Anda di halaman 1dari 21

4

Macroscopic turbulence theory

In the preceding chapters we considered the dynamics of an individual system.


Starting from a smooth state, ne structures develop, which, in general, become
unstable at some point. After the onset of instability the structure of the ow
is very complex and irregular and, most importantly, the further behavior is
unpredictable in the sense that minimal changes would soon lead to a completely different state. Such a behavior is commonly called turbulent. Though a
direct view of the continuously changing patterns is certainly most eyecatching
and fascinating, a pictorial description of these structures is not very suitable
for a quantitative analysis. On the other hand, it is just this chaotic behavior
which makes turbulence accessible to a theoretical treatment involving statistical methods. While individual shapes and motions are intricate and volatile,
the average properties of the turbulence described by the various correlation
functions are, in general, smooth and follow rather simple laws. A well-known
paradigm is the turbulent behavior in our atmosphere. We try to predict the
short-term changes, called weather, in a deterministic way for as long as is
feasible, which, as daily experience shows, is not very long, while predictions
of the long-term behavior, called climate, can be made only on a statistical
basis.
Dividing the elds into mean and uctuating parts, we derive equations for the
average quantities, the generalized Reynolds equations, which contain secondorder moments of the uctuating parts, the turbulent stresses. A rigorous treatment would require one to consider also the equations for these stresses, which
introduces third-order moments and, when it is continued, leads to the wellknown innite hierarchy of correlation functions, the approximate treatment of
which, the problem of closure, is the central task in turbulence theory. In this
chapter we consider only the Reynolds equations, using physical arguments to
obtain phenomenological models of the turbulent stresses, eddy viscosity, and
eddy resistivity. This level of approximation, called one-point closure theory,
65
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

66

Macroscopic turbulence theory

is the basis for describing the global properties of a turbulent system, i.e., the
actual way in which turbulence is excited by boundary effects and the turbulent
transport properties, such as the drag on a fast-moving car, which depends on
the shape of its body, or, to give a magnetic example, the transport of angular
momentum in an accretion disk.
The second part of the chapter deals with macroscopic self-organization, the
formation of large-scale coherent structures, which has a fundamental inuence
on the development of MHD turbulence, in contrast to NavierStokes turbulence, for which no such processes seem to exist. Self-organization is intimately
connected with the existence of several ideal invariants and the difference in
their dissipation rates, which gives rise to selective decay processes, either the
tendency toward a force-free magnetic state or the tendency toward an aligned,
or Alfvenic, state. The latter process is based on the Alfven effect in MHD
turbulence. The decay of the turbulence energy is essentially determined by
these selective decay processes.

4.1 One-point closure


Most applications are primarily concerned with the properties of the mean
quantities, such as the mean ow averaged over the turbulent uctuations, for
which the turbulence enters only through enhancement of transport effects. We
split the velocity and magnetic eld into mean and uctuating parts,
v = v +
v,

B = B +
b.

(4.1)

Since, however, the mean quantities are in general neither homogeneous nor
stationary we are, for instance, interested in the evolution of the mean velocity
prole of a turbulent shear ow , the distinction regarding which scales should
be averaged and which should be regarded as mean proles is to a certain
extent arbitrary. Such a two-scale approach, which is frequently applied in
turbulence theory, relies, strictly speaking, on there being a spectral gap between
the small scales, over which the average is taken, and the large scales, which are
treated explicitly. Though such a gap rarely exists, one may nonetheless in many
cases rather clearly distinguish between the smaller inertial-range scales and
the macroscales of the system. However, in so-called large-eddy simulations,
which are very convenient in practical turbulence computations, the break is
assumed to occur within the inertial range at some wavenumber kc , at which
only the larger scales kc l > 1 are computed directly, while the effect of the
smaller scales kc l < 1 is accounted for by phenomenological expressions,
which depend explicitly on the cutoff kc .
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.1 One-point closure

67

4.1.1 Reynolds equations for MHD


We restrict our consideration to incompressible motions (effects of compressibility will be discussed in Chapter 9), but treat the MHD equations in the
original velocity form (2.7) or in terms of the Elsasser elds (2.106), instead of
the vorticity form (2.28). The equations for the mean velocity and the magnetic
eld, the Reynolds equations for MHD, are
v
v
b
b + 2 v, (4.2)
t v + v v = P + B B 


v
b + 2 B,
(4.3)
t B = v B + 
with v =
v = 0. The equations are written in Alfven time units assuming
that we have a homogeneous density. The effect of the turbulence is contained
in the turbulent-stress tensor

vi
vj
bi
b j  = Ri j ,

(4.4)

which consists of the Reynolds tensor RiVj = 


vi
v j  and the Maxwell tensor
M


Ri j = bi b j , and the turbulent electromotive force
 = 
v
b,

(4.5)

the negative of the electric eld induced by the turbulence,  = E apart


from resistive effects. The electromotive force can also be written in tensorial
form:
!i = i jk S jk ,

Si j = 
vi
v j
bj
bi .

(4.6)

Here i jk is the completely antisymmetric unit tensor. Expressed in terms of the


Elsasser elds, the mean-eld equations (4.2) and (4.3) become
t z  + z  z  = P 
z  + 12 ( + ) 2 z 
z
+ 12 ( ) 2 z ,

(4.7)

where the effect of the turbulence appears in the tensor,

z

z i
j  = Ri j ,

(4.8)

from which Ri j and Si j can readily be obtained as the symmetric and antisymmetric parts. The stress tensors Ri j and Si j are special cases of the twopoint correlation functions 
v(x, t)
v(x  , t  ) etc., the equations for which are
readily derived by multiplying the equations for the uctuations
v(x, t) by

v(x  , t  ) and averaging, and similarly for the other correlation functions. Since
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

68

Macroscopic turbulence theory

at this point we are not concerned with these equations, we only point out their
structure,
t uu   = L2 uu   + L3 uu  u  ,

(4.9)

where u represents any component


vi ,
bi , and L2,3 are differential operators
depending on the mean elds v and B. Hence, to calculate the second-order
correlation functions, we need to know the third-order functions and so forth,
which leads to the problem of closure in turbulence theory, which is discussed
in more detail in Chapter 6.

4.1.2 Turbulent transport coefcients


In the one-point-closure approximation expressions for the turbulent stresses,
the one-point correlations, are obtained by phenomenological modeling. Let us
rst discuss the case of hydrodynamic turbulence. One introduces the concept of
the eddy viscosity t (index t for turbulent) by writing the turbulent Reynolds
stresses in a form analogous to the viscous-stress tensor (2.5),

RiVj = 
vi
v j  = 13 
v 2 i j + t i v j  + j vi  .
(4.10)
The rst term on the r.h.s. is the isotropic part, which can simply be added to
the pressure  p, while the second is the deviatoric part (for the denition, see
the footnote preceding equation (2.5)). Equation (4.10) implies isotropy of the
small-scale turbulence, since one expresses the ve independent components
of RiVj essentially in terms of two quantities, the mean turbulence energy E =
1

v 2  and the eddy viscosity t , in a way similar to what we assumed for the
2
effects of collisions in molecular viscosity. In the simplest model one assumes

that t depends only on E and the energy-dissipation rate ! = i j (i
v j )2 .
2
From dimensional arguments E and ! have the dimensions [E] = L /T 2 and
[!] = L 2 /T 3 we obtain the form
t = C E 2 /!,

(4.11)

where C is a free numerical parameter. The quantities E and !, which are


functions of the large-scale coordinates, are obtained from additional coupled
advectiondiffusion equations. The Ansatz (4.11) is usually called the K !
model, since K is the traditional notation for the kinetic energy in hydrodynamic
turbulence.
A more physically motivated model of the eddy viscosity is based on the
concept of the mixing length lm , a turbulent mean free path,
t =
vlm ,

(4.12)

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.1 One-point closure

69

where
v represents a typical velocity-uctuation amplitude. The mixing length,
which was introduced by Prandtl (1925), is, roughly speaking, the scale over
which momentum is transported during an eddy turnover time and is hence of the
order of the size of the dominant eddies. In applications lm is usually considered
as a free parameter to be adjusted to t the observations. (For turbulence close
to a wall lm has to be a function of the distance x from the wall, lm x.)
The uctuation amplitude can be estimated by assuming that the gradient of
the velocity uctuation compensates the mean velocity gradient driving the
uctuation, |
v| dv/d x; hence
dv
,
(4.13)
dx
which is called the mixing-length estimate of the turbulence level. From (4.12)
and (4.13) we now obtain




2  dv 
t = lm
,
(4.14)
dx 

v lm

where the vertical bars indicate that the eddy viscosity is to be taken positive,
since the effect of the turbulence should be dissipative. For practical applications
these concepts are used in large-eddy simulations, which will be discussed in
Section 4.1.3.
We now return to the MHD case and give expressions for Ri j , (4.4), and Si j ,
(4.6), generalizing the hydrodynamic Ansatz (4.10) (Yoshizawa, 1990),1




Ri j = 13 
v 2
b 2 i j + tV i v j  + j vi  tM i B j  + j Bi  ,




Si j = t i jk Bk  + tM i B j  j Bi  tV i v j  j vi  .

(4.15)
(4.16)

The last equation can also be written in the more familiar vector form,
 = t B tM  j + tV ,

(4.17)

with j = B, the current density, and = v, the vorticity. (Similar


expressions have been derived by Zhou and Vahala (1991.) Let us rst discuss
some general properties of the various eddy-viscosity-type coefcients appearing in these expressions. Since  and j are vectors, while B and are axial
vectors, tV and tM are scalars, while tM , tV , and t are pseudoscalars (a pseudoscalar reverses sign under coordinate reection). The physical meaning of the
coefcients tV and tM can easily be understood. tV is an eddy viscosity similar
1

t is conventionally denoted simply by , but we prefer to add the subscript to distinguish this
quantity from the characterizing the eddy viscosity in an accretion disk, Section 11.2.2.

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

70

Macroscopic turbulence theory

to t in the hydrodynamic relation (4.10), while comparison of (4.17) with the


simple version of Ohms law E = j shows that tM is a turbulent resistivity
(remember that  = E); hence both tV and tM are expected to be positive.
The tM -term in (4.15) results from the magnetic tension force exerted on v
by the uctuation
b, which explains the minus sign of this term. Its counterpart
V
is the t -term in (4.16) representing the effect of the velocity uctuation
v on
the magnetic eld B. The coefcients tV and tM are primarily connected
v2 +
b2 , while tM and tV are connected
with the turbulence energy E = 12 
C
with the cross-helicity (2.65), H = 
v
b. Yoshizawa found the members
of both pairs of analogous quantities to be of the same order, tV  tM and
tM  tV ; however, as numerical simulations (Section 4.1.3) indicate, the latter
two coefcients are signicantly smaller than the former and can be neglected
to a good approximation.

4.1.3 Large-eddy simulations of MHD turbulence


Numerical simulations have become an indispensible tool for studying turbulent phenomena. In basic turbulence theory, in which one deals with scaling
properties and small-scale structures, one usually needs the exact information
down to the dissipative scales. Hence, in order to cope with the limitations
imposed by the available computer power, approximations are made at the large
scales, such that the computational system is restricted to a small section of the
entire turbulence eld, where conditions are approximately homogeneous. Direct numerical simulations of homogeneous turbulence will be discussed in the
following chapters.
For practical, or even technical, applications, in which the primary interest is
in the large energy-containing turbulent eddies, it is natural to make approximations at the small scales. In such computations (called large-eddy simulations
(LESs)) one introduces a cutoff wavenumber K , such that a function v(x)
is split into two parts, a low-pass-ltered part v <
K (x) and a high-pass-ltered
(x),
part v >
K


v<
vk eikx , v >
vk eikx ,
(4.18)
K (x) =
K (x) =
kK

k>K

>
<
1
v = v<
is
K + v K , of which only v K representing the large scales l > K
1
computed directly, while the effect of the smaller scales l < K contained in
v>
K is treated in some phenomenological way called subgrid-scale modeling.
The average ow v in (4.1) is now the ltered velocity eld v < , which is
actually resolved numerically, and the role of the mixing length is played by
the grid spacing # related to the cutoff wavenumber, K = /#. In the most
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.1 One-point closure

71

common form, which as introduced by Smagorinsky (1963), the eddy viscosity


becomes

1/2

2
<
< 2
1
( v + j vi )
,
(4.19)
t = (C S #)
2 i j
ij

which has the same basic form as the simple mixing-length Ansatz (4.14).
Comparison of LESs with corresponding experiments determines the parameter
C S ; one nds C S  0.1 for turbulence far away from boundaries, whereas in
boundary-layer turbulence C S should decrease on approaching the wall (for
a more detailed discussion see Yakhot et al., 1989). A review of the physical
aspects of LES in hydrodynamic turbulence can be found in an article by Canuto
(1994). Chapter 12 of Lesieurs book (Lesieur, 1997), and a recent monograph
by Sagaut (2001) present comprehensive introductions to the eld.
LES is not as well established in MHD turbulence as it is in hydrodynamics.
On the one hand the physics is more complex, since the MHD equations contain
two elds, which introduces considerably more freedom into the dynamics, for
instance the presence of both direct and inverse spectral cascades (Section 5.2).
On the other hand experimental verication of the validity of a subgrid-scale
model and the calibration of free parameters are very difcult, since MHD
turbulence is mainly observed in astrophysical systems in situ measurements
are possible only in the solar wind not in controllable laboratory experiments.
At present, the most reliable way to check the accuracy of LESs of MHD
turbulence is by comparison with high-resolution direct numerical simulations.
Yoshizawa (1991) presented a systematic derivation of a subgrid-scale theory,
but the corresponding expressions are too unwieldy for actual computations and
their physical meaning remains unclear. Let us therefore briey describe several models based more on physical reasoning and numerical practicality. One
possibility is a direct generalization of Smagorinskys Ansatz (4.19), considering only the dissipative effects tV and tM in (4.15) and (4.16) (e.g., Theobald
et al., 1994),

1/2

V
V
2
<
< 2
1
( v + j vi )
, tM = (C M #)2 | j < |. (4.20)
t = (C #)
2 i j
ij

It appears, however, that this Ansatz is too dissipative (Agullo et al., 2001). One
indication of such behavior is that, while the dynamics of MHD turbulence is
suppressed when either b or v vanishes, the turbulent diffusivities (4.20) are not
directly affected. A possible remedy is a combination of the effects of tV and
tM , and tM and tV , respectively. In addition, to account for the inverse cascade
of magnetic helicity, and along with it also of magnetic energy, one should allow
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

72

Macroscopic turbulence theory

negative values of the turbulent resistivity. Muller and Carati (2001) studied the
following subgrid model:

1/2


2
<
<
<
< 
t = (C#)  (i v j + j vi )(i b j + j bi ) ,
 ij

t = (C  #)2 s| j < < |1/2 ,

(4.21)

where s is a sign factor, which has been chosen to be s = sgn( j < < ). This
model is found to give good agreement of the evolution of kinetic and magnetic
energies E K and E M with previous direct numerical simulations. The indenite
sign of t in (4.21) substitutes in some way for the t -term in (4.16) and (4.17),
the dynamo effect, which is neglected in most LES studies.
If no direct numerical simulations of sufciently high Reynolds number
are available, which is the case in most practical applications, in particular
for inhomogneous systems, the free coefcients in the subgrid model can be
determined by a dynamic procedure (Germano, 1992; see also Meneveau and
Katz, 2000), which exploits the self-similarity of the turbulence. Here one
introduces a second ltering process, the test lter, with the cutoff K t = K ,
< 1, which is applied to the resolved elds. Since the elds are known
explicitly in the range K t < k K , one knows also their contribution to the
stress tensors. If the turbulence is self-similar, the same form of the subgrid
model can be chosen both on the original grid and on the coarser test grid with
the scale parameter # suitably adjusted. This gives a relation for the stress
tensors on the test-grid level, which should hold if the subgrid model is exact.
Optimal values of the free parameters of the subgrid model chosen, such as C
and C  in (4.21), are then obtained by minimizing the actual error occurring in
this relation.

4.1.4 Mean-eld electrodynamics


On inserting the t -term into Faradays law (4.3), we nd that t B is proportional
to j. Since the most interesting applications of the dynamo effect are found in
axisymmetric systems in spherical shells such as in planetary and stellar dynamos and in the toroidal plasma column of a reversed-eld pinch , it is natural
to divide B into a poloidal (or meridional) and a toroidal (or azimuthal) component. The relation t B = t j then shows that a poloidal eld, corresponding
to a toroidal current, amplies the toroidal eld (for the appropriate sign of
t ), while the latter, corresponding to a poloidal current, in turn amplies the
poloidal eld, which is the basis of the dynamo mechanism. To see how the
dynamo process is driven by turbulence, the simplest approach is to consider
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.1 One-point closure

73

the kinematic phase, in which the magnetic eld is still sufciently weak to
allow us to neglect the Lorentz force in the equation of motion, such that the
velocity can be regarded as an independent given quantity in Faradays law. In
this case we treat only the latter, i.e., the equations for the mean magnetic eld,
(4.3), and for the eld uctuation,


b (v
b) = (
v B) +
v
b 
v
b + 2
b.
t
(4.22)
It is also assumed that the uctuations are small compared with the mean
eld,
b  B, such that the nonlinear term in (4.22) can be omitted. This is
called the quasi-linear approximation, whereby only the nonlinear term in the
mean-eld equation (4.3) is retained. These equations are referred to as meaneld electrodynamics and were developed in a pioneering paper by Steenbeck
et al. (1966) (for a review of the theory, see Krause and Radler, 1980). To
obtain a simple explicit expression for the magnetic uctuation, one assumes,
in addition, that the spatial variation of the mean velocity is weak, such that
the second term on the l.h.s. of (4.22) reduces to a simple advective term.
Equation (4.22) now reads
t
b + v
b = (
v B)

(4.23)

neglecting resistive effects. By integrating along the orbit of a uid element,


(4.22) can formally be solved, giving
 t
  


dt 
v(x , t ) B .
(4.24)
b(x, t) =

By inserting
b into the electromotive force (4.5) one obtains
 t

!
=
v (
v B) .
dt 

(4.25)

Here the mean eld can be regarded as constant, since its variation over a
correlation time of the uctuations is small. Expression (4.25) is most easily
evaluated by considering vector components  = {!i }. The -operator acts
either on
v or on the average eld; hence !1 contains B1  and the derivatives
2 B3  and 3 B2 ,
 t


!1 =
v2 1
dt  
v3  
v3 1
v2  B1 

 t


v2
dt  
v2  2 B3  
v3
v3  3 B2  .
(4.26)

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

74

Macroscopic turbulence theory

Since the effect of the mean magnetic eld on the turbulence is neglected, the
velocity uctuations may be assumed isotropic, which implies invariance of
(4.26) under cyclic permutations,
v3  
v3 1
v2  = 
v2 (1
v3 3
v1 ) = 13 
v
v ,

v2 1
v1  = 
v2
v2  = 
v3
v3  = 13 
v
v .

v1

(4.27)
(4.28)

When these expressions are inserted into (4.26), comparison with (4.17) gives
 t
1
dt  
v
v  = 13 
v
v = 13 H K ,
(4.29)
t = 3

tM


=

1
3

dt  
v
v  = 13  
v2  = 23  E K ,

(4.30)

where and  are velocity-correlation times,   , H K is the kinetic helicity


and E K is the kinetic energy of the turbulence. The third coefcient tV in
(4.17) is not included in this derivation, since derivatives of the mean velocity,
in particular the vorticity , are neglected in (4.23).
Expression (4.29) forms the basis of the theory of magnetic-eld generation
in most cosmic objects, especially in stars, where a natural cause of helical
uid motions is thermal convection in a rotating system. Here one estimates
the magnitude of t by a mixing-length approach. The helicity H K is, roughly,
proportional to the two vectors characterizing convective turbulence in such
systems, the angular frequency (an axial vector) and the density (or temperature) stratication ln . Hence by simple dimensional arguments one nds
HK
vlm ln , where the mixing length is assumed to be of the order of
v,
the density scale height, lm | ln |1 . Since the correlation time is lm /
we obtain from (4.29)
t lm2 ln .

(4.31)

One should, however, keep in mind that the expression (4.29) is valid only
in the kinematic limit, i.e., for sufciently weak magnetic elds. Estimates of
the reaction on the uid motions show that the nonlinear decrease of t sets in
when the spectral intensity |Bk |2 becomes of the order of the kinetic energy |vk |2 ,
which is rst reached at the small scales and is connected with the Alfven effect

considered in Section 4.2.2. Yoshizawa (1990) derived that t 
v
b j;
hence t may be strongly reduced for |Bk |2 |vk |2 . This nonlinear quenching
justies neglecting the t -term in the subgrid-scale models used for LESs of
MHD turbulence discussed in Section 4.1.2.
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

75

4.2 Self-organization processes


Self-organization means the spontaneous generation of large-scale coherent
structures. Since the term structure implies already that its appearance is
coherent, we will henceforth talk only of structures. The qualier spontaneous is, however, important, since most systems of fully developed turbulence exhibit large-scale structures, which reect the properties of the turbulence drive, for instance the large-scale vortices in the wake behind some object
or the primary vortices of the KelvinHelmholtz instability in a turbulent jet,
which are continuously regenerated simply by the geometry of the system. By
contrast, the structures generated by self-organization arise spontaneously out
of a sea of homogeneous turbulence. Such processes are intimately connected
with the presence of an inverse spectral cascade, which will be considered in
more detail in Chapter 5, whereas this section is focussed on the macroscopic
aspects of self-organization.

4.2.1 Selective decay


It is often said that self-organization in turbulence is due to the presence of
several ideal invariants. The classical example is 2D hydrodynamic turbulence,
which is characterized by the ideal conservation both of energy and of the
mean-square vorticity, or enstrophy. Here self-organization leads to formation
of large-scale vortices, such as the cyclones and anticyclones in the atmosphere,
whose size is limited only by the variation of the normal component of the
Earths rotation vector caused by the curvature of the Earth. By contrast, fully
3D hydrodynamic turbulence does not seem to exhibit self-organization. In
MHD theory there are three (quadratic) invariants, the energy E, the crosshelicity H C , and the magnetic helicity H M . Let us give again the conservation relations written in Alfven time units, assuming that incompressibility
holds and ignoring surface terms by assuming, for instance, periodic boundary
conditions,



d
dE
2
2
1 2
(v + B ) d V =
j d V 2 d V,
(4.32)

2
dt
dt


d
d HC

v B d V = ( + ) j d V,
(4.33)
dt
dt


d
dHM

A B d V =
j B d V.
(4.34)
dt
dt
Dissipation of turbulence occurs primarily at small scales. Since the dissipation
terms contain different orders of spatial derivatives, the decay rates of the ideal
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

76

Macroscopic turbulence theory

invariants are, in general, different; in particular, H M is expected to decay


more slowly than E. In addition, the integral on the r.h.s. of (4.33) is not
positive denite, such that also H C will, in general, decay more slowly than
the energy. (H C and H M can, in principle, even increase with time.) There
are hence two selective decay processes, which govern the decay of MHD
turbulence. On the one hand, turbulence may relax to a minimum-energy state
under the constraint of constant helicity H M (if H M = 0; in the special case
of a reectionally symmetric state with H M = 0 a different behavior occurs,
as discussed in Section 4.2.3), which is described by the following variational
principle (Woltjer, 1958; Taylor, 1974):



2
1 2
1

(v
+
B
)
d
V

B
d
V
= 0.
(4.35)
2
2
Variation with respect to A gives the equation
B B = 0,

(4.36)

while variation with respect to v gives v = 0. Hence the minimum-energy


state (properly speaking, the extremum-energy solution; the minimum property
has to be proved separately) is a static constant- force-free eld, called a
linear force-free eld,2 where , the Lagrange multiplier
 2 in theM variational
equation, is determined by the value of H M , = Bmin
d V /H , and Bmin
itself depends on through (4.36). Such force-free states play an important
role in reversed-eld-pinch plasmas (see, for instance Biskamp, 1993a). They
also serve as model elds in low- astrophysical systems, for instance coronal
loops.
Competing with this relaxation to a static force-free eld, there is a further process of self-organization, which leads to an alignment of v and B.
This behavior can be associated with the slow decay of the cross-helicity compared with that of the energy, such that the relaxed state is the minimumenergy state for a given value of H C , which follows from the variational
principle



2

1 2
v B d V = 0.
(4.37)

(v + B ) d V
2
Variation with respect to v or B yields the equations
v  B = 0,
B  v = 0.
2

A magnetic eld is called force-free if the Lorentz force vanishes, j B = 0, such that B =
B. In general, varies in space, satisfying only the condition B = 0, which follows from
B = 0.

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

77

Hence the Lagrange multiplier saties 2 = 1, such that


v = B,

(4.38)

which is valid locally at each point in space. The relaxed state (4.38) is a socalled pure Alfvenic state since it corresponds to a nite-amplitude Alfven
wave. Since in an aligned state (4.38) the dynamics is turned off, as can be
seen directly in the Elsasser formulation of the MHD equations (2.106) (in
an aligned state either z + or z vanishes), this is also the nal state to which
the turbulence decays, apart from the very slow collisional diffusion. Which
of the two relaxation processes just discussed dominates depends on the initial
values of H M and H C . In a strongly helical system, the nal state is the forcefree eld, whereas for sufciently large initial alignment the system ends up in
an Alfvenic state. Numerical computations by Stribling and Matthaeus (1991)
conrmed this picture.

4.2.2 Alfven effect and dynamic alignment


We now show, following Dobrowolny et al. (1980), that the relaxation to an
aligned state is indeed likely to occur. For this purpose it is convenient to use the
Elsasser elds presented in Section 2.7, which describe Alfven waves propagating along a guide eld (2.107). This guide eld need not be an external static
eld, but can also be the eld in the large-scale energy-containing eddies. A
remarkable property of the dynamic equations for the Elsasser elds, (2.106),
is the absence of self-interactions in the nonlinear term, which just couples z +
and z . Hence only Alfven waves propagating in opposite directions along the
guide eld interact. This is the basis of the Alfven effect introduced independently by Iroshnikov (1964) and Kraichnan (1965b), which may play a crucial
role in MHD turbulence, assuming that the cascade dynamics is mainly due to
scattering of Alfven waves.
We distinguish between two dynamic time scales, the Alfven time A = l/vA
and the time for distortion of a wave packet, or eddy, zl+ of scale l by a similar
eddy zl and vice versa (a more precise denition of the amplitudes zl is given
in Section 5.3.2),
l = l/zl ,

(4.39)

where in general A  l . Since the interaction time of two oppositely propagating wave packets is A , the change of amplitude #zl during a single collision
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

78

Macroscopic turbulence theory

of two wave packets is small,


#zl
A
=  1.
zl
l
Because of the diffusive nature of the process, N (zl /#zl )2 elementary
interactions are needed in order to produce a relative change in amplitude of
order unity. Hence the energy-transfer time is
Tl N A (l )2 /A .

(4.40)

Since both E = E + + E and H C = 12 (E + E ) are ideal invariants, so


are E + and E , where E = 14 (z )2 d V . The spectral densities (zl+ )2 and
(zl )2 are therefore cascading quantities as discussed in Section 5.2, and the
corresponding energy uxes are
!l (zl )2 /Tl (zl+ )2 (zl )2 A /l 2 .

(4.41)

Since these uxes are constant across the inertial range, we write !l = ! ,
where ! are the dissipation rates of E , the symmetry in (4.41) indicates that
! + = ! , and hence the cross-helicity is not dissipated in this picture,
d H C /dt = 14 (! + ! ) = 0.

(4.42)

(Strictly speaking, the relation ! + = ! holds only in the local approximation in


wavenumber space, and will be somewhat modied by considering interactions
within a certain band of wavenumbers, as discussed in Section 5.3.)
Let us now consider the consequences of (4.42) on the dynamics of decaying
turbulence, assuming that initially zl+ > zl . In this case the energy uxes
are time-dependent ! (t) but still preserve the property ! + = ! , which
follows from the symmetry of (4.41) and does not require global stationarity.
The energy-transfer times are, however, different. From (4.40) one nds
that
Tl+ /Tl = (zl+ /zl )2 > 1,

(4.43)

which means that energy transfer, and hence damping, of the minority eld
zl is more rapid, leading to a continuous increase of the ratio E + /E , until
the dynamics is nally switched off in a pure Alfvenic state. Thus dynamic
alignment is a direct consequence of the MHD equations, valid both in 3D and
in 2D. The process has been studied quantitatively in the framework of closure
theory (Grappin et al., 1982 and 1983) as well as in direct numerical simulations
in 2D (Pouquet et al., 1988; Biskamp and Welter, 1989).
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

79

4.2.3 Energy-decay laws


In the preceding sections we used only the property that the turbulence energy
decays more rapidly than does the cross-helicity or magnetic helicity, but we
have not yet specied the energy-decay law. Whereas the rate of decay a of
smooth laminar eld or ow is proportional to the resistivity or the viscosity,
2
E k e2k t for a mode k, in fully developed (3D) hydrodynamic turbulence
both experimental results and numerical simulations indicate that the decay
of energy is essentially independent of the Reynolds number, and this property seems to be valid also in MHD turbulence even in 2D (in contrast to 2D
hydrodynamic turbulence, which was discussed in Section 8.1).
The decay properties of turbulence not only constitute an intriguing academic
problem but also have considerable practical consequences. Think of the turbulence in the wake of a large airplane, which can produce a violent disturbance
on a smaller plane crossing this wake. It is therefore clearly important to know
how fast the turbulence decays. Or consider the interstellar magnetic eld in a
galaxy, which is presumably highly turbulent. The rate of decay of this turbulence gives a measure of the lifetime of the galactic eld and helps us to decide
whether this eld can be of primordial origin.
Decay of hydrodynamic turbulence
Self-similarity of the decay of turbulence, if this property is indeed satised,
suggests that there should be a power-law behavior E t , where the exponent
is a characteristic property of the turbulent system.3 A theory of the decay
of hydrodynamic turbulence was rst derived by Kolmogorov (1941b)
 on the
basis of the assumed invariance of the Loitsiansky integral " = 0 v (x)v
(x + r )r 4 dr , where v = v r /r . The basic idea is rather simple. Let us dene
the integral scale length L by the relation
" = E L 5 = constant.

(4.44)

On the other hand, the integral length should be independent of the Reynolds
number and hence depend only on the integral quantities of the system, the
turbulence energy E and the energy-decay rate !; hence, by simple dimensional
arguments,
L E 3/2 /!,
3

with ! = d E/dt.

(4.45)

In the following analysis, which is mainly built on dimensional arguments, the integral quantities
such as E and H M refer to a xed volume, which we can assume to be the unit volume, such
that for instance E = V 1 12 v 2 d V has the same dimensions as the integrand, [E] = L 2 T 2 ,
or the energy-dissipation rate [!] = L 2 T 3 .

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

80

Macroscopic turbulence theory

Combining (4.44) and (4.45) gives the differential equation


d E/dt E 17/10 ,

(4.46)

which has the asymptotic solution E t 10/7 . Unfortunately, however, the


Loitsiansky integral is not preserved, as noted by Batchelor and Proudman
(1956), which invalidates Kolmogorovs approach. Lesieur and Schertzer
(1978) considered the decay problem in the framework of closure theory. The
authors relate the decay of energy to the principle of permanence of big eddies,
the invariance of the energy spectrum at small wavenumbers,
E k = Ck s = constant,

for

k < kin ,

(4.47)

where kin L 1 is the injection wavenumber and s is restricted to 1  s 4.


The energy-decay law can be obtained using the self-similarity of the spectrum
during the decay of turbulence. Thus we may write the energy spectrum E k in
terms of an invariant function f (k L),
E k (t) = E(t)L(t) f (k L),

(4.48)

which, by use of (4.47), yields


E k  E L s+1 = constant

for

k L < 1,

(4.49)

generalizing relation (4.44). Using (4.45) as in Kolmogorovs theory, one now


nds the exponent of the energy-decay law
= 2(s + 1)/(s + 3),

(4.50)

while the integral length increases, L t p ,


p = 2/(s + 3).

(4.51)

Since, as has been assumed, the shape of the small-k spectrum is invariant for
s 4, the energy-decay law depends on the initial spectrum. For instance, one
has = 1 for s = 1 or = 65 for s = 2. If the spectrum is steeper, s > 4,
initially, it is found to relax rapidly to s = 4, which hence plays a special role.
For s = 4 (4.50) seems to indicate the Kolmogorov value = 10/7  1.43,
but in this case the parameter C in (4.47) is no longer constant but varies in
time, which is related to the noninvariance of the Loitsiansky integral mentioned
above and leads to a slightly lower decay exponent,  1.38.
Experimental observations of the decay of turbulence are mainly performed
on grid-generated turbulence in a wind tunnel by measuring the turbulence
level at various distances x using Taylors hypothesis to interpret the variation
with time at a xed position in terms of the spatial variation t = x/U , where
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

81

U is the mean velocity. A decay exponent close to 1.38 has indeed been found
by Warhaft and Lumney (1978), but in other studies also signicantly lower
values are observed, typically 1. Since it is difcult to measure the low-k
spectrum, the relevance of this theory cannot yet be established.
Decay of MHD turbulence
We now consider the decay of energy in MHD turbulence. Since for a high
Reynolds number the magnetic helicity is well conserved, we can use similar arguments to those in Kolmogorovs theory given above. Let us therefore rst discuss the case of nite H M , which is characteristic of most astrophysical plasmas, since magnetic turbulence usually occurs in rotating systems, wherein the combined action of Coriolis and buoyancy forces naturally
leads to twisted eld lines. We dene a magnetic integral scale L M of the
turbulence by
E M L M = H M = constant,

(4.52)

where E M is the magnetic energy, E = E M + E K . On the other hand, we


have the dynamic scale length L, (4.45). Assuming that self-similarity of the
decay holds, in particular E E M E K , and combining (4.52) and (4.45)
identifying the two scales L M L, one obtains
d E/dt E 5/2 ,

(4.53)

which has the asymptotic solution E t 2/3 . (This result was originally derived by Hatori (1984), who, unnecessarily, assumed the existence of a specic
inertial-range spectrum.) Since experiments on MHD turbulence are difcult
to perform, the only practical way, at present, to test this prediction is by direct numerical simulations. Several numerical studies have been reported, in
particular by Biskamp and Muller (2000a, 2000b), who used a relatively high
spatial resolution and hence high Reynolds number, which is important in order to insure that H M remains constant. Figure 4.1 shows E(t) from a typical simulation run. It is, however, difcult to test the theoretical prediction
for the decay exponent directly from a loglog plot of E(t) as given in
this gure, since the exact solution of (4.53) is E = E 0 /(t t0 )2/3 , which
contains the integration constant t0 of the order of the initial eddy-turnover
time and thus depends on the initial conditions, while the asymptotic behavior t becomes visible only at sufciently large times t  t0 . To check the
validity of the theoretical assumptions, it is therefore simpler and more reliable to compare the simulation results with the differential decay law (4.53),
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

82

Macroscopic turbulence theory


1.0
E

t 1/2
0.1

10

Figure 4.1: A loglog plot of the energy E(t) from a simulation run with magnetic
M
helicity H M = 0.68Hmax
. The dashdotted line indicates the asymptotic law t 1/2 .
(From Biskamp and Muller, 2000b.)

which can be written in the form


E 5/2
= constant.
!HM

(4.54)

However, simulations show that expression (4.54) is not constant but increases
with time, indicating that the decay is slower than t 2/3 . The origin of this discrepancy lies in the assumption of self-similarity, which is actually not satised;
in particular, the energy ratio $ = E K /E M is not constant but decreases. Let
us therefore generalize the analysis by including the variation with time of the
energy ratio $. Assuming that the most important nonlinearities in the MHD
equations arise from the advective terms, we write

dE
E
= ! v E (E K )1/2
dt
L

(4.55)

instead of (4.45). Making the substitutions L L M from (4.52) and E K =


E$/(1 + $) gives
$ 1/2
E 5/2
= constant.
M
! H (1 + $)3/2

(4.56)

Figure 4.2 shows that this ratio is indeed constant for t > 2, after turbulence has
become fully developed, with little spread of the curves obtained from a number
of runs with different values of H M and Rm, which indicates that (4.55) is rather
generally valid and also shows that the decay of turbulence is independent of
the Reynolds number.
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

83

E 5/2 /(H ) [/(1+) 3 ] 1/2

10

Figure 4.2: Expression (4.56) for a number of simulation runs with different values of
the Reynolds number and of H M (from Biskamp and Muller, 2000b).

Relation (4.55) does not yet give the energy-decay law, since it contains
the energy ratio $, which varies with time. The simulations show that $ is
proportional to the total energy,
$ E/H M .

(4.57)

By inserting this relation into (4.56) we obtain the differential equation for E(t),
which in the limit $  1 reduces to

dE
E3
.

dt
(H M )3/2

(4.58)

It has the similarity solution t 1/2 ; hence energy decays more slowly than the
t 2/3 decay predicted for a self-similar turbulence decay. Note, however, that
this behavior is reached only asymptotically, whereas for nite $ the energy
decays more rapidly, as can also be seen in Fig. 4.1. Using this result and the
asymptotic behavior $ = E K /E M  E K /E, (4.57) gives the decay law of the
kinetic energy
E K t 1 .

(4.59)

The decay of the energy ratio is consistent with the relaxation of the turbulence to a static force-free state (4.36). In fact, relation (4.58) has a simple
interpretation in terms of the inverse cascade of the magnetic helicity. As will be
shown in the following chapter, for nite magnetic helicity turbulence continuously develops larger magnetic structures, whose size is limited only by external
constraints. Physically this process occurs by a sequence of coalescence events
of adjacent helical ux tubes, which preserve the total helicity, while there is no
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

84

Macroscopic turbulence theory


1.0
E

t 1

0.1

10

Figure 4.3: A loglog plot of the energy for a simulation run with H M = 0. The dash
dotted line gives the power law t 1 .

correponding process for velocity structures. Hence the decrease of the energy
ratio is a consequence of the inverse magnetic cascade. Using (4.52) relation
(4.57) can be written as
L(t = 0)
EK
,

M
E
L

(4.60)

where the size of the magnetic eddies increases as L t 1/2 .


We now consider the case of small helicity, H M  0, for which the derivation
of the decay law given above fails. Since a nonhelical eld is less constrained,
one expects a faster decay of the energy. This is indeed numerically observed;
one nds the differential decay law !/E 2 = constant and hence the asymptotic
behavior E t 1 , as illustrated in Fig. 4.3. In contrast to the nite-H M case, the
energy ratio is approximately constant. The behavior for H M = 0 is reminiscent
of the decay of energy in 2D MHD; see Section 8.2.2. Since, however, no
macroscopic mechanism such as a selective decay process seems to exist for
H M = 0, the decay law need not be universal, but could depend on the initial
conditions, similarly to the behavior in 3D hydrodynamic turbulence discussed
before.
The preceding analysis of the decay of turbulence energy was restricted to
weak velocitymagnetic-eld alignment |H C |  E. Effects of nite initial
alignment and their competition with the selective decay due to nite magnetic helicity have been studied extensively for low-order 3D MHD systems by
Stribling and Matthaeus (1991), though the relevance of the results for highReynolds-number turbulence remains unknown. The simulations by Biskamp
and Muller (2000a) indicate that, for an initial correlation |C | > 0.1 0.2,
Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

4.2 Self-organization processes

85

|C (t)| increases, implying that H C decays more slowly than the energy. The
correlation is dened by4
C = H C /E.

(4.61)

Since the alignment reduces the v B nonlinearity in the magnetic-eld equation, the turbulence dynamics is weakened and hence the decay of energy is
slowed down. In contrast to magnetic-helicity-dominated turbulence, the energy ratio $ does not decrease, which is to be expected, since the system tends
toward an aligned state with $  1.
4

Since, as we have seen, the energy ratio may become very small, E K  E M , this denition
does not give a direct measure of the mean alignment. In this case one should dene instead
C = v B/(v B) = H C /(4E K E M )1/2 .

Downloaded from http:/www.cambridge.org/core. National Chiao Tung University, on 29 Sep 2016 at 03:42:06, subject to the
Online Cambridge University
Press, 2009
Cambridge Core terms ofCambridge
use, available Books
at http:/www.cambridge.org/core/terms.
http://dx.doi.org/10.1017/CBO9780511535222.005

Anda mungkin juga menyukai