PHYSICAL REVIEW E 72, 056314 共2005兲

Enstrophy dissipation in two-dimensional turbulence

Marco Baiesi and Christian Maes*

Instituut voor Theoretische Fysica, K.U.Leuven, Celestijnenlaan 200D, B-3001, Belgium
共Received 21 June 2005; published 11 November 2005兲

Insight into the problem of two-dimensional turbulence can be obtained by an analogy with a heat conduc-
tion network. It allows the identification of an entropy function associated with the enstrophy dissipation and
that fluctuates around a positive 共mean兲 value. While the corresponding enstrophy network is highly nonlocal,
the direction of the enstrophy current follows from the Second Law of Thermodynamics. An essential param-
eter is the ratio Tk ⬅ ␥k / 共␯k2兲 of the intensity of driving ␥k ⬎ 0 as a function of wave number k, to the
dissipation strength ␯k2, where ␯ is the viscosity. The enstrophy current flows from higher to lower values of
Tk, similar to a heat current from higher to lower temperature. Our probabilistic analysis of the enstrophy
dissipation and the analogy with heat conduction thus complements and visualizes the more traditional spectral
arguments for the direct enstrophy cascade. We also show a fluctuation symmetry in the distribution of the total
entropy production which relates the probabilities of direct and inverse enstrophy cascades.

DOI: 10.1103/PhysRevE.72.056314 PACS number共s兲: 47.27.Gs, 44.10.⫹i

I. INTRODUCTION dimensional heat conduction problem provides ingredients to

Three-dimensional turbulence displays an inertial range,
in which energy is transferred from the spatial scales at
which it is introduced into the system down to small scales,
where it is finally dissipated by viscous forces. The standard
picture of turbulence in two dimensions is qualitatively dif-
ferent. Following the pioneering works of Kraichnan 关1,2兴,
Leith 关3兴, and Batchelor 关4兴 共KLB兲 on the two-dimensional
Navier-Stokes equation for the fluid velocity field, one ex-
pects an inverse energy cascade from the forcing scales to
large scales, and simultaneously a direct enstrophy cascade
from the forcing scales to small scales. The enstrophy is the
variance of the vorticity; namely, the ensemble average of
the squared curl of the velocity.
Two-dimensional turbulence has been a very active area
of theoretical, numerical, and experimental investigation
关5,6兴, not only as an easier test case but also relevant to
certain real quasi-two-dimensional situations. Examples in-
clude oceanic currents and atmospheric and geophysical
flows 关7兴, but two-dimensional flow is also realized in labo-
ratory situations 关5,6兴. However, the picture of two-
dimensional turbulence remains not fully understood and in
fact, there are some limitations to the above classical KLB
scenario. For example, the standard enstrophy cascade disap-
pears when considering a bounded domain where only a
monoscale forcing is applied 关8兴. Moreover, the mechanism
of the direct enstrophy cascade and the determining factor
for the direction of the enstrophy current has not been fully
understood as a consequence of a more general principle.
There have been recent clarifications, going into the details
of the physical mechanism 共e.g., Refs. 关9,10兴兲, but it seems
interesting and natural to connect the situation also with bet-
ter understood scenarios and to be able to see the enstrophy
dissipation as the result of a more generally valid principle.
In the present paper, we address the issue of the enstrophy
current and its direction. A very close analogy with a two-
understand the enstrophy cascade in its full qualitative be-
havior. It turns out, as will be shown later, that the stochas-
tically driven Navier-Stokes equation for the vorticity can be
mapped to a problem of heat conduction: at each wave num-
ber, k a thermal reservoir is attached with temperature Tk
= ␥k / 共␯k2兲, where ␥k is the forcing strength and ␯ is the vis-
cosity. From the Second Law of Thermodynamics, which
will be derived in its detailed version, it follows that enstro-
phy is dissipated as heat flows: from higher to lower tem-
perature, or here, when ␥k is peaked around some small
mode k, from small to large wave numbers. In other words,
the origin and the direction of the enstrophy flux is simply
and directly a consequence of the Second Law applied to the
enstrophy. We also go beyond the study of the average en-
strophy current and discuss a symmetry in its fluctuations.
That estimates the probability of going backwards; i.e., the
probability of an inverse enstrophy cascade. At the same
time, we obtain a steady state fluctuation theorem in the con-
text of turbulence.
In the next section, we start by reminding the reader of the
standard picture of two-dimensional turbulence. In Sec. III
we present the analogy with heat conduction. From it follows
the final analysis of the enstrophy dissipation in Sec. IV. The
main general consequences and conclusions are discussed in
Sec. V.
The paper will describe the arguments and analog in a
formal way, avoiding, however, a fully rigorous mathemati-
cal analysis. The main goal is indeed to point out a useful
picture and analogy that is sufficiently powerful to specify
the enstrophy cascade. To add the mathematical details and
hypotheses is not believed to be extremely difficult; only a
few remarks are added to guide the mathematically inclined.
II. NAVIER-STOKES EQUATION

The Navier-Stokes 共NS兲 equation 关11,12兴 for the velocity

*Electronic address: christian.maes@fys.kuleuven.be field uជ 共t , r兲 is

1539-3755/2005/72共5兲/056314共7兲/$23.00 056314-1 ©2005 The American Physical Society

M. BAIESI AND C. MAES PHYSICAL REVIEW E 72, 056314 共2005兲

⳵uជ
⳵t
ជ 兲uជ = ␯⌬uជ − ⵜ
+ 共uជ · ⵜ ជ p + ជf , 共2.1兲 ⌽ j,ᐉ ⬅
j 2ᐉ 1 − j 1ᐉ 2
4␲
冉 1
兩j兩 2 −
1
兩ᐉ兩2
冊.

where p is the pressure, ជf is the external force, and ␯ is the These ⌽ j,ᐉ = 0 if and only if either j 储 ᐉ or 兩j兩 = 兩ᐉ兩. Alterna-
viscosity. This is supplemented by the incompressibility con- tively,
dition
ជ · uជ = 0,
F k共 ␻ 兲 = 兺 ␾k,ᐉ␻ᐉ␻k−ᐉ 共2.5b兲
ⵜ 0⫽ᐉ⫽k

and in our case by periodic boundary conditions for a finite with

冉 冊
ជ
spatial region V. Similar equations arise for the vorticity ␻ k 1ᐉ 2 − ᐉ 1k 2 1 1
ជ
⬅ ⵜ ⫻ uជ , by taking the curl of 共2.1兲 ␾k,ᐉ ⬅ − = ␾−k,−ᐉ .
4␲ 兩ᐉ兩2 兩k − ᐉ兩2
ជ
⳵␻ ជ 兲␻ ជ 兲uជ = ␯⌬␻
+ 共uជ · ⵜ ជ − 共␻
ជ ·ⵜ ជ + gជ 共2.2兲 In that notation,
⳵t
␾k,ᐉ + ␾k,k−ᐉ = 0
ជ ⫻ ជf . The energy of the system is given by the total
with gជ = ⵜ
kinetic energy represents the so called triad relation of 关13兴. Another alter-

冕
native, which will be useful later, is
u2
E= , 1 k 1ᐉ 2 − ᐉ 1k 2
V 2 F k共 ␻ 兲 = − 兺
2␲ ᐉ 兩k − ᐉ兩2
␻ᐉ␻k−ᐉ . 共2.5c兲
while the enstrophy is defined as

冕
Finally, we must specify the forcing gk. A translationally
␻2 invariant and stationary turbulent state can be achieved by
⍀= .
V 2 imposing a force that is homogeneous in space and time. A
Gaussian random field with zero mean is the simplest ex-
Its role will become clearer later on.
ample: in that case the force ជf in 共2.1兲 is a Gaussian noise
Consider Cartesian coordinates r = 共x1 , x2 , x3兲 = 共x , y , z兲;
that is white in time and colored in space, completely deter-
the two-dimensional case is conveniently represented by set-
mined by its covariance
ting the third component of the velocity equal to zero: uជ
ជ = 共0 , 0 , ␻3兲 is better represented by
= 共u1 , u2 , 0兲. Therefore, ␻ 具f i共s,r兲f j共t,r⬘兲典 = Cij共r − r⬘兲␦共t − s兲,
a pseudoscalar ␻ = ␻3. Equation 共2.2兲 thus acts on a single
component, and since uជ and ␻ ជ are now perpendicular to each where ⳵iCij = 0 共incompressibility兲. Equation 共2.4兲 then turns
other, it is further simplified by the vanishing of the term into the stochastically driven NS
共␻ ជ 兲uជ , as
ជ ·ⵜ d␻k共t兲 = − ␯k2␻kdt + Fk共␻兲dt + 冑2␥kdWk共t兲, 共2.6兲
⳵␻ ជ 兲␻ = ␯⌬␻ + g, in which dWk = dW̄−k represents a standard Wiener process
+ 共uជ · ⵜ 共2.3兲
⳵t 关14兴. That driving pumps vorticity into the system at wave
number k with intensity ␥k 艌 0, while the viscosity ␯ ⬎ 0 en-
where g = ⳵ f 1 / ⳵y − ⳵ f 2 / ⳵x. The pressure has disappeared but
ters in the first term on the right-hand side of 共2.6兲 to dissi-
Eq. 共2.3兲 is still nonlocal because uជ = K␻ for some Biot-
pate the vorticity. Equation 共2.6兲 is the starting point of our
Savart kernel K.
analysis.
We take our system bounded in a rectangular domain,
where it is useful to consider the Fourier transform
A. Mathematical assumptions

␻k =
1
2␲
冕 V
e ik·r
␻共r兲. The point of departure 共2.6兲 is a stochastic differential
equation to be understood in the Itô sense 关14兴. Solutions are
Markov processes but note that they are infinite-dimensional.
These modes satisfy ␻
¯ k = ␻−k which will always be under-
In general, the resulting diffusion is not elliptic because some
stood. ␥k can be made zero. That brings us to the problem of un-
Upon Fourier-transforming 共2.3兲, we thus get, for k derstanding the assumptions on the strengths ␥k and on the
苸 Z2 \ 兵0其, viscosity ␯ so that there is a unique invariant probability
⳵␻k measure ␮. A lot of mathematical work has been devoted to
− F k共 ␻ 兲 = − ␯ k 2␻ k + g k , 共2.4兲 that problem in recent years. For example 关15兴, if ␯ is suffi-
⳵t
ciently large as a function of 兺␥2k , then ␮ is unique. In addi-
where tion, 关16兴, ␮ is unique when there is ␬ ⬎ 0 so that ␥k ⬃ 兩k兩−␣
for some ␣ and for every 兩k兩 ⬎ ␬. Even 关17–19兴 when ␥k
F k共 ␻ 兲 = 兺
j+ᐉ=k
⌽ j,ᐉ␻ j␻ᐉ 共2.5a兲 ⫽ 0, for every 兩k兩 艋 N where N is some number that depends
on ␯ and on 兺␥2k , then ␮ is unique. We refer to Ref. 关20兴 for
with coefficients given by even stronger and more recent results.

056314-2
ENSTROPHY DISSIPATION IN TWO-DIMENSIONAL… PHYSICAL REVIEW E 72, 056314 共2005兲

In what follows we simply assume, with no further ado,

that ␮ is unique and has smooth local densities. Another
assumption in the technical manipulations is to start from a
Jin 冑
k = 2␥k Re 冕 −␶
␶
␻
¯ k ⴰ dWk

finite-dimensional analysis. In other words, we choose a fi- as the current going, respectively, “out” and “in” the system
nite but arbitrary N and consider Eq. 共2.6兲 only for k2 艋 N at mode k with respect to the external enstrophy reservoir.
with ␥k ⫽ 0 there. That cutoff will take care of convergence The difference,
problems in what follows and it allows us to speak of ␮共␻兲
as the density of ␮ with respect to the flat measure d␻. Jk ⬅ Jout in
k − Jk ,

is the net enstrophy current that leaves the system 共enters the
B. Euler equation environment兲 at mode k.
On the other hand, the local conservation law reads
The vorticity and the corresponding enstrophy play an
important role in two-dimensional turbulence because of the 兩␻k兩2共␶兲 兩␻k兩2共− ␶兲
appearance of an extra conservation law. The Euler equation −
2 2
corresponding to 共2.6兲 is

d␻k共t兲 = Fk共␻兲dt = − ␯k2 冕

−␶
␶
兩␻k兩2共t兲dt + Re 冕 −␶
␶
␻
¯ kFk共␻兲dt

冕
关cf. 共2.5a兲, 共2.5b兲, and 共2.5c兲兴. It is easy to see that ␶
+ 冑2␥k Re ␻
¯ k ⴰ dWk共t兲
兺k ␻¯ kFk共␻兲 = 0, 共2.7兲 −␶

k + 兺 Jᐉk + Jk ,
= − Jout 共2.8兲
in

so that enstrophy is conserved: ᐉ

d⍀ 1 d兩␻k兩2 which defines

=兺
冕
= 0. ␶
dt k 2 dt 1 ᐉ 1k 2 − k 1ᐉ 2
Jᐉk ⬅ Re dt ␻−k␻k−ᐉ␻ᐉ
2␲ −␶ 兩ᐉ − k兩2
As a consequence, the enstrophy in 共2.6兲 is changed by the
injection 共at rate ␥k兲 and the dissipation 共with intensity ␯兲 but the net current from mode ᐉ to mode k 关here we used 共2.5c兲兴.
is transported without dissipation over the various modes via Note the asymmetry Jkᐉ = −Jᐉk.
the nonlinear and highly nonlocal terms in Fk. That invites As said before, the redistribution of enstrophy due to in-
the definition of various enstrophy currents. teractions between different modes globally does not change
the total amount of enstrophy in the system: 兺k兺ᐉJkᐉ = 0 关see
C. Currents 共2.7兲兴.
One of the main problems for the cascade picture is to
The net enstrophy current Jk that leaves the system at understand the direction of the flow of the Jkᐉ. That is basi-
wave number k is obtained from investigating the sources cally determined by the stationary 具Jk典. At the end of the
and sinks to the enstrophy. The total enstrophy dissipation paper, we also discuss its fluctuations.
over the time interval 关−␶ , ␶兴 is computed from
D. Spectral distribution
1
⍀共t兲 = 兺 兩␻k兩2共t兲 Heuristically, the reason the enstrophy flows towards
2 k small scales 共large wave number k兲 is because, at these small
scales, the dissipative term ␯⌬␻ in 共2.3兲 dominates over the
and ជ 兲␻. A more refined argument, started by
advection term 共uជ · ⵜ

⍀共␶兲 − ⍀共− ␶兲 = 兺 − ␯k2

k
冋 冕 ␶

−␶
兩␻k兩2共t兲dt
Kraichnan 关1兴, derives the two-dimensional cascade picture
共the so-called direct cascade for the enstrophy and the in-

册
verse cascade for the energy兲 by investigating the energy

+ 冑2␥k Re 冕 −␶
␶
␻
¯ k共t兲 ⴰ dWk共t兲 ,
spectra. The Fourier spectrum of energy embodies the KLB
picture by showing a power-law regime for each of the two
cascades. Since the enstrophy spectrum is simply related to
the energy one, from the inspection of the energy spectrum
where the last integral is in the Stratonovich sense 关14兴, and
one can argue where energy and enstrophy are transferred or
Re stands for real part. The expression thus evaluates the
dissipated. In a way the cascade of enstrophy to small scales
change of enstrophy during a time interval 关−␶ , ␶兴 for a his-
is the two-dimensional analog of the energy cascade in three
tory 关␻k共t兲兴 and for a realization of the noise 关dWk共t兲兴. It is
dimensions. We skip here the details of that Kolmogorov-
therefore natural to set
Kraichnan theory as they have been excellently reviewed by

冕 ␶ many. We refer to Ref. 关21,22兴.

k = ␯k
Jout 兩␻k兩2dt,
2 The purpose of the present paper is to try an alternative to
−␶ that analysis by opening an analogy with heat conduction. It

056314-3
M. BAIESI AND C. MAES PHYSICAL REVIEW E 72, 056314 共2005兲

is interesting that in this way a natural enstrophy dissipation ⳵␳

function appears, the thermodynamic entropy production, as L+c ␳ = − 兺 Fk共␻兲 共3.5a兲
k ⳵␻k
will be explained in Sec. IV.
and
III. FORMAL ANALOGY WITH HEAT DISSIPATION ⳵ ⳵ 2␳ ⳵Xk
L+d ␳ = ␯ 兺 k2 共 ␻ k␳ 兲 + 兺 ␥ k 2 = 兺 ␥ k ,
Remember our starting equation 共2.6兲. Let us first forget k ⳵␻k k ⳵␻ k k ⳵␻ k
about the coupling between the various modes, so that the
共3.5b兲
system is reduced to the stochastic dynamics
where we made use of the shorthand
d␻k共t兲 = − ␯k2␻k共t兲dt + 冑2␥kdWk共t兲 共3.1兲
2/2 ⳵ ␤ 兩␻ 兩2/2
describing an ensemble of uncoupled oscillators labeled by X k ⬅ e −␤k兩␻k兩 共e k k ␳兲
⳵␻k
the wave number k. While in the original NS equation the
viscosity represents an irreversible loss, here it balances re- with ␤k ⬅ ␯k2 / ␥k.
versibly with the stochastic forcing. The dynamics 共3.1兲 has For the stationary measure ␮, we have L+␮ = 0, and in
a reversible equilibrium measure particular,
2 ¯ k典 − ␯k2具兩␻k兩2典 + ␥k = 0.
具Fk共␻兲␻ 共3.6兲
e−兩␻k兩 /2Tk
␮ 共d␻兲 = 兿
0
d ␻ kd ␻
¯k 共3.2兲 From now on we use that notation 具·典 to denote a stationary
k Zk
average according to ␮. Equation 共3.6兲 gives, for every time
that we will use as a reference. interval 关−␶ , ␶兴,
The parameter Tk ⬅ ␥k / 共␯k2兲 can be viewed as a kind of
1
“temperature” of the reservoir attached to wave number k; it 具Jk典 = ␯k2共具兩␻k兩2典 − Tk兲. 共3.7兲
is, of course, no physical temperature. Thus, our approach is 2␶
different from previous attempts to use a thermodynamical This equation is the detailed enstrophy balance equation in
formalism in turbulence, identifying variables like ␻2k with a stationarity; summing over k gives the somewhat more fa-
temperature 共see Ref. 关23兴 and references therein兲. miliar
The reversibility of the dynamics 共3.1兲 is taken with the
usual kinematical time reversal that reverses the sign of the ␯ 兺 k2具兩␻k兩2典 = 兺 ␥k ,
velocity field: the dynamical time reversal of a history k k

␰ = 共␻共t兲,t 苸 关− ␶, ␶兴兲 共3.3兲
in a given time interval 关−␶ , ␶兴 is
⌰␰ = 共− ␻共− t兲,t 苸 关− ␶, ␶兴兲. 共3.4兲
When we add the Fk共␻兲 to 共3.1兲 to obtain 共2.6兲 the oscil-
lators become coupled, in fact, in a nonlinear and nonlocal
way. That coupling does, however, preserve the enstrophy
very much like a Hamiltonian coupling that conserves the
energy. The picture that thus emerges is formally equivalent
to a heat conduction network wherein the vertices of the
network are represented by the modes k.
The Euler equation represents the conservative part of the
time evolution. That is changed by the addition of the Lange-
vin forces that represent “thermal” reservoirs at each of the
k; thus obtaining our equation 共2.6兲. Observe that the “fric-
tion” depends on the “location” k of the oscillator. Standard
thermodynamics then teaches us that there will be a “heat
current” from higher to lower temperature. That “heat cur-
rent” is in our present setup played by the enstrophy current.
Hence, if the driving makes Tk a decreasing function of 兩k兩,
e.g., by having ␥k ⬃ k−␣ for some ␣ ⬎ 0, then the enstrophy
should be transported from small 兩k兩 towards larger 兩k兩. That
picture will be detailed in the following sections.
The forward generator L+ corresponding to the Markov
diffusion 共2.6兲 can be split into a “conservative” and a “dis-
sipative” part, L+ = L+c + L+d , with
but at the same time and as a new interpretation of 共3.7兲, we
recognize how the net current into the enstrophy reservoir at
mode k is like a heat current into a thermal reservoir as
determined by the difference between what now plays the
role of a local kinetic temperature 具兩␻k兩2典 and the reservoir
temperature Tk.
IV. ENSTROPHY DISSIPATION
Continuing with the analogy above, a quantity is now
brought to the forefront which we call the entropy current S.
Since the net enstrophy current leaving the system at each
mode k is Jk = Jout in
k − Jk and the corresponding “effective tem-
perature” is Tk, we set
1
S⬅兺 Jk 共4.1兲
k Tk
as the variable entropy current. It is a function of the history
共3.3兲 over 关−␶ , ␶兴. The entropy current S is the entropy pro-
duction in the environment associated with the enstrophy
dissipation; it is the usual sum over all dissipative currents
divided by the respective temperatures. In the stationary
state, the average 具S典 is the total change of the entropy in the
universe over the time interval 关−␶ , ␶兴.
We will now show what is suggested thermodynamically
by the previous analogy: S should measure the irreversibility

056314-4
ENSTROPHY DISSIPATION IN TWO-DIMENSIONAL… PHYSICAL REVIEW E 72, 056314 共2005兲

and 具S典 艌 0 as a consequence of the Second Law of Thermo-

dynamics.
Remember that ␮ is the stationary measure of the NS
Re 冋冕 −␶
␶
F̄k共␻兲d␻k − Re 册 冋冕 ␶

−␶
册
F̄k共␻兲d␻k ⴰ ⌰

dynamics 共2.6兲; we denote by ␲␮ its time reversal. In a

given time interval 关−␶ , ␶兴, each history 共3.3兲 is realized in
= lim Re
⌬t→0
冋兺 j
F̄k„␻共t j−1兲…关␻k共t j兲 − ␻k共t j−1兲兴 册
the system with a probability that comes from the path-space
measure P␮␶共d␰兲; i.e., the stationary Markov diffusion process − lim Re
⌬t→0
冋兺 F̄k„− ␻共t j兲…关− ␻k共t j−1兲 + ␻k共t j兲兴 册
冋兺 册
associated with the stationary measure ␮ and the stochastic j

dynamics 共2.6兲. F̄k„␻共t j兲… − F̄k„␻共t j−1兲…

We compute the logarithmic density 关see also 共3.3兲 and = − lim Re 关␻k共t j兲− ␻k共t j−1兲兴2
共3.4兲兴 ⌬t→0 ␻
¯ k共t j兲 − ␻
¯ k共t j−1兲

冋冕 册
j

␶
⳵F̄k共␻兲
P␮␶ = − Re dt = 0,
R ⬅ ln 共4.2兲 −␶ ⳵␻¯k
P␲␶ ␮⌰
because ⳵F̄k / ⳵␻
¯ k = 0 关see 共2.5b兲兴.
as a measure of irreversibility. It gives the ratio between the As a consequence, 共4.3兲 becomes
probability of a history ␰ and the probability of the time-
reversed history ⌰␰. We show that R coincides with S up to R = S + ln ␮关␻共− ␶兲兴 − ln ␮关␻共␶兲兴 共4.4兲
a temporal boundary term. Moreover, taking stationary aver- with

再冋 冎
ages, 具R典 = 具S典. Since by construction 具R典 艌 0, it also follows
that 具S典 艌 0.
To compute R, it is useful to compare the path-space mea-
S=兺
k
1
Tk
␻2k 共− ␶兲 ␻2k 共␶兲
2
−
2
+ 册冕 ␶

−␶
Re关␻kF̄k共␻兲兴dt .
sure with the reference path-space measure of the uncoupled
case 共3.1兲, denoted by P␮0,0␶ 共this is stationary and reversible兲. From 共2.8兲, this expression coincides exactly with 共4.1兲, as
Thus, first we compute the action promised.
When, instead of the stationary ␮, we had taken some
initial density evolving as ␳t , t 苸 关−␶ , ␶兴, the analysis above
P␮␶ would be essentially unchanged. In that case the source of
A␮ ⬅ ln 0,␶
P ␮0 irreversibility is
1
and similarly A␲␮ ⴰ ⌰, to finally estimate 共4.2兲 as the source R=兺 Jk + ln ␳−␶共␻−␶兲 − ln ␳␶共␻␶兲, 共4.5兲
k Tk
of time-reversal breaking
where the only difference with 共4.4兲 resides in the last two
R = A␮ − A␲␮ ⴰ ⌰. 共4.3兲 terms, the temporal boundary
The comparison of the two measures P and P0 is made by 关− ln ␳␶共␻␶兲兴 − 关− ln ␳−␶共␻−␶兲兴.
means of the Girsanov formula 关14兴, obtaining

A␮ = 兺
1
2␥k 再冕 冋 ␶
1
␯k2 Re关␻kF̄k共␻兲兴 − 兩Fk共␻兲兩2 dt
2
册 A. Mean entropy production
From the definition 共4.2兲 it directly follows that

册冎
k −␶

+ Re 冋冕 ␶

−␶
F̄k共␻兲d␻k + ln ␮关␻共− ␶兲兴 − ln ␮ 关␻共− ␶兲兴.
0
具e−R典 = 1
共it is essentially the normalization condition of the path-
space measure P␲␶ ␮⌰兲. Hence, by a convexity inequality, the
Substituting ⌰␰ gives stationary enstrophy dissipation 具S典 = 具R典 艌 0. We can, how-
ever, be more explicit concerning that point by deriving an

A ␲␮ ⴰ ⌰ = 兺
1
再冕 冋 ␶
1
− ␯k2 Re关␻kF̄k共␻兲兴 − 兩Fk共␻兲兩2 dt 册
expression for 具S典 that is explicitly non-negative. In fact, we
will show that

冓冉 冊冔
2␥k 2

册 冎
k −␶

冋冕
⳵ 2
␶ 具S典 = 兺 ␥k exp关− Vk共␻兲兴 exp关Vk共␻兲兴 , 共4.6兲
+ Re F̄k共␻兲d␻k ⴰ ⌰ + ln ␮关␻共␶兲兴 − ln ␮0关␻共␶兲兴. k ⳵␻k
−␶
where
Here, Itô stochastic integrals are performed and one should Vk共␻兲 ⬅ 兩␻k兩2/共2Tk兲 + ln ␮共␻兲.
remember that these are themselves not time-reversal sym-
metric 关14兴. As an example for computing 共4.3兲, we see that From 共4.6兲, 具S典 ⬎ 0 strictly, as we can only have

056314-5
M. BAIESI AND C. MAES PHYSICAL REVIEW E 72, 056314 共2005兲

⳵ ␤ 兩␻ 兩2/2 B. Enstrophy network

关e k k ␮共␻兲兴 = 0
⳵␻k The situation can now be summarized as follows. Locally,
in the stationary measure, we have
for all k when ␮ = ␮0 of 共3.2兲.
Here comes the proof of 共4.6兲. Denote by E␳−␶ the expec-
tation in the process P␳␶ started from ␳−␶ . We assume that at
兺ᐉ 具Jᐉk典 = 具Jk典 = 具␻¯ kFk共␻兲典
−␶
time ␶ the evolved measure is described by a density ␳␶ . We
have 共4.5兲, in expectation, and globally,

E␳−␶关R兴 = 兺 ␤kE␳−␶关Jk兴 + S共␳␶兲 − S共␳−␶兲, 共4.7兲 兺k 具Jk典 = 0. 共4.12兲

k

where S共␳兲 ⬅ −兰d␻ ␳共␻兲ln ␳共␻兲 is the Shannon entropy of For the enstrophy dissipation,
the density ␳. Another formulation is
具S典 = 兺 ␤k具Jk典 ⬎ 0.
冕 ␶ 共4.13兲
E␳−␶关R兴 = Ṙ共t兲dt k
−␶
We have here formally the same situation as for a heat con-
with, similar to 共3.7兲, duction network, as considered, e.g., in Ref. 关24兴. The rela-

冋冕 册
tions 共4.12兲 and 共4.13兲 do not, of course, uniquely determine
1 d
Ṙ共t兲 ⬅ ␯ 兺 k2␤k
the mean enstrophy currents, but their direction or sign is
d ␻ 兩 ␻ k兩 2␳ t共 ␻ 兲 − + S共␳t兲.
k ␤k dt thermodynamically determined by analogy with heat con-
duction.
共4.8兲 Let us first consider the typical case in which the strengths
␥k are nonzero only for a neighborhood of k = 0, say ␥k = 1
The previous considerations thus identify the mean dissipa-
when 兩k兩 艋 ␩, and outside that large wavelength regime:
tion rate at time t 共in the transient regime兲 with Ṙ共t兲. ␥k ↓ 0 , 兩k兩 ⬎ ␩. In terms of heat conduction, it would mean
To see the relation with 共4.6兲, we start by evaluating the that the temperatures Tk = 1 / 共␯k2兲 are decreasing outward in
time-derivative of the Shannon entropy: the disk for 兩k兩 艋 ␩ and fall to Tk = 0 outside 共兩k兩 ⬎ ␩兲. Clearly

冕 冕
then, there will be a heat current toward increasing 兩k兩 or,
dS d␳
共␳兲 = − d␻ ln ␳ = − d␻共L+␳兲ln ␳ . 共4.9兲 here, an enstrophy current towards smaller wavelengths. In
dt dt other words, the enstrophy current is a kind of nonlocal heat
current the direction of which is determined by the Second
Using the invariance of the Shannon entropy under the con- Law. Because of the nonlocality of the term Fk共␻兲, the cur-
servative 共Euler兲 part of 共3.5a兲, we get
rent will not stop at the boundary of the disk but will be more

冕
and more suppressed when regarding Jkᐉ for k inside and ᐉ
dS
共␳兲 = − d␻共L+d ␳兲ln ␳ outside the disk. For really large ᐉ there is no longer a visible
dt local heat current. That seems compatible with the observa-

= 兺 ␥k
k
冕 d␻ Xk
⳵ ln ␳
⳵␻k
tions 关8兴 that the enstrophy cascade remains pretty localized
around the forcing window.
In general, however, when all ␥k ⬎ 0 are active, we have a

= 兺 ␥k 冕 冉d␻ Xk
Xk
␳
− ␤ k␻
¯k 冊 “temperature” profile ␥k / 共␯k2兲 that can, of course, be com-
plicated. If the ␥k depend only on 兩k兩, we have in essence a

冓冋 册 冔
k one-dimensional heat conduction problem 共along the radial

= 兺 ␥k
k
Xk
␳
2
− ␯ 兺 k2
k
冕 d␻ ␻
¯ kX k . 共4.10兲
direction兲.

C. Fluctuations
Minus the second term, it reads Looking back at 共4.7兲 and 共4.8兲, we found the mean en-

冕 冕 冉 冊
tropy as the change of entropy in the environment S plus the
⳵␳
␯ 兺 k2 ¯ kX k = ␯ 兺 k 2
d␻ ␻ d␻ ␻
¯k + ␤ k␻ k␳ change of 共Shannon兲 entropy due to the stochastic dynamics
k k ⳵␻k in the system. Its stationary mean 共具R典 ⬎ 0兲 is strictly posi-

冕 冉 冊 tive. We will now look at its fluctuations. More precisely, we

1
= ␯ 兺 ␤ kk 2 d ␻ ␳ 兩 ␻ k兩 2 −
k ␤k
Substituting 共4.11兲 into 共4.10兲 and then 共4.10兲 into 共4.8兲, we
immediately obtain the desired identity 共4.6兲.
. consider the R of 共4.4兲 and ask for its probability distribu-
tion. Since by construction 共4.2兲
共4.11兲
P␮␶共␰兲 = eR共␰兲 P␲␶ ␮共⌰␰兲,
we have that

056314-6
ENSTROPHY DISSIPATION IN TWO-DIMENSIONAL…
冕 f共⌰␰兲dP␲␶ ␮ = 冕 f共␰兲e−R共␰兲dP␮␶共␰兲 共4.14兲
is exactly valid for all observations f and for all times ␶. The
relation 共4.14兲 is called a fluctuation symmetry, 共see, e.g.,
Ref. 关25兴兲, because it generates the so-called fluctuation theo-
rem for the entropy production as first formulated in Refs.
关26–28兴. Remember that R equals the S up to a temporal
boundary term 关see 共4.4兲 and 共4.5兲兴.
One of the consequences of the fluctuation symmetry
共4.14兲 is that
Prob关R ⬍ 0兴
= 具e−R兩R ⬎ 0典, 共4.15兲
Prob关R ⬎ 0兴
which is sometimes easier to check numerically and experi-
mentally. Roughly speaking, this last relation tells us that the
probability of observing the inverse cascade for the enstro-
phy is exponentially smaller than the probability of observ-
ing the direct cascade.
V. CONCLUSIONS
The main conclusion is derived from the analogy with a
heat conduction network. Above and beyond all detailed
physical mechanisms that give rise to the direct enstrophy
cascade in two-dimensional turbulence stands the Second
Law of Thermodynamics for the entropy 共4.1兲, which gives a
direction to the enstrophy flow. The relevant parameter is the
ratio ␥k / 共␯k2兲, which plays the role of an effective tempera-
关1兴 R. H. Kraichnan, Phys. Fluids 10, 1417 共1967兲.
关2兴 R. H. Kraichnan, J. Fluid Mech. 47, 525 共1971兲.
关3兴 C. E. Leith, Phys. Fluids 11, 671 共1968兲.
关4兴 G. K. Batchelor, Phys. Fluids 12共II兲, 233 共1969兲.
关5兴 P. Tabeling, Phys. Rep. 362, 1 共2002兲.
关6兴 H. Kellay and W. I. Goldburg, Rep. Prog. Phys. 65, 845
共2002兲.
关7兴 D. G. Dritschell and B. Legras, Phys. Today 46共3兲, 44 共1993兲.
关8兴 C. V. Tran and T. G. Shepherd, Physica D 165, 199 共2002兲; C.
V. Tran and J. C. Bowman, Physica D 176, 242 共2003兲; Phys.
Rev. E 69, 036303 共2004兲.
关9兴 G. L. Eyink, Physica D 91, 97 共1996兲.
关10兴 S. Chen, R. E. Ecke, G. L. Eyink, X. Wang, and Z. Xiao, Phys.
Rev. Lett. 91, 214501 共2003兲.
关11兴 M. Lesieur, Turbulence in Fluids, 3rd ed. 共Kluwer, Dordrecht,
1997兲.
关12兴 U. Frisch, Turbulence 共Cambridge University Press, Cam-
bridge, UK, 1995兲.
关13兴 R. H. Kraichnan, Adv. Math. 16, 305 共1975兲.
关14兴 R. S. Lipster and A. N. Shiryayev, Statistics of Random Pro-
cesses I, II 共Spinger-Verlag, New York, 1978兲.
关15兴 J. C. Mattingly, Commun. Math. Phys. 206, 273 共1999兲.
关16兴 J. P. Eckmann and M. Hairer, Commun. Math. Phys. 212, 105
共2000兲.
056314-7
PHYSICAL REVIEW E 72, 056314 共2005兲
ture of an enstrophy reservoir to which each mode k is
coupled. If the forcing is restricted to a finite window, then
the temperature outside is effectively equal to zero. The con-
servative part in the enstrophy conduction is nonlocal but
does not contribute to the dissipation.
We have identified a general entropy function 关共4.1兲 and
共4.4兲兴, also in the transient regime 关see 共4.7兲 and 共4.8兲兴. We
have shown that the stationary entropy production is strictly
positive. It provides the general mechanism driving the direct
enstrophy cascade. The fluctuations in the entropy satisfy the
symmetry 共4.14兲, which gives an estimate 共4.15兲 of the rela-
tive probabilities of direct versus inverse cascades.
An important open question remains however. The above
analogy is silent about the inverse energy cascade. We have
not found a heat conduction analog that would reveal the
inverse cascade for the energy dissipation in two-
dimensional turbulence. Of course, as energy and enstrophy
are entangled and spectrally related, the direct enstrophy cas-
cade has direct consequences in the form of the inverse en-
ergy cascade. That point follows from the standard treat-
ments 共see also Sec. II D兲, as in Ref. 关1兴, but has not been
clarified in the present paper. In fact, a naive extension of the
present formalism but for the energy would find the inverse
cascade quite surprising as it seems to reduce entropy. We
have not investigated whether the combination of dissipative
currents, enstrophy, and energy, would still lead to a total
positive entropy production, as expected thermodynamically.
Clarifying that entropy balance remains one of the most in-
triguing problems of two-dimensional turbulence.
关17兴 E. Weinan, J. C. Mattingly, and Y. Sinai, Commun. Math.
Phys. 224, 83 共2001兲.
关18兴 S. Kuksin, A. Piatnitski, and A. Shirikyan, Commun. Math.
Phys. 230, 81 共2002兲.
关19兴 J. Bricmont, A. Kupiainen, and R. Lefevere, Commun. Math.
Phys. 230, 87 共2002兲.
关20兴 M. Hairer and J. C. Mattingly, C. R. Math. 339, 879 共2004兲.
关21兴 A. Kupiainen, lectures given at the workshop Random Media
2000, Madralin, 19–26 June 2000.
关22兴 D. Bernard, e-print cond-mat/0007106.
关23兴 E. Aurell, G. Boffetta, A. Crisanti, P. Frick, G. Paladin, and A.
Vulpiani, Phys. Rev. E 50, 4705 共1994兲.
关24兴 C. Maes, K. Netǒcný, and M. Verschuere, J. Stat. Phys. 111,
1219 共2003兲.
关25兴 C. Maes, in Poincaré Seminar 2003, edited by J. Dalibard, B.
Duplantier, and V. Rivasseau 共Birkhäuser Verlag, Basel, 2004兲,
pp. 145–191.
关26兴 D. J. Evans, E. G. D. Cohen, and G. P. Morriss, Phys. Rev.
Lett. 71, 2401 共1993兲.
关27兴 G. Gallavotti and E. G. D. Cohen, Phys. Rev. Lett. 74, 2694
共1995兲.
关28兴 G. Gallavotti and E. G. D. Cohen, J. Stat. Phys. 80, 931
共1995兲.