Physica A 199 (1993) 453-475

North-Holland
PHYSICA
SDI: 0378-4371(93)E0295-P

Self-organization and fractal dynamics in

turbulence
A. Bershadskii, E. Kit and A. Tsinober
Department of Fluid Mechanics, Faculty of Engineering, Tel Aviv University,
Tel Aviv 69978, Israel

Received 9 June 1993

Results of analysis of the field of helicity, obtained in three different turbulent laboratory
flows (grid-flow, boundary layer and jet) and a simple helical fracton model has been used in
order to provide a quantitative explanation of anomalous turbulent diffusion in the tropo-
sphere and in the ocean. It is shown that Kolmogorov turbulence is critical in respect to the
localization effects of subregions with large helicity (helical fractons) and it breaks up into
helical fractons under the condition Df ~ 2 , where Df = 2D/D W is the so called fracton
dimension (D is the fractal dimension of the turbulent fractal and D w is the dimension of
random walks on this fractal). For strictly Kolmogorov turbulence Df = 2. We study the
internal structure of helical fractons and demonstrate that they are characterized by D r = 4/3.
Finally, we look at the influence of helical fractons on diffusion of a passive scalar in
turbulence. It is shown that their influence is manifested in the scaling law for the turbulent
diffusivity in the form ~ _ ~ 8 / ~ in both three-dimensional and quasi-two-dimensional
situations. This (anomalous) law is in a very good agreement with a large number of
experimental data of different authors in the troposphere and in the upper ocean.

I. Introduction

Geophysical turbulent flows are characterized by rather large Reynolds

numbers. Therefore, it has been a common expectation that universal relations
(such as the energy spectrum E(k) ~ k -5/3, diffusivity Yg"- (4/3) should be valid
in such flows as well as their "two-dimensional" analogs in quasi-two-dimen-
sional situations. However, it appears that this is not always the case. For
example, it has been shown [3] that large scale instabilities in quasi-two-
dimensional turbulence in the stratosphere result in considerable differences in
spectral and diffusion relations both from "pure two-dimensional" and from
" t h r e e - d i m e n s i o n a l " u n i v e r s a l l a w s ( e . g . E ( k ) ~ k -7/3 a n d n o t k -5/3 OF k - 3 ) .
The governing parameter in this c a s e is n o t t h e m e a n rate of energy (of
entrophy) dissipation (transfer), but rather the mean rate of helicity "dissipa-
t i o n " ( g e n e r a t i o n ) . It is r e a l i s e d o n h e l i c a l t r a v e l i n g w a v e s r e s u l t i n g f r o m t h e

0378-4371/93/$06.00 (~ 1993- Elsevier Science Publishers B.V. All rights reserved

454 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

instability of two-dimensional turbulence in three-dimensional space. This is

one of the examples of topological instability leading to essential changes of
relations in turbulence.
We study in this paper another example of such an instability which results in
renormalization of proper Kolmogorov turbulence in three dimensions.
Analysing the experimental data obtained in laboratory for turbulent grid,
boundary layer and jet flows [1,2] we show that Kolmogorov (homogeneous)
turbulence is unstable in respect to local s t a t e s - f r a c t o n s (cf. [4]), which
appear to be the subregions with large helicity. These self-organized states arise
spontaneously in subregions of turbulent flow with essential breaking of
reflexional symmetry (in which the helicity is large). The governing dimension-
al parameter for helical fractons is different from the Kolmogorov one (see
section 3). In particular, the diffusivity Y{ in fractons follows the relation

Y~"~ e 8/v (a)

and not the law of Richardson-Kolmogorov,

ffLr_~ 4/3 " (b)

When the number of helical fractons is large enough, the mean diffusivity (over
the whole flow region) will follow the relation (a) too. However, the energy
spectrum can be not compatible with (a) in the generally accepted sense due to
the complicated structure of helical fractons, which are able to trap and detain
the passive scalar within their interior.
Fig. 1, adapted from [5], contains a collection of a large number of
experimental results on cloud width versus travel time. The straight line
corresponds to the relation (a) and the dotted lines correspond to the relation
(b) and Y{- ,. The striking feature of the results shown in fig. 1 is an extremely
broad range of universal scaling behaviour. Since these results have been
obtained in essentially different external conditions it is naturally to assume
that the processes responsible for such universal behaviour are realized on a
spatially localized (and compact) carrier with universal dynamics (we call this
process the fracton transfer of a passive scalar, see below).
The relation of ~ versus ~ is shown in fig. 2 on the basis of the results
obtained in the ocean by Okubo [6]. Okubo arrived to an empirical relation
~2 t2.34, whereas the relation (a) results in ~ 2 t2.33. Recent drifter data in
the ocean [7-9] appeared to be consistent with the patch dispersion observa-
tions by Okubo [6]. It is explained in [7-9] how one can establish the
correspondence between the patch dispersion and the drifter data, and a
relation has been found between the patch dispersion data and the fractal
characteristics of drifter trajectories.
 A. Bershadskii et al. I Self-organization and fractal dynamics in turbulence 455

iO J

,~,v t 7/6 I ~ x~'xx

io~ 107 _

-Z
0)
E
t- u K ~,~e/;' /
~ i0 ~

P,
S ///
~E I0 ~
L)

to c
.
p//" 10 3

, I , I
/ I i I ~ I
I0-~00 102 104 106 I03 10 5 iO T
t, travel time (sec) ~,cm
Fig. 1. Fig. 2.

Fig. 1. Observations of widths (horizontal standard deviation) of diffusing tracer as a function of

downwind travel time. Different symbols correspond to the results of different authors. The figure
is adapted from [5].
Fig. 2. Eddy diffusivity versus scale ~ in the ocean. Adapted from [6].

Of course, in the atmosphere and in the ocean there are observed relations
of 5~(~) different from (a) and the geophysical conditions leading to the fracton
transfer of a passive scalar are not clear yet.

2. Regions with large helicity

Moffatt ([11, p. 123) was the first to draw the attention to "blobs of
vorticity" with nonzero helicity in an inviscid flow. Helicity is defined as an
integral over some volume of helicity density h = u-to, where u is the fluid
velocity and to = curl u its vorticity. The particular role of regions with large
helicity in turbulent flows became later on a controversial subject which up to
now has not been resolved (see [12] and references therein). The attractiveness
of helicity stemmed from the following properties: (i) it is an inviscid invariant
and therefore it was speculated that in regions with large helicity the nonlinear
processes may be reduced and thereby such regions may be selected out to
survive for much longer periods than other structures, (ii) it characterizes the
topological structure of vorticity field - one of the key quantities in turbulent
flows. On the other hand, the existing evidence (experimental and numerical)
456 A . Bershadskii et al. / Self-organization and fractal dynamics in turbulence

is mostly of qualitative nature and the additional difficulty is that helicity of an

arbitrary volume (which is an integral of u . t o over this volume) generally is
not an inviscid invariant. Nevertheless, on the basis of theoretical speculations
and experimental evidence we argue that helicity is a quasi-invariant in regions
where it is large in turbulent flows. In order to show this as well as to study the
quantitative characteristics of such regions we have used two approaches.
The first one is the well-known m e t h o d of moments (for the definition of
moments of some quantity ~, see (30), section 6). In fig. 3 we show some even
moments of helicity density u . to from experimental data obtained at the
centerline of a circular jet at Re x = 880 [2]. One can easily see that moments
up to the order 60 (!) exhibit a very stable scaling behaviour in the range of
scales spanning over two decades. Since the major contribution to higher
moments comes from regions with large helicity it is natural to relate the above
scaling behaviour to some particular properties of these regions and to look for
an explanation of such behaviour consistent with experimental data. One of the

0 o 0
o~,,~ q-4_ oo q-40
-I
-20

-3 -40 -28.i

= -60t

I
-5i
-80
E I 1 I -I00
0 I 2 3 4 I 2 3 4

o
_ o',~ q-6 oo q-20
-I0
-4
-20 - 13.4
rj
-8 -30
c-

-40
I 1 I
-120 I 2 3 4 0 I 2 3 4
.-

q
E
o 0
-4
-8
q.,o-,oI ~ Oo q,6o_]
-12 -80

-16 -120
-20 I I i \
0 I 2 3 4 0 I 2 3 4

togr,arbitrary units
Fig. 3. Moments of helicity density for x / D = 30 at the axis of a turbulent jet flow (arbitrary units)
[2].
 A. Bershadskii et al. I Self-organization and fractal dynamics in turbulence 457

prominent features of the regions with large helicity is their spatial localization.
In order to describe this phenomenon of spatial localization we use the
so-called fracton approach (see, for example, [4,10]), which has not been used
in turbulence research so far. This approach allowed us to study the phenom-
enon of spatial localization in disordered systems (both theoretically and
experimentally) and to obtain some quantitative characteristics of such
systems. It should be emphasized that we do not transfer the techniques from
the theory of fractons literally. We use only the basic ideas and from the very
beginning study some fracton properties of turbulence independently of the
methods of the theory of fractons. The coincidence of quantitative results
obtained in such a way with those from the fracton theory seems to be not
accidental and most probably is due to the common nature at least of some
aspects of both phenomena. However, this coincidence plays an auxiliary role
only and the main test of our fracton approach is the comparison with
experimental results for turbulent flows. From now on we call regions with
large helicity helical fractons.

3. Scale invariance and generalized Kolmogorov turbulence

Formally the "2/3" law of the Kolmogorov turbulence follows from the
requirement of scale invariance of the rate of mean energy dissipation (E) =
1 I d ( u z ) / d t [ . Indeed, the scaling transformations r---~Ar and t-+ a t result in
u - + ( h / a ) u and ( E ) - + ( h 2 / a 3 ) ( e ) for the velocity and the m e a n rate of
dissipation, respectively. The requirement of scale invariance of ( e ) , i.e.
(e)--~ (e) gives the relation A2/a 3= 1. Consequently u--~ h2/3u, i.e. the "2/3"
law of Kolmogorov.
However, if the trajectories of fluid particles are singular on a set dense
everywhere the time measure dt will not remain homogeneous under the
transformation t-+ at, i.e., generally will not be transformed as d ( a t ) ~ a dt. If
the singularities of the fluid particle trajectories have the property of (self-)
similarity (fractal), then d(at)--~ a z dt, where z is some constant and generally
z ~ 1. In this case the operation of taking the mean ( ) is not commutative with
the operation of time derivative d / d t , since the averaging means smoothing out
the singularities of the trajectories (i.e. the discontinuities of their derivatives)
of fluid particles, i.e u--> (ha-Z)u and

(~ > = Id(u ~ > ldtl-.-> ~.2o~-(~=+')(E).

The requirement of scale invariance of (e) leads to the relation/~2Og--(2z+l) = 1.
In particular
458 A . Bershadskii et al. / Self-organization and fractal dynamics in turbulence

!/2----~ ~ 2/(2z + 1)U2 ,

which for z = 1 turns into the " 2 / 3 " Kolmogorov law.

In order to find the relation between the exponent z and the fractal
properties of fluid particles let us consider the fraction of the observation time
that some fluid particle is found in the vicinity of the origin of the observation.
Suppose that the flow region is divided into a set of small cells. During the time
t after the start from some cell the fluid particle will depart from its original
position at a distance ( ~ t ~/Dw, where D w is the dimension of random walks on
a fractal [4,10,13]. Then the number of cells visited by a particle in a sphere of
radius ~ is estimated as N ~ ~ D ~ t D / D w , where D is the fractal dimension of a
turbulent fractal (it will be shown in the next section that D is a fixed quantity
within the structures we call helical f r a c t o n s , cf. [14, 22]). The mean time that
the fluid particle spends in some cell is estimated as

t' ~ t / N ~ t 1-D/Dw ~ t 1-Df/2 .

We will call t' the f r a c t i o n a l time. The quantity Df = 2 D / D w (which defines t')
is called the fracton dimension [4,10,13]. The relation for the corresponding
measures becomes d t ' - t -De/2 dt and consequently under the scaling trans-
formation t---~ a t ,

dt'---~ a z-Df/e dt' . (1)

It is seen from the relation t ' - t 1 - D f / 2 that the fractional time t' is decreasing
with t if Df ~ 2, and for Df < 2 the fractional time is increasing with t. In other
words, the case Df < 2 can be associated with the localization of the motion of
fluid particles, whereas the case Df > 2 is related to the internal intermittency
(see below). Thus, it is plausible to assume that for the proper Kolmogorov
turbulence Df = 2 (obviously in this case t' - t). It follows from this assumption
together with z = 1 that the fractional time measure is scale invariant in
Kolmogorov turbulence, i.e. dt'--~ dt'.
Following the above considerations let us define the generalized Kolmogorov
turbulence as possessing two i n d e p e n d e n t ~1 scale invariants,

(,)
and

dt' --~ d t ' .

#1 It is noteworthy that for proper Kolmogorov turbulence these invariants are not independent.
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 459

The relation between z and Df follows from the last condition, namely,
z = Df/2, and together with the first condition gives the generalization of the
" 2 / 3 " law of Kolmogorov,

U2_._> A2(Df-1)(Df+I)-lu2 .

Specifically, for Df = 2 (and z = 1) one obtains the " 2 / 3 " , i.e. the proper
Kolmogorov turbulence.
In order to facilitate the formal obtaining of scaling relations for other
characteristics in generalized Kolmogorov turbulence (spectra, diffusivity, etc.)
it is convenient to introduce some parameter g such that

( U 2) ~ & 2 ( D f - 1 ) ( O f + l ) - I .

The parameter ~ can be chosen as having the following dimensionality #2.

[ ~ = [L]2[T]-(I+DO , (2)
and it can be seen as the renormalized rate of dissipation. Then, for instance,
one can find from dimensional arguments the spectral density of turbulent
energy

E(k) ~ gXk-*, (3)

where

x = 2/(1 + D,) and 3' = 3 - 4/(1 + O f ) . (4),(5)

At D e = 2 one obtains from (4) and (5) x = 2 / 3 and 3 / = 5 / 3 , i.e. the

Kolmogorov spectrum.
It is noteworthy, that Df = 2 is a critical value in the theory of fractons in the
sense that "in the presence of disorder all states are localized when D~ ~ 2
independently of the fractal dimension D " [4]. In order to study this and related
questions for turbulent flows let us look at the relation between helicity and the
fracton dimension of turbulent flow.

4. Quasi-invariance of helicity in helical fractons and the fracton dimension

It is known that for any volume V (characteristic scale r) of inviscid

incompressible fluid bounded by a surface S the following relation is valid [12]:

,2 Though there is some arbitrariness in such a choice it does not change the final results.
460 A . Bershadskii et al. / Self-organization and fractal dynamics in turbulence

dY(/ dt = ~s w.Q ds , (6)

where

f hdv, h=u'to, Q=luZ-p/p, (7)-(9)

v

and to = curl u, p is the pressure and p is the fluid density.

Let us consider the situation where the fracton dimension from the outside
of the surface S (including S) is D f, while inside S, i.e. in the volume V, the
fracton dimension is D if and

D~ > Dif. (10)

We will discuss the physical meaning of this relation below.

The characteristic time T of change of ~ can be estimated as follows [15]:

T-I 11daVdtl-1 (11)

In order to find the dependence of T on the characteristic scale r of volume
V let us estimate the quantities I el and Ida/dr[ -1 using dimensional argu-
ments, the relation (7) with Dif and the relation (6) with Dr. The result is as
follows:

[dYg/dt[-' - gZr~, (12)

where

z=3/(l+Df) and a=a-6/(l+Df), (13)

and

I 1- e'Z'r~', (14)
where

z'=2/(l+Di0 and a'=a-n/(l+Dif). (15)

Finally, from these estimates and (11)

T - - r ~'-~ ' (16)

where a and a ' are given by (13) and (15).

 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 461

Let us compare now the characteristic time T of change of helicity ~ with

the characteristic time of turbulent processes at scale r represented by the
turnover time r [16],

r- [k3E(k)]-'/2, (17)

where k is the modulus of the wave number, E ( k ) is the spectral density of

turbulent energy. If

E(k) -- k -~ , (18)

then using k - r -1 and (17) one obtains that

r -- r (3-~')/2 (19)

Consequently from (16) and (19) the ratio

T/r ~ r e , (20)

where

/3 = ( a ' - a ) - (3-7)/2. (21)

If fl < 0 then T/r---~oo when r---~ 0. In other words, in the case when/3 < 0 the
helicity Y of volume V can be considered as a quasi-invariant [15] since it is
changing very slowly on the characteristic time of turbulent processes r.
It is straightforward to see from (5), (13) and (15) that/3 < 0 if

D~ > Dif , (22)

both for either Df = Dif or D f = D~ in (5).

H e n c e Yg is a quasi-invariant both in respect o f turbulent processes inside the
v o l u m e V a n d outside o f it p r o v i d e d that the condition (22) is satisfied.
It is important that this property of quasi-invariance is generally not valid for
any subvolume of V which is not coinciding with V. In reality - since Y( is
quasi-invariant only - the surface S is surrounded by a boundary layer so that
in every subvolume, bounded by surfaces belonging to this boundary layer, Y(
will have the property of quasi-invariance (see also [17]).
In order to gain some insight into the physical meaning of the above
property of quasi-invariance let us compare the surface integrals in (6) for the
surface S e with Df = D~ and any inner surface S i with D r = Dif, i.e. let us
estimate the ratio of two integrals I = ~s % Q ds over these surfaces. It follows
that
462 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

I~ I I i ~ r* , (23)

where

~b : 6(D~ - Dif)/[(1 + D f ) ( 1 + Dif)]. (24)

It is obvious that if D f > Dif, i.e. if the condition (22) is satisfied, the ratio
I~/Ii---~O with r-->0.
In the conclusion of this section we would like to comment on another aspect
of the quasi-invariance of ~ . The condition tonis = 0 , i.e. S is a vorticity
surface, is sufficient for the strict invariance of ~ . In fact, vanishing of the
integral in the right hand side of (6) (see (23), (24)) would guarantee the
invariance of Y(. In this latter case the vorticity lines do not have to lie on the
surface S and can cross S in a rather arbitrary manner. In the case under
consideration in this section the vorticity lines are almost not crossing the
surface S~ (or more precisely, are crossing the surface S~ on a measure tending
to zero), i.e. the surface S~ is a quasi-surface of vector to.

5. Structure of helical fractons

We have already mentioned in section 2, that Kolmogorov turbulence (with

Df = 2) comprises a critical state in the sense of formation of fractons from the
general point of the theory of fractons [4]. It is noteworthy that the critical
character of Kolmogorov turbulence from a different standpoint was noted
before [18,19]. We consider this question in some detail in this section using the
results from the previous one.
Suppose that D f = 2 (i.e. strictly Kolmogorov turbulence is realized at the
surface S). Using the definition Df = 2 D / D w and using that for Kolmogorov
turbulence D = 3 (see, for example, [20]) one obtains that D~w = 3. A fluid
particle approaching the surface S from the outside (and D fi < D r ) gets into the
boundary layer discussed in section 3. The characteristic time of motion across
this layer ( - - T ) is much larger than the turnover time ~- and consequently the
fluid particle stays in this layer for a long time of order T. Actually D ~ D~ft =
2, since the particle is moving effectively on a surface. On the other hand the
topological constraint does not change the dimension D w and, hence, in the
boundary layer

Dit = 2 D ~ f f / D ~ ~ 4/3 . (25)

In order to leave the boundary layer further into the subvolume Vs the particle
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 463

should pass through a state in which Dett = 3, i.e. to cross again a "Kol-
m o g o r o v " surface with Df = 2. Thus the boundary layer should be limited by
such a surface from inside too. When the particle crosses this surface the
picture is repeated. Hence a volume bounded by a Kolmogorov surface (with
D~ = 2 ) can be seen as a sandwich of layers described above separated by
Kolmogorov surfaces. It should be emphasized that an important ingredient of
this model is the assumption that the helical fracton has a homogeneous
effective fracton dimension Dit ~ 4 / 3 . We do not consider other options here
since the experimental evidence favours clearly the model given above. It is
noteworthy also that in the general fracton approach the value Df = 4/3 is
directly related to (statistically) stable spatially localized states - fractons (the
so-called hypothesis of A l e x a n d e r - O r b a c h , see [4,10]).
Let us estimate the thickness 6 of the layers in the fracton. Assuming
3 ~ r ( 7 / T ) it follows that

6 - r 1-8 , (26)

w h e r e / 3 is given by (21).

6. Method of moments and experimental verification of the fracton approach

In order to find a criterion of fracton identification let us estimate the ratio of

helicities of two volumes of the same scale r but with different D t. It follows
from (14) and (15) that

1 11/1 21-r' , (27)

where ~b = 4[(1 + D f 2 ) -1 - (1 + D f l ) - l ] .
Thus 1~11/1~21~o~ for r--~0 if ~b < 0 , i.e. if Of 2 > D f l .
We have seen that the dimension Df of helical ffactons is m 4 / 3 , while
outside fractons Df ~ 2. Hence, one can expect that the helicity taken over the
volume of a ffacton will be a local maximum, i.e. will be larger than the
helicity of any subvolume of the ffacton as well as the helicity of a volume
enclosing the fracton. This property allows us to identify the helical fractons in
the turbulent flows.
One of the effective methods of identification of (sub) volumes with maximal
values of some field is the method of moments (see, for example, [21-23]).
Suppose that a region of scale L occupied by turbulent flow is divided into a
set of subvolumes 0 i with characteristic scale r and define a local mean of some
scalar field h(x) over each subvolume
464 A . Bershadskii et al. Self-organization and fractal dynamics in turbulence

hi(r ) = r -a f h(x) (28)

D

a normalized local mean

~i(r) ": hi(r)/max hi(r ) , 0 <- ~i(r) ~< 1, (29)

and a q-moment as

(~q) = (L/r) -a ~ ~q(r). (30)

i

Denote Nj(r) the number of terms in the sum (30) satisfying the condition
(j - 1 ) / M <~~i(r) <~j/M, where j = 1 . . . . , M and M is some integer. The mean
over the jth interval is defined as

(~7) : [Nj(r)]-I ~ ~q (r),

aj

where the sum is taken over all the terms satisfying the above condition. Now
the q-moment can be represented in the following form:

(~q) : ~ [(L/r)-dNj(r)l~, (31)

J

and the quantity (L/r)-aNj(r) is the discrete analog of PDF (probability density
function) p({, r) of the continuous variable ~, i.e. (31) can be represented as

({q) = f p(~, r) ]~q(r) dE. (32)

0

In case of scaling behaviour

Nj(r) = cjr -dj . (33)

If there exists the limj__,=(M__,) dj = d= then for large enough j

-d
N] = c j r ~ , (34)

and for ~---> 1

p( ~, r)---> c( ~ ) r d-d~ . (35)

 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 465

It has been found from the analysis of the experimental data that d~ = 0 for all
the three flows. The relation (35) can be used for an asymptotic estimate of the
scaling behaviour of the integral in (32) for large q,
1

( ~ q ) ~ A q r d-d~ , A q = J c ( ~ ) ~q d ~ . (36),(37)
o

It follows from (29) and (36) that

( h 7) ~ A q r a - d ~ [ m a x hi(r)] q (38)

Introducing the generalized dimension Dq in the usual way [22] via the
relations

( h q) ~ r -~q , ~Jl,q = (d - O q ) ( q - 1), (39)

and using the relation

q) l/q
(v.
lim,,_, h i : m a x hi(r ) (40)
q"~ k i i

it is straightforward to obtain that

max
i
hi(r ) ~ r ~ - d
~
where D~ =lim D_ .
q---~~ ~t
(41)

Finally using (41) and (38) we arrive at the asymptotic representation

( h q ) ~ r d-d~-(d-D~)q (42)

or

tXq = (d= - d ) + (d - D = ) q (43)

for large q.
A similar asymptotic relation has been observed for the fields of enstrophy
and dissipation [23] with different D~ for each of these fields.
It is noteworthy that for the set of subregions in which hi(r ) is large (see (28)
and (41))

If h(x) d x ~ r .
Oi
(44)
466 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

In this work we are interested in two aspects. First, D~ is characteristic of

subregions with large [fal h ( x ) d x I . Secondly, the quantity D~ can be found
from the relation (44) and experimentally determined moments, i.e. /Zq,
provided that the moments exhibit a scaling behaviour (42). We have already
shown (fig. 3) the even moments of helicity density obtained from experimen-
tal data at the centerline of a turbulent jet at Re~ = 880. These moments
exhibit a surprisingly stable scaling behaviour in the range of scales spanning
over more than two decades. The scaling behaviour of moments for the usually
measured quantity (Ou/Ot): is expressed considerably weaker.
In fig. 4 we show the values of ~q(q) obtained from fig. 3. The asymptotic
behaviour (43) is well expressed and the value D~ ~ 2.28 - 2.29.
Let us compare this value with the one for fractons following from (14) and
(15) with Dif = 4/3,

lyfl - r 16/7 (45)

Since 16/7 = 2.286, the agreement with the experimentally obtained value is
very good.
It is noteworthy that similar results have been obtained for grid turbulence
[1] for much lower Re~ = 70. In fig. 5 we show some of the even moments of
helicity density for this flow, which also exhibit a rather stable scaling
behaviour though the scaling range is somewhat shorter than for the jet flow.
The exponents IXq(q) obtained from these moments are shown in fig. 6 together

4C
/..Lq

2C

8 ~ I/ , I , I8
6 -- 20 40

4
C
/ q2 . ~ "D~ 229

o , I ' L , I ' I
2 4 6 8 I0
q
Fig. 4. The intermittency exponent ],/~q v e r s u s order of moment q for the data shown in fig. 1.
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 467

0 0
t,,-
q-4

-2

R
'~ -4
-12
I I I "-'-t
~ 0 I 2 :5 4 0 I 2 3 4

0
q-6 q"lO
-4

-8
-12

-16
I I I
0 I 2 3 4. 0 I 2 :3 4
tog r , o r b i t r o r y units
Fig. 5. Moments of helicity density for a turbulent grid flow (arbitrary units) [1].

y5

4 //'//// ,,-f /
::k

2 "
ZL zz

i ,11 , I , J , i , I
0 j 2 4 6 8 I0
q
Fig. 6. The intermittency exponent/xg ( x ) for jet (Re, = 880), (O) for grid flow (Re A= 70), (+)
for boundary layer (Re, = 290).

with the results for the jet center at R e x = 880. One can see that the difference
is pretty small, D~o ~ 2.30 for the turbulent grid flow. Similar results have b e e n
obtained also for the outer part of b o u n d a r y layer flow ( y > 0 . 2 ~ ) and
R e x = 290, where 6 is the nominal thickness of the b o u n d a r y layer. A n
example of these results is also shown in fig. 6.
468 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

7. Stability of fractons

In this section we discuss some questions related to the stability of helical

fractons and the difference between helical fractons and regions with large
helicity density.
In fig. 7 we show PDFs of the cosine of the angle between vorticity and
velocity. The curve a corresponds to the whole field, while the curve b
corresponds to fractons only, i.e. to the parts of the signal maximizing the
quantity hi(r ) = ISai h(x) dx I with r chosen in the middle of the scaling range of
the moments of helicity. Finally the curve c corresponds to the largest helicity
density u . to with the same number of points in the signal as the curve b. The
P D F corresponding to the whole field (curve a) is almost flat with a very weak
asymmetry. On the contrary the P D F corresponding to fractons (curve b)
exhibits a strong asymmetry, i.e. there is an anomalous polarization of helical
fractons. It is noteworthy that the sense of polarization is the same as the
(small) asymmetry of the whole field. The same behaviour was observed in grid
turbulence in which the asymmetry was of the opposite sign and also in
boundary layer. This p h e n o m e n o n can be explained on the basis of the
difference between the time scales T and ~- (see the relation (20)) within
fractons and their surroundings respectively. Since T >> ~- the tendency to
isotropisation in the fractons will be much slower than in the surrounding
regions and thereby the large scale anisotropy can be preserved in fractons
down to much smaller scales than in their environment. On the other hand, the

2 c
t
o.8 i
0.4-

- 0.8 - 0.4 0 0.4 0.8

cos ( u ~ )
Fig. 7. P D F of cos O, where curve a represents the full signal (80 000 points), curve b represents
helical fractons (1000 points) and curve c represents 1000 points with the largest helicity.
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 469

process of fracton creation seems to be of fluctuative nature. T h e r e f o r e , the

fractons will arise with largest probability in regions with largest asymmetry in
parity, i.e. regions with large helicity of particular sign. Due to the large time
life of fractons we are able to observe a rather strong breaking of reflexional
symmetry (fig. 7, curve b).
The above discussion pertains directly to the question of stability of fractons.
We have seen that the characteristic time T of change of helicity of the fracton
is of the order of the "transversal" time of the layers in fractons and this time is
much larger than the "longitudinal" time r in the layers, i.e. T >>z. It is this
anisotrophy which may be the reason for rather long life time of the helical
fractons, since the characteristic time of turbulent diffusion of a fracton is of
the order of its "transversal" time. For the above reason a fluctuatively created
fracton with locally larger Reynolds number than its environment will survive
for a much longer time than a strong but quasi-isotropic turbulent patch with
the same local Reynolds number. It is noteworthy that since the ratio T / z (see
[20]) grows with decreasing r, there may exist a critical scale rcr at which the
fracton becomes quasistable.
One can look at the problem of fracton stability from a statistical point of
view in the sense that a state of some system is considered stable (unstable) if it
corresponds to maximum (minimum) of P D F of some quantity under consid-
eration [24]. It follows from fig. 7 (curve b) that the most (statistically) stable
state of helical fractons in our jet flow corresponds to cos 0 = - 1 (0 is the angle
between u and to). Similarly in the grid flow and boundary layer it is at
cos 0 = 1. The subsystem corresponding to points with large helicity (curve c)
has two statistically stable states corresponding to cos 0 = ---1. It is noteworthy
that the curve b transforms into the curve c in fig. 7 when r (the integration
range in (7)) is taken to the left of the inertial interval and is decreasing. This
can be interpreted as an indication of the existence of helical structures of
smaller scale than those in the inertial range (i.e. helical fractons). One of the
features of these smaller scale structures is that they are polarized in the sense
opposite to the helical fractons. It is quite possible that the tendency to a
scaling behaviour in higher moments (see fig. 3) at small r is closely related to
these small structures. In order to gain more understanding of this problem
viscosity has to be taken into account. This question is, however, out of the
scope of this paper and will be addressed elsewhere.

8. Velocity structure functions versus fractons

It is argued here that the asymptotic behaviour of velocity structure functions

[25-28] is closely related to the breaking of reflexional symmetry on fractons.
470 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

We start with considering Lagrangian structure functions of the longitudinal

velocity component ((Aur)q), where A u r = lu(7 + T) - u(~-)l. Obviously
"r+T
Ou
(ur)= f --ff{dt . (46)
T

If the characteristic time scale of Ou/Ot is much smaller than T, the integral in
(46) can be represented as

l f -Ou
-~ ~ dt = riP07) dT1, (47)

where p(7/) is the PDF of Ou/Ot and a = max{l~Tmaxl, Inm~,[)-

Putting

P(n) = P + ( ~ ) + P - ( n ) , (48)

where p + 07) = 1 [P0?) + P(-~7)I, P - 07) = [P0?) - P ( - n ) l , one gets the follow-
ing representation of Aur:

=r i rip-(n) dn , (49)
--a

where ( )f has the meaning of an ensemble average over ffactons. Since for
flows with reflexional symmetry p _ ( , / ) = 0 it follows from (49) that nonvanish-
ing Au r in the interval (t, t + T) are related with nontrivial antisymmetric
function p _ ( ~ ) # 0. In other words large values of A u r (and in particular max
Aur) are expected to be found at intervals with strongly broken reflexional
symmetry, i.e. in ffactons (see, also [29-33]).
Assuming a scaling behavior

((AUT)q) ~ T ~q , (50)

it is easy to show - in the same way as in section 5 - that the asymptotic

behaviour of ~q has the following form:

~q ~ 1 + o'q, and max A u r ~ T ~ , (51)

i.e. (again as in section 5) the asymptotic scaling behaviour of structure

functions is controlled by their scaling behaviour at intervals with maximal
Aur.
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 471

Following the above arguments maximal A u T are found in fractons. There-

fore on dimensional grounds

max A u r -- ~ST~ , (52)

with

=1, tr = ( D r - 1). (53)

If the fraction dimension D~ = 4 the exponent o - = ~ , i.e. the asymptotic

behaviour of ~'q is given by

ffq ~ 1 + l q . (54)

Applying similar dimensional arguments to Eulerian structure functions Au,

[16], one obtains that

max Au r ~ gar~ , (55)

with

8=(l+D,) -1 and tr=(Df-1)(l+Df)-'. (56)

For the fracton dimension D e = 4 one obtains tr = 1 and the asymptotics of the
exponents of Eulerian structure functions are given by

ffq ~ 1 + l q . (57)

It is noteworthy that both asymptotics (54) and (57) are rather close. This
facilitates the comparison with experimental data in view of the general
difficulties of relating Eulerian and Lagrangian structure functions and the
validity of Taylor hypothesis [16]. Values of ffq obtained from two experiments
at rather large Reynolds numbers are shown in fig. 8. It is seen that the
experimental data are in better agreement with the Eulerian asymptotics.
However, at smaller Reynolds numbers one can expect better agreement with
the Lagrangian asymptotics; the results obtained in [28] on the basis of very
long velocity samples (about two order of magnitude larger than usually) at
Rea = 530 exhibit very good agreement with the Lagrangian asymptotics as can
be seen from fig. 9. The same is true for much smaller Reynolds numbers [1,2].
472 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

4--

3-

2-
_~~ \t,~- ~ o w3

I I I , I , I , l
0 4 8 12 16 20
P
Fig. 8. The intermittency exponent for velocity structure functions: (x) jet, Re~ = 852 [25]; (O)
large wind tunnel in Modane, Re~ = 2720 [26,27].

tg a --1/6

I I I I I I , , I [
0 I0 20

P
Fig. 9. Scaling exponent ~'q of structure functions of the streamwise velocity component, (O)
Re~ = 530. Adapted from [28].

9. Eddy diffusivity in fracton turbulence

T h e r e n o r m a l i z e d m e a n rate of dissipation ~ is a g o v e r n i n g p a r a m e t e r
allowing to find the e d d y diffusivity ~(~e) f r o m dimensional a r g u m e n t s ,

i d(e2>
- 6 dt ~-z~. (58)

In a similar w a y as in ( 3 ) - ( 5 ) the result is

z = (1 + D r ) - ' , /3 = 2Dr(1 + D r ) - ' . (59)

In the case of K o l m o g o r o v t u r b u l e n c e D r = 2, which t o g e t h e r with (59) gives

/3 = 4 / 3 , i.e. the R i c h a r d s o n - K o l m o g o r o v diffusion law. F o r the ffactons
 A . B e r s h a d s k i i et al. / Self-organization a n d fractal d y n a m i c s in t u r b u l e n c e 473

Df = 4/3 and it follows from (59) that/3 = 8/7, i.e. the fracton diffusion law (a)
mentioned in section 1, or more precisely

if/" ~ g3/7~8/7 . (60)

If the number of fractons in the turbulent flow is large, they can have an
essential influence on the processes of diffusion of a passive scalar. This
influence can extend to the scales far beyond the scales of the fractons. Such a
possibility is rooted in the properties of fractons (described above) enabling
them to trap the passive scalar inside them for a very long time. Therefore,
after some initial period most of the passive scalar will be located within the
fractons. On the other hand, the interaction of fractons with their environment
is controlled primarily also by the parameter g, i.e. this parameter controls the
statistical properties of the stochastic trajectories of fractons. In other words,
the statistical properties of the fractons trajectories will be determined mainly
by the properties of the fractons themselves and to a much lesser degree by the
properties of their environment. This brings us to the conclusion that the
relation (60) can be valid not only on the scales of the order of scales fractons
but also in a range of much larger scales. It is natural to call this range the
fracton range of scales.
It is plausible that these properties of fractons form the basis of the
extremely broad range of universal behaviour of the dependence ((t) in the
troposphere (fig. 1). Indeed, it follows from (58) and (60) for the fracton range

~ t 7/6 (61)

Looking at fig. 1 - adapted from [5] and containing data of many authors in
different conditions - one is amazed that all these results are well described by
a single universal relation (61) in the range of scales (horizontal standard
deviation) from one meter to one h u n d r e d kilometers. As seen from fig. 1 the
universality of the relation (61) is manifested not only by the single exponent
(7/6) but also by the universal constant in this relation. Apparently in all these
experiments the fractons have been of the same type and the scales of cloud
width fell into the fracton range. The relations (60), (61) are valid also for
quasi-two-dimensional turbulence (large horizontal scales in the troposphere,
fig. 1, and in the ocean, fig. 2) since fractons - which are three-dimensional
formations of a rather small scale - can be effective in the transport of a
passive scalar on much larger quasi-two-dimensional scales for the same
reasons as argued above. The properties of drifter trajectories in the ocean
have been studied in [7-9] and it has been shown in [7] that in order to
determine the eddy diffusivity from the data on drifter trajectories one has to
take into account that "the relative speed along the trajectory will not be
474 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence

stationary, rather it will accelerate". This in turn allowed to show [7] that the
observations in [7-9] are consistent with the patch dispersion observations in
[6] and thereby with the relation (60).

Acknowledgements

The authors are grateful to A. Arneodo, E. Bacry and J.F. Muzy for
information and H. Vaisburd's essential help in data processing.
This work was supported in part by Wolfson and Gordon Foundations as
well as by the Israeli Ministry of Absorption.

References

[1] A. Tsinober, E. Kit and T. Dracos, J. Fluid Mech. 242 (1992) 192.
[2] E. Kit, A. Tsinober and T. Dracos, Appl. Sci. Res. 51 (1993) 185.
[3] A. Bershadskii, E. Kit and A. Tsinober, Proc. R. Soc. A 441 (1993) 147.
[4] S. Alexander, Physica A 140 (1986) 397.
[5] F.A. Gifford, in: Sixth Symp. on Turbulence and Diffusion (Boston, USA, 1983) pp.
300-304.
[6] A. Okubo, Deep-Sea Res. 18 (1971) 789.
[7] B.G. Sanderson and D.A. Booth, Tellus Ser. A 43 (1991) 334.
[8] A.R. Osborne, A.D. Kirwan Jr., A. Provenzale and L. Bergamasco, Tellus Ser. A 41 (1989)
416.
[9] A.R. Osborne and R. Caponio, Phys. Rev. Lett. 64 (1990) 1733.
[10] S. Havlin and D. Ben-Avraham, Adv. Phys. 36 (1987) 695.
[11] H.K. Moffatt, J. Fluid Mech. 36 (1969) 117.
[12] H.K. Moffatt and A. Tsinober, Annu. Rev. Fluid Mech. 24 (1992) 281.
[13] B. O'Shaughnessy and I. Procaccia, Phys. Rev. Lett. 54 (1985) 455.
[14] J.C. Vassilicos and J.C.R. Hunt, Proc. R. Soc., A 435 (1991) 505.
[15] G.K. Batchelor, The Theory of Homogeneous Turbulence (Cambridge Univ. Press, Cam-
bridge, 1953).
[16] A.S. Monin and A.M. Yaglom, Statistical Fluid Mechanics, vol. 2 (MIT, Cambridge, MA,
1975).
[17] B. Levich, E. Levich and A. Tsinober, in: Turbulence and Chaotic Phenomena in Fluids, T.
Tatsumi, ed. (North-Holland, Amsterdam, 1984) p. 309.
[18] A.G. Bershadskii, Sov. Phys. Usp. 33 (1990) 1073.
[19] A.J. Chorin, J. Stat. Phys. 69 (1992) 67.
[20] U. Frisch, in: Turbulence and Predictability in Geophysical Fluid Dynamics and Climate
Dynamics, M. Ghil, R. Benzi and G. Parisi, eds. (North-Holland, Amsterdam, 1985) p. 71.
[21] A.G. Bershadskii and A. Tsinober, Phys. Lett. A 165 (1992) 37.
[22] K.R. Sreenivasan, Annu. Rev. Fluid Mech. 23 (1991) 539.
[23] A.G. Bershadskii and A. Tsinober, Phys. Rev. E 48 (1993) 282.
[24] G. Nicolis and J.W. Turner, in: Bifurcation Theory and Applications in Scientific Disciplines,
Ann. NY Acad. Sci. 316 (1979) 251.
[25] F. Anselmet, Y. Gagne, E.J. Hopfinger and R.A. Antonia, J. Fluid Mech. 140 (1984) 63.
 A. Bershadskii et al. / Self-organization and fractal dynamics in turbulence 475

[26] Y. Gagne, Th6se, Etude Experimentalle de l'Intermittence et des Singularites dans le Plan
Complexe en Turbulence Developee (Universit6 de Grenoble, 1987).
[27] J.F. Muzy, E. Bacry and A. Arneodo, Phys. Rev. Lett. 67 (1991) 3515.
[28] W. van de Water, Experimental study of scaling in fully developed turbulence, in: Turbulence
in Spatially Extended Systems, R. Benzi, C. Basdevant and S. Ciliberto, eds. (North-Holland,
Amsterdam) in press.
[29] S.F. Edwards, Progr. Theor. Phys. Suppl. 69 (1980) 1.
[30] E. Levich and A. Tsinober, Phys. Lett. A 96 (1983) 292.
[31] E. Kit, E. Levich, L. Shtilman and A. Tsinober, Phys. Chem. Hydr. 10 (1988) 615.
[32] L. Shtilman, R. Pelz and A. Tsinober, Comput. Fluids 16 (1988) 341.
[33] A. Tsinober, E. Kit and M. Teitel, Spontaneous symmetry breaking in turbulent grid flow,
Presentation at the 17th IUTAM Congress, Grenoble, August 21-27 (1988).