PHYSICAL REVIEW E 85, 056315 (2012)

Inverse energy cascade and turbulent transport in a quasi-two-dimensional magnetized

electrolyte system: An experimental study

L. Bardóczi
Budapest University of Technology and Economics, Budapest, Hungary

M. Berta
Széchenyi István University, EURATOM Association, Győr, Hungary

A. Bencze
Wigner RCP, RMKI, and EURATOM Association, Budapest, Hungary
(Received 27 May 2011; revised manuscript received 4 October 2011; published 30 May 2012)
We present an experimental study of the inverse energy cascade, spectral condensation, and turbulent particle
transport in an electromagnetically driven thin layer of NaCl electrolyte. The presence of the bottom friction
provides an energy sink at large scales for the turbulent flow. This energy sink crucially contributes to the
balance of the forcing and dissipation which makes the inverse cascade steady. The present work provides
an estimation of the linear dissipation rate on an experimental basis. We also show how the dissipation rate
affects the characteristic features of the velocity spectrum and the dynamics of the spectral condensation. A
quantitative study of the turbulent diffusion shows a significant decrease of the radial transport during the spectral
condensation process.

DOI: 10.1103/PhysRevE.85.056315 PACS number(s): 47.27.De

I. INTRODUCTION rigid bottom wall of the container. The central role of this linear
dissipation in the development of spectral characteristics of the
The concept of an inverse energy cascade originates from
direct enstrophy cascade has been recently demonstrated by
the famous groundwork of Kraichnan [1] who first proposed
Boffetta and co-workers, showing a systematic departure from
that energy and enstrophy can cascade in two dimensions. This
Kraichnan’s E(k) ∝ k −3 prediction [11]. We study here the
dual cascade was found to be the consequence of the presence
counterpart of Boffetta et al.’s experiments, namely, the effect
of two quadratic invariants: the mean-squared velocity and
of the bottom friction on the energy spectrum in the inverse
the mean-squared vorticity. The Kraichnan phenomenology
energy cascade range as well as on the spectral condensation
predicts for the inverse cascade a self-similar range of scales
regime.
with energy spectrum scaling as E(k) ∝ k −5/3 . In Kraichnan’s
Recently Shats and co-workers [12] have conducted a
view, the inverse energy cascade cannot be stationary in a
similar experiment to clarify the role of three-dimensionality
pure two-dimensional (2D) system, and a sink of energy at
in such thin magnetized electrolyte layers. They found that
large scales is required to reach a stationary state; moreover,
the relative importance of the z component of the velocity
Kraichnan also conjectured that in finite systems the energy
scales linearly with the normalized electrolyte depth. As
will condense in the lowest accessible k mode. This implies
three-dimensionality increases, the induced 3D turbulent
that the inverse cascade can be considered as a transient state
motions make the turbulent viscosity stronger, resulting in the
which evolves towards the condensed state. Since then, this
inhibition of spectral condensation. Since the electromagnetic
conjecture got strong support from both computer simulations
forcing directly introduces 3D motions, we show in the present
[2] and laboratory experiments [3,4].
experiment that this effect can be avoided using weaker forcing
It was first mentioned by Shats and Punzmann [5] that
for thicker layers.
besides the fundamental scientific importance of understand-
ing the spectral condensation dynamics in 2D turbulence, it
may have practical impact, e.g., in the field of magnetically II. EXPERIMENTS
confined fusion, namely, the understanding of the low-to-high
(L-H) confinement transition and the appearance of coherent A. Experimental setup
mesoscopic flows (zonal flows) [6–9]. From this point of Forced quasi-two-dimensional turbulence has been studied
view the change in the turbulent transport in the presence experimentally in thin layers (3–8 mm) of NaCl electrolyte
of large-scale coherent flows is a crucial issue [10]. in a static electromagnetic field. The experimental setup is
In the present paper we focus on two aspects of the similar to the ones already presented in the literature [4,13]
condensation process: first the dynamics of the transition from and the sketch is shown in Fig. 1(a). The electrolyte is placed
disordered turbulent flow to ordered coherent global flow is in a plastic container of 12 × 12 × 2 cm3 . A spatially periodic
considered, and second a quantitative description is proposed magnetic field is generated by 5 × 5 cylindrical neodymium
for the turbulent transport in the presence of large-scale magnets that are placed horizontally below the container with
flow. One of the main control parameters which determines alternating poles (the maximum magnetic field is 0.3 T). The
the final state of such laboratory systems as our quasi-2D transverse dimension of the magnets is 0.5 cm, and the spacing
magnetized electrolyte is the linear friction imposed by the between magnets is 1 cm, resulting in an injection wave

1539-3755/2012/85(5)/056315(6) 056315-1 ©2012 American Physical Society

L. BARDÓCZI, M. BERTA, AND A. BENCZE PHYSICAL REVIEW E 85, 056315 (2012)

6 5 integration of the Biot-Savart law provided that the permanent

(a) (b) magnets can be replaced by solenoids with identical geometry.
The numerical scheme has been tested in analytically well-
y (cm)

y (cm)
0 0 known geometries (e.g., straight conductors, conducting loops,
single solenoids, etc.), and it was found that Maxwell’s
equations are satisfied within a 10−7 relative error. Knowing
-6 -5 the electric and the magnetic fields, [∇ × (E × B)]z can be
-6 0 6 -5 0 5
x (cm) x (cm) determined [see Fig. 1(d)].
-4 According to the k spectrum of this theoretically expected
5 x10 5
(c) 15 (d) vorticity field [see Fig. 1(e)], the injection wave number
appears to be 628 m−1 , which is consistent with the k value
y (cm)

10 y (cm)
0 0 determined by the distance between the magnets. Interestingly,
5 a local maximum appears at kinj = 236 m− 1 due to the finite
number of magnets [see Figs. 1(c) and 1(e)]. Before discussing
-5 -5 the experimental results, we briefly describe the measurement
-5 0 5 -5 0 5
x (cm) x (cm) techniques used to obtain the temporal evolution of the random
velocity field.
1 (e)
Power (norm.)

0.8 B. Overview of experiments

0.6
Experiments were done with different thicknesses (h) of the
0.4
electrolyte layer; this allows us to control the linear dissipation
0.2
0 rate. A constant electric current of 1 A was driven through the
0 500 1000 system while all parameters were kept constant.
k (m-1 ) In order to keep the dissipation rate at reasonably low levels
and at the same time assure the quasi-two-dimensionality of
FIG. 1. (Color online) (a) Schematic view of the experimental
the flow, the electrolyte layer depth was chosen between 3 and
setup, (b) initial state with 16 central vortices in a 3-mm-thick
electrolyte layer, (c) numerically calculated magnetic field in T,
8 mm.
(d) numerically calculated vorticity field, and (e) k spectrum of the As the electric current was kept constant, the current density
numerically calculated vorticity field. scales as j ∼ 1/ h; this allows the forcing to be adjusted in
such a way that the externally induced 3D turbulent viscosity
remains weak and the inverse energy cascade can take place.
number (kinj ) of 628 m−1 . The electric field is generated by
The length of the measurements was 1 min. This was
two electrodes immersed into the electrolyte at the edges of
limited by the dissolving metals of the electrodes, which
the container. The direction of the field is orthogonal to the
cause the undesirable growth of viscosity and density in the
magnetization axis of the magnets (E = Ey = 1.6 V/cm).
thin-layer experiments at 1 A current. This effect can be
Observations have been made by a commercial digital
identified by the slow decay of the total kinetic energy during
camera placed above the experimental device. This camera
the stationary regime.
records a video stream with a spatial resolution of 640 × 480
To sum up the experimental circumstances, the following
pixels at a rate of 30 frames per second. From the geometrical
parameters were varied by varying the thickness: (1) the
and optical constraints, the spatial resolution at the image plane
dissipation rate (2) the forcing, and (3) the two-dimensionality
is 0.25 mm/pixel.
of the flow. All the other parameters were fixed.
Figure 1(b) shows a typical initial configuration of the flow
1 s after the onset of the electric current. The frame presented is
averaged over 1 s of the flow evolution. As can be seen, 4 × 4 C. Determination of the velocity field
counter-rotating vortices are generated in the central part of The primary quantity in the present experiment is the
the tank. This array of 16 vortices is bounded by a series of time-evolving velocity field v(x,y,t). Other quantities of
weaker counter-rotating vortices. physical interest can be derived mathematically from v(x,y,t).
It can be shown that the vorticity field in an electromagneti- The determination of the velocity field has been done using
cally driven system at t = 0 is proportional to [∇ × (E × B)]z . the particle image velocimetry (PIV) technique which is
This fact gives us the possibility to design the initial vorticity based on two-dimensional cross-correlation functions [14].
configuration of our system by numerically calculating the For visualization of the turbulent flow, uniformly distributed
magnetic field and keeping the electrostatic field homoge- semolina grains were used as tracer particles with a typical
neous. Following the temporal evolution of the flow from size of 0.5 mm ≈ 2 pixels. Those particles which were sitting
different exotic initial vorticity fields, such as, e.g., dipole on the bottom of the container were carefully excluded from
or quadrupole vortex lattices, we can study the interaction of the data processing; therefore only the free surface motions
coherent structures in turbulent media. This will be the subject were analyzed.
of a subsequent presentation. Here we just describe in some Each recorded movie frame is represented by a matrix
detail the field calculation process. Mt (x,y). The Mt (x,y) matrix covers the fluid domain de-
In Figs. 1(c)–1(e) we show the results of a numerical termined by [x,y] ∈ [−2.5,2.5] × [−2.5,2.5]. This matrix is
calculation of the magnetic field based on the numerical divided into submatrices Si,j,t (x,y) and the two-dimensional

056315-2
INVERSE ENERGY CASCADE AND TURBULENT . . . PHYSICAL REVIEW E 85, 056315 (2012)

In our description the friction due to the bottom wall is taken

kin. energy (arb. units)

enstrophy (arb. units) (a)
to be proportional to the magnitude of the flow velocity. From
this model it is clear that at t = 0 the dissipation rate [ε(t),
(b) the total dissipated energy per unit time] is zero. In a given
experiment the electromagnetic forcing rate is ∂t j = ∂t Ėin =
0 (where Ėin is the rate of the energy inflow). Therefore in
the initial state the increment of the total kinetic energy (Ėkin )
0 10 20 30 40 50 0 10 20 30 40 50 is determined only by the energy inflow [Ėkin (t = 0) = Ėin ].
time (s) time (s) This means that the time dependence of the kinetic energy in
diss. rate (arb. units)
averaged quantities

(c) (d) the initial state is linear, as can be seen from the experiments
[see Fig. 2(b)]. Moreover, taking the stationary state (Ėkin = 0)
one can say that the energy inflow is balanced by the dissipation
enstrophy rate:
energy

3 4 5 6 7 8 3 4 5 6 7 8 ε(stationaryphase) = Ėin = Ėkin (t = 0). (3)

h (mm) h (mm) Therefore the dissipation rate in the stationary phase can be
determined experimentally as the slope of the total kinetic
FIG. 2. (Color online) Time evolution of (a) enstrophy and
energy in the initial phase.
(b) energy; (c) mean of total kinetic energy and enstrophy in the
Figure 2(d) shows that the dissipation rate decays as the
stationary regimes versus thickness (in relative units); (d) dissipation
versus the thickness.
thickness of the electrolyte increases.

cross-correlation function is computed for the corresponding B. Time evolution of vorticity fields
submatrices of consecutive frames. The displacement field is The vorticity field was numerically computed as the curl
determined by the shift of the cross-correlation maximum with of the velocity field. The measured flow evolution shows
respect to zero spatial shift. By dividing the displacement three different phases. In the initial phase [∼(0,7) s] structures
field by the sampling time, one can get the velocity field. with linearly growing energy [see Fig. 2(a)] are forced by
This method allows the calculation of the velocity field at the electromagnetic field. According to our observations one
59 × 59 points with 30 Hz sampling frequency. The typical can speculate that above a critical velocity the initial vortices
length of our experiments (1 min) allows us to measure are destabilized and their energy is accumulated into lower
the stationary regime for several tens of seconds. The total wave numbers. This transient phase is called an inverse
statistical uncertainty was estimated to be less than 10%, and energy cascade [∼(7,15) s]. As a result of this inverse energy
originates from the uncertainty in the determination of time cascade, in the stationary phase, a large stable vortex is formed
resolution and from the error caused by the PIV technique. (condensate) [∼(15,50) s]; its size is comparable with the size
of the flow. Figure 3 shows the vorticity field in the initial,
III. EXPERIMENTAL RESULTS transient, and stationary regimes at each thicknesses.
The measured vorticity in the initial state shows a reason-
A. Time evolution of the energy and enstrophy ably good agreement with the numerical calculation presented
The total kinetic energy and enstrophy are defined by the in Fig. 1(d). This indicates that the forcing of the surface is
following relations: dominated by the electromagnetic field.
 In the next section we present a detailed description of the
E(t) = v2 (x,y,t)dxdy, (1) inverse energy cascade phase in terms of different spectral
quantities.

Z(t) = [∇ × v(x,y,t)]2 dxdy. (2) C. Inverse energy cascade
The evolution of enstrophy and total kinetic energy at different In the x direction, one-dimensional k spectra of vx (x,yi ,tj )
thicknesses can be seen in Figs. 2(a) and 2(b). were calculated at fixed yi and tj and then averaged along the
The plots show that at the beginning of each experiment y coordinate. As a result we got spectra computed from 59
the energy and the enstrophy grow linearly. Afterward, these points and 59 spectra were used for the averaging procedure.
quantities turn to stationarity. For the thinnest layer (h = This was repeated for each tj , resulting in the E(kx ,t). The
3 mm) the total kinetic energy starts to decay after 25 s, which same procedure was done for E(ky ,t). In order to decrease the
indicates the effect of the dissolving metals; therefore we statistical noise, a time average over 0.5 s was also performed.
processed only this time interval from the stationary regime. The injection wave number appears to be 597 m−1 [see
For thicker layers this problem can be neglected until 45 s; Fig. 4(a)]. This observation is consistent with the k spectrum of
therefore we processed corresponding signals until 40 s. The the forcing calculated on the base of the numerically calculated
mean of the total kinetic energy and enstrophy in the stationary magnetic field [see Fig. 1(e)]. It is also clear that the energy
regime show monotonic dependence on the layer thickness; see content of spectral components at both injection wave numbers
Fig. 2(c). The slopes of the linear parts of the curves also show decreases with the increasing electrolyte thickness [see Figs. 4.
monotonic dependence on the thickness and can be used to and 5(a)]. It has to be mentioned that not only does the absolute
estimate the dissipation rate from the measurement. value of the energy content decrease with increasing thickness,

056315-3
L. BARDÓCZI, M. BERTA, AND A. BENCZE PHYSICAL REVIEW E 85, 056315 (2012)

initial transient final

(a)
time: 0 s time: 3 s time: 15 s 10
1 1
0.5
3 mm 0
-0.5 0.1
-1 0.1 k/k 1

E(k) (10 8arb. units)

inj
0.6 (b)
0.3 10
5 mm 0
-0.3 1
-0.6
0.1
0.4 0.1 1
6 mm 0.2 k/k inj
0
-0.2 (c)
-0.4
10
0.2 1
8 mm 0.1
0 0.1
-0.1 0.1 1
-0.2 k/k inj

FIG. 3. (Color online) Evolution of vorticity fields. Columns: FIG. 5. Experimental k spectra of (a) initial, (b) transient, and
initial, transient, and stationary regimes. Rows: 3, 5, 6, and 8 mm (c) stationary regimes.
thickness. Units are the same in each plot; color scales are adjusted
to the maximum vorticity of the stationary states. it was found to be less than 5% for the fitting procedure. If
we consider the errors caused by the measurement technique
but also the ratio of the energy content of the two injection and the data processing methods (10%), the total estimated
scales changes. average error of the spectral exponents is less than 12%. Data
In the transient and stationary phases at k = are listed in Table I and plotted on Fig. 6.
[200,500] m−1 , the energy spectra follow a power-law depen- Another important rule has been established which de-
dence, which is a signature of the presence of inertial range in scribes the dynamics of the transition to the condensate state.
the energy cascade. It is also important to note that the spectral The time evolution of the maximum energy in the k space
exponent varies as the fluid thickness changes, which means [see Fig. 7(a)] was computed by fitting a parabola around
that the presence of linear dissipation in a quasi-2D turbulent the maximum. During the inverse energy cascade, the energy
system greatly affects the spectral exponent [see Figs. 5(b) and accumulation toward lower wave numbers is linear in log-log
5(c)]. It was found that around 5 mm thickness the spectrum scale, described by the following rule:
follows the −5/3 law and in the condensate state this power
is around −3.23, which is in excellent agreement with the kmax ∝ t −β . (4)
previous experiment [10], where it was found to be −3.3. The
relative error of the spectral exponents has been estimated, and The final average wave number (which is approximately the
inverse average size of the condensate) and the duration of the
3 mm final spectral rearrangement versus the dissipation rate are plotted
5 mm
(arb. units)

10 10 in Figs. 7(b) and 7(c). These plots show that with increasing
Time (s)

Time (s)

transient 0.1 fluid thickness (corresponding to decreasing dissipation) the

1 1
0.05 duration of the condensation process also increases [see
initial
0.1 0.1
0
0 2 4 6 8 10 0 2 4 6 8 10 TABLE I. Measured spectral exponents for different layer thick-
k (10 2 m -1) k (10 2 m -1) nesses in different states of the flow.
6 mm 8 mm
(arb. units)

10 10
Transient state Stationary state
Time (s)

Time (s)

0.1
1 1 h Spectral Rel. error h Spectral Rel. error
0.05
0.1 0.1
(mm) exponent (%) (mm) exponent (%)
0
0 2 4 6 8 10 0 2 4 6 8 10 3 −1.38 13 3 −2.86 11
k (10 2 m -1) k (10 2 m -1) 5 −1.73 11 5 −3.20 10
6 −2.42 11 6 −3.45 11
FIG. 4. (Color online) Time evolution of k spectra at 3, 5, 6, and 8 −4.12 10 8 −4.46 10
8 mm thicknesses.

056315-4
INVERSE ENERGY CASCADE AND TURBULENT . . . PHYSICAL REVIEW E 85, 056315 (2012)

log10[rad.disp.(arb. units)]

log10[rad.disp.(arb. units)]
-1
transient
stationary 0 0
-2 -1 α ~ 1.7 -1 α ~ 0.9
Spectral exponent

-2 3 mm
-2 3 mm
5 mm 5 mm
-3 -3 6 mm
-3 6 mm
-4 8 mm
-4 8 mm

0 1 0 1
-4
log10[time steps] log10[time steps]

-5 FIG. 8. Radial displacement (a) in the transient state and (b) in

the final state.
2 3 4 5 6 7 8 9
h (mm) some observations have been reported about the effect of
condensates on radial particle transport [5,10,15], in this
FIG. 6. (Color online) Measured spectral exponents as functions section we report a quantitative study to describe the effect of
of thickness h. spectral condensation on particle transport, and this can also
be considered as an independent experimental verification of
Fig. 7(c)] and at the same time bigger structures build up the transport reduction using a different quantitative method
[see Fig. 7(b)]. through three transport parameters. The trajectories of about
2300 virtual particles were computed in the measured velocity
D. Particle motion fields by the following expression:
As the spectral energy is transferred to lower wave ri (tj + t) = ri (tj ) + vi (ri (tj ),tj )t, (5)
numbers, the appearance of condensates cause a significant
decrease in the tracer particles’ diffusion level and triggers where ri (t) is the location of the ith particle at time tj , and
ordered motion along the azimuthal coordinate. Although vi (ri (t),t) is the velocity of this particle. The motion of tracers
was followed for T = nt, where T was long enough for
particles in the condensate state to get from one edge of the
k max(10 2 m -1)

6 (a) container to the other. All the parameters of this computation

were fixed at each experiment.
The convective motion was measured with a high accuracy,
2 while the diffusive motion of the particles is integrated over
4 × 4 mm2 areas, defined by the size of the interrogation
1.2 windows of the PIV method.
0.1 1 10 According to the theory of Brownian motion the average
time (s) squared radial displacement of particles is d(t) = [ri (t) −
ri (t = 0)]2 i ∝ t α . The motion is subdiffusive if α  1, dif-
fusive if α = 1, and superdiffusive if α  1 [16]. In the
1.5 (b)
k min (10 2 m -1)

transient regime the value of the diffusion exponent α indicates

1.4 superdiffusive radial transport; see Fig. 8(a). Since α drops
1.3 from 1.7 to 0.9 in the stationary state, the radial transport
becomes subdiffusive [see Fig. 8(b)]. We can also conclude
1.2
that the diffusion exponent does not depend on the control
1.1 parameter (h).
0.5 1 1.5 According to the parameter α, which globally characterizes
ε (arb. units) turbulent transport in terms of diffusivity, two more specific
quantities can be defined, in order to describe the particle
25 (c) transport in the azimuthal and radial directions:
20
1 
N
tIC (s)

Lm (t) = maxj,k [|ri (tj ) − ri (tk )|], (6)

15 N i=1

10 N 
1  n
σR(t) =  [ri (tj ) − ri (t)t ]2 . (7)
0.5 1 1.5 N i=1 j =1
ε (arb. units)
σR(t) measures fluctuations around the most likely radial
FIG. 7. (a) Time evolution of the most energetic wave number, position, and Lm (t) characterizes the dynamics of buildup of
(b) average of the most energetic wave number in the condensed the ordered azimuthal motion of the fluid.
state (inverse condensate size), and (c) duration of the spectral Figure 9 shows the clear effect of the condensation process
rearrangement versus dissipation. in each experiment. During the inverse energy cascade the

056315-5
L. BARDÓCZI, M. BERTA, AND A. BENCZE PHYSICAL REVIEW E 85, 056315 (2012)

electromagnetic fields. We have concentrated our analysis

L m(arb. units)

σR(arb. units)
on the changes of the spectral behavior and the particle
3 mm transport.
5 mm From the spectral point of view we can conclude that the
6 mm
8 mm condensation process builds up larger structures in thicker
0 10 20 30 40 0 10 20 30 40 layers where the dissipation rate is lower, and the transition
time (s) time (s) time from the disordered state to the condensed state was
found to be longer. During the spectral energy redistribution
FIG. 9. (a) Averaged maximum displacement and (b) averaged process, the most energetic wave number follows the power-
variance of the radial coordinate. law behavior kmax ≈ t −β . The measured spectral exponents
show monotonic dependence on the fluid thickness. Around
variance of the radial coordinate [σR(t) , Fig. 9(b)] decreases 5 mm thickness we see a good agreement with the Kol-
to ≈60% of its maximum. Saturation of the length of the mogorov turbulence theory and some previous experimental
maximum displacement indicates that the particles make an results [13,17].
ordered azimuthal motion around the center. A quantitative method has been developed for determina-
We can conclude that the formation of a condensate reduces tion of the turbulent particle transport behavior in different
the radial transport and orders the particle motion along the states. The result of the analysis shows that the transport
azimuthal coordinate. turns to subdiffusive in the stationary regime from the su-
perdiffusive transport of the transient state. The condensation
IV. CONCLUSIONS triggers ordered motion along the azimuthal coordinate and
In the present work we have studied the formation causes significant suppression of the radial particle transport
of condensed quasi-2D flow placed in externally driven (≈60%).

[1] R. H. Kraichnan, Phys. Fluids 10, 1417 (1967). [10] M. G. Shats, H. Xia, H. Punzmann, and G. Falkovich, Phys. Rev.
[2] L. M. Smith and V. Yakhot, J. Fluid Mech. 274, 115 Lett. 99, 164502 (2007).
(1994). [11] G. Boffetta, A. Cenedese, S. Espa, and S. Musacchio, Europhys.
[3] J. Sommeria, J. Fluid Mech. 170, 139 (1986). Lett. 71, 590 (2005).
[4] J. Paret and P. Tabeling, Phys. Fluids 10, 3126 (1989). [12] M. G. Shats, D. Byrne, and H. Xia, Phys. Rev. Lett. 105, 264501
[5] M. G. Shats, H. Xia, and H. Punzmann, Phys. Rev. E 71, 046409 (2010).
(2005). [13] J. Paret and P. Tabeling, Phys. Rev. Lett. 79, 4162 (1997).
[6] A. Bencze, M. Berta, S. Zoletnik, J. Stockel, J. Adamek, and [14] M. Raffel, Ch. E. Willert, S. T. Wereley, and J. Kompenhans,
M. Hron, Plasma Phys. Control. Fusion 48, S137 (2006). Particle Image Velocimetry: A Practical Guide (Springer, Berlin,
[7] P. H. Diamond, S-I. Itoh and T. S. Hahm, Plasma Phys. Control. 2007).
Fusion 47, R35 (2005). [15] M. G. Shats and H. Xia, Plasma Fusion Res. 4, 012 (2009).
[8] A. Fujisawa, Nucl. Fusion 49, 013001, (2009). [16] R. Balescu, Phys. Rev. E 51, 4807 (1995).
[9] Andrew P. L. Newton and Eun-Jin Kim, Phys. Plasmas 18, [17] J. Paret, M. C. Jullien, and P. Tabeling, Phys. Rev. Lett. 83, 3418
052305 (2011). (1999).

056315-6