tokyo 153-8902, Japan}
\date{\today}
\begin{abstract}
By studying a nonequilibrium Langevin system, we find that a
simple condition determines the decomposition of the coarse-grained
force into a dissipative force, an effective driving force
and noise. From this condition, we derive a new universal
inequality, $D \ge \gamma \mud^2 T$, relating the diffusion
constant $D$, the differential mobility $\mud$, the bare friction
constant $\gamma$ and the temperature $T$. Due to the general nature
of the argument we present, we believe that our idea concerning this
decomposition can be applied to a wide class of systems far from
equilibrium.
\end{abstract}
\pacs{05.40.-a, 02.50.Ey, 87.16.Nn}
\maketitle
%%%%%% introduction %%%%%
%%% general
The nature of a force depends on the scale on which it is observed.
For example, consider a force exerted by water molecules on a
colloidal particle. Such a force can be described by mechanical
laws on molecular time scales, while it is described as a
dissipative (frictional) force and thermal noise on time scales
of the order of $10^{-3}$ sec. In analogy to this example, for a
wide range of systems, including bio-mechanical systems \cite{ex1}
and granular systems \cite{ex2.1}, it might be expected
that a fluctuating force obtained through some coarse-graining
procedure can be decomposed into a dissipative force and other
components. In the regime near equilibrium, such a decomposition
is uniquely determined by fluctuation-dissipation relations (FDRs).
However, no rule is known that determines such a decomposition far
from equilibrium \cite{fn:proj}. We wish to discover a rule of
this kind for a class of systems exhibiting fluctuating forces.
%%% model
With the above-stated purpose, in the present work, we study a
Langevin equation describing the motion of a Brownian particle
with a tilted periodic potential in a one-dimensional space.
Although we study the simplest system realizing nonequilibrium
steady states (NESSs), the arguments below can be applied to a
wide class of Brownian motors
\cite{Butt,fla,pul}. The Langevin equation that
we analyze is described by
\begin{equation}
\gamma\der{x}{t}=f-\der{ U(x)}{x}+\xi(t).
\label{model1}
\end{equation}
Here, $\gamma$ is a friction constant,
$U(x)$ is a periodic potential of period $\ell$, $f$ is a
constant external driving force, and $\xi$ is Gaussian white
noise satisfying
\begin{equation}
\bra\xi(t)\xi(t^{'})\ket=2\gamma T\delta(t-t^{'}),
\label{noise}
\end{equation}
where $T$ is the temperature of the environment, and the Boltzmann
constant is set to unity. We consider a description of large-scale
motion, which is obtained by taking an average over a time interval
$\delta t$ that is chosen to be sufficiently longer than the
characteristic time of the system. In this description, the finite
time average of the force $-dU/dx$ acts on the particle as a
fluctuating force. We conjecture that this fluctuating force can be
decomposed into a dissipative force, an extra driving force and
random noise.
%%% time average
In order to investigate time averaged quantities, including that
of $-dU/dx$, we introduce the finite time average of an arbitrary
quantity $Z(t)$:
\begin{equation}
\overline{Z}_n\equiv \frac{1}{\delta t}\int_{t_n}^{t_{n+1}} dt Z(t),
\end{equation}
where $t_n=n \delta t$, $n=0,1,2 \cdots$. Then, the finite time
average of $-d U/d x$ that we consider is given by
\begin{equation}
-\overline{\der{U}{x}}_n = -\frac{1}{\delta t}\int_{t_n}^{t_{n+1}}
dt\left. \der{U}{x} \right\vert_{x=x(t)}.
\end{equation}
We hypothesize that this can be decomposed into a dissipative
component $A (x_{n+1}-x_n)/\delta t$, where $x_n\equiv x(t_n)$, and
a non-dissipative component. That is, we assume the form
\begin{equation}
-\overline{\der{U}{x}}_n=A \frac{x_{n+1}-x_n}{\delta t}+B_n,
\label{decom}
\end{equation}
where $A$ is a constant and $B_n$ is a fluctuating quantity whose
statistical average $\bra B_n\ket$ takes a non-zero value $\fp$:
\begin{equation}
B_n=\fp+\delta B_n.
\end{equation}
The quantities $\fp$ and $\delta B_n$ correspond to an extra
driving force and
random noise, respectively. Substituting (\ref{decom}) into an
integrated form of (\ref{model1}), we obtain
\begin{equation}
(\gamma-A) \frac{x_{n+1}-x_n}{\delta t}=f+ \fp +\delta B_n
+\overline{\xi}_n.
\label{model3}
\end{equation}
Then, with the definitions
\begin{eqnarray}
\Gamma &\equiv& \gamma-A,
\label{gamma} \\
F &\equiv& f+\fp ,
\label{Force} \\
\Xi_n &\equiv& \delta B_n+\overline{\xi}_n ,
\label{Noise}
\end{eqnarray}
we express (\ref{model3}) as
\begin{equation}
\Gamma \frac{x_{n+1}-x_n}{\delta t}=F +\Xi_n,
\label{model2}
\end{equation}
where $\Xi_n$ is expected to exhibit a Gaussian distribution for
large $\delta t$. This equation is regarded as an effective model
of (\ref{model1}).
The main claim of the present Letter is that the simple condition
\begin{equation}
\lim_{\delta t\to\infty} \delta t \bra \delta B_n \overline\xi_n \ket =0
\label{hcon}
\end{equation}
uniquely determines the constant $A$ in (\ref{decom}). Note that
a correlation of time averaged quantities is proportional to
$\delta t^{-1}$ in general and (\ref{hcon}) indicates that the
proportional constant for the case $\bra \delta B_n \overline\xi_n
\ket$ becomes zero.
The condition (\ref{hcon}) implies that $A (x_{n+1}-x_n)/\delta t$,
which fluctuates in time, can be distinguished from $B_n$ by the
condition $(\ref{hcon})$ \cite{fn:decom}. After presenting the proof
of this claim, we remark on three important topics related to it:
a general inequality obtained as a direct application of (\ref{hcon}),
energetic considerations related to (\ref{hcon}), and the relation
between (\ref{hcon}) and a time reversal symmetry in the stochastic
sense.
%%%%%% end of introduction %%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\paragraph*{Preliminary consideration:}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Before presenting the proof of the main claim, we first consider
how the parameters of the effective model (\ref{model2}) can be
expressed in terms of the steady state velocity $\vs$ and the
diffusion constant $D$, defined as
$\vs \equiv \lim_{t \to \infty} \bra (x(t)-x(0))/t \ket$ and
$D \equiv \lim_{t \to \infty} \bra (x(t)-x(0)-\vs t)^2/2t \ket$.
Because $\vs$ and $D$ are independent of the scale on which we
describe the system, the same values of $\vs$ and $D$
should be obtained from (\ref{model2}). This implies the relations
$F=\Gamma\vs$ and
\begin{equation}
\bra \Xi_n \Xi_m \ket =2 \delta_{nm} D\Gamma^2 (\delta t)^{-1}.
\label{noise:eff}
\end{equation}
From these expressions, all the parameters of the effective model
(\ref{model2}) are given in terms of $\vs$ and $D$ when $\Gamma$ is
determined.
%% equilibrium case
In the equilibrium case ($f=0$), $\Gamma$ should satisfy the
relation $D\Gamma^2=\Gamma T$, which is referred to as the FDR of
the second kind \cite{kubo}. {}From this, $\Gamma$ is expressed as
\begin{equation}
\Gamma=\frac{T}{D}.
\label{2nd}
\end{equation}
Furthermore, we can prove
\begin{equation}
D=\mu T,
\label{1st}
\end{equation}
where $\mu $ is the mobility, defined as
$\mu= \lim_{f \to 0} \vs(f)/f$. The two FDRs (\ref{2nd})
and (\ref{1st}) lead to
\begin{equation}
\Gamma=\mu^{-1}.
\label{eqgam}
\end{equation}
However, for NESSs far from equilibrium, (\ref{1st}) is violated
\cite{HSIIIa}, and (\ref{2nd}) does not hold in general.
Therefore, for treatment of such systems,
it is necessary to find a guiding principle to determine $\Gamma$.
%% potential variation method in NESS
In a previous work \cite{HSIIIa}, we studied NESSs and proposed a
natural method to determine $\Gamma$ by considering the response
of the particle to a slowly varying potential $V(x)$ in space.
Below, we present a heuristic argument from which the result
obtained there can be understood. For detailed presentation of
the systematic perturbation method used to derive this result,
see Ref. \cite{HSIIIa}.
Because the gradient of the slowly varying potential $V(x)$ can be
regarded as a modulation of the external force $f$, the large-scale
motion in the modulated system is described by
\begin{eqnarray}
\frac{x_{n+1}-x_n}{\delta t}&\simeq& \vs \left(f-\frac{d V(x_n)}{dx_n}
\right)+\frac{\Xi_n}{\Gamma} \nonumber \\
&\simeq& \vs(f) -\mud\frac{d V }{dx_n}+\frac{\Xi_n}{\Gamma},
\label{heu}
\end{eqnarray}
where $\mud$ is the differential mobility defined by
\begin{equation}
\mud\equiv \frac{d \vs(f)}{df}.
\label{mud}
\end{equation}
Then, assuming that $-d V/dx_n$ is the force acting on the particle
even in this effective description, from (\ref{heu}) we obtain the
result
\begin{equation}
\Gamma=\mud^{-1}.
\label{Gamma}
\end{equation}
Note that this expression represents an extension of (\ref{eqgam})
to the presently considered non-equilibrium case.
{}From (\ref{gamma}) and (\ref{Gamma}), we find that the
constant $A$ appearing in (\ref{decom}) should satisfy
\begin{equation}
\gamma-A=\mud^{-1}.
\label{gamA}
\end{equation}
%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%
\paragraph*{Proof:}
%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%
We demonstrate that the decomposition condition (\ref{hcon})
uniquely determines the constant $A$, yielding (\ref{gamA}).
%
%%% proof 1
%
Using (\ref{decom}) and (\ref{model3}), we rewrite the condition
(\ref{hcon}) as
\begin{eqnarray}
& &\lim_{\delta t\to\infty} \delta t
\bra \left(\overline{\der{U}{x}}_n + A \frac{x_{n+1}-x_n}{\delta t}
\right) \ketp\nonumber \\
& & \brap \left(\gamma \frac{x_{n+1}-x_n}{\delta t}
+\overline{\der{U}{x}}_n \right) \ketc =0,
\label{con2}
\end{eqnarray}
where $\bra \ \ketc$ represents the cumulant. Through the definition
\begin{equation}
G \equiv -\lim_{\delta t\to\infty}\delta t \bra \overline{\der{U}{x}}_n
\cdot \frac{x_{n+1}-x_n}{\delta t} \ketc,
\end{equation}
it is easy to obtain
\begin{equation}
\gamma-A= \frac{2\gamma T}{2\gamma D-G}.
\label{final}
\end{equation}
%%% proof 2
In order to connect (\ref{final}) with the
differential mobility $\mud$ defined by (\ref{mud}), we express
it in terms of $D$ and $\vs$. We start with the path integral
representation
\begin{eqnarray}
\bra x(\delta t)-x(0) \ket &=& \int \cD x
(x(\delta t)-x(0))
\nonumber \\
&\phantom{=}&\e^{-\frac{1}{4\gamma T}\int_{-\infty}^{\delta t} dt
\left( \gamma \dot x-f+\der{U}{x} \right)^2}.
\end{eqnarray}
Then, differentiating both sides with respect to $f$, we derive
\begin{equation}
\der{}{f}\bra x(\delta t)-x(0) \ket =\frac{D}{T} \delta t
-\frac{G}{2\gamma T} \delta t+O(\delta t^2)
\end{equation}
for large $\delta t$. This leads to the relation
\begin{equation}
\mud=\frac{2\gamma D-G}{2\gamma T}.
\label{mud-rel}
\end{equation}
{}Comparing (\ref{final}) and (\ref{mud-rel}), we have arrived at
(\ref{gamA}).
%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%
\paragraph*{Inequalities:}
%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%
%% inequality
As a simple application of the decomposition condition
(\ref{hcon}), we derive several useful inequalities. {}From the
square of both sides of (\ref{Noise}) and the condition (\ref{hcon}),
we obtain
\begin{equation}
2 \Gamma^2 D=\delta t \bra (\delta B_n)^2 \ket +2 \gamma T,
\end{equation}
where we have used (\ref{noise:eff}). This immediately leads to an
inequality relating the intensity of the force noise in the original
system and the quantity representing its effective value in the
coarse-grained system:
\begin{equation}
\Gamma^2 D \ge \gamma T.
\label{ine1}
\end{equation}
Because $\Gamma=\mud^{-1}$ (see (\ref{Gamma})), (\ref{ine1})
can be written as
\begin{equation}
D \ge \gamma \mud^2 T.
\label{ine2}
\end{equation}
{}We believe that this inequality holds in other Brownian motors
\cite{Butt,fla,pul}
because the path integral expression is valid even for cases
with a time dependent potential and the expressions
(\ref{con2})-(\ref{mud-rel}) given in the proof are the same for
those models.
The inequality (\ref{ine2}) involves only directly measurable
quantities and therefore can be tested experimentally.
%% remark on inequality
In a related work, Sasaki
conjectured that the inequality $D \ge \mud T$ holds generally
for Brownian motors \cite{SSK}. If we define
the effective temperature $T_{\rm eff}$ using the FDR violation
factor \cite{HSIIIa}, this conjecture is equivalent to the assertion
that $T_{\rm eff}$ is not less than the temperature of
the environment, $T$. Because it has been observed that
$T_{\rm eff} > T$ in glassy systems \cite{glass2} and driven
many-body systems \cite{drive1,HSIV}, the inequality
$D \ge \mud T$ does seem plausible. However, Sasaki reported that
this inequality is violated for the model (\ref{model1}) with
an appropriate choice of $U(x)$ \cite{SSK2}. This result leads us
to believe that, in the present context, if there exists a generally
valid inequality among measurable quantities, perhaps it involves
the intensity of the force noise and its effective one, not the
temperature and its effective one.
%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%
\paragraph*{Energetics:}
%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%
%% introduction
In equilibrium systems, heat is distinguished from work according
to the second law of thermodynamics. However, obviously, heat
can be considered as a mechanical work done by a force at a
microscopic scale. This tempts us to investigate how heat and
work come to be expressed in different ways through a coarse-graining
procedure. We treat this problem on the basis of the decomposition
of the force $-\overline{dU/dx}_n$ given by (\ref{decom}).
%% basics of energetics and question
In Langevin systems, the heat absorbed from a heat bath is
interpreted as the work done by a force $-\gamma dx/dt+\xi$
exerted by the heat bath \cite{KS}. With this interpretation,
the heat absorbed during an interval $t_n\le t\le t_{n+1}$ can
be expressed as
\begin{equation}
q_n=-\int_{t_n}^{t_{n+1}} \left(\gamma \frac{dx}{dt}-\xi\right)
\circ dx(t),
\label{hm}
\end{equation}
where the symbol $\circ$ indicates that the integral here is
the stochastic Stieltjes integral in the Stratonovich sense \cite{KS}.
Then, the energy balance equation for the model (\ref{model1}) is
derived as
\begin{equation}
U(x_{n+1})-U(x_n)=q_n+f(x_{n+1}-x_n).
\label{b}
\end{equation}
Through similar considerations applied to the effective model
(\ref{model2}), we define the heat (absorbed from an ``effective
heat bath'') as
\begin{equation}
Q_n=-\left(\Gamma\frac{x_{n+1}-x_n}{\delta t}-\Xi\right)(x_{n+1}-x_n).
\end{equation}
With this, the energy balance equation for (\ref{model2}) is
obtained as
\begin{equation}
Q_n+F(x_{n+1}-x_n)=0.
\label{mb}
\end{equation}
The difference between the two quantities $q_n$ and $Q_n$ becomes
obvious when their steady state averages are compared; we have
$\bra q_n \ket=-f v_s\delta t$ and $\bra Q_n \ket=-F v_s\delta t$,
while it is known that outside the linear response regime, in
general $F\ne f$ (see Fig. 2 of Ref. \cite{HSIIIa}).
%% interpretation
In order to understand the difference between $q_n$ and $Q_n$, we
consider a decomposition of the work done by the force $-dU/dx$
during a time interval $t_n\le t\le t_{n+1}$, which is written as
\begin{equation}
-\overline{\frac{d U}{dx}\frac{dx}{dt}}_n\delta t = \Qp_n + \Wp_n.
\label{enerdec}
\end{equation}
Although we conjecture that $\Qp_n$ and $\Wp_n$ correspond to ``heat''
and ``work'', respectively, no rule is known that distinguishes
``heat'' from ``work'' in this case. However, in the present system,
it seems natural to assume
\begin{equation}
\Wp_n=\fp (x_{n+1}-x_n),
\label{wh}
\end{equation}
because the extra driving force $\fp$ was determined from the
decomposition condition (\ref{hcon}). With this assumption, we
can derive the relation
\begin{equation}
Q_n=q_n+\Qp_n,
\label{addi}
\end{equation}
which provides a clear interpretation of the difference between
$q_n$ and $Q_n$.
%% perspective
The argument above leads to the following question: Is there a
simple rule of energetics from which we can obtain the decomposition
(\ref{enerdec}) along with (\ref{wh}) without the decomposition
condition (\ref{hcon})? This fundamental question
will be studied in the future. In addition to their role in such
fundamental problems, energetic considerations applied to different
time scales may be useful when we attempt to interpret the efficiency
of motor proteins \cite{YNG} within stochastic models \cite{TH}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\paragraph*{Adjoint dynamics:}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% definition of adjoint dynamics
As a final topic here, we consider the relation between the
decomposition condition (\ref{hcon}) and time reversal symmetry.
In order to represent this symmetry explicitly, we consider the
path probability density $P$ for a discrete time series $[x]_N=
(x_0,x_1,\cdots,x_N)$ generated within the model under
consideration. The time-reversed path probability density $P^*$ is
defined by $P^*([x]_N)=P(\tilde{[x]}_N)$, where $\tilde{[x]}_N$
represents the time reversed trajectory of $[x]_N$, that is,
$\tilde x_{n}=x_{N-n}$. When $P^*$ is obtained from the frequency
distribution of trajectories in a steady state for some stochastic
dynamics, these dynamics are called ``adjoint dynamics''.
% result and proof
When the constant $A$ appearing in (\ref{decom}) is chosen correctly,
(\ref{model3}) can be regarded as an effective Langevin model.
In this case, we find that adjoint dynamics are described by
\begin{equation}
(\gamma-A) \frac{x_{n+1}-x_n}{\delta t}= -f- \fp +\delta B_n+
\overline{\xi}_n.
\label{model42}
\end{equation}
Then, note that the condition (\ref{hcon}) is equivalent to
\begin{equation}
\bra(-\delta B_n+\overline\xi_n)^2 \ket
=\bra (\delta B_n+\overline\xi_n)^2 \ket.
\label{con}
\end{equation}
Because both $\delta B_n$ and $\bar \xi_n$ exhibit Gaussian
distributions for sufficiently large $\delta t$, the condition
(\ref{con}) allows us to replace $\delta B_n+ \bar \xi_n$ in
(\ref{model42}) by $-\delta B_n+ \bar \xi_n$. Thus, the adjoint
dynamics can be expressed as
\begin{equation}
(\gamma-A) \frac{x_{n+1}-x_n}{\delta t} = -f-B_n+\overline{\xi}_n.
\label{model41}
\end{equation}
Then, because (\ref{model3}) can be rewritten as
\begin{equation}
(\gamma-A) \frac{x_{n+1}-x_n}{\delta t} = f+B_n+\overline{\xi}_n,
\label{model50}
\end{equation}
we find that the decomposition of $-\overline{dU / dx}_n$ given by
(\ref{decom}) is characterized by parity with respect to time reversal
in the stochastic sense. More specifically, the dissipative force
$A (x_{n+1}-x_n)/\delta t $ remains, while the other contribution,
$B_n$, changes sign in the adjoint dynamics.
%% Bertini et. al. %%
In a previous study related to adjoint dynamics, Bertini et al.
succeeded in deriving the large deviation functional $S$ of the
density profile for a special model of the hydrodynamic equation
$\partial_t\rho=\cD(\rho)$, which is obtained as the continuum
limit for a nonequilibrium lattice gas \cite{Ber}. In their analysis,
the equation describing the adjoint dynamics $\partial_t\rho=
\cD^*(\rho)$ was rigorously derived in the form
\begin{equation}
\cD^*(\rho) = \frac{1}{2}\nabla\left(\chi(\rho)\nabla
\frac{\delta S}{\delta \rho} \right )-\cA(\rho)
\label{ber2}
\end{equation}
for the case in which $\cD(\rho)$ is given by
\begin{equation}
\cD(\rho) = \frac{1}{2}\nabla\left(\chi(\rho)\nabla
\frac{\delta S}{\delta \rho}\right) +\cA(\rho),
\label{ber}
\end{equation}
where $\chi(\rho)$ is the current noise intensity. They called the
relation $\cD(\rho)+\cD^*(\rho)=\nabla\left(\chi(\rho)\nabla
(\delta S/\delta \rho)\right)$ the ``fluctuation dissipation
relationship'' for NESSs far from equilibrium, because it reduces
to one expression among the linear response relations for states near
equilibrium, where the relation $\cD(\rho)=\cD^*(\rho)$ holds due
to detailed balance. We remark that (\ref{ber2}) and (\ref{ber})
are similar to (\ref{model41}) and (\ref{model50}) in the model we
study.
%%% many-body systems %%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\paragraph*{Conclusion:}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In conclusion, we have found that the decomposition condition
(\ref{hcon}) leads to the relation $\Gamma=\mud^{-1}$, which was
obtained in our previous study \cite{HSIIIa}. The condition
(\ref{hcon}), which is related to time reversal symmetry, yields
the new inequality (\ref{ine2}) and leads to an interesting question
regarding the decomposition of work given in (\ref{enerdec}).
%%
%%
%%
One of the authors (K. H.) acknowledges her mother for encouragement
over the past decade. This work was supported by a grant from the
Ministry of Education, Science, Sports and Culture of Japan
(No. 16540337).
\begin{thebibliography}{99}
%% intro %%
\bibitem{ex1}
R. Merkel et al., Nature {\bf 397}, 50 (1999).
\bibitem{ex2.1}
E. Longhi and N. Easwar, Phys. Rev. Lett. {\bf 89}, 045501 (2002).
%\bibitem{ex2.2}
%B. N. J. Persson, {\it Sliding Friction: Physical Principles and
%Applications} (Springer, Berlin, 1998).
%\bibitem{ex3}
%force vs time : turbulence
\bibitem{fn:proj}
A projection operator formalism cannot provide such a rule
without serious considerations of the choice of a suitable
projection operator. See e.g. K. Kawasaki, J. Phys. A {\bf 6},
1289 (1973).
\bibitem{Butt}
M. B{\" u}ttuiker, Z. Phys. B {\bf 98}, 161 (1987).
\bibitem{pul}
A. Ajdari and J. Prost, C. R. Acad. Sci. Paris II {\bf 315}, 1635
(1992); M. O. Magnasco, Phys. Rev. Lett. {\bf 71}, 1477 (1993).
\bibitem{fla}
R. D. Astumian and M. Bier, Phys. Rev. Lett. {\bf 72}, 1766 (1994);
J. Prost et al., Phys. Rev. Lett. {\bf 72}, 2652 (1994).
\bibitem{fn:decom}
The condition (\ref{hcon}) takes a different form with the
orthogonality condition that determines a random force in a
projection operator formalism ( H. Mori, Prog. Theor. Phys. {\bf 33},
423 (1965) ). See also \cite{fn:proj}.
%%% preliminary consideration
\bibitem{kubo}
R. Kubo, M. Toda, N. Hashitsume, {\it Statistical Physics II:
Nonequilibrium Statistical Mechanics } (Springer, Berlin, 1991).
\bibitem{HSIIIa}
K. Hayashi and S. Sasa, Phys. Rev. E {\bf 69}, 066119 (2004).
%% inequality %%
\bibitem{SSK}
K. Sasaki, J. Phys. Soc. Jpn. {\bf 72}, 2497 (2003).
\bibitem{glass2}
For a review, see A. Crisanti and F. Ritort, J. Phys. A {\bf 36},
R181 (2003).
\bibitem{drive1}
L. Arrachea and L. F. Cugliandolo, cond-mat/0407427.
\bibitem{HSIV}
K. Hayashi and S. Sasa, cond-mat/0404453.
\bibitem{SSK2}
K. Sasaki, private communication (presentation at the annual
meeting of the JPS, March, 2004).
%% energetics %%
\bibitem{KS}
K. Sekimoto, J. Phys. Soc. Jpn. {\bf 66}, 1234 (1997).
\bibitem{YNG}
M. Nishiyama, H. Higuchi and T. Yanagida, Nat. Cell. Biol. {\bf 4}, 790
(2002).
\bibitem{TH}
T. Harada, cond-mat/0310547.
%% adjoint %%
\bibitem{Ber}
L. Bertini et al., J. Stat. Phys. {\bf 107}, 635 (2002).
\end{thebibliography}
\end{document}