You are on page 1of 45

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/230603184

Modeling magnetohydrodynamic turbulence by low-dimensional dynamical


systems

Chapter · January 2008

CITATIONS READS

0 79

8 authors, including:

Vincenzo Carbone Fabio Lepreti


Università della Calabria Università della Calabria
407 PUBLICATIONS   7,192 CITATIONS    108 PUBLICATIONS   1,112 CITATIONS   

SEE PROFILE SEE PROFILE

Giusy Nigro Luca Sorriso-Valvo


Università della Calabria Italian National Research Council
48 PUBLICATIONS   344 CITATIONS    191 PUBLICATIONS   3,178 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

survey of Southern Italy View project

Sediment-Turbulence Complex Interactions View project

All content following this page was uploaded by Giusy Nigro on 28 May 2014.

The user has requested enhancement of the downloaded file.


Research Signpost
37/661 (2), Fort P.O., Trivandrum-695 023, Kerala, India

Anomalous Fluctuation Phenomena in Complex Systems: Plasmas, Fluids, and Financial Markets,
2008: 57-100 ISBN: 978-81-308-0255-8 Editors: Claudia Riccardi and H. Eduardo Roman

Modeling magnetohydrodynamic
3 turbulence by low-dimensional
dynamical systems
Vincenzo Carbone1, Fabio Lepreti1, Giuseppina Nigro1
Luca Sorriso-Valvo2, Antonio Vecchio1 and Pierluigi Veltri1
1
Dipartimento di Fisica and CNISM Universitá della Calabria, Via P. Bucci Cubo
31C, 87030 Rende (CS), Italy; 2Liquid Crystals Laboratory, CNR, Via P. Bucci
Cubo 33B 87030 Rende (CS), Italy

Abstract
This chapter will be focused on the scaling
properties of low-frequency magnetohydrodynamic
turbulence, as observed both in laboratory and
astrophysical plasmas. The presence of anomalous
scaling laws are interpreted in terms of deviation of
turbulent fluctuations from a Gaussian statistics, that
is the presence of intermittency, one of the basic
properties of fully developed turbulence. This
particular stochastic process can be often investigated
through low-order dynamical models, usually called
Correspondence/Reprint request: Dr. Vincenzo Carbone, Dipartimento di Fisica and CNISM Universitá della
Calabria, Via P. Bucci Cubo 31C, 87030 Rende (CS), Italy. E-mail: carbone@fis.unical.it
58 Vincenzo Carbone et al.

shell model that is able to describe the dynamical behavior of the energy
cascade, and can describe the main statistical and dynamical properties of
intermittency of real plasma systems.

1. Introduction
Turbulence is a highly stochastic state of fluid flows, where by fluids we
mean continuously movable and deformable media. Liquids, gases and
plasmas are considered to be fluids when the scale of observation is much
larger than the molecular mean free path. Historically the theory of turbulence
rises in context of fluid dynamics (Frisch, 1995). Everyone knows that
turbulence has to do with vortex production and interaction (cfr. Fig. 1). This is
even embedded in the Latin etymology of the word “turbulence”: turba for
crowd and turbo for vortex. Namely, a turbulent flow can be described as a
crowd of vortices in nonlinear interaction.
Many generations of scientists have struggled valiantly to understand both
the physical essence and mathematical structure of turbulent fluid motion. In
1507 Leonardo da Vinci named the phenomenon he observed in swirling flow
la turbolenza, described by the following picture: Observe the motion of the
surface of the water, which resembles that of hair, which has two motions, of
which one is caused by the weight of the hair, the other one is caused by the
direction of the curls; thus the water has eddying motions, one part of which is
due to the principal current, the other to the random and reverse motion. Two
aspects of Leonardo’s observations remain with us today: the separation of the

Figure 1. Turbulence observed from Crooked Bridge in Mostar (Bosnia– Herzegobina),


where one can note different wakes downstream some stones (picture taken by V.C.).
Magnetohydrodynamic turbulence in low-dimensions 59

flow into a mean and a fluctuating part, anticipating the analogous Reynolds
decomposition (Reynolds, 1895) of the fluid velocity, and the Leonardo’s
identifications of eddies as intrinsic elements in turbulent motion.
In the early to mid 19th Century, Navier and Stokes introduced differential
equations which describe the rate of change of momentum at each point in a
viscous fluid, that in the incompressible case read

(1)

where v(r, t) is the velocity field fluctuations, P (r, t) is the pressure, ρ is the
density (constant in the incompressible case), and ν is the kinematic viscosity,
namely the viscosity divided by the density. This equation (when multiplied by
ρ to get force per unit volume) is simply Newton’s law for a fluid particle.
Incompressibility means that the velocity field is divergenceless, say ∇ ⋅ v = 0.
Using experiments on pipe flow, Reynolds (1895) identified a single
dimensionless parameter, now called the Reynolds number, and denoted by Re,
that characterizes the flow behavior in turbulent regimes. The Reynolds
number is the ratio of the nonlinear term, responsible for the flow instability, to
the linear dissipative damping, which converts kinetic energy into thermal
energy in the Navier-Stokes equations

(2)

where L represents the typical dimension of the physical system under study
and uL a typical value of the velocity. The Reynolds number defines the
dynamical regime for fluid flows, namely fluids with the same value of the
Reynolds number behaves in the same way. Fully developed turbulence
corresponds to flows where nonlinear convection is dominant, i.e. is larger
than linear dissipation just by a factor of the order of Reynolds number. We
will focus on fully developed turbulence, namely the limit of very large
Reynolds numbers, which corresponds either to very large velocities (strong
advection), and/or very small viscosity (weak dissipation, which tends to a
constant as the Reynolds number tends to infinity), and/or very large turbulent
scales.
Taylor (1938) proposed a probabilistic/statistic approach based on
averaging over ensembles of individual realizations, although he soon replaced
ensemble average by time averages at a fixed point in space. Taylor used the
idealized concept (originally introduced by Lord Kelvin in 1887) of
statistically homogeneous, isotropic turbulence. Homogeneity and isotropy
imply that spatial translations and rotations, respectively, do not change the
average values of physical variables.
60 Vincenzo Carbone et al.

Richardson (1926) proposed a pictorial description of turbulence in gen-


eral called Richardson’s cascade in which the large eddies characterized by
large scales (where the energy is injected) interacting nonlinearly, fragment
themselves into other eddies of smaller and smaller scale, down to scales
where the dissipative mechanism become efficient, interrupting the nonlinear
energy transfer. The presence of an energy cascade can then be thought as
follows. Looking at the Navier–Stokes Eq. (1) we can realize that the Reynolds
number can be written as the ratio between two different characteristic times,
namely Re = τD/τL. Here τD ∼ L2/ν is the time required for the dissipative term
to be active at the scale L, and represents the time required for the
nonlinear term to be active at the scale L. At that scale the energy (per unit
mass) injected can be estimated as while the energy
(per unit mass) dissipated is This means that εL/εD ∼
Re, namely, in the case of fully developed turbulence where Re >> 1, the
energy dissipated at the scale L is negligible with respect to the energy injected
at that scale. In other words in a turbulent flow we inject as much energy as
that the system can dissipate through the dissipative term. The exceding energy
must be eliminated in some way, and a turbulent system “chooses” to transfer
energy to smaller scales where the dissipation is more active. Then in this
description energy per unit mass, introduced at large scale, is cascading down
the hierarchy of eddies, and is removed by dissipation at the dissipative scales.
Kolmogorov (1941a,b,c) published three papers that provide some of the
most important quoted results of turbulence theory. These results comprise what
is now referred to as the K41 theory. To construct this theory Kolmogorov
supposed that the injection of energy is confined to the large scales, while the
dissipation of energy is confined to the small scales. This assumption allows the
definition of an intermediate range of scales, called the inertial range, where the
flow behaves in a conservative manner and completely separated from both the
injected and dissipative range. Kolmogorov looked at turbulence as a su-
perposition of three regions: i) a large scale energy-containing eddies which are
not universal but reflect both the mechanism that gives rise to them and the
boundary conditions that contain them; ii) an intermediate inertial-range of scales
where eddies transfer the energy in a loss-less fashion; iii) a small scale range
where eddies are dissipated. He also supposed that the Navier-Stokes solutions
tend to a stationary state at large time and that in this state the mean dissipation
rate remains finite in the fully developed turbulence regime.
Generic properties of turbulence may be described through the statistics of
velocity differences between two locations separated by a distance l, denoted
as δv(r, r + l) = v(r) − v(r + l). Generally, since measurements are mainly
available in one spatial dimension, it is used the longitudinal velocity increment
at the scale A, defined as
Magnetohydrodynamic turbulence in low-dimensions 61

(3)

(the scale A being defined A = |l| and eA is the unit longitudinal vector). Since
are random variables, their properties are defined when their statistical
properties are known. Then we can define the so called structure functions,
namely the moments

(4)

where p is the order and the angle brackets denote an ensemble average, that is,
an average over many statistically independent realizations of the flow, which
coincides with time average, when an ergodic hypothesis can be assumed. By
definition the structure functions are related to the probability density functions
of fluctuations. In the case of locally homogeneous and isotropic fully
developed turbulence, in which it is possible to consider a probabilistic/statistic
approach as that proposed by Taylor, Kolmogorov derived the famous four-
fifth’s law. This law describes an exact relationship for the third-order structure
function in the inertial range, namely (Kolmogorov, 1941a)

(5)

This is probably the only exact and nontrivial result of turbulence (Frisch,
1995). Here εD, namely the energy-dissipation rate per unit mass, remains
finite when ν vanishes (Frisch, 1995). The above relation means that the
probability density function is asymmetric, that is an energy cascade is at
work, and the negative sign means that the energy cascades towards smaller
scales. Actually, relation Eq. (4) is the exact proof that turbulence
phenomenology can be described by an energy cascade towards smaller scales.
During the last fifty years, physicists explored the case of statistically
stationary, homogeneous and isotropic turbulence, and under which
hypotheses, assuming periodic boundary conditions, it is possible to represent
turbulent fields in terms of Fourier modes. In fact, to look at such a complex
phenomenon it is often useful to investigate the behavior of the Fourier
coefficients of the fields. Under the assumption of periodic boundary
conditions the α-th component of velocity field can be Fourier decomposed as

(6)

where k = 2πn/L and n is a vector of integers. When Fourier transforms are


used in the Navier-Stokes equation we get
62 Vincenzo Carbone et al.

(7)

where the coupling coefficients are defined as

(8)

While the dissipation term is optimally represented in the Fourier space,


because Fourier modes diagonalize the Laplacian operator (for periodic
boundary conditions), the nonlinear convective term is very complicated in
Fourier space where it becomes a convolution sum, i.e. all Fourier modes are
involved. The quadratic nonlinearities of the equations correspond to a
convolution term involving wave vectors k, p and q related by the triangular
relation p = k − q. Fourier coefficients locally couple to generate an energy
transfer from any pair of modes p and q to a mode k = p + q.
Even if most progress has been made, turbulence remains yet an unsolved
problem because our traditional conceptual and technical tools are inadequate.
For instance, classical Hamiltonian mechanics describes steady states of
conservative systems, but turbulent flows are non–stationary and dissipative.
Classical dynamics only solves systems with a few degrees of freedom, while
fully developed turbulent flows have a very large, perhaps even infinite,
number of degrees of freedom. Classical statistical theories deal with closed
reversible systems in thermal equilibrium, but turbulent flows are open
irreversible systems out of thermal equilibrium. Classical mathematical
methods solve linear differential equations, but cannot (apart from a very few
cases) integrate analytically the nonlinear partial differential equations
encountered in the study of turbulence. In fact, even the existence and
uniqueness of solutions of the Navier Stokes equations describing the fluid
motions is an unsolved problem when nonlinear advection becomes dominant,
i.e. in the fully developed turbulent regime.

1.1 Turbulence in astrophysical phenomena


When a fluid is electrically conducting, turbulent motions are accompanied
by magnetic field fluctuations. Usually we deal with ordinary fluids, since conduct-
ing fluids (plasmas) are rare in our terrestrial world, while they are dominant in all
Universe. In fact magnetic fields permeate the whole Universe: they are found in
planets, stars, accretion discs, galaxies, clusters of galaxies, and the intergalactic
medium. While there is often a component of the field that is spatially coherent at
the scale of the astrophysical object, the field lines are in general tangled
chaotically and there are magnetic fluctuations over a wide range of scales. The
cause of this disorder is the turbulent state of the plasma in the systems.
Magnetohydrodynamic turbulence in low-dimensions 63

The plasma turbulent state has a very large diffusion in the Universe:
many astrophysical objects are characterized by this state. A turbulent state is
also found for gas of stars, interstellar dust, and plasma in galaxies and inter-
galaxy space. The evidence for the turbulent nature of the interstellar medium
has been known for more than half a century (Münch, 1958). Turbulence in the
solar wind is known since 1960s (Coleman, 1968) and in the 1970s and 1980s,
impressive advances have been made in the knowledge of turbulent
phenomena in the solar wind (Bruno & Carbone, 2005). In the same years the
idea of a turbulent state in the solar and stellar corona starts to appear in
scientific publications. Moreover, turbulence is one of the most efficient keys
for a better understanding of different observations as the generation of large-
scale magnetic fields accompanying many celestial objects by the turbulent
dynamo effect (Biskamp, 2000), the dynamics of stellar winds and their
interaction with planetary magnetospheres (Goldstein & Roberts, 1999), the
discrepancy between observed and predicted life-time of star-forming
molecular clouds in the interstellar medium (Zweibel, 1999), and the angular
momentum transport within accretion disks prone to magnetorotational
instability (Hawley, 1999).
We can try to understand turbulence in astrophysical phenomena as
follows. In many astrophysical plasmas the conductivity is very high, so that
the magnetic field lines are entrained by (frozen into) the fluid motion. The
fields are stretched and bent by the turbulence. The turbulent advection of the
magnetic field’s back reaction together gives rise to the statistically steady
state of fully developed magnetohydrodynamic (MHD) turbulence. In this
state, energy and momentum injected at large scales (object–size scales) are
transferred to smaller scales and eventually dissipated.
Though magnetic field fluctuations occur on all scales, magnetic fields are
most important at macroscales, i.e., mean wavelengths exceeding the internal
plasma scale lengths, such as the ion gyroradius. In this regime the MHD
approximation provides the appropriate framework. This approximation holds
when the characteristic spatial scales of the system are much larger than the
mean free path of its microscopic constituents and the shortest resolved time
scales are those of the fast magnetosonic wave. Moreover, MHD regards the
plasma as a quasi–neutral, electrically conducting, single fluid featuring vis-
cosity and resistive dissipation. Actually, in this framework, the dissipation
processes, independently of their nature, serve only as energy sinks which cut
off the spectrum of turbulent fluctuations at small scales, but do not affect the
main turbulent scales.
Since MHD turbulence is related to hydrodynamic turbulence, by
following similar equations one may apply, and generalize, the formalism
developed for the latter. However it is important to keep in mind that, though
there are many analogies between hydrodynamics and MHD, the development
64 Vincenzo Carbone et al.

of turbulence in the presence of magnetic fields differs significantly from the


nonmagnetized case because of the inherent anisotropy introduced by magnetic
field (Iroshnikov, 1963; Kraichnan, 1965; Golreich & Sridhar, 1995, 1997;
Bruno & Carbone, 2005).

1.1.1 MHD theory


The MHD equations can be derived in a formally consistent way from the
conservation laws of mass, momentum and electric charge in combination with
Maxwell’s equations (Braginskii, 1965). This theory is also called one fluid
theory, because one considers the fluid constituted by only ions while the
electrons contribute only to the current. We can remember that the two fluid
equations (the equations of mass continuity, motion and energy) could be
obtained from Boltzmann equations for electrons and protons distribution
function. Then the zeroth, first and second moments give the equations of
continuity, momentum and energy for each species, while the fluid equation of
motion and generalized Ohm’s law follow from the sum and the difference of
the momentum equations for electrons and protons.
MHD equations (Krall & Trivelpiece, 1986) can be written as follows

(9)

(10)

(11)

(12)

where ρ is the mass density of the plasma, v is the fluid velocity, B is the
magnetic field and P is the pressure. Equation (9) represents the continuity
equation, while Eq. (11) is the momentum equation. In this equation, ρν and ζ
are the first and the second viscosity coefficients, respectively. The equation
that describe the evolution of the magnetic field, namely Eq. (12), has been
derived by the usual induction equation in absence of the displacement current,
the electric field being derived from the generalized Ohm’s law
(13)

where η is the magnetic diffusivity, j is the current density, E is the electric


field. Finally
Magnetohydrodynamic turbulence in low-dimensions 65

represents the diffusion term for the magnetic field. After few algebra we can
write the MHD equations for an incompressible plasma

(14)
(15)

(16)

(17)

The above equations imply the existence of magnetohydrodynamic waves


(Alfvén and magnetosonic waves) when linearized around a uniform and
constant magnetic field (Landau & Lifshitz, 1984). It is also interesting to
notice, for what follows, that Eqs. (14)-(17) admit exact nonlinear solutions in
the form of Alfvén plane waves of arbitrary amplitude, where
v and B are components transverse to the direction of the propagation of the
wave (for example the x-axis) and the functional dependence B(x ± cAt) is
arbitrary (4πρ ) is the Alfvén velocity). These solutions require
that the total pressure P + B2/8π is constant. In the case of an incompressible
fluid where ρ = const, sometimes one uses the rescaled magnetic field

The dynamic state of a MHD flow is characterized, together with the


kinetic Reynolds number (see Eq. (2)), by the non–dimensional parameter

(18)

called magnetic Reynolds number, this involves the magnetic diffusivity λ of


the MHD fluid. In Eq. (18) cA = B0/(4πρ)1/2 is the Alfvén velocity of the largest
scales, and B0 is the magnetic field of that scale. The above parameter
represents the ratio between the diffusive time and the Alfvén time. If both Re
and S became sufficiently larger than unity, an MHD flow undergoes a
transition from the laminar state, characterized by stationary stream–line
topology, to turbulence, where the fluid motion is unpredictable.
One can consider both the kinetic density energy (per mass unit)
66 Vincenzo Carbone et al.

and magnetic density energies (per mass unit) of the flow

(V denotes the volume of the system). If Ekin  Emag, the magnetic field is
passively advected by the fluid, while if Ekin  Emag the strong magnetic field
forces the fluid motion into quasi two–dimensional geometry.
In the ideal case (ν = 0, λ = 0), Eqs. (16) and (17) conserve both the total
energy of the system E = Ekin + Emag, and the cross-helicity (Biskamp, 1993),
defined as

The cross-helicity gives a measure of the degree of correlation between the


magnetic field and the velocity of the fluid. There is a third ideal invariant
conserved by MHD equations. In 3D the conserved quantity is the magnetic
helicity, defined as:

where a is the vector potential. Magnetic helicity provides a measure of the


degree of structural complexity of the magnetic field (Moffat, 1978). In a 2D
geometry, the third ideal invariant is the squared vector potential, defined as:

where ψ is defined by the condition and is an axis orthogonal to the


2D plane.
The incompressible MHD Eqs. (14)-(17) may be written in a more symmetric
form by using the Elsässer variables (Elsässer, 1950) defined by z± = v ± b, that is
(19)

(20)

This form is similar to the Navier–Stokes equation, or more precisely, this is


similar to the equation for a passive field transported by the velocity field.
Note that in this form it is easy to realize that an exact nonlinear solution is the
Magnetohydrodynamic turbulence in low-dimensions 67

Alfvénic fixed point of the equation, namely either z+ = 0 and z− ≠ 0, or vice


versa (Dobrowolny et al., 1980). The conservation of the total energy and
cross–helicity is now equivalent to the conservation of the pseudo-energies

(21)

Note that, since the incompressibility condition, the pressure field Ptot = P +
b2/2 can be obtained by solving the Poisson equation

(22)

Even in the MHD case we can investigate turbulence by using differences of


fields at two positions separated by a distance A, namely

(23)

being ψ either the Elsässer variables or the velocity or magnetic field variables.
Even in the MHD framework an exact relationship can be obtained, assuming
local homogeneity and isotropy. In fact, it can be shown that the mixed third–
order moment of Elsässer differences is related to the separation A through the
relation (Politano & Pouquet, 1998a,b)

(24)

being =± the pseudo–energies dissipation rates per unit mass defined through

where ∂|| is the gradient along the parallel (streamwise) direction and are
the increments of the parallel component of Elsässer variables. Relation Eq.
(24) is similar to the Yaglom relation for a passive scalar (Yaglom, 1949;
Antonia et al., 1997), this is because Eq. (20) is also similar to an equation for
the passive scalar. Even in fluid flows the Yaglom relation gives an equation
that is more general than the 4/5–law by Kolmogorov. In fact, using the
Navier–Stokes equation (Antonia et al., 1997; Danaila et al., 2001), it can be
shown that

(25)

from which, being (Pope, 2001)


68 Vincenzo Carbone et al.

the Kolmogorov relation Eq. (5) can be obtained.


Our approach to the MHD turbulence corresponds to a statistically station-
ary, homogeneous and isotropic state of turbulence. Under these hypotheses,
and assuming periodic boundary conditions, we may represent turbulent fields
in terms of Fourier modes, for which the α-th component of the Elsässer field
can be decomposed as

(26)

where k = 2πn/L and n is a vector of integers. When Fourier transforms are


used in the MHD equations, we get

(27)

where the coupling coefficient Mαβγ has been already defined in Eq. (8). The
incompressibility condition requires that k ⋅ z± (k) = 0.

2. Phenomenology of turbulence
In analogy with hydrodynamics, turbulence in hydromagnetic fluids can be
characterized by the statistics of fluctuations or eddies at the scale A. Here we
denote fluctuations by zA, assuming that this is a characteristic value for differences
δψA at the scale A. We can choose, for example, the r.m.s. value
Then, considering a state of fully developed turbulence where the nonlinear terms
are dominant, we may associate the eddy turn-over time at the scale A

(28)

The eddy turn-over time represents the typical time for a structure of size A to
undergo a significant distortion due to relative motion of its components.
Moreover, τA is the typical time for the transfer of excitation (energy) from
scale A to smaller ones. The energy flux per unit mass from scales A to smaller
scales is then estimated as follows

(29)
Magnetohydrodynamic turbulence in low-dimensions 69

In a stationary state, the energy flux through the scales within the inertial range
must be equal to the finite mean energy dissipation rate εD, so that from the
equality IIA ∼ εD we immediately get a scaling law for fluctuations,

(30)

where z0 represents the fields fluctuation at large scale A0. It is worthwhile to


note that MHD Eqs. (19), or the fluid equations, in absence of dissipative terms
are scale invariants. In fact, the scale transforms ∇ → r∇′ and z± → rhz±′
(where r is a scaling parameter), leave unaltered the MHD equations for each
value of h. This means that we expect solutions in the form of scaling laws.
The Kolmogorov phenomenology assumes that the Elsässer fields are scale-
invariant with exponent h = 1/3.
The Kolmogorov relation Eq. (30) means that we can predict the statistics
of fluctuations. In fact, if we assume that a scaling law is observed in the
inertial range, that is if we define the scaling exponent ζp through the relation

then the structure functions scale linearly with p

(31)

that is ξp = p/3. Using the Fourier transform we can write immediately the
relation between the second-order structure function and the spectral density at
the wave vector k ∼ 1/A,

so that from Eq. (31) we immediately obtain the Kolmogorov spectrum

(32)

which is the most famous relation within fully developed turbulence. In spite
of the simple arguments used to derive Eq. (32), and due to the fact that it is
relatively easy to measure the spectral energy, the Kolmogorov spectrum has
been observed in all experiments on turbulence. It is widely (and to a certain
extent not exactly correct) used as a testing benchmark for turbulence.
Remarkably, the 5/3–Kolmogorov spectrum is observed throughout in nature,
from laboratory on earth (Frisch, 1995) to astrophysical flows (Bruno &
Carbone, 2005).
70 Vincenzo Carbone et al.

As far as the stationary properties of cascade are concerned, we may write


the following relation for the characteristic eddy-turnover time,

(33)

When we assume that the viscous and resistive coefficients are equal, we may
define, using Eq. (27), a dissipative time as

(34)

where kD is the wave vector corresponding to the dissipative scale AD. Imposing
the condition that the dissipative time is equal to the eddy turn-over time at the
dissipative scale AD, using Eqs. (33) and (34) and the definition of the Reynolds
number, we obtain the relation

(35)

and the relation between the eddy turn-over time of large scales and the dissi-
pative time

(36)

Equation (35) indicates that the range of wave vectors where the Kolmogorov
spectrum is verified, that is the inertial range, is a positive power of Re. The
larger Re the larger becomes the inertial range.

3. The problem of intermittency in turbulence


Figs. 2 and 3 show examples of heliospsheric turbulence. The panels
represent the longitudinal velocity differences and the magnetic field intensity
differences, computed from solar wind data for three different values of the
scale (reported on the figures). The bottom curve is the large scale case, and
the signal looks like a brownian motion. As the scale decreases the signal
becomes more and more intermittent. We can observe the emergence of
localized regions whose fluctuations are stronger as the scale decreases. The
intermittent events at small scales are nicely visible on both figures. We can
expect that they play a key role in the statistics of turbulence.

3.1 Data analysis on turbulent flows


Here we give some basic concepts on data analysis of some turbulent
fields. In general, measurement results in some time series of fields, for example
through a hot wire placed at a certain height inside the turbulent boundary of the
Magnetohydrodynamic turbulence in low-dimensions 71

Figure 2. Fluctuations δuτ = u(t+τ)−u(t) as a function of time t (in days of the year) for
three different scales τ as reported on the figure, of the bulk velocity field. Fluctuations
are calculated through a turbulent sample from Helios 2 satellite in the solar wind.

Figure 3. Fluctuations δBτ = B(t + τ) − B(t) as a function of time t (in days of the year)
for three different scales τ as reported on the figure, of the magnitude of the magnetic
field. Fluctuations are calculated through a turbulent sample from Helios 2 satellite in
the solar wind.

atmosphere, we get the vector speed v(t), u(t) being the streamwise component,
and the temperature field T (t). In situ satellite measurements of both velocity
and magnetic fields B(t) can be obtained in interplanetary space, while a
turbulent magnetic field can be also measured at the edge of plasma devices in
72 Vincenzo Carbone et al.

the laboratory, mainly on toroidal Reversed Field Pinch configurations where


high–amplitude magnetic fluctuations can be detected. Quantities of interest
are the differences at a certain time scale τ of the streamwise velocity vector,
namely δuτ = u(t + τ) − u(t), and of the temperature δTτ = T(t + τ) − T(t). The
quantities δuτ are analogous to spatial differences, that is using the Taylor’s
hypothesis (Taylor, 1938), A = − τU (U being the average speed of the flow
in the streamwise direction, that in the solar wind coincides with the
direction of motion of the spacecraft, namely the radial out–of–Sun
direction). As far as the magnetic field is investigated, we use the magnitude
of the field, namely thus computing the differences δBτ
= B(t + τ) − B(t). However, we can compute also differences for a single
component of the magnetic field, or for the vector field itself (Carbone et al.,
2004b).
Scaling exponents are obtained from the p–th order structure functions
Sp(τ) = 〈(δψτ)p〉 in the inertial range (ψ represents one of the measured vari-
ables). In order to recover scaling laws we use the Extended Self–Similarity
(ESS) analysis, a data analysis technique introduced by Benzi et al. (1993).
Using the third–order structure function as a generalized scale, we can plot the
p–th order structure functions against the 3–rd order one. In this case, the range
of scales where a linear relation Sp(τ) ∼ [S3(τ)]ξp/ξ3 exists, is extended thus
allowing a better determination of the normalized scaling exponents (Carbone
et al., 1996). It is worthwhile to report the fact that scaling exponents obtained
in fluid flows using ESS, are the same as scaling exponents obtained in the
same flow without ESS (Benzi et al., 1993).
From data analysis the structure functions are calculated simply using time
averaging, but a word of caution must be given concerning the values of the
order p which can be used in the calculation. In fact the more formal definition

(P represents the probability distribution function (PDF)) clearly indicates that,


in order to get a value Sp(τ) which is meaningful, we have to make sure that
the function F (δψp(τ)) = δψp(τ)P(δψp) be integrable. In general, this defines
an upper bound on the maximum value of p, which we can reliably calculate
with a certain number of points N in our data set. As a rule of thumb, when we
have a data set of N points at our disposal, the maximum allowed pmax is of the
order of pmax  log N (for stretched–exponential PDF’s).
Typical scaling exponents of structure functions for both velocity and
magnetic field in the heliosphere, are reported in Table 1. Data were obtained
from the Helios 2 spacecraft by sampling low–speed streams. In the same table
we report, for comparison, the scaling exponents for velocity and temperature
Magnetohydrodynamic turbulence in low-dimensions 73

as obtained in a wind tunnel experiment. We report normalized scaling


exponents ξp/ξ3 calculated through the Extended Self-Similarity analysis
(Benzi et al., 1993).
Scaling exponents of magnetic structure functions, obtained from
laboratory plasma experiments on a Reversed–Field Pinch at different
distances from the external wall (Carbone et al., 2000), are reported in Table 2.
Structure functions are computed from time series of the magnetic field B(t) at
different distances r from the external wall. In laboratory plasmas it is difficult
to measure all components of the vector field at the same time. Here we report
scaling exponents obtained using differences as calculated from the radial
component of the magnetic field in the toroidal device. As it can be seen the
degree of intermittency increases going towards the external wall. This appears
to be similar to what is observed in channel flows where intermittency
increases going towards the external wall (Pope, 2001).

Table 1. Normalized scaling exponents ξp/ξ3 for velocity and magnetic variables in the
solar wind. Errors represent the standard deviations of the linear fitting. As a reference
we reported the scaling exponents of structure functions for velocity and temperature,
as calculated in a wind tunnel.

Table 2. Normalized scaling exponents ξp/ξ3 for magnetic fluctuations in a laboratory


plasma. Exponents have been calculated from different time series B(t) taken at
different distances r/R (R  0.45 cm being the minor radius of the torus in the
experiment) from the external wall, inside the toroidal device. Errors represent the
standard deviations of the linear fitting.
74 Vincenzo Carbone et al.

Scaling exponents of structure functions for Alfvén variables, velocity and


magnetic variables have been calculated also for high resolution 2D MHD
numerical simulations (Politano et al., 1998). These scaling exponents are
reported in Table 3. Note that, even in numerical simulations, the magnetic
field behaves like a passive field; intermittency for magnetic variables is
stronger than for the velocity field, even if the difference between scaling
exponents for velocity and magnetic fluctuations in numerical simulations is
smaller than the differences we found in the solar wind observations.
The linear behavior of the structure functions scaling exponents with the
moment order as deduced from K41–like phenomenology representing the
Kolmogorov scaling, are not observed in experimental data except at second
order for which the K41 theory worked quite well. But when the more and
more accurate experimental techniques permitted the investigation of higher
moments, the need for a different interpretation arised (Frisch, 1995). In Fig. 4
we report the normalized scaling exponents ζp = ξp/ξ3 for high–order moments
of velocity and magnetic field increments as measured in the Solar Wind
plasma from Table 1. As a comparison, in the same Fig. 4 we show the scaling
exponents for velocity and temperature (considered as a passive field) as
obtained in a wind tunnel experiment. As it can be seen the departure from the
Kolmogorov linear scaling is similar in all experiments, that is ζp < p/3 for p >
3 while ζp > p/3 for p < 3. What is interesting, is the fact that both velocity
fields display the same degree of intermittency (calculated as the distance
between the linear law p/3 and the actual values of scaling exponents); there is
no difference between scaling of flows on Earth and in space (Carbone et al.,
1995). This gives us the idea of a kind of universality of turbulence. The
magnetic field is much more intermittent than the velocity field, that is, as far
as the intermittent properties are concerned, the magnetic field behaves more
like the passive temperature field (Carbone et al., 2004b; Bruno & Carbone, 2005;

Table 3. Normalized scaling exponents ξp/ξ3 for Alfvénic, velocity and magnetic
fluctuations obtained from data of high resolution 2D MHD numerical simulations.
Scaling exponents have been calculated from spatial fluctuations; different times, in the
statistically stationary state, have been used to improve statistics.
Magnetohydrodynamic turbulence in low-dimensions 75

Figure 4. The normalized scaling exponents ζp as a function of the moment order p are
reported, along with the linear value p/3 (full line) expected from the K41 Theory. Data
refers to the bulk velocity (black circles) and the magnitude of the magnetic field (white
circles), as measured by the Helios satellite in the inner heliosphere at 0.9 Astronomical
Units during slow wind streams. Reported for comparison are the normalized scaling
exponents for longitudinal velocity field (stars) and the temperature field (passive
scalar) in usual fluid flows measured by Ruiz Chavarria et al. (1995).

Bershadskii & Sreenivasan, 2004). Of course this cannot mean that the
magnetic field is actually passive, statistical features are different from dynamical
properties.

3.2 Probability density functions (PDFs) of fluctuations


The presence of scaling laws for fluctuations in general is a signature of
selfsimilarity for the phenomenon at hand. In fact, the observable δψA, which
depends on a scaling variable A, is invariant with respect to the scaling relation
A → λA, when there exists a parameter µ(λ) such that

The solution of this last relation is a power law δψA ∼ Ah with scaling exponent
h = − logλ µ. Then the ratio of fluctuations at two scales, namely

depends only on the value of h, that is we cannot define any characteristic


scale. This means that PDFs of scaling variables are related through the
relation
76 Vincenzo Carbone et al.

Let us consider the standardized variables

It can be easily shown that when h is unique, say in a pure self-similar situation,
PDFs are such that P (yA) = P (yλA), namely by changing scale PDFs collapse.
In Fig. 5 we report PDFs for the normalized fluctuations δuτ /〈(δuτ)2〉1/2 of
velocity, as observed in atmospheric flow, and fluctuations δBτ /〈(δBτ)2〉1/2 at
three different scales τ, for three different data sets, namely a set within the solar
wind, inside a laboratory plasma and in 2D numerical simulations. It appears
evident that the global self-similarity in turbulence is broken. PDFs at different
scales do not collapse, their shape seems to be strongly dependent on τ. In
particular, at large scales PDFs look almost gaussian, but they become more
stretched as τ decreases. At the smallest scale PDFs are stretched exponentials.
This scaling dependence of PDFs is a different way to say that scaling exponents
of fluctuations are anomalous, which is a different definition of intermittency.
Note that the wings of PDFs are higher than a Gaussian function. This implies
that large fluctuations have a probability of occurrence greater than what they
would have if they were normally distributed. In other words, anomalously large
stochastic fluctuations are less rare than we should expect from the point of view
of a Gaussian approach to the statistics of turbulence.
Intermittency generates rare and large events, which can be seen as
coherent structures present on all dynamically interesting scales. The times tj
of the occurrence of the maxima for the time evolution of both |δυτ| and |δBτ|,
can be extracted from time series using a threshold (Boffetta et al., 1999). Then
for each scale τ we can get a set of waiting times ∆t = tj+1 − tj. The distributions
P (∆t) for both velocity and magnetic variables, at a given scale, are reported in
the bottom panels of Fig. 5. As can be seen, a power law is recovered

with different values for the scaling exponents β. The presence of a power law
instead of an exponential decay, is a signature that the process underlying the
formation of rare bursts does not follow Poisson statistics implying a certain
degree of memory.

3.3 Intermittency cannot simply means on-off


It is perhaps interesting to spend a few words about the different meaning
of the word intermittency which can be usually encountered in the literature. In
Magnetohydrodynamic turbulence in low-dimensions 77

Figure 5. In the first three panels we report PDFs of fluctuations, at three different
scales τ, for three different experiments, namely the atmospheric flow (we report
fluctuations of velocity δuτ), the solar wind and a laboratory plasma (we report the
fluctuations of magnitude of magnetic field δBτ ). Note the same scaling behavior of
τ

PDFs, even if scales τ are completely different for each experiment. The three bottom
panels refer to probability distribution functions of waiting times ∆t between structures
at the smallest scale for each experiment. The PDFs of waiting times behave like P(∆t)
∼ (∆t)−β, values of β for each experiment are reported on the figures.

fact, the same word usually refers to some different (and often contrasting)
kinds of phenomena, and confusion may arise. Intermittency in fully developed
turbulence is what we have just described in the previous sections, that is the
departure from a global self-similarity in fluid or magnetofluid systems (cfr.
Frisch (1995) for references). Intermittent transition to chaos happens when
the time evolution of some chaotic systems show alternation between laminar
periods and stochastic periods. This happens when the tuning parameter is set
close enough to the critical value which define the order-to-chaos transition.
As the parameter becomes closer to that critical value, the durations of laminar
periods decrease as a power-law (Pomeau & Manneville, 1980), with universal
behavior recognized in some systems. Intermittency in self-organized systems
78 Vincenzo Carbone et al.

appears in some out-of-equilibrium systems that relax through bursts showing


typical 1/ƒα spectra. These systems can be simulated by cellular automata, and
from a theoretical point of view they can be seen as self-similar systems at the
borderline of chaos in a self-organized critical state. These systems, when
continuously perturbed, show a typical sequence of intermittent isolated
events. The classical example is the sandpile model (Bak et al., 1987), where
the critical state is represented by a critical slope of the pile. As a perturbation
is introduced, in the form of random addition of sand at random sites, and the
local slope of the pile exceeds a critical slope, the system relaxes through a
burst of activity, that is an avalanche. The ensemble of successive avalanches
generates an intermittent sequence of chaotic bursts (Bak et al., 1987). Since
the critical slope is an attractor, avalanches are Poissonian events, the waiting
times between avalanches being distributed according to an exponential
function (Boffetta et al., 1999). Finally On-off intermittency is observed in
chaotic systems externally driven through, for example, stochastic
perturbations. The system, which has been set to lie normally in a laminar state
(an attractor), occasionally is pushed away from this state, thus generating
intermittent turbulence bursts. The waiting times between successive bursts are
distributed as power laws (Platt et al., 1993), with universal scaling exponents.
For a nice and very simple example, one can study the behavior of the logistic
map xn+1 = anxn (1 − xn) with variable parameter as, for example, an = 1/2 with
probability p (say p = 0.34) and an = 4 with probability 1 − p.

3.4 The multiplicative cascade model


To understand the way the phenomenology must be changed to include
intermittency in turbulent fields, the picture of the Richardson cascade can be
modified. One of the main points on which the Kolmogorov K41 theory was
based is the fact that the actual spatial statistics of the energy dissipation rate ε
does not come into play. The idea of universality implied by the model suggests
a uniform distribution. However only the global mean value of the energy
transfer rate is constant through the cascade, while its local value can be a
(stochastic) fluctuating function, with both bursty and quiet zones alternatively.
In this framework the presence of strong activity regions must be scale-dependent,
showing the concentration of active structures on definite positions in space, and
such concentrations become more and more evident as the scale decreases.
Kraichnan (1974) pointed out that in order to obtain phenomenological
information about inertial range quantities, the energy dissipation rate
which appears in the phenomenology must be replaced by the local energy
transfer rate, namely As a consequence the p-th order structure
function will scale taking into account the scaling of fluctuations of the energy
transfer rate, namely
Magnetohydrodynamic turbulence in low-dimensions 79

Then, by assuming a scaling law for the energy transfer rate the
correction to the scaling exponents of the structure functions ξp = p/3+τp/3 due to
intermittency, comes from the scaling behavior of ε. This opens a “Pandora's
box” of possibilities (Kraichnan, 1974) to model the energy transfer rate, and to
compare the scaling exponents of the model with that of real experiments. The
most common way to recover a model is to interpret the energy cascade as a
multiplicative process, according to the Richardson's picture. In this framework
the energy transfer rate at a generic scale εn = εAn, where An = 2−n L0, is viewed as
a stochastic variable, computed as the result of a multiplicative process,

The statistics of the scaling exponents ξp depend on the statistics of multipliers


βi, say assuming that all multiplicative factors be derived from the same
process, we get

Then, giving a model for the cascade we can try to work out an expression for
the statistics of β, and then obtain a model for ξp. The model can be fitted to
the various data sets in order to derive the values for the parameters of the
model. The most common models encountered in the literature are reported in
Frisch (1995).

3.5 The multifractal approach


An elegant way to describe the occurrence of intermittency in turbulence
has been worked out by Frisch & Parisi (1985), who introduced the
multifractal model for intermittency. The multifractal idea is to consider a
continuous spectrum of possible values of the scaling exponent h. That is, the
space is described as made of an infinite number of subsets G(h), each one of
fractal dimension D(h), and described by the scaling exponent h(x) (Frisch,
1995). Summing together all the subspaces contributions to the scaling of the
fields yields,

The structure functions can now be considered as the superposition of infinite


power-laws, each one representing the set where its exponent is valid. Since (A/L0)
 1 , the leading exponent for each value of the order p is the minimum, so that
80 Vincenzo Carbone et al.

Here, µ(h) is the measure representing the probability distribution function of


the exponents h. The integral can be solved with the saddle point method and
leads to

This expression can be inverted for a given value of p. Using the model, at a
fixed value of p, we select singularities of order h within a set of fractal
dimension D(h), where the scaling δυA ∼ Ah holds.

3.6 Scaling behaviour of PDFs


Several models can be introduced to capture also the scaling behavior of
PDFs of increments of fields. The fundamental relation linking such scaling is
of course

(37)

which needs some further assumptions on the energy cascade in order to get a
model for PDFs scaling.
Within the multifractal framework, a quantitative analysis of the
continuous scaling departure of PDFs from a Gaussian can be performed. In
fact, in order to describe the PDFs at a given scale A, two ingredients are
needed: the parent distribution at the large scale A0 and the distribution of the
energy transfer rate. In fact, a correspondence exists between the energy
transfer rate = and the variance σ of the conditioned PDFs. The dependence on
scales of the PDFs can be eliminated by looking at the PDFs conditioned to a
given value of the energy transfer rate at the scale A, so that for each scale A the
field can be decomposed into a set of Gaussian curves, each one corresponding
to a given value (or bin of values) of the energy transfer rate. The energy
transfer rate distribution can thus be represented using a distribution for the
variances of the conditioned PDFs. The PDF of the field increments is then
seen as a superposition of curves (the conditioned PDFs) whose standard
deviations are distributed according to a given phenomenological law (Sorriso
et al., 2000).
This can be done by computing the convolution

(38)
Magnetohydrodynamic turbulence in low-dimensions 81

Both the large scale PDF P0 and the distribution of the variances Lλ (σ) of the
above general relation, could in principle be determined experimentally. In a
turbulent flow it is even observed that the parent distribution is clearly gaussian,

(39)

Moreover, it is well known that large scale PDFs of turbulent increments are
gaussians, both in fluids (Frisch, 1995; Castaing et al., 1990) and in plasmas
(Tu & Marsch, 1995; Sorriso et al., 1999).
The function Lλ (σA) could be determined by computing the variances of
the conditioned PDFs, and then observing the PDF of such variances. Unfortu-
nately, a very large amount of data would be necessary in order to have enough
values of variances for their PDF to be computed. It is then useful to approach
the problem using models for the energy transfer rate. In the paper by Castaing
et al. (1990) (see also Sorriso et al. (1999)) a log–normal ansatz has been tried

(40)

but other possibilities could be easily investigated. In Eq. (40), is the


variance of the ln σA distribution, and σ0,A is the most probable value of σ at the
scale A.
When the model is used on actual turbulent data set1, a power law is
observed (Castaing et al., 1990; Sorriso et al., 1999) for the scaling behavior of
λ2, namely Perhaps this solves the difficulties encountered with the
log–normal model when used to simulate the energy transfer rate (Frisch, 1995).

4. Dynamical models for turbulence


In a 3D numerical simulation describing different behavior of a fluid
(magnetized and/or ordinary fluid), the minimum number of grid points which
are necessary to obtain information on the fields at these scales is given by N ∼
(L/AD)3 ∼ Re9/4. This rough estimate shows that a considerable amount of
memory is required when we want to perform numerical simulations with high
Re. Typical values of Reynolds numbers at present reached in 2D and 3D
numerical simulations are respectively of the order of 104 and 103. At these
values the inertial range spans approximately one decade or little more.
Because of the situation described above, the question of the best
description of the dynamics which results from the original equations, using

1
Note that the space scaling parameter A, using the Taylor hypothesis, becomes a time
scaling τ when the model is used on experimental data sets.
82 Vincenzo Carbone et al.

only a small amount of degrees of freedom, becomes an important issue. This


can be achieved by introducing turbulence models which are investigated using
tools of dynamical system theory (Bohr et al., 1998). Dynamical models then
represent minimal sets of ordinary differential equations that can mimic the
gross features of the energy cascade in turbulence. These studies are motivated
by the famous Lorenz’s model (Lorenz, 1965), which, containing only three
degrees of freedom, simulates the complex chaotic behavior of turbulent
atmospheric flows, becoming a paradigm for the study of chaotic systems.

5. Shell models of turbulence cascade


As stated in the Introduction (Sect. 1), fully developed turbulence involves
a hierarchical process, in which many scales of motion are involved through an
energy cascade. In this framework the shell model approach to describe
turbulence is viewed as a consistent and relevant tentative to approach the
energy cascade of turbulence (Bohr et al., 1998). These models mimic the
gross features of the time evolution of spectral Navier–Stokes or MHD
equations. The 3D hydrodynamic shell model is usually quoted in literature as
the GOY model, and has been introduced some time ago by Gledzer (1985)
and by Ohkitani & Yamada (1989) (for a recent review see Biferale (2003)).
The MHD shell model, which coincides with the GOY model when the
magnetic variables are set to zero, has been introduced independently by Frick
& Sokoloff (1998) and Giuliani & Carbone (1998) (see also Giuliani (1999)).
Here we derive and discuss the basic features of the GOY MHD shell model,
the hydrodynamical version of which can be obtained by imposing zero
magnetic variables. A typical shell model can be built up through four different
steps. First of all, we introduce discrete wave vectors and assign to each shell
discrete scalar variables. Then, we introduce a dynamical model which
describes nonlinear evolution and fix as much as possible the coupling
coefficients.
As a first step, we divide the wave vector space into a discrete number of
shells whose radii grow according to a power kn = k0λn, where λ > 1 is the inter
shell ratio, k0 is the fundamental wave vector related to the largest available
length scale L, and n = 1, 2, ..., N. Of course the maximum number N of shells
depends on the Reynolds number we want to investigate. The dissipative wave
vector is then given by kD = k0λnD ∼ ν 3/4, so that the required number of modes
grows as ln Re. To appreciate the reduction of modes, and then of computer
time required to numerically solve our equations, that number must be
compared to the usual power Re9/4.
Each shell is assigned two or more complex scalar variables υn(t) and bn(t),
Elsässer variables These variables describe the chaotic
dynamics of modes in the shell of wave vectors between kn and kn+1. It is worth
Magnetohydrodynamic turbulence in low-dimensions 83

noting that the dynamical variables υn(t) and bn(t), mimicking the average
behavior of Fourier modes within each shell, represent characteristic
fluctuations across eddies at the scale That is, the fields have the
same scalings as field differences, for example
in fully developed turbulence. In this way we ruled out the possibility to
describe spatial behavior within the model. We can only get, from a dynamical
shell model, time series for shell variables at a given kn, and we loose the fact
that turbulence is a typical temporal and spatial complex phenomena.
Looking at Eq. (27) a model must have quadratic nonlinearities among op-
posite variables and and must couple different shells, that is in
general,

(41)

where represent arbitrary real coupling coefficients (the symbol *


being complex conjugation). We exclude from the sum in Eq. (41) terms where
j = m and terms where j = 0 or m = 0. The first constraint comes from the fact
that in the original equations the coupling coefficient vanishes for equal wave
vectors. The second constraint comes from the Liouville theorem, which states
that the nonlinear term in the primitive equation must conserve volumes in
phase space. This means that We introduce the shell
model where couplings happen among next and next–nearest shells, that is
both j = ±1, ±2 and m = ±1, ±2.
The last step is not standard. A numerical investigation of the model might
require the scanning of the properties of the system when all coefficients are
varied. Coupling coefficients can be fixed by imposing that the system Eq. (41)
must satisfy the quadratic conservation laws of the original equations, namely
the pseudo-energies E±(t)

Then, by imposing that equation (41) must satisfy dE±/dt = 0, we get

(42)
84 Vincenzo Carbone et al.

In terms of velocity and magnetic shell variables υn(t) and bn(t), from Eq. (42)
we can write down immediately a set of equations as

(43)

(44)

Conservation of pseudo–energies E±(t) implies that these equations conserve


equivalently both total energy E(t) and cross–helicity HC (t), say

The equations we recover describe the nonlinear evolution of the shell model.
In the following, we will add to the right–hand-side of Eq. (42) the dissipative
and forcing terms that restore turbulence.

5.1 Two–dimensional and three–dimensional shell models


As we mentioned above, shell models cannot describe spatial geometry of
nonlinear interactions in turbulence, so that we lose the possibility of
distinguishing between two–dimensional and three–dimensional turbulent
behavior. The distinction is however of primary importance, for example as far
as the dynamo effect in MHD is concerned. The shell model Eq. (42) yet
contains two free parameters which can be fixed by introducing a third ideal
invariant, H(t), which can be later identified as a surrogate of the magnetic
helicity or of the squared vector potential.
As can be easily verified from Eq. (42), the invariant which can be
conserved takes the form

(45)

By imposing that Eq. (44) must satisfy dH/dt = 0, we get two classes of
models. The first class identifies a shell model where the third invariant Eq.
(45) is positive definite. When we choose α = 2, H(t) can be dimensionally
identified with the squared magnetic potential, so that this model mimics a
Magnetohydrodynamic turbulence in low-dimensions 85

kind of 2D MHD turbulence. The second class of models, obtained by


imposing that α must be complex, identifies a shell model where the third
invariant is not positive, and defined by

(46)

When the real part of α, namely αR is equal to unity, the invariant H(t) can be
dimensionally identified as the magnetic helicity and the shell model mimics a
kind of 3D MHD turbulence. Finally the most common choice λ = 2 for the
inter–shell ratio (Bohr et al., 1998) fixes the free parameters of the MHD shell
model to the values a = 5/4 and c = −1/3 for the 2D case, a = 1/2 and c = 1/3
for the 3D case.
The MHD shell model evolves in a phase space built up by considering
(υn, bn) as coordinates. When bn = 0, the phase space of the system reduces to a
subspace described by the GOY hydrodynamical shell model (Gledzer, 1985;
Ohkitani & Yamada, 1989).

5.2 Basic properties of shell models


Taking into account both the dissipative and forcing terms the GOY MHD
model can be written as

(47)

for the shell n, where F± is the forcing term, which injects energy. In the
following we will consider only the case where the dissipative coefficients are
the same, i.e. ν = µ.
The existence of a cascade towards small scales is expressed by an exact
relation which is equivalent to the 4/5–law for MHD turbulence. Using Eqs.
(47) the scale–by–scale pseudo–energy budget is given by

The second and third terms on the right–hand side, represent respectively the
rate of pseudo–energy dissipation and the rate of pseudo–energy injection. The
86 Vincenzo Carbone et al.

first term represents the flux of pseudo–energy along the wave vectors, respon-
sible for the redistribution of pseudo–energies on the wave vectors, and is
given by

(48)

Using three classical assumptions, namely: i) the forcing terms act only on the
largest scales; ii) the system can reach a statistically stationary state; iii) in the
limit of fully developed turbulence, ν → 0, the mean pseudo–energy
dissipation rates tend to finite positive limits ε±, it can be found that

(49)

This is an exact relation which is valid in the inertial range of turbulence. Then
it can be used as an operative definition of the inertial range, that is the inertial
range of the energy cascade in the shell model is defined as the range of scales
kn where the law Eq. (49) is verified.
It is worthwhile to point out that in the case of the hydrodynamical model,
apart from kinetic energy, an exact relationship exists also for the flux of
kinetic helicity (Biferale, 2003). No similar result exists for the magnetic
helicity in the MHD shell model.
The shell models contain some interesting fixed points, defined as
solutions of the nonlinear term of Eq. (42). From Eqs. (43) and (44), the main
fixed point can be cast as a scaling law for Elsässer variables In
fact by using this scaling law in Eq. (42), it is found that scaling exponents
must be related by In the case Elsässer variables have the same
+ −
scaling h = h = h, this reduces to the Kolmogorov’s scaling h =1/3, which is
in fact the only fixed point of the GOY hydrodynamical shell model.
As far as the MHD FSGC shell model is concerned, a new interesting fixed
point appears. In fact a trivial solution of Eqs. (43) and (44) is υn(t) = ±bn(t) for
each shell. This corresponds to have and (or vice versa). This
solution, even if trivial, is particularly interesting because it shows that an
Alfvénic fluctuation is an exact nonlinear solution of the MHD equations.

5.3 Numerical simulations


Here we investigate the basic behavior of the energy cascade described by
shell models. We use numerical simulations of the equations, carried out
through a 4–th order Runge–Kutta integrator. When dissipative coefficients are
Magnetohydrodynamic turbulence in low-dimensions 87

set different from zero, and we want to investigate the smallest scales of the
turbulent cascade, the total number of shells N must be carefully chosen
according to the condition kN > kD, where kD ∼ ν−3/4 is the wave number at
which dissipative effects start to be effective. Here we present results for N =
18 shells, the kinematic viscosity and magnetic diffusivity are set to ν = µ =
0.5 × 10−7, and the system is run for about 2 × 104 large scale turnover times.
We use the FSGC model, the only thing which marks off is the behavior of
both models versus the forcing term. Unless the GOY model, FSGC shell
model critically depends on the kind of forcing term we use.

5.3.1 Dynamical properties of GOY MHD shell models


The GOY MHD model has remarkable properties which closely resemble
those typical of MHD phenomena. One of these is the magnetic dynamo
action, that is the amplification of a seed of magnetic field and its maintenance
against losses of dissipation. Starting from a well developed turbulent velocity
field, a seed of magnetic field is injected and the growth of the magnetic
spectra monitored in time. The energy is injected at the shell n = 4 with k0 = 1
through a forcing which corresponds to injecting only
kinetic energy at large scales. We used both the 2D and the 3D model2.
As shown in Giuliani & Carbone (1998) a kind of dynamo effect is visible
in the MHD shell model, say the magnetic energy grows rapidly in time and
forms a spectrum which on average is of the same order of magnitude to that
for the kinetic energy, the spectral index being closed to the Kolmogorov
spectrum. This kind of dynamo effect is absent in the 2D version of the model
of Giuliani & Carbone (1998). Since the smallness of bn, its back reaction on
the velocity field in this case is negligible, thus the kinematic part of the model
evolves independently from the magnetic one and the scaling
clearly emerges. This scaling immediately follows from a cascade process of
the quantity which is the 2D hydrodynamic invariant conserved by
the kinematic part. Let us stress that the sign of the third ideal invariant is
essential as far as the growth of a seed of magnetic energy is concerned.
The presence of a constant forcing term in the shell model induces a
dynamical alignment. In fact it is found that, unless the model is forced
appropriately, it evolves invariably towards a state in which velocity and
magnetic fields are strongly correlated, that is where and In
this state the nonlinear terms vanish, the energy cascade is stopped and
magnetic and kinetic spectra become steeper. We can say that the Alfvénic

2
Since the 2D shell model belongs to a family of shell model which do not present energy
cascade a large–scale viscosity ν′ has been used to remove energy injected by the
forcing. The term has been added to the equation for the velocity field.
88 Vincenzo Carbone et al.

state is a strong attractor for the system. When we want to investigate


statistical properties of turbulence described by MHD shell models, this should
be avoided. In fact the Kolmogorov transient and the aligned states can be
mixed during the averaging procedure, thus leading to unreliable results for
scaling laws. It is possible however to replace the constant forcing term by an
exponentially time–correlated Gaussian random forcing which is able to
destabilize the Alfvénic fixed point, thus assuring the energy cascade. The
forcing is obtained by solving the following Langevin equation

(50)

where µ(t) is a Gaussian stochastic process δ–correlated in time, 〈µ(t)µ(t′)〉 =


2Dδ(t′ − t). In this case the fixed point is destabilized, the system spends some
large–scale turnover times around one of the Alfvénic attractors, jumping from
one to the other at irregular periods. This kind of forcing will be used to
investigate statistical properties.

5.3.2 Statistical properties of GOY MHD shell models


In this section we investigate the statistical properties of MHD shell
models, by using the forcing obtained in the previous section, in order to get a
statistically stationary state. This stationary state is reached by the system, as
shown in Giuliani & Carbone (1998), with a well defined inertial range, say a
region where relation Eq. (49) is verified. In Figs. 6 and 7 we report the spectra
for both the velocity |υn(t)|2 and magnetic |bn(t)|2 variables, as a function of kn
obtained in the stationary state. Fluctuations are averaged over time. The
Kolmogorov spectrum is also reported as a solid line.

Figure 6. We show the kinetic energy spectrum |υn(t)|2 as a function of log2 kn for the
MHD shell model. The full line refers to the Kolmogorov spectrum
Magnetohydrodynamic turbulence in low-dimensions 89

Figure 7. We show the magnetic energy spectrum |bn(t)|2 as a function of log2 kn for the
MHD shell model. The full line refers to the Kolmogorov spectrum

Intermittency in the shell model is due to the time behavior of shell


variables. It has been shown (Okkels, 1997) that the evolution of GOY model
consists of short bursts travelling through the shells and long period of
oscillations before the next burst arises. In Figs. 8 and 9 we report the time
evolution of the real part of both velocity variables υn(t) and magnetic
variables bn(t) at three different shells. It can be seen that, while at smaller kn
variables seem to be Gaussian, at larger kn, variables present very sharp
fluctuations in between very low fluctuations.
The time behavior of variables at different shells changes the statistics of
fluctuations. In Fig. 10 we report the probability density functions P (δυn) and
P(δBn) of standardized variables

for different shells n. Typically, we see that PDFs look differently at different
shells. At small kn fluctuations are quite Gaussian distributed, while at large kn
they tend to become increasingly non–gaussian, by developing fat tails. Rare
fluctuations have a probability of occurrence larger than a Gaussian distribution.
This is the typical behavior of intermittency as observed in usual fluid flows
and described in previous Sections.
The same phenomenon gives rise to the departure of scaling laws of
structure functions from a Kolmogorov scaling. Within the framework of the
shell model the analogous of structure functions are defined as
90 Vincenzo Carbone et al.

Figure 8. Time behaviour of the real part of velocity variable υn(t) at three different
shells n reported on the figures.

Figure 9. Time behaviour of the real part of magnetic variable bn(t) at three different
shells n reported on the figures.

For MHD turbulence it is also useful to report mixed correlators of the flux
variables

Scaling exponents have been determined from a least square fit in the inertial
range 3 ≤ n ≤ 12. The values of these exponents are reported in Table 4. It is
interesting to notice that velocity, magnetic and Elsässer variables are more
intermittent than the mixed correlators. This could be due to the cancellation
effects among the different terms defining the mixed correlators.
Magnetohydrodynamic turbulence in low-dimensions 91

Figure 10. In the first three panels we report PDFs of both velocity (left−hand panels)
and magnetic (right−hand panels) shell variables, at three different shells An. From the
top An = 2−20, An = 2−12 and An = 2−2, respectively. The bottom panels refer to probabilty
distribution functions of waiting times between intermittent structures at the shell n =
12 for the corresponding velocity and magnetic variables.

Table 4. Scaling exponents for velocity and magnetic variables, Elsässer variables, and
fluxes. Errors on are about an order of magnitude smaller than the errors shown.

Time intermittency in the shell model generates rare and intense events.
These events are the result of the chaotic dynamics in the phase–space typical
of the shell model (Okkels, 1997). The chaotic dynamics is characterized by a
certain amount of memory, as can be seen through the statistics of waiting times
between these events. The distributions P(∆t) of waiting times is reported in the
92 Vincenzo Carbone et al.

bottom panels of Fig.10, at a given shell n = 12. The same statistical law is
observed for the bursts of total dissipation (Boffetta et al., 1999).

6. Further modeling of turbulence


In this Section we briefly outline some further variants of shell models, or
different modeling of turbulence, that can be encountered in literature.

6.1 A simple variant of the GOY model


Let us consider a phase transformation of the Elsässer variables

(51)

and let us consider the fact that these variables are random. Then assuming that
the statistical properties of pair correlations remain invariant under the phase
transformation Eq. (51), we get

Since phases might be random this must imply that for all n ≠ m.
On the contrary, it can be immediately verified that relations Eq. (42) are
invariant under these transformations providing that the following two
relations between phases hold

.
Owing to the presence of these phase relationships, correlations are present for
variables at two different shells. In particular, apart from the pseudo–energies
there is a quadratic form which is different from zero, namely
This phase invariance can be exploited by defining a slightly
different shell model where the spectrum of possible correlations is reduced.
The model reads (Biferale, 2003),

(52)

As it can be verified immediately, the model Eq. (52) has the same phase
invariance, except that in that case the following phase relations hold
Magnetohydrodynamic turbulence in low-dimensions 93

.
Owing to these new relations, the only quadratic forms different from zero are
the pseudo–energies. Even in this case, when bn = 0 we recover a further
version of the hydrodynamical model (Biferale, 2003).

6.2 A shell model to describe hydromagnetic turbulence in solar


arcades
The dynamics of low–frequency plasmas described by MHD, is very
interesting because it represents the first approach to the study of a wide
variety of phenomena in both laboratory and astrophysical plasmas (Biskamp,
1993). In some fusion plasmas, or in the solar corona, the plasma parameter β
(that is the ratio between the kinetic and magnetic pressure) is low, β  10−2,
and the MHD equations can be simplified into the so–called Reduced MHD
(Rosenblut, 1992; Zank & Matthaeus, 1992). This approximation is valid, for
example, for a plasma column with a low aspect ratio a/L  1 (a and L being
respectively the radius and the length of the cylinder), with a strong magnetic
field B0 along the axis of the column.
Let us consider a plasma inside a cylinder, of diameter a and length L,
whose axis is directed along the x–axis. A constant background magnetic field
B0 is assumed to be in the x–direction. This is the typical situation to describe
the behaviour of turbulence in a solar arcade (Stix, 1991). The plasma is
described by the velocity field v(r, x, t) and the magnetic field b(r, x, t) =
b(r, x, t)/(4πρ)1/2 (r = (y, z) being directed in the plane perpendicular to the x–
axis). In the (y, z) plane we Fourier transform the fields, in a 2D box of size a

(53)

(54)

where e(k) is a unitary vector in the direction of k = 2πm/a, and m = (my, mz)
is a vector of integers. After some algebra, it can be shown that in this approxi-
mation, for each value of x, the MHD equations (omiting for simplicity the
time and x dependence of the Fourier amplitudes) reduce to

(55)
94 Vincenzo Carbone et al.

where c(k, p, q)=(pxqy − pyqx)/2kpq is a geometrical factor. The sum on the


r.h.s. of Eq. (55) is related to the Kronecker symbol

which means that the sum is extended over all wave vectors p and q which
satisfy the triad–interaction relation k = p + q.
Nigro et al. (2004), cfr. also Veltri et al. (2005), investigated the
occurrence of turbulence in a solar arcade by introducing a shell model that
can be obtained from the reduced MHD equations Eq. (55). They used a
geometry in which B0 is along the x–axis, and shells are defined only in the
perpendicular plane of the wavevector space, namely kn = 2nk0 is defined only
in the (ky, kz) plane. Complex shell variables are introduced in the form
and these variables are allowed to satisfy the following equations
(Nigro et al., 2004)

(56)

The first term on the r.h.s. couples different shells at various values of x, while
the nonlinear term is the usual shell model where coupling coefficients are spe-
cialized to a 2D–like case. Rather than through a usual forcing term, turbulence
is restored through the boundaries at x = 0 and x = L. In particular, according
to Eq. (56), propagates in the negative x direction with the Alfvén speed,
while propagates in the positive x direction. At the lower boundary x = 0
only the value of is imposed, while at x = L only is imposed. Using
the linear relation between and the velocity shell variables υn, an
expression for the boundary values of the entering variables can be easily
obtained. Then Nigro et al. (2004) chose to inject energy only through the
lower boundary, while the velocity at the upper boundary is set to zero. As a
consequence of the first term on the r.h.s. of Eq. (56), the energy injected at the
boundary is distributed along the x direction, while turbulence develops for a
fixed x. Numerical results (Nigro et al., 2004; Veltri et al., 2005) show
sequences of dissipative bursts at different heights x along the loop, as
previously stated by Boffetta et al. (1999). These bursts are interpreted as
sequences of microflares within a solar arcade. The model is obviously able to
Magnetohydrodynamic turbulence in low-dimensions 95

reproduce power law statistics for energy, duration and waiting times between
events (Nigro et al., 2004).

6.3 Dynamical models derived from reduced MHD


Equations (55) have quadratic invariants (Biskamp, 1993), namely the
total energy

and the cross-helicity

When the background magnetic field is set to zero B0 = 0, also the mean–
square of the vector potential

is conserved. An infinite number of non–quadratic invariants also exist


(Biskamp, 1993). However, quadratic invariants, also called rugged invariants,
play a key role, because they survive each finite number of triads wave vectors
(k, p, q) which satisfy the triad–interaction relation. In other words, even if
the r.h.s. of Eq. (55) contains an infinite number of nonlinear interactions, E(t),
HC (t) and HA(t) remain invariant for each triad of wave vectors k, p and q,
such as: k + p + q = 0. When we consider a box of size a = 2π each wave
vector k turns out to be represented by a couple of integers. In this way it is
relatively easy to consider a “Pandora box” of models by enumerating
wavevectors in the plane (kx, ky).
For example, De Bartolo & Carbone (2006) consider only three wave vec-
tors, namely k1 = (1, 1), k2 = (2, −1), k3 = (3, 0), thus obtaining a set of 12
ODE’s for the complex Fourier modes of both velocity
and magnetic field The system can be further reduced
through a projection of equations on a subset of the phase space, that is by
considering only real fields. This can be seen by writing the fields in the form
and bj = |bj|eiβ, and by defining real fields through Vj (t) = |υj| cos
αj and Bj (t) = |bj| cos βj (subject to the conditions sin αj = 0 and sin βj = 0). In
this case a set of 6 ODE’s for Vj and Bj, namely
96 Vincenzo Carbone et al.

can be found (Carbone & Veltri, 1992). For its simplicity the model represents
the basic system to investigate the structure of nonlinear interactions in 2D
MHD, and to study the role played by the rugged invariants during the
dynamical evolution. The system of 12 ODE’s is called a 3–mode model, while
M –mode models (M > 3) can be investigated as Lorenz–like examples of tran-
sition to chaos. This has been investigated in the fluid framework (Boldrighini &
Franceschini, 1979; Lee, 1987). As a different approach Servidio & Carbone
(2005) investigated a Galerkin approximation of Eq. (55), namely a system with
a reduced number of modes. Rather than dealing with usual fast Fourier trans-
forms, the system has been investigated entirely in the spectral space, thus the
invariants are exactly conserved. In this way authors investigated the occurrence
of a kind of self–organization in MHD turbulence. This process has been
experimentally investigated in laboratory plasmas (Servidio & Carbone, 2005).

7. Conclusions
In the late 1957, the first space flights showed that the interplanetary space is
permeated by a tenuous plasma, called the solar wind, and that this plasma is in a
state of fully developed turbulence. The solar wind is widely used, 50 years later,
as the largest laboratory to investigate MHD turbulence (Bruno & Carbone, 2005),
and these studies, along with numerical simulations, have enormously enhanced
our knowledge on MHD turbulence. However, modeling turbulent flows remains a
promising field of research, mainly motivated for rapidly answering some
questions concerning both statistics and dynamical behaviour of turbulence. The
reader should be aware that accurate numerical simulations reach Re  104, even
with actual supercomputers. Simulations of high Reynolds numbers turbulence (in
astrophysical context are of the order of Re  108 or more) are possible only with
simpler modeling of turbulence. The price we must pay is a rough description of
local properties of turbulence, while both statistical and dynamical properties of
turbulence are in general captured by the GOY shell model. Of course using the
classical GOY model we lose all spatial description of turbulence. It is worthwhile
to remark that GOY shell models have been used as an alternative to Self–
Organized Criticality (SOC) models in describing complex dissipative phenomena
(Bak et al., 1987). From a computational point of view shell models are as simple
Magnetohydrodynamic turbulence in low-dimensions 97

as sandpile models but, at variance to SOC models, GOY models are able to
capture all statistical features of observations, both in space and in laboratory
plasmas (Boffetta et al., 1999; Carbone et al., 2002).
In the present Chapter we reviewed some topics related to the statistical
behaviour of real turbulent flows, and we described in detail the GOY MHD
shell model that is able to reproduce the gross features of turbulent flows. We
also give a brief overview of different modeling of turbulence in different
contexts. As the reader can recognize, modeling turbulence is not a trivial thing.
As the accuracy of real experiments increases, leading to a growing knowledge
on turbulence properties, modeling requires even more accuracy. However
modeling turbulence remains a crucial aspect, as far as topics related to complex
situations are concerned. This is true as far as some astrophysical situations are
concerned, as for example understanding turbulent properties of the solar
atmosphere or the interstellar medium (Bertin et al., 2004), and the competition
between linear and nonlinear dynamics (Carbone et al., 1987, 1990).
Further approaches, not described here, are related to transport processes and
particle accelerations. Turbulence is a spatio–temporal process, and this is a
crucial feature when we try to investigate transport properties or stochastic
particles acceleration. Spatio–temporal data on turbulence are rare, they are
limited to observations of convective turbulence in the solar photosphere
(Vecchio et al., 2005), and to few experimental analysis in laboratory plasma.
From the data analysis point of view, modern multi–CPU computers can be used
to develop the Proper Orthogonal Decomposition (POD) (Holmes et al., 1996),
thus investigating and modeling different features of spatio–temporal turbulence
(Vecchio et al., 2005). In general spatio–temporal models are used to investigate
diffusive processes in turbulence, where standard or anomalous diffusion of test
particles have been found. In general scientists investigated the statistical and
dynamical properties of test particles injected into a turbulent fluid, assumed both
stationary or, more recently, evolving in time (Falkovich et al., 2001; Annibadi et
al., 2002; Pettini et al., 1988; Castiglione et al., 1999; Nicolleau & Vassilicos,
2003; Carbone et al., 2003, 2004a). Further models have been used to investigate
the occurrence of stochastic acceleration of test–particles, originally investigated
by Fermi (1949), in turbulent fields (Lieberman & Lichtenberg, 1972; Bouchet et
al., 2004; Veltri & Carbone, 2004; Dmitruk et al., 2004; Perri et al., 2007).
Finally realistic spatio–temporal modeling of turbulence, with statistical
properties that are quantitatively in agreement with experiments, have been
introduced as “synthetic models” of turbulent flows (cfr. e.g. Lepreti et al. (2006)
and references therein).

Acknowledgments
We acknowledge valuable discussions with Annick Pouquet, Alain
Noullez, Vanni Antoni, Roberto Bruno and Paolo Giuliani.
98 Vincenzo Carbone et al.

References
1. S.V. Annibaldi, G. Manfredi, and R.O. Dandy: Phys. Plasmas 9, 791 (2002)
2. R.A. Antonia, M. Ould–Rouis, F. Anselmet, and Y. Zhu: J. Fluid Mech. 332, 395
(1997)
3. P. Bak, C. Tang, and K. Wiesenfeld: Phys. Rev. Lett. 59, 381 (1987)
4. R. Benzi et al.: Phys. Rev. E 48, R29 (1993)
5. A. Bershadskii, and K. Sreenivasan: Phys. Rev. Lett. 93, 064501 (2004)
6. G. Bertin, D. Farina, and R. Pozzoli Eds.: Plasmas in the Laboratory and in the
Universe: New insights and new challanges, AIP Conference Proceedings 703
(2004)
7. L. Biferale: Ann. Rev. Fluid Mech. 35, 441 (2003)
8. D. Biskamp: Nonlinear Magnetohydrodynamic, Cambridge Univ. Press, Cam-
bridge, U.K. (1993)
9. D. Biskamp: Magnetic Reconnection in Plasmas, Cambridge Univ. Press, Cam-
bridge, U.K. (2000)
10. G. Boffetta, V. Carbone, P, Giuliani, P. Veltri, and A. Vulpiani: Phys. Rev. Lett.
83, 4662 (1999)
11. T. Bohr, M.H. Jensen, G. Paladin, and A. Vulpiani: Dynamical system approach to
turbulence, Cambridge Univ. Press, Cambridge, U.K. (1998)
12. C. Boldrighini, and V. Franceschini: Comm. Math. Phys. 64, 159 (1979)
13. F. Bouchet, F. Cecconi, and A. Vulpiani: Phys. Rev. Lett. 92, 040601 (2004)
14. D.I. Braginskii: Rev. Plasma Phys. 1, 205 (1965)
15. R. Bruno, and V. Carbone: Liv. Rev. Sol. Phys. 2, 4 (2005)
16. P. Castiglione, A. Mazzino, P. Muratore–Ginanneschi, and A. Vulpiani: Physica A
134, 75 (1999)
17. V. Carbone: Phys. Rev. Lett. 71, 1546 (1993)
18. V. Carbone, G. Einaudi, and P. Veltri: Sol. Phys. 111, 31 (1987)
19. V. Carbone, P. Veltri, and A. Mangeney: Phys. Fluids A2, 1487 (1990)
20. V. Carbone, and P. Veltri: Astron. Astrophys. 259, 359 (1992)
21. V. Carbone, P. Veltri, and R. Bruno: Phys. Rev. Lett. 75, 3110 (1995)
22. V. Carbone, R. Bruno, and P. Veltri: Geophys. Res. Lett. 23, 121 (1996)
23. V. Carbone, L. Sorriso-Valvo, E. Martines, V. Antoni, and P. Veltri: Phys. Rev. E
62, R49 (2000).
24. V. Carbone, R. Cavazzana, V. Antoni, L. Sorriso-Valvo, E. Spada, G. Regnoli, P.
Giuliani, N. Vianello, F. Lepreti, R. Bruno, E. Martines, and P. Veltri: Europhys.
Lett. 58, 349 (2002)
25. V. Carbone, F. Lepreti, and P. Veltri: Phys. Rev. Lett. 90, 055001 (2003)
26. V. Carbone, F. Lepreti, and P. Veltri: Phys. Plasmas 11, 103 (2004)
27. V. Carbone, R. Bruno, L. Sorriso–Valvo, and F. Lepreti: Planet. Space Sci. 52, 953
(2004)
28. B. Castaing, Y. Gagne, and E.J. Hopfinger: Physica 46D, 177 (1990)
29. P.J. Coleman: Astrophys. J. 153, 317 (1968)
30. L. Danaila, F. Anselmet, T. Zhou, and R. Antonia: J. Fluid Mech. 430, 87 (2001)
31. R. De Bartolo, and V. Carbone: Europhys. Lett. 73, 547 (2006)
32. P. Dmitruk, W.H. Matthaeus, and N. Seenu: Astrophys. J. 617, 667 (2004)
33. M. Dobrowolny, A. Mangeney, and P. Veltri: Phys. Rev. Lett. 45, 144 (1980)
Magnetohydrodynamic turbulence in low-dimensions 99

34. W.M. Elsäasser: Phys. Rev. 79, 183 (1950)


35. E. Fermi: Phys. Rev. 75, 1169 (1949)
36. U. Frisch, and G. Parisi: in Turbulence and Predictability in Geophysical Fluid
Dynamics and Climatology, ed. M. Ghil, R. Benzi & G. Parisi (Amsterdam: North
Holland), 84 (1985)
37. U. Frisch: Turbulence: The legacy of A.N. Kolmogorov, Cambridge Univ. Press,
Cambridge, U.K. (1995)
38. G. Falkovich, K. Gawedski and M. Vergassola: Rev. Mod. Phys. 73, 913 (2001)
39. P. Frick, and D.D. Sokoloff: Phys. Rev. E 57, 4155 (1998)
40. P. Giuliani, and V. Carbone: Europhys. Lett. 43, 527 (1998)
41. P. Giuliani: Shell models for magnetohydrodynamic turbulence. In: Nonlinear
MHD waves and Turbulence, ed by T. Passot, P.L. Sulem, Lecture Notes in
Physics (Springer, Berlin Heidelberg New York 1999)
42. E.B. Gledzer: Sov. Phys. Dokl. 18, 154 (1985)
43. P. Goldreich, and S. Sridhar: Astrophys. J. 438, 763 (1995)
44. P. Golreich, and S. Sridhar: Astrophys. J. 485, 680 (1997)
45. M. Goldstein, and L. Roberts: Phys. Plasma 11, 4154 (1999)
46. S.A. Hawley: Phys. Plasma 12, 4444 (1999)
47. Holmes, J.L. Lumley, and G. Berkooz: Turbulence, Coherent Structures,
Dynamical Systems and Symmetry, Cambridge Univ. Press, Cambridge, U.K.
(1996)
48. P.S. Iroshnikov: Astronomicheskii Zh. 40, 742 (1963)
49. A.N. Kolmogorov: Dok. Akad. Nauk. SSSR 30, 9 (1941a)
50. A.N. Kolmogorov: Dok. Akad. Nauk. SSSR 31, 538 (1941b)
51. A.N. Kolmogorov: Dok. Akad. Nauk. SSSR 32, 19 (1941c)
52. R.H. Kraichnan: Phys. Fluids 8, 1385 (1965)
53. R.H. Kraichnan: J. Fluid Mech. 62, 305 (1974)
54. N.A. Krall and A.W. Trivelpiece: Principles of Plasma Physics, San Francisco
Press, Inc. (1986)
55. L.D. Landau, E.M. Lifshitz, and L.P. Pitaevskii: Electrodynamics of Continuous
Media, Pergamon Press, 2nd edition (1984)
56. J. Lee, Physica 24D, 54 (1987)
57. F. Lepreti, V. Carbone, and P. Veltri: Phys. Rev. E, (2006)
58. M.A. Lieberman, and A.J. Lichtenberg: Phys. Rev. A 5, 1852 (1972)
59. E.N. Lorenz: J. Atm. Phys. 18, 154 (1965)
60. H.K. Moffat: Magnetic Field Generation in Electrical Conducting Fluids, Cam-
bridge Univ. Press, Cambridge, U.K. (1978)
61. G. Münch: Rev. Mod. Phys. 30, 1035 (1958)
62. F. Nicolleau, and J.C. Vassilicos: Phys. Rev. Lett. 90, 024503 (2003)
63. G. Nigro, F. Malara, V. Carbone, and P. Veltri: Phys. Rev. Lett. 92, 194501 (2004)
64. K. Ohkitani, and M. Yamada: Prog. Theor. Phys. 89, 329 (1989)
65. F. Okkels: The intermittent dynamics in turbulent shell models, PhD Thesis,
unpublished (1997)
66. S. Perri, F. Lepreti, V. Carbone, and A. Vulpiani: Europhys. Lett. 78, 40003 (2007)
67. M. Pettini, A. Vulpiani, J.H. Misguich, M. De Leener, J. Orban and R. Balescu:
Phys. Rev. A 38, 344 (1988)
100 Vincenzo Carbone et al.

68. N.A. Platt, E.A. Spiegel, and C. Tressar: Phys. Rev. Lett. 70, 279 (1993)
69. H. Politano, and A. Pouquet: Phys. Rev. E Rapid Comm. 57, R21 (1998)
70. H. Politano, and A. Pouquet: Geophys. Res. Lett. 25, 273 (1998)
71. H. Politano, A. Pouquet, and V. Carbone: Europhys. Lett. 43, 516 (1998)
72. Y. Pomeau, and P. Manneville: Comm. Math. Phys. 74, 189 (1980)
73. S.B. Pope: Turbulent Flows, Cambridge University Press (2001)
74. O. Reynolds: Philos. Trans. R. Soc. London Ser. A, 186, 123 (1895)
75. L.F. Richardson: Proc. R. Soc. London Ser. A, 110, 709 (1926)
76. M.N. Rosenblut, D.A. Monticello, and H.R. Strauss: Phys. Fluids 19, 1987 (1976)
77. G. Ruiz Chavarria, C. Baudet, and S. Ciliberto: Europhys. Lett. 32, 319 (1995)
78. S. Servidio, and V. Carbone: Phys. Rev. Lett. 95, 045001 (2005)
79. L. Sorriso–Valvo, V. Carbone, P. Veltri, H. Politano, and A. Pouquet: Europhys.
Lett. 51, 520 (2000)
80. L. Sorriso–Valvo, V. Carbone, P. Veltri, G. Consolini, and R. Bruno: Geophys.
Res. Lett. 26, 1801 (1999)
81. M. Stix: The Sun: An Introduction, Springer–Verlag (1991)
82. G.I. Taylor: Proc. R. Soc. A 164, 476 (1938)
83. C.-Y. Tu, and E. Marsch: Space Sci. Rev. 73, 1 (1995)
84. A. Vecchio, V. Carbone, L. Primavera, F. Lepreti, L. Sorriso–Valvo, P. Veltri, G.
Alfonsi, and T. Straus: Phys. Rev. Lett. 95, 061102 (2005)
85. Veltri, and V. Carbone: Phys. Rev. Lett. 92, 143901 (2004)
86. P. Veltri, G. Nigro, F. Malara, V. Carbone, and A. Mangeney: Nonlin. Proc.
Geophys., 18, 245 (2005)
87. A. Yaglom: Dokl. Akad. Nauk SSSR 69, 743 (1949)
88. G.P. Zank, and W.H. Matthaeus: J. Plasma Phys. 48, 85 (1992)
89. E. Zweibel: Phys. Plasma 5, 1725 (1999)

View publication stats

You might also like