Professional Documents
Culture Documents
L.
1. lohn: Partial Diffenmtial Equations, 4th cd.
-. UI
~
,
..
-,
34. Kevorkian/Cole: Perturbation Methods in
. ..
-~~
.
3. Hale: Theory of Functional Differential 35. CO": Applications of Centre Manifold Theory.
- ?nA .d 'tit. - ....... • • .,
-
4. Pucus: Combinatorial Methods. Data Assimilation Methods.
S. \Ion Mises/Friedrichs: Fluid Dvnamics. 37. Saperslone: Semidynamical Systems in Infinite
6. FreibergerlGrenander: A Short Course in Dimensional Spaces.
Computational Probability and Statistics. 38. Uchtenberg/LiebemuJn: Regular and Chaotic
7. Pipkin: Lectures on Viscoelasticity Theory. Dynamics. 2nd cd.
8. Giacoglia: Perturbation Methods in Non-linear 39. PiccinilStampacchialVidossich: Ordinary
Systems. -umerenuw Jj(}uauons in K·.
9. Friedrichs: Speclfal Theory of Operators in 40. Naylor/Sell: Linear Operator Theory in
. -
. ..... . ... ---... .......... - . .
. ...........-..-
J1UDCfl ~pm;c:. --
- - . .~
~. • 'I'L I ___ - ~.~
Mathematical Theory.
.L". ure,...,...... CiU~IU . - • n
.
UI & . . . ."au
. I:'~
Porous Media.
01. ' J . • ,..
. . . . . . .a ..J ..... hI' .Ii
... 01 .. :.1 01___ __.I C'
,
n#
-
~ ftftd ..,.,..:. Ci4 fU", ......... , ilV'lIr IUfilinll Pnint Tnenrv
-- - - - - - - - - - - -
With 45 Illustrations
~
•
~
~nrinapr
"C '0
- _..............J, '-J--'~ v"S" ~v
USA
Series Editors
loA' ,'L . 1 n I
..
• ("!I:
.,.. . (1 QQl).
, , 17£'T.1. ROTRJ; R?MdO
I
532~0527-dc20 93-4311
the written permission of the publisher (Springer-Verlag Inc., 175 Fifth Avenue, New
York NY 10010 USA) excent for brief excernts in connection with reviews or scholarlY
analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or
hereafter developed is forbidden.
The use of descriptive names, trade names, trademarks, etc., in this publication, even if
tne former are not espeCIally laentmea, IS not to be taKen as a SIgn that SUCh names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used
~ . ,. 1
AI. c;~:a'y lJ'y -or
February 1993
- ---~-
,..,. 1",_
v
1
~
. . . . . . . . . &&~ '& ~~
-
1.1. The Euler and Navier-Stokes Equations
_... .. -..- ~.. -... . ·. · ··· 5
· . · ···· '"
~
q.~.
,.,. .....,
.1.
.
4.1. Statistical Equilibrium .
ueft
, , .. ,.. · . ·· · · · ...·
~
., ···. TTY
·· ·
,y
lUlU In vvave l'lUUlUer npace
· ·· · · ·
,..
67
(~
4.3. The
·
Combinatorial Method: The Approach to Equilibrium
and Negative Temperatures · · · · · · . · · · . ···· 74
4.4. The Onsager Theory and the Joyce-Montgomery Equation. 77
4.fi Thp. n . T.hnit. A.nt"J thp. Rfllp. of Tl Vii, • HO
4.6. The Approach to Equilibrium, Viscosity, and Inertial Power
T
. . ..········· ·.··· ··· · ·.·
OA
u-z
... 'T.
,.., .1 • .n. ...
u. Y VI. loll;;:;A u"'" 01.L
"'"
5.1. Vortex Lines Stretch · · · · · · · · · · · · · · · · ·· ··· 91
5.2. Vortex Filaments · · · . · · · . . · · · · . . · · · .·· ·· 94
5.3. Self-Energy and the Folding of Vortex Filaments ······ 96
5.4. Fractalization and Capacity
···· ·.· ········ 99
5.5. Intermittency · · · · · · · ·
··· ··.···· · ··· 101
·······.·····
I:: Q '1:1 £'1. 0 1l\Q
V.V.
···· ··
YVJ. "'VA ..... vv
5.7. Enstrophy and Equilibrium · . · · ··· ·.·· ······ 108
···. · ·· ··
. . . . u~
~
1
1"l7 ~~
"
157
Index 169
Introdllction
ticity field. I hope that by the end of the book the reader will believe that
these subjects are identical, and constitute a special case of fairly standard
. ical mechanics with both eauilibrium and non-eauilibrium asnects.
'rho '1 ,l.'
nf fl,11rl hll·hll1An('A ~rA rlnn tn tlu> .1 .
- .L- r -r
imposed by the Euler and Navier-Stokes equations, which include topolog-
ICal
,
• IISI.D1.111 .S
1 '
III LlIt: LIUt:t:-
, 1 • 1
t;~t:•
rrn
~
.
1lt: Uli:1.11l
,.
Ul
A
-
with certain properties of vortices in superfluids arises in a natural way,
,. ., , "" .
ana It POIntS OUt Slml1antleS as weu as mrrerences oetween quantum ana
. .
classical vortices,
The book is rather concise, but I have tried to make it self-sufficient. I
assume that the reader is familiar with the bread-and-butter techniques of
,1- . 11' .r:!roAn 'f;l c_
. T.'_ . ~n~Jvf;liQ -" !,- .
,
r -" -"
and with basic incompressible hydrodynarnics, I have provided introduc-
. , .,- . , .-
'.
_ l' _1- 1-'" '-
lIIVl.li:j lJV V.l r 'J lJl.l'VVJ. ~ r
book has been taught in a graduate course at Berkeley with students drawn
.
\ ' ,
a first-year graduate student in anyone of these fields.
ate tur u ence to statlstlca mec
interestin issues that do not contri ute 0 • ted.
For example, there is no discussion of correlations in time nor of the re-
others.
Introduction 3
begun. The Kolmogorov theory, for all its brilliant intuition, is imprecise
and, I believe, partly misguided. It does however provide a useful frame-
work for later analysis. Its conclusions are revisited in Chapter 7. Much
more can be said about the two-dimensional case than I said in Cha ter
4; the interested reader is directed to the references. My main interest is
doubt.
My major conclusion is that turbulence can no longer be viewed as
Inc mp
of phenomena; e.g., compressible turbulence differs from incompressible
turbulence, quantum turbulence differs from classical turbulence, etc. Yet
there is a reasonable and consistent eneral a roach to the roblem one
that is in harmony with the equations of motion and with results in other
already formed.
In many important respects, and certainly in point of view, this book
is a second version of m Lectures on Turbulence Theo of 1975. The
differences are due to the extensive progress made in the last 20 years in
is at hand, giving enough information to enable the reader to find the full
. . . . ..-.. ",..Y
CItatIOn In toe r
. . . " .. . .
ali l;ne ena or I;ne OOOK.
errors, however, is mine alone. I would like to thank the Institute for
A _,. ., C'f, .I. J"'" • £.".' .1.'.1. • "f (\{\1 {\O I. _ a_~4-
lJL uuy (l,L ~ LUL LIJO ... "LLL .L07v.L v~, 'fV uv.u .... £ .... uu
draft of this book was written, and the Basic Energy Sciences program of
the U.~. Department of ~nergy tor Its support over the past decade.
of constant density, in several forms that will be usefullaterl In particular,
the vorticity and vortex magnetization are intro uce ,and a rst iscussion
of s ectral variables is iven. Useful results whose roofs are avallableTri
elementary textbooksI-w11T'be merely ~statf!f1...
,
t = 0) located at a, one will note that it will move in time. The equation
o 1 S raJec ory IS
dx(t)
dt
(1.3)
aq dXj 8q
--+-=
j J
where V = 8
8
Xl
, !.:>8 '8
UX2
8
:1:3
is the differentiation vector. We shall often
write 8 1 or 8X1 for 8 . The operator (at + U • ~) will be denoted"by D/ Dt
(even though it is in fact identical to
, it). '
The left-hand side is the acceleration, the right-hand side the force. The
to them:
1.4
II
known as the Navier-Stokes equations. If R- 1 - 0 they are known as
boundary av of aboundeclaomafn
are:
u - 0 when R- 1 =I 0 ,
(1.5) n·n 0 when R- 1 = 0
where n is the normal to av.
If V is a bounded domain, and w a sufficiently smooth vector, w can be
where div u = 0, u·n = 0 on avo The vectors u and grad cP are orthogonal
in function space:
1.6)
In
If
(1.7)
,
account. On the other hand, there is no way for intern81 energy to become
conver e In 0 Ine IC energy SInce, y e nl lOn, an Incompressi e ui
does not allow for changes in the fluid's specific volume, and an inspection
of the thermodynamic formula for the work done by a fluid shows that the
·s m I lar tem erature of the fluid
the one that can be measured by a thermometer, can have only a limited
.
vice versa.
1.2. Vorticity Form of the Equations 9
e,
~
To find an equation for take the curl of equation (1.4). Some manip-
ulatIon ot vector IdentItIes YIelds
f
fc = lc u(x,t) ·ds.
Let Ct = cPt (C) be the image of C under the flow map. The quantity
r(/) r n(y
.. , J<'/t ' ., f).rh:.;~
., . ~~nt.;n timp (thiQ;<;:
, t.hp "(';r('111~t;rm t 1. ")
I
quantity r = IE e.
dE) where E is a cross-section of the tube; r is constant
in p y ivergence eorem, an w en ,i is a so a
constant in time (by the circulation theorem).
The circulation theorem shows that vorticity cannot be created ab nihilo.
This conclusion remains true in the viscous case R- 1 o. In two s ace
dimensional inviscid flow vorticity is merely transported by the velocity
, ,
circulation theorem the magnitude of { can change. Indeed, DelDt -
- . st_r~tc III term , IS t e
rate of change of u along a vortex line.
e
If = 0 at t = 0, it can become non-zero by generation of vorticity at
external forces.
1 (x - x') x {(x')
47f Ix - x'13
where x is a cross product. This is the Biot-Savart law. The convolution
f *9 = f(x')g(x - x')dx' .
(1.10)
where K is the operator -(41rlxI 3 )-lXX. K can also be written in the form
1
1.2. Vorticity Form of the EQuations 11
T
J.U lIWU .r
:/. . ,U
T/"
.L \. ...
...
<,.,
.1-
tv l1'GL 'G
1 I ~ \
(1.11) K= 2: ~ a:Jf. ) log Ixl .
f
E = ~ I u.f.dx.
J
"T •• l:lcr~in 11 rourl A. . , An,:>
1. .
·0 .....
r1
LJ
1
2 J
ru"
L\
~\
"T
1- I
L A.l"1
" \ 1 J
UA.
e
If has compact support, (Le., if it is zero outside a bounded set), U
decays for larQ"e Ixl as Ixl- 2 .ffrom (1.10) abovel and A decays as lxi-I .
Since the area of a sphere is proportional to Ix1 2 , the divergence term can
h" 1'. .1 ;",l-", n 1' . . ,1 I- 1-. .... 4- •• 1. nl- • n • • l-1-. ..",
""'" u ... u""' .... . 0 ....... uu~v .... u u , u .........~
(1.12) E=!
2
J A.{ dx=..!..-
87f
JJ dx dx' t;(x)-e(x') ,
Ix - x'i
where the formula for A above has been used. Formula (1.12) will be a key
,. 1
Ll.. .1
LV.\. U.\.U.\.U .\..\..\. \JUV Dv'! uv,\,.
- Jr
J.k - ~.
.1-.. , •
tx)ax ,
.
I'; -
~
1,~,
~
ol, •••
1t = .I e· u dx
I= e
x x dx , x = cross product sign,
is a constant of the motion for both viscous and inviscid flow. The meanin
of I will be discussed in detail below.
,
vortex ring of outer radius R and inner radius p (Figure 1.1):
(1.14) u = r(log(BRjp) + C)j41rR+ O(pjR)
where r = lel 1rp2 and the constant C depends on the distribution of vor-
i=l
where it does not vanish; Le. ~i = 0 for all x not in its support (Figure
1.2 . A special, useful choice of functions ~i x is
8 small,
argument K = K o.
port, neglecting the deformation of that support by the flow; their velocities
14 1. The Equations of Motion
........'"
......
-, dXi , , ~ _. , ,-
~l.lO)
dt
- UtXi) - L l\cHXi Xj)1 j
j::f'i
(1.17)
dXil 1 '\' rj (Xj2 - Xi2)
. ,
dt 21f "--' r~·
j::f'i 13
where rii = V(Xil - Xjt}2 + (Xi2 - Xj2)2 = IXi - xii and (Xil,Xi2) are the
of x.:.
Introduce the Hamiltonian H
, .. "1'7\ , .... n \ '-
= - 4~ L:i Ej~i firj logrij. Equations
\.J..J.,) -\J..J.0j
-'-
• r ............
~_/T" \
"'0....\ ... 2/
1 :1: T' ....
.......... 1. ...
f\ ~~...J
v .............. ...
1 Ll.
-v
. ......."1 .
n .. ~....
T.\ /1 1 ' 7 " 1 H!\
\ ........ I \ . ' /
then become
dX~i _ 8H dx~. 8H
(1.19) . --;--
at,
- ---.
UX2i
-" , ~
at
" - -",
U:.Lili
,
:,,; ..v .......
h n J•• I',...".YY> ... U ... .....,.:' . A .
, ....... ,1 • • ...01...,,'
.
........ , ..................... "'J~ ,.......... .
shows that dd~ = O.
3See , e.g. A. Chorin, 1972; A. Chorin and P. Bernard, 1973; O. Hald, 1979; J. T.
Beale and A. Majda, 1982. For recent reviews, see K. Gustafson and J. Sethian, Vortex
Methods and Vortex Flows, SIAM Books, 1991, and G. Puckett, 1992.
1.3. Discrete Vortex Representations 15
(1.20)
differs from E by a finite constant. Note however that the energy must be
. ..
POSl lve y e m IOn) w I e can e p I v
the signs of the r i and the values of Tij.
When the flow is confined to a finite domain 'D on whose boun ary t e
= 0 is rescribed the formulas above under 0 a sli ht mod-
ification. The Green's function used to relate vorticity to velocity must
take into account the boundary; it can be written in the form G(x, x') =
16 1. The Equations of Motion
ei
where the supports of the are small. One can try to make these supports
spherical, and leave the connectivity constraints to be satisfied weakly.
in three space dimensions are integral lines of a smooth vector field and
, ,
the identity dive = 0.) This kind of construction is useful numerically4
resent ro ems smce t e constramts are
very important.
One possible choice is to assume that supp eiis a closed vortex tube.
e
The tube has a strength f i , the integral of across a cross-section, that is
that surrounds the tube without surrounding any other vortex tubes. To
ave a ni e energy, e u emus ave a ni e cross-sec e
have N such tubes. At a point x far from the tubes, the relation (1.10)
becomes
where the inte ral is alon the center-line of the tube and ds has the obvious
meaning of a vector element of length. For x near x', the structure of the
dSi • ds·
with obvious notations. This integral must be smoothed near points where
J
Much more detail will be given below. For the convergence of this type of
approximation as N --+ 00, see e.g. Greengard. 5
the USe of vorticee;Hamiltonian fm llIutations aie not ·l1niqlle, once one ftftS
been found others can be derived from it.
In three space dimensions one specific Hamiltonian formulation that
7
m=u+ at t = °.
Obviousl at t = 0, u = Pm, with the ro 'ection P defined above. It is
not required that div m = 0. We have { = curl u = curl m. If one thinks
(1.23)
(with summation over multiple indices, and u = Pm). The claim is that
the resulting u is identical to the solution of Euler's equations if u(x, 0) =
at u + P( (u . V) u) =0 ,
,
m and u always differ by a gradient._
e now ave an equa ion or e evo u ion 0 p
gradient. We shall put this gradient to good use.
e
Suppose = curl u has support within a ball B of finite radius p. In
a imen i n teri r of a s here is sim I connected and
thus outside B one can write u = -grad ij for some ij. Put q = ij in (1.22).
The resulting m has support in B. m can thus be "localized", and this
. .
e
Suppose has support in a small sphere B t5 ; calculate the resulting m
so that m also has support in Bs:
(1.25)
N
m=
i=l
1.4. Magnetization Variables 19
N
(1,;/ = U(Xi) = L U(i)(Xi)
J.
(1.26) ,
.; .1
J
where u(j) is the velocity (1.25) due to the 1-th "maQ"net". The coefficients
M(i) are not constants; from equation (1.23) one finds
dMi(k) (k)
(1.27) .J
Uf,
= -M-j OiUj(Xk),
.1..
thp.re is summation over repeated indices and the u,j are the compo-
nents of u = L: u(k) .
Onp. rRn now .'- ., t.hHt thp. Bow of t.hp~p "mn.1-' I~I.:"\" iR Hfolmi' .
~
with
J..T rv 1
:l.J
f rn • 11 nv - !
:I.
f
J"
(11...... (frAn n \ • 11 nv
<:> :.1./
-!
:l.J
f 11 2 rlv
.
... 'CO.,
.1-
"'U'I;;
,.. .J. •
~'-lU: '-JJ
.
"0
, ,
a.
YT. ., . J:. .1
..v.. ....... 'CO
Jl
UV YV
'J:
.... VU'CO UO'l;;O
.1
....... '1;;
appropriate variables.
One can check that the equations
dXjk 8H dM(j)
k oH , r. I. \1
L.....J \ ....J1, .... J"/" WJ;jJJ ,
dt 8M(j) , dt 8Xjk
k
are exactly equations (1.26) and (1.27).
Thp. " " Mrh ... havp. a
. .1 . . Onp. rH.n .1 ., hv
• ~
" • •
II
v
a painful but elementary calculation,9 that the velocity field (1.25) is the
.... . ., l::",1~
........ ~~
• ..J ..J l-...•• ~ ~~~11
IJJ ... YV~ V~""~UJ
.. 1~~~ ~~ .l-l..~ .r~_~ .C D~
.. vvp V.L un"" "V.L~~" v .. .L ~O""I.\,j
,,
.L • .L,
with p small, M perpendicular to the plane of the loop, and IMI = rll'R2 =
e
have thus approximated by a sum of small vortex loops.
ere IS an ana ogy e ween magne os a lCS an Ul ynamlcs, III
which the current corresponds to vorticity and the magnetic induction cor-
responds to velocity; the magnetostatic variables are related by the Biot-
avart law ·ust iiI<! the fiui 'a . I rm
to the magnetization, hence the name. Since the name is awkward, we
"
a Buttke 109P representation. Consider a large vortex loop C of circulation
ans . The non-uniqueness of ~ corre-
s onds to the non-uni ueness of and m. Construct a coordinate s-stem
on ~ in terms of some parameters, say 81, 82. In each small rectangle R
, , " ,
struct a Buttke loop of strength M = rt5s1bs2, oriented in an orthogonal
.re .Ion 0 n on .IS en y. e sum 0 ese u e 00 up
the original loop.
The converse problem, how to reconstruct macroscop~c vortex loops from
an arra of Buttke 100 s is much harder. How does one decide whether
small objects can be viewed as parts of a larger whole? An answer to
this kind of question will be given below in Chapter 6, in the context of
r x x ds = 2Ar ,
Xk xM
.the fact that a Buttke loop has a velocity that is determined by the
6·
addition of a velocity proportional to M(k) to dXk/dt does not destroy
1.5. Fourier Representation for Periodic Flow 21
C k arbitrary constants,
k
vergence of the discrete approximation to the continuum limit is unaffected,
since as N --t 00 t e "sel -velocity' the e ect 0 the k-th loop on its own
velocity) is a shrinking fraction of the total velocity.
Stokesuf:<l-uatiQn§ is
u = Pm .
(1.29) k· Uk =0,
k'
22 1. The Equations of Motion
, ,
] and i - yCI (and thus not an index). Elimination of Pk
W] k~
".
Or ,
k'
an ,as e ore, Uk = u_ k . e componen soan opera or
that projects the vector k6Q6 on a plane tangent to a sphere centered at
the origin; Le., the Po -, are the components of P, the Fourier transform of
. .
The energy of the flow can be written, via the Parseval identity, as
Define the m-th "energy shell" as the portion of k-space such that
and thus
00
m=O
00
for some function E(k). In the present setting this is the case only if the
. .
, ,
implies that the average 'energy per unit area tends to zero. We shall see
1.5. Fourier Representation for Periodic Flow 23
k2
~ '" ~
d
'-- -........
'""\
'"\ \ \ \ ,
•
•
•
••
• •••••••••••
• ••••••••••••• ~~
• •••••••••••••
....
• •• kl
• • (m + !)~
~
~- 211'"
L
be made finite by throwing out, in some clever way, all the wave numbers
above some large bOund. l\.max ~ tne Cleverness is need.ed. to mInImIZe tne
effect of this truncation on the components that remain). The two types of
finite calculations are not equivalent. A finite number of vortices gives rise
to . ,
. and · . f1plrl~ W~HJ:-'I Fourier series have an infinite
Vl
.1
omitted any mention of what should be done when R-: 1 =f 0, Le., the dis-
. . . . ...-. .-.. ..
cusSIon was ror InVISCIQ now OnlY. Tne omISSIon wrrn:m repmre-rr III \7 -r
2 below. In the spectral case, the viscous term presents no particular prob-
lem and was therefore included as a matter of course.
fine homogeneous random flow, its energy, vorticity and dissipation spectra,
and its Fourier transform.
00
n=l n
An E B'Vn.
J1, is called a probability measure if J1,(n) = 1 (the probability of something
.. . . ii
space. We shall denote a probability measure by P.
dx winO.
e -valued function
defined for w E O. Let 1/ satisfy the following condition: For every Borel
, . .,
probability to the event that 1/(w) has a numerical value in a certain set].
1/(w) is called a random variable, i.e., a variable whose value depends on an
experiment, with. probability assigned to the event that it should assume
'n alue. The inte al of w if it exists is called the ex ected value
or ~ean of 1/ and is denoted by (1/):
(1/) = 1/(w)dP .
xF~(x)dx
An infinite set of
where 8 1 ,82 are Borel sets on the line, and 8 1 x 8 2 is the corresponding
" "
,
Furthermore, if "l1 and "l2 are independent,
28 2. Random Flow and Its Spectra
.,
T.o~ .... ho a -' •
.~
1 1
.~~.
Tho """'n'1l how-co
~66~
r f f
(~.l) .t;l11'·J =J 11" d .P = J xlOd.J4~(x) = J X" ff}(x)dx
if F~ = if} exists, are called the moments of 11. The moments of 11- (i) = "1e
~ l'
(W.'V
11 ..J L 1.
"J.J.~
1
VJ. J.
T_
~.u ..-
'.1
,
.
~TJC)
')
,...-
V (TJ~) = standard deviation of TJ .
f+OO fOO
{G(TJ)) - G(1])dFn > G(TJ)dFn
J-oo Ja
Thus
\ ., (G("1))
nl
.L
.....
VI ::::. u).:::: G(a) .
In particular, let G("1) = "1 2, and "1 = Ig- (g)l where 9 is a random variable;
UTt> finn
T"'> II ...... r 11 , \
var(g)
I UY .CJL9jJ ~ U) ~
a2
where var(g) is the variance of g.
Let "11,1]2, ... ,TJn be independent random variables, each with the same
..J. ,L '1 . ~ .
. anrl ... "'~"" I" • 'YYI Qnn 1T
2 T,t>t '1'1
P ( I:~=l1]i (1]1)
~) (]'2
e > 0:
n > e < ne'""'.
\ J
1
uy
11 .
t.UC J.VJ.lllUJ.o. o.UVYC.
_1
-'- rn1
~UUi:),
lim P
n-+oo
l/ "n
~i=1
n
'It - (1]1) >e)=o.
'\
2.1. Introduction to Probability Theory 29
'T'h;...
~ ~~~~
"F;~ h
~u~~
"nyo •
~~~
.. ~
'r1 no I"\f "
£~~,
l) I"\YO " -~ ~ o ~
l)
• 1<',... yo 1'" ~ oyo
~ ~~ ~~. ,~~
n·...'"
~~~
'-'
p( -L."lin- m
,
> ka)
- <-
- ",Iii
1
- k2
where k is a constant and we wrote E = ka / For n = 1, P(I17 - m·1 > vn.
ka) < 1/k"J.; a variable is not likely to diffe~ from its mean by more than a
few standard deviations. The standard deviation Q"ives a rouQ"h estimate of
tIu:>
. -l I"\f (<Y'J
\. , I'Y'I\ ,
'-' "II I
For later use we wish to single out a particularly important class of
1
. VUl
. .. •
..--.
I
r>
LlIe
. to .
i'"W , , t F= (;,-=:.-n,,~lij',./2 uu
•
.
l.:rO',m~:C)
V21fa2 J -00 e
A random variable "1 which admits Ga,m as its distribution function is called
Gaussian. It is easv to verifv that
(r]) =m
I \ II \2\ 2
Va.l~' /}
'VI Ill, J I u
L;entral LImIt Theorem LJet De maepenaent ranaom varl- 'f}l, ... ,'f}n
ables with common distribution, mean m and variance a 2 < +00. Then
F(xl,' .. , x n ) =
Xl
...
fxn F'(Xl, ... ,Xn)dxl, ... dx n ,
J 00 J 00
~n
F'= v F.
8 XI .. ·8xn
30 2. Random Flow and Its Spectra
entries
Le., for each x E 'D, the velocity u(x) is a random variable; u(x) = u(x,w),
wEn, with (n, 8, P) a probability space. The knowledge of the distribu-
tion of u(x) for each x does not provide much information about the flow.
, , . .. .
statement does not allow one to distinguish between a flow of the form
u = 11 x constant,
where "1 is a Gaussian random variable, and a flow in which the variables
h os ible sam les of the
first random flow are very smooth, while those of the second are very wild.
2.2. Random Fields 31
finite number n of points XI, X2, ... , X n , the joint distribution function of
1 , •.• , n
Let FX1 ••••• Xn (YI, ... Yn) be the joint distribution function of U(Xl, ... ,xm );
and Yjp is the p-th component of the vector Yj' For simplicity, we shall
merif:} The famtty of functions FX1;"x n {til ' Y"nlmust be distributioll func-
. .
requirements:
I F ... =F
and
FX1 " ' Xn (YI ... Yn) satisfying conditions (i) and (ii) above are given.
fined by
u is complex, we define
11 XI, Xl _ ,
2
(iii) IR(Xb x2)1 < R(XI' xdR(X2, X2),
iv For every n, any Xl, ... ,Xn , and any complex numbers Zl, ... ,Zn,
n n
j=lk=l
32 2. Random Flow and Its S ectra
j k
2
(2.3) (U(Xj) - m(Xj»Zj >0.
Le., flows which fill out all of space and whose statistical properties are
e en 0 e par icu ax pom in s'pace one conSI ers. i cu ies
with boundary conditions (and boundary layers) are thus bypassed. A
discussion of the reasons why a flow can be viewed as random will appear
in the next section.
A flow field U is said to be homogeneous if for any n, any Xl ••• X n and
, n
.
distribution of U(XI + X}, ... , u(xn + X). In particular, for a homogeneous
flow field
ml(X) = ml = constant;
The properties (i), (ii), (iii) and (iv) of R(Xl, X2) become:
1 X =
(ii) n(O) > 0
(iii) IR(x)1 < n(O)
,
2
flow field. u (x) is the "energy" at the point x of the realization u(x), Le.,
2.2. Random Fields 33
.. "x
lim 2X~ u 2 (x)dx ,
X-too -x
if it exists will be called the snatial mean enerl!V of u(x) = u(x. w). For
the satre·of exposition, we shall first take the spatialluean of u 2 and then
i~o ~ ... n""
~,~ ,~~, •
rr',.1.· ~
~h"" ..
•• 1 g ...o~ ohr.n1.r1 nr.~ h"" ~r.r.
~ ~ ~
1 " '.
,
;~
~v
where the nk are given numbers and the ak, O'.k are random variables. u(x)
is a random field, fully determined by the joint distribution of the ak and
(l",. The set of freauencies nk- is called the snectrum of u(x). To simnlifv
where Ck = ak +ibk , and ak, bk are real independent random variables. The
mp.::I.n ~v
of thiH ..J fip.lc1 iH if thp. 1.1 arp. . , IsmaA-
.....
able,
\ J -A I \ J -A k r I
\ "',. ,
k=f.r
., " .1')
34 2. Random Flow and Its Spectra
F(k) =
(Khinchin): For a/unction 'R(x) , -00 < x < +00, to be the correlation
junction of a field which has translation invariant means and correlation
. . '0
(Iu(x + h) - u(x)1 2 ) -+ 0 as h -+ 0 ,
z zs necessary an
+00
'R(x) = exp(ixk)dF(k) .
Given F(k) [or ¢>(k)], 'R.(x) can be reconstructed with the help of Fourier
Integrals.
More generally, let u(x) be a vector-valued random field in a multi-
dimensional space. Let u(x) be stationary in the sense that (u(x)) is a
vector independent of x, and the functions 'R.ij(r) = (Ui(X)Uj(x + r)) are
. .1 11'
. -' nf v ~nr1 - ..
~rl3 thl3 nf 1i'J1rt' , lp.t
r>. ( " ••
\ - .1- -/
r
J3iilrJ~~_J- e~p(~~'E)d!i~t®-l_
where Fij is such that the matrix with entries dFij (k) is nonnegative defi-
nite, and
dFij(k) = q>ij(k)dk .
'l'hp for thi~ . Rrp. rn~inlv (R) thHt it. i~ ('onvenipnt anrl fb)
, ~
that it is in agreement with both experiment and our intuitive ideas about
Ll- I 1 • _1 , , .0_
u;y
, rpt.. _ " ,. "
YY 111".11 lIUl
. _1 .L
llVVV
~~
10 r
1.111,., r 0 • .1. .11'"
in the literature contain assumptions which are no more justifiable than the
bald statement above. Note that it
r
(Ui(X)Uj(X + r)) = Rij(r) = } exp(ik· r)q>ij(k)dk ,
then,
thus ~q>ii is the density, in wave number spa~e, of the contributions to the
;.... +1-..... ,. ..... ....... "r"'~r.n '
.
" ' ......... & bJ' ... "'., U& •. ' " '&""'&bJ -oS .t-
We define
r
E(k) = ~ ~ q>ii(k)dk
Ikl=k
36 2. Random Flow and Its Spectra
We shall now consider random fields u x w which for each w Le. for
each experiment that produces them), satisfy the Navier-Stokes equations.
u depends also on the time t; we shall usually not exhibit this dependence
thus
Let e(x) be the vorticity vector, e= curl u; the components of eare given
au
i = Cijk a~j ,
o i,j, k are not all different
where Cijk = 1 if i,j, k is an even permutation of 1,2,3
-1 if i,i, k is an odd permutation of 1,2,3 .
'-
obtain
(e(x) . e(x)) = 1 00
Z(k)dk ,
where
vve now want to calCUlal;e line aI8Slpalll0n lSpecuum, I.e., l,;Ile CUntflDUl,;lOn
of various wave numbers to the total energy dissipation.
The Navier-Stokes equations read, in component form,
..
where Ui = Ui{X}, lJi = Ji'!;. t and the summation conventIOn IS used. Let
X' = x + r, and write u~ = 'Ui (x') for brevity. The equations at Xl read
the forces f have been omitted because they play no role in the argument.
A also that lu\ O.
Multiply the equation at x by uj and the equation at x' by Ui; the result
. \
III ".\
\ ..... ' /
, !:J I!:J , !:J I 0-1 'A. .
re· ., ,
u J
(2.8)
Now average the equations, Le., take the expected value of each term and
!1rlrl thll:\ ' . nntll:\ th~t
{)
1 •
2.4. Random Fourier Transform of a Homogeneous Flow Field 39
C'l . ·_OJ
,:
..
..: ~ .1
J' ....U u u.pp"" U.lO
.1 • +1..._
UU\.i - .
-
~-+- •
·0
. 1...._
UJ p ....... uo
+-
uu ~'1.~-
-~.
where A r = ~8;.. All the terms on the right-hand side except for the last
one vanish at r ."0 by homogeneity, thus
. - -
.Q '0 .. \ a 1...2\ ~ A '0 .. \
R -["' "1I1i11"
~ I 1.2 tn . .11,\ ,.11,
....,;. "'11 II'" =u
R7 dt \~ I =u ~..,
1 .
'1 .- - . .
~l(, -/C-CJ!ii IS tne COntflDution or tne motion wltn wave n
. ..
arouna K
tQ. the total dissipation,_ Integrating this expressi()n on the sphere of radius
k = Ikl first, we obtain
. ..nn
8t'Riilr=o = ~ (u ) = - io 2
D(k)dk ,
where D(k) = 2R- 1 k2 E(k) is the dissipation spectrum. The vorticity and
diSSIpatIOn spectra are proportIOnal to each other. It does not follow that
dissipation and vorticity are themselves proportional. Two very different
functions, with different supports, can have the same spectrum. We have
noted before that the spectrum provides only partial information about a
1 1"- . .
ut x , W) - Jr9{X, S)Ptas ),
where g(x, s) is a non-random kernel and p(ds) is a random quantity with
simnle nronerties. The snecia! case o(x. s) = eix .s will be the random
(or generalized) Fourier transform. The condition for its existence is that
1..2/.., .\\ l..~ ~ •
, .... II .... v
\~,....,
~~_ ~~~l..r : _
........ v ...." ....
+l.~4- 4-l. ... ~4- ~
... v., u .... ""'u ......v .... """"'0" " ...." .. oJ ""'u .... r
~, ....
h""
. ~
finite (as one could expect in fluid mechanics). By contrast, for the usual
li'Ourler transtorm one must reqmre that the energy III tlfe wnoTe space De
finite.
40 2. Random Flow and Its Spectra
Skorokhod. 2
et e a pro 2 , , e space 0
random variables defined on 0 that have finite first and second 'moments,
I(1J) I, (1J2) < +00. The inner product of two such variables is defined as
. 2
that
(i) peA) E L 2 (0, B, P), p(0) = 0;
.. .
,
meA) is non-negative by (iii):
also have m
This ro erty is called ''finite additivity". One can show (and we shall not
do so) that a family peA), A c A can be constructed.
XA(X) =
o X¢ A.
2.4. Random Fourier Transform of a Homogeneous Flow Field 41
, ... , n
whole space V, and construct the function
n
constants.
. q
such q(x) one associates a random variable
(2.9)
by judicious use of intersections one can use the same Ak's in both defini-
with
(rl) = J
q2(x)m(dx) , m(dx) = (lp(dx)1 2 ) •
sense, i.e., u -
lu-J9p(dSW =0.
g(xl,s)p(ds) g(x2,s)p(ds)
,
in V. We shall not prove this converse here, but shall use it anyway.
As a first application, consider a homogeneous flow field. For the sake
of economy in notation, u and x wi e wntten as sc ar. e now
2.4. Random Fourier 'I'ransform of a Homogeneous Flow Field 43
1T't.
.L .1£\;"
r. . DIJ_\
.L V" J
_1 ~
..... .....
~~
vu
.. \..
U.L.1'-
1~
u~~.....
~ ·HT••
. ~ • Uv
(
!:I \ ..... , i3
.\
J ...isx.,
m(ds) = dF(s)j then
u(x,w) =
J
f eikxp(dk) , (lp(dk)1 2 ) = dF(k) .
This is the random Fourier transform which exists whenever tlu 2 (x w)l~" is
finite, and generalizes the usual Fourier transform. The energy spectrum
:~
40.> 04
. V.U
l-..~ •
L.IJ
E(k)dk = Ir \ Jk<lkl<k+dk
Ip 2(k)lm(dk) \
/
,
can take the positive or the negative square root at different values of k.
Tuen
Jr ? I
t
ql X lax <...
, •
-roo , JrCJ2 t ? I
X lax
, •
<... -roo ,
v
-k reikxql(X + a)dx =
~II oJ
-~
v""',
r
oJ
eik(x-a)ql(X)dx = e-ika(ir(k) .
Thus
'R(XI, X2) = j hex - xl)h(x - x2)dx ,
A
"..f rr
;:1 \ ' I
0\
'}
- J,.{o
'\'
,...\
-} ,
m(ds) - ds ,
44 2. Random Flow and Its Spectra
yields
u(x) = h(x - s)p(ds) ,
-00
sion of random fields in Section 2.2 and because the results will be needed
(ii) If 0 < tl < t2 < t3 < t 4, then the random variables W(t2) - w(td
a 4 - 3 r
(iii) For all s, t > 0, the variable w(t + s) - w(t) is a gaussian variable
with mean 0 and variance s/2.
(iv) With probability one, w( t, w) is a continuous function of t for each
One has to show that a process that satisfies these conditions exists, which
is not 0 vious. or examp e, i iv were rep ace y t e con ition t at
w t w be differentiable then there would be no wa to construct the cor-
responding process. For a proof of existence, see, e.g., Lamperti. 3 A sum-
" t
....
x
...h
FIGURE 2.1. Brownian walk in time.
these particles to perform independent Brownian motions. The consequent
density of the particles at time t will be v(x, t).
....... . . . . . . 'I. • • .. • ..
l'IUW, gIve Lne panIcIe:s, In aaOUlOIl LO Lne ranaOIIl mOLlun, a y
velocity a (a "drift"); symbolically,
(2.13) dx = adt+dw;
in a step of length ilt, ilt small, a particle will move from x to x + ailt +
w, wnere W IS a gaussIan variable with mean U and. varIance D.t/"I.. J:!jqua-
tion (2.13) was written in a peculiar form in terms of differentials, since, as
we have seen, dw/ dt does not exist as a regular function. (It does exist as
R. t· '1 • SI.nrl iR ('Allpn " t· nOiRp." .) rrhp. • . ~ .J bv the
• , &
.
'V" •
~
(2.14)
ll'" .~ 4
V
fi 11.
if'.: hprp t() ii...l·
v
;~t thp Ii.
•
II~· ~~ ()f w t,() whHt i~ .J.J
.. 01 wIllen we SHau neea oelOw omy tne prmCIpte OU{; nOt tne
III ,
details. The key observation is that the diffusion term in the N avier-Stokes
equation can be approximated by simply giving vortex particles random
pushes of the right amplitude. the amplitude decreasing with the Revnolds
number R:
length h, to the right with probability ~ and to the left with probability ~,
with one step per time interval At. We shall call such a walk a Brownian
walk. In the (x, t) plane, the walk will look as in Figure 2.1. There are
obV~lJll~ ~.nRlo(Tl1P_': in two ~n~.CP rlimpnsions with four directions to choose
~
from at each step, and in three space dimensions, with six directions to
- , &..
u. V.lU..
A J. n L'L..._
.11 __ L _ _ 'nl .
, .l.l1U.l.l
In.
, OJJ
~o /I, V, '0' lJV (11
the central limit theorem, which asserts that sums of random variables
converge to gausslans. The llmit IS wlt) With a tactor that depend.S on fit
and h.
InFigufe 2~2'we dIsplay a Btowlltun walk Oil 31two=dimeftsi6ft8:ll~iGe,
, ,
with th~ t11'Y\~ ~vic;:l ' Tn ~11 th~ .1' .L'
th~ tirnp. Rxis
.. -..- c
'T .T1Q
cities; walking down the street one feels gusts of wind on a human scale.
, . ",
All these scales of motion are strongly coupled, Le., to calculate any of
50 3. The Kolmogorov Theory
them or at least a larger fraction of them than may seem reasonable a priori.
forces which change appreciably over a time T. The time it takes the fluid
to respon to tee ange in the orce is compara e to ; t us) t e ana YSIS
of the res onse of the lar e-scale motion to the external forces re uires a
solution of the equations of motion and does not have a universal charac-
such adjustment can take place in a short time compared to the time it
takes the flow to decay, and thus it can conceiva y give dse to eatures
inde endent of the articular roblem at hand.
Let us temporarily identify Fourier components of a velocity field with
. .
"
Assume u(x) is periodic) with Fourier series
3.2. Kolmogorov Theory: Dimensional Considerations 51
-1
, A_
independence of the small scales and the large scales to assert that the
scales and large scales can be compatible with the strong coupling between
sea es.
. .
3.2. o mogorov ImenSlona erations
onsl er a omogeneous ow WI 1 energy spec rum i an Isslpa Ion
1 2
spectrum 2R- k E(k). The homogeneity assumption is not very severe
for small-scale motion since it is plausible that any flow can be viewed as
homogeneous on a small enough scale. It is customary in the discussion of
statistics invariant under rotation. We are content here to assume that the
ow is su cien y i eren rom wo- imensiona e
latter's constants of motion.
When R is very large one must wait for a very large k before the dissi-
ation s ectrum is si nificant. Thus the ran e of wave number k for which
2
k E(k) is significant may be quite large because of the lack of damping.
2
52
EX.k)
"
believe that all that happens in the inertial range is that energy passes
• «
1
range") to the region around k2 where the energy is dissipated ("the dissi-
patlon range. at cou In t IS mterme la e regIOn epen on. t
is plausible that it depends only on k and on the rate of energy dissipation
f = .it u 2 • The units of E k are L3 T 2 whffe* fis a uiitror1eil-"'tliafia""
3 • • • • • •
J
T a unit of time [remember (u2 ) = E(k)dk], k has units L-l, and f units
2 •
fJ. The only combination of these parameters that has the dimension of
engt IS 1/ f. = ",; ", IS t e o mogorov sca e. t IS P aUSI e, an In-
de~~ verifiec:lby<experiIIlent~thatml] is the a.pproximate order of magw.tuQQ
of the scales on which dissipation becomes important. Thus k 2 l/'fJ, and r-.J
interested in the inertial range only, we can assume that the limit v -? 0
, 2
in 0 < k < Ok l , 0 = some constant> 1, and an inertial range for k > Ok l •
Pic a large constant K sue that k 1 K is small, and perform the
change of variables k' = k/K. In tenns of k', E(k') (u 2 )b(k') + r-.J
smoothing near k' = 0 leads to a smoothing near r' = 00 (r' = the variable
54 3. The Kolmogorov Theory
the elaborate analysis of Fourier transforms in this specific case, but the
e mer-
the range being irreversible and all in one direction. We shall present this
y
Divide wave-number space into energy shells, as in Figure 1.3; for ex-
ample, let the n-th energy shell be 2n < k < 2n . Let Un be a typical
velocity associated with this shell (for example, the square root of an av-
r--J -1
typical length in the shell, and let En r--J u; be the energy associated with
,
magnitude. The characteristic time for motion in the shell, Le., for motion
t at correspon s to wave num ers In t e s e , IS n Un = Tn. uppose
that in a time Tn the motion in the shell gives up its ener to other shells;
the rate of energy dissipation is then
((u(x + r) - U(X))2), we have derived the Kolmogorov 2/3 law (3.3) and,
by Fourier transformation, the 5/3 law (3.1). This cascade is "locar' be-
(i) If all the energy moved from one shell to thenext, theenergywoutct
n n - ,
what has been deduced. If u~ decreases with n, were does the extra
energy go?
(ii) If the energy moves from one shell to the next, and the energy
sect u h no a s the time of transmission To must be the
same for all shells, again contrary to what has been deduced.
separately, with energy moving from shell to shell. In each shell for a given
n- I
is thus proportional to the fraction of time the energy spends in each shell:
where Tn = in/VE and T is the time it take the energy to flow across
t e Inertia range. y a ouner
h this is the truth the
Kolmogorov law is well established experimentally) but only demonstrate
. .
Kolmogorov picture can exist and lead to a different spectrum.
a case e picture rna es sense, one pro a y mus ave a comp ex
interplay between' distant shells. On the other hand, nothing in the Kol-
mogorov theory forbids one to consider the establishment of the spectrum
as being irreversible, like the spreading of water that has initially been
. ,
been established being nearer to a sloshing to and fro, with some leakage
turbulent flow. In other words, it may well be that once the inertial range
3.3. The KolmogoTOv Spectrum and an Energy Cascade 57
has been formed, energy goes back and forth across the spectrum, with the
5/3 law was derived), and the energy range should be independent of the
inertial range see ection 3.1 . The Euler equations are invariant under
the time reversal t -7 -t u -+ -u. The formation of the inertial ran e is
irreversible, in the following sense: if one starts with a very smooth flow
,
the Euler equations, is incompatible with the cascade picture, unless the
w o e ru. u ,
1
once u has developed, if time reversal changes the inertial range the cas-
cade picture is plausible, and if time reversal does not change the inertial
ran e then the cascade icture is im lausible.
The even moments of u 1 are invariant under the reversal u -+- -u but
where C3i are constants and r becomes a shorthand for O(r) if the flow is
. .
distribution derivative of
i ,
and the third moments of u[ are approximately zero through all of the
__ ___ ~~ ana SIS supports e a u pIC ure.
ever little experimental support for the notion that Kolmogorov scaling
. .
ingful.
hd C = lim Hmin
p-O
which hd(C) = 00, = least upper bound of d for which hd(C) = 0, exists
5B. Mandelbrot, Fractals: Form, Chance and Dimension, Freeman, 1977.
3.4. Fractal Dimension 59
, ,
three segments of length 1/3 and throwaway the middle one; take each
o e remaImng pIeces an row away e ml e 1r, an so on. e
"remamder Is the Cantor set. Asstlme that the HaHsdorff mCilasureo£ C is
finite and non-zero. C consists of two subsets of equal measure, each subset
f h hal reduc d b a factor 1 3 thus
hD(C) = 2 . 3D . hD(C),
XN = 'r/i,
i=l
60 3. The Kolmogorov Theory
(the Biot-Savart law); the energy is thus defined and has a spectrum.
3.5. A First Discussion of Intermittency 61
Pick a sample flowj and calculate {(x) .{(x + r). If x does not belong to a
r z r ( sn on 0 vor ex eg,
same is true. If both x and x + r belong to vortex legs the product is non-
zero. Average it. Ifx and x+r belong to different legs {{(x) ·{(x+r) = 0,
since the steps in the walk are independent. Thus {{(x) . {(x + r)) = 0 for
.
"'.. t-o",10
. .
constant (tending to infinity when CT - . 0) in that neighborhood. As h ~ 0,
By comparing our two examples, we see that when D has shrunk from
3 to 2, I has decreased from 3+ to O. This is quite reasonable, since as
Oluau.v1., n I n u
transform flattens out. Relation (3.2) shows that this is a general property
o t e Fourier trans orm, since it sows that as the support 0 a unc-
tion shrinks the su ort of its transform rows. However the scalin
given by (3.2) does not change power laws {since (I(k)-' = K-'k-' =
. -," . .
forms is also the mathematical form of the Heisenberg uncertainty principle
things being equal; it is not clear what the last phrase really meanSj indeed,
One can show that Brownian motion is self similar (part of it, appropri-
ate y own-up, IS I entlca to t Ie woe . spectrum 0 t le orm ,
even for I = 0, is also self similar; C k K -"/ = C k-' where C is a
1 1
6H. Dym and H. McKean, Fourier Series and Integrals, Academic, 1972.
62
ior are well documented experimentally. Many authors give values of the
flatness (defined for a random variable 'fJ as ('fJ4) / ('fJ2) 2 ) and the skewness
'fJ 1] of quantities such as au ox, or more general y n u x n ,
derivatives of a com onent of u. For a aussian variable the flatness is
always 3 and the skewness O. For a quantity such as au/ax· the flatness is
,
the energy transfer between wave numbers.
pro a 11 y IS rI U Ions a genera e e 0
skewness.
It has been suggested that intermittency should affect the value of
th x ne t in the I r k rv 2/3 -"/. the dimensional
argument would be modified by the introduction of an additional length
scale that characterizes intermittency and the cascade argument would be
of concentrations.
in a turbulent flow. One needs to find out not where ef; 0 .but where f
flow field (fixed w). Pick a finite region A of space, and look for a subset
e e
Ae of A such that for all f, A dx > (1 - f A dx Le., Ae contains
E
. " "
it is not clear how a smallest member of a family of sets is to be pIcked.
f, e,
and furthermore, that D is independent of w. Then D is the dimension
of the essential support of the vorticity, or, more loosely, the dimension
of the set that contains all but a negligible fraction of the vorticity. The
'nte tin case occurs when D 3' this can ha en onl if R- 1 = 0 or in
the limit R- 1 ~ O.
,
ox(u 2 ) = O. For fixed initial data, this equation does not generate random
e aVlOr, ere ore we assume e a a are ran om a = : u x, -
g(x,w), g(x,w) smooth for each w. Each flow will develop shocks; if we
e e
define = ox u, will have the form ~(x,w) = ¢>(x,w) + EAj 8(x - Xj),
.. .
and the sum has a finite number of terms that do not vanish in a bounded
D =1= 3 lies elsewhere. Turbulent flow contains many scales, has large fluc-
. .,
and all the scales are strongly coupled. Similar situations are encountered
in the theory of "critical phenomena" (see Chapter 6), where the statistics
are characterized by "critical exponents" similar to 'Y. In that other con-
0< a < 3,
e on A then
each set of dimension D' < D = dim AE , having zero measure in dimension
, 2
,
for a > D. If F( a) has a different form, one has a "multifractal" vorticity
IS n u Ion. n IS case, su se s 0 0 varIOUS ImenslOns con am non-
trivial parts of the vorticity. For a multifractal vorticity distribution, a
conjecture that 'Y depends on dimension must be interpreted as stating
that T is a functional of F. 9 '
8 A.
Chorin, 1988.
For more references to multifractals, see, e.g., U. Frisch and G. Parisi, 1985,
C. Meneveau and K. Sreenivasan, 1987.
3.5. A First Discussion of Intermittency 65
.III LUn:~e ~ (
..
As a final remark, note that we have only considered inertial ranges
. . ilia
• n
llUW.
,... • II . k!
......
. ..
(1.UU
. . ar-
guments based on the idea that enstrophy cascades through an inertial
range in two space dimension have been offered, and we wait until the next
chapter to dismiss them.
•
ImenSlons
i = 1, ... 3N.
See, e.g., L. Landau and E. Lifshitz, Statistical Physics, Pergamon, 1980j D. Chan-
dler Introduction to Modern Statistical Physics, Oxford, 1987.
68 4. Equilibrium Flow in Spectral Variables and ...
the exclusion of its macroscopic motion, i.e., the motion of its container) ..
en nergy i e y
OULl'''U.LL''
for N large,
2 E
shows that for a perfect gas T is proportional to the energy per particle.
uppose we s ow y a partIc es to t IS ISO ate system, WIt out c anglllg
the energy. It is hard to imagine how one would do that in a classical
mechanical s stem but it will turn out below that this is exact! what
one does do in a three-dimensional vortex system. The result of such an
number.
One may well wonder how, and indeed whether, equilibrium is reached if
one starts away from equilibrium. Suppose we consider a collection of sim-
i s e i a small nei h-
borhood on H = E. If equilibrium is reached, these systems will move to
problem of showing that this indeed happens is difficult and on the whole
.
e ow in e con e 0 vo ex ynamlCS.
When the probability on an invariant portion of phase space is constant,
all microscopic states (those described by the 6N variables) are equally
. . n that n
be observed in physical space) is proportional to A, and S, the entropy, is
. ,
Consider a small subsystem of an isolated system in equilibrium at an
energy ; w a IS e pro a 11 y a esu sys enl as energy a, lI«
E? For the question to be meaningful, one must be able to assign an energy
to the subsystem, therefore its coupling to the rest of the full system must
. r
the whole can be overlooked. It is not always easy to quantify what the
last sentence must mean.
4.1. Statistical Equilibrium 71
erms,
(4.3)
Space
(4.4)
Set to zero all amplitiudes Uk with k > K max • One can heuristically justify
this truncation by assuming that one is trying to solve the ~orresponding
2 _ 2
x
viscosity decreases all the Fourier coefficients that correspond to k large
equations. Since u is real, U-k = uk' Assume that there is no mean flow,
Le., t e average 0 u over a perio is zero, an t us Uo = u x = .
One can readil check that the ener E:
E=
, ,
real, ao = f30 = 0, with U-k = Q!k -if3k' and E = :E(a~ +tf~). The solution
o can e represen e as a pain in a - imen i n p
where N is the integer proportional to K max , K max = 2{ N, where L is
the period of the flow.
(4.5) --+--=0.
8CXk dt 8(3k dt
smoothness properties, since they are described by the Euler and Navier-
Stokes equations. These properties are not taken into account in the trun-
cated system, an as a resu t, samp e ows WIt Uk = con~tant are very
unsmooth. Fluid flows also have a number of inte ral invariants kdx
in two space dimensions, circulation around arbitrary contours in inviscid
seen in the previous section that invariants playa major role in the theory
,
ing smoothness and invariance properties into account will produce spectra
3See e.g. P. Lax, Hyperbolic Conservation Laws and the Mathematical Theory -of
Shock Waves, SIAM, 1972.
4E. Hopf, 1952.
74 4. Equilibrium Flow in Spectral Variables and ...
w=
where the term in arentheses is the number of wa s of dividin N ob·ects
into M groups of nt, n2, etc.; it is assumed that ni » 1 for all i, M « N.
system.
with E = 'EniEi the total energy. For a vortex system, in which the
interactions are long range, this is a questionable assumption, and we shall
argument that led to the canonical ensemble and so that the conclusions
can e compare wi e canonica en em e. u ni i = ni =
We now wish to maximize W or S subject to these constraints; the
result will be the most probable partition and it should be a statistical
e uilibrium. For lar e n n! rv n: n Stirlin 's formula and thus
The
(E = (E)), assuming S
dS
same holds for the absolute equilibrium of Section 4.2 and in fact for most
,
can perfectly well imagine systems such that for E moderate there are.
76 4. Equilibrium Flow in Spectral Variables and ...
-
'd1- nat - ·-
-..- . - ± -ill- -dE +a/!]2
- -Itt >0
1 crt .
n
H l
, ,
Ei = (Ei) in earlier notations. The sum of the momenta EmiUi must be
zero 1 t e sys em as a woe oes no move In 1 S coor ma e sys em. e
1
entropy 8i in the i-th box is a function of E i only, and if d8i /dEi = T- < 0
the entropy increases when energy moves from the E i to the Ui . Thus a
system ~t T < 0 should be expected to have large scale motion even at
6
4.4. The Onsager Theory and the Joyce-Montgomery Equation 77
1,· .. ,XN
-N
1
(4.8) (Ec ) = --N(N - 1) dx'log Ix - x'i +B
on the average. Clearly, one can produce a larger E by bunching vortices
to ether and thus if S has no local maxima other than 4.6 T- 1 =
dSjd(E) < 0 for (E) > (Ec ). This is Onsager's observation. 8 If T > 0,
vv .
\ni!'" nIt!) \n 1 !· .. n M!7--n
,...,.
~ ne .. rJ
.
lIS u
. , ~- •
,.. . lUg
'FYF
vv t
, . .
Lne -c
. --ur
"
now be assumed that the vortex system is part of a larger isolated system,
with the remainder of the system acting as a "heat bath", Le., a source of
interactions that allow E to vary with 1 fixed, but do not change the area
available to the vortex system or the total vorticity. The maximization of
S leads to the equation
J G(x - x')e(x')dx', where Q(x) =- 2~ log Ixl + C. One can easily check
4.4. The Onsager Theory and the Joyce-Montgomery Equation 79
x.
N
2d= -=--~
(4.11) and (4.12) are vortex versions of the canonical distribution. One
. .
1, 2
the vortex Hamiltoniain involves on Xl, X2 as conjugate variables, and one
does not have to worry about the distribution of particles in boxes in a
four-dimensional osition momentum s ace ii the stream function and
the Hamiltonian are related, (iii) the vorticity is proportional to the one-
. .
has been used.
the stationary average density of the vorticity. Specific flows may depart
om IS average, u we expec e epar ure 0 e sma. e reasons
can be found in the references given earlier in this section; the theory we
have iven is a "mean field thea "with small fluctuations, as can be shown
to be appropriate.
solutiOns. In the latter case the solutions are non-unique; the solutions have
. ,
but smooth peale Equation (4.11) has a double peak when the entropy is
maximum and -161r N < < 0, one positive pe and one negative pea .
For < -81rN Le. "hotter" than T = -1 81rN the Jo ce-Mont orne
equation-with f2: 0' tIas"TIo ctasslcatsolutioIt mid-in fact does mtt 688et'ibe
,
with r = +1 in a bounded region V. 1O
The canonical distribution applies
to small systems as well as to large ones, and therefore applies here. Fix
one vortex, an conSI er POSSI e pOSItions 0 t e ot ere t e se ara-
tion of the vortices is r the ener of the air is E = --10 r; the Gibbs
factor exp(-E/T} is exp(+(,B/47T)logr) = r/3/41r ; the partition function is
41r = 2+ 411"
that when
. ITI
. = 00, the energy E is proportional to N2 for large N, and
.
p
4.5. The Continuum Limit and the Role of Invariants 81
/'
//
/ \
/ \
/ e
/
I
\
\
FIGURE 4.1. Scaled entropy and energy.
. S
hm N = seD)
N-HX>
.<> ~ .~ • ~. • ~
eXIst. 1 nus .J!j - IV -e, '" - IV s. II tne sI;rengtn or tne vonices IS 1 r 1,
r
E = N 2 2 e; we consider for simplicity the case fi = r > 0 for all i. One
expects s( e) to be the maximum of the entropy over all probability density
fundions which nroduce the Q'iven scaled enerQ'V e. All these exnectations
are fulfilled\!.:) s = s( e) can be calculated, in particular numerically, and
1 ~ 1. :_ n ' A 1
<AU u 0 .
The asymptotic slope on the right is -81L Note that e is growing when
1 IS, II one auows IOr tne pecuuar ract tnat 1 < U IS notter tnan 1 > U.
One can define a "scaled" temperature T by T- l = ~~, T = T(e). Then
e = e(r), s = set). The usual temperature T- 1 = ~~ = Nds/(N 2 r 2 de)
satisfies T = Nr 2 T; if one chooses Nand r so that Nf = (0 = constant,
• ., •• " .,. ,ry, f . ' ) " .,., ,.;; ,
KeepIng Lue vunex uen:sILy con:SLanL illS lV -+ (X) , LUeU.1. - ~c;.OllV).1. anu
ITI -+ 9as 1! -+ 00. The collapse of the Joyce-Montgomery equation occurs
when {3 = 'I ·1 = -81L Thus T is a function of N, but the physics remain
the same if T is constant. The sign of T is the sign of T and depends
on s( e) only; it is determined by the distribution in V of the normalized
•.
"
f(y)'
.. , f
r f(y)r1.y
JV'" ,
-_ e-P;j, 1 -iJ~
-~"p - -- = -;:;;-e ,
J'D e-{j'l/Jdx Z
setting y = eH 1 we obtain
4EAZ v'2Eq(1 Ae V2Eq )-2
11(0)
{3
I')
As expected, the solution exists as a smooth function only for {3 > -811". It
\. ~
~~u.:J a
.'-
,.:u~~ p
~~1
pvau. au
L L\.
u~~..,
.. .
~
solute" equilibrium and of the vortex equilibrium diverge from each other.
truncated.
e
of area h2 , and make constant on each. Consider only the states that can
eo ame rom a smg e s a e y permu mg e squares, no wo s ares
ever coming to rest on top of each other ("self-avoiding configurations", in
the language of Chapter 6 below). Clearly all the I k are the same in all
these confi urations. The 10 of the number of distinct confi urations for
a given E is the entropy. One can construct the whole theory in terms of
these permutations; the results are unchanged in their main features. Thus
1 ,
The canonical ensemble has been consistently used in this section. This
raises questions 0 princlp e. e SImp e erivatlon 0 t e canomca en-
semble from the microcanonical ensemble in Section 4.1 does not a 1 to
vortex systems, as can be seen by contrasting the treatment of the energy
in equilibrium with an external "heat bath" rather than with the remain-
er 0 an i 0 e v e , w v
temperature, h~w does it interact with the vortex system, and is its tem-
perature T or T? One could argue that only the microcanonical ensemble
makes sense for vortex systems, and thus one has to establish the equiva-
15
questions remain. One can encounter situations with T < 0 where the two
6
A comparison of this chapter with Section 3.1 shows that in two space
Imenslons t e unlVersa eqm I nUID pos u ate ere IS sImp y s atls lea
equilibrium. The independence of small and large scales reduces to the
statement that small scales are asymptotically insignificant.
14J.Miller, 1991.
See e.g. G. Eyink and H. pohn, 1992.
16M. Kiessling, private communication, 1992.
84 4. Equilibrium Flow in Spectral Variables and '"
periment (Figure 4.1). It has been suggested that certain large-scale vorti-
ca s rue ures seen III wo- ImenSlOna ows, or examp e, Upl er s rea
Red Spot, are examples of vortex equilibria in nature. 17
. Statistical equilibria are of interest only if they are reached from most
that is the sum of their individual energies. To make up for the loss of
(c
t =196 t =374
~ ~
.-, ,.,
tlme= 0.0 tlme= 2.0
I/) I/)
N N
I/)
..:
-
"H1t1 I/)
..:
......
PH
00
_ '
I/)
f
~
I f1\
~ CD
(I)'
_ 0
,• 1'l.
~
~ I \ I i.P'
~ ~ ~ ~
..... ~
-~
>-d• >-d•
~
~
yo
~
,~ ~
yo
~ :j
...:..... • i-I I H.
..
CII
~
--. -
CII
~
'*'
• •
111 111
I
~ to)
• -3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5
• -3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5
X axis X axis
111 I/)
(')
on
...
"""'"
..
.u
(')
UIIIQ- -
II.U
on
N N
on II!
...:
\,
yo Al"l'Pr I-i'fI'l,.
Prloo ~
~O'
~
.w to
>-0
on
~
~Qd
--
~
ta;,
~~
(I)
_0
)(
CD
'" on
>-0•
•
"A ~
~
:@>
:rJ~ ~~ ,
,
H. dJ,J'
• ill::
to
..: ~ II!
• iji;ii •
on
N
I
..on•
'
on on
(') C'i ..
-3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5 -3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5
X axis X axis
lI'I
..:
I/) lI'I
"
( I0
)' (I) •
_ _0
)(
~ «1
>-d I
>-ci I
lI'I lI'I
.,.: ..:
....
I I
V)
I
-3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5 -3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5
X axis X axis
I/)
(')r:-----------------, t'ir----------------,
I/)
I/) I/)
.,.: ..:
I/) lI'I
(I)' U) •
_ 0 _0
)( )(
WI/) ml/)
>-0I >-0 I
/ /
/
-3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5 -3.5 -2.5 -1.5 -0.5 0.5 1.5 2.5 3.5
X axis X axis
duced from the invariance of the energy and the enstrophy in spectral
arm: = cons an , = cons an . some energy
moves towards the large k's (small scales), then even more energy must
move towards the small k's (large scales). On the whole, there is an energy
u» ,
If the initial ~ is complicated, and has many maxima and minima, one
r , .n wine r y ir u r pa a 00
on small scales, then slowly migrate towards each other and consolidate if
.he curdles can never truly merge, since
the flow rna is one-to-one. At ~~~h-~ta·eaf This curd in t e nearT\r
circular patches are nearly independent, with whatever correlations their
one can readily imagine that arbitrary large-scale structures have "univer-
m -sca e ea ur s. ere, in wo i en in, n'v
grow to large scales, and an imposed shear or an imposed boundary mass
interferes wi~h them. It is readily imagined however that the curdling pro-
cess will simply stop when it ceases to be compatible with the conditions
One can wonder about the effect of a small viscosity von the processes
that has just been imagined is to couple weakly the notional fluid at the
to reduce the latter. If T < 0, the cooling of the vortex system brings
one c oser to t e = 00 eqm Istn utlOn so utlOn) In agreement WIt t e
intuitive idea that random pushes should interfere with the formation of
concentrMed¥Oftiees;~-Afteraloag QaGllgh time one~urould cDd .upwith
It has been argued that two-dimensional flow has inertial ranges with
"power lawn spectra of the form E(k) k-', with 3 < 'Y < 4. If the
('V
The main evidence for this kind of power law comes from numerical
ca cu atlons, w IC ave t eir own reasons or pro ucing power aws, an
from meteorolo ical observations which encom ass henomena more com-
plex than mere incompressible flow. The arguments for power laws in two
23
Pil 'tT
'" II i..... ll
T'.1npc::: n'1.1·....... 1I I
.. ,,..
.lll.1V~ oy
.,. . .. . AI,
where ( , ) denotes the ordinary inner product and W = ATA is a pos-
, W.llIU
'"
A2,
,
l:or- .1-~C:L.l,~ II-!.. .. A3 W.lU
responding normalized orthogonal eigenvectors el, e2, e3. The flow being
incompressible, the Jacobian of the Bow map is 1 (Section 1.1) and thus
3
oxO =L ox?ei'
,;. . 1
3
loxOl 2
= ~(OX~)2,
i=l
, '., I ~
= L{Ai)(OX?)2.
i=l
1..
with equality holding only if all the Ai = 1 with probability one-a trivial
..J
......... ,---, ,-
tI~9'OY rnl- ,. IIIA~tI2\\;~ •
'''u~, II:J .....
•
...,
except for trivial turbulence in which the derivative is zero. On the aver-
age, over tne prODaOU1ty space ana over au InitIal arrectlOns, nne lengtn
5.1. Vortex Lines Stretch 93
increases. Lines in some directions may well shrink while others lengthen;
i .
directions and
d
dt
or ex ines are s eci .nes, an cons 1 u e a neg Igl e ac ion 0 a
lines (there is one vortex direction at each point, but an infinite number of
others). All arguments that involve averages with respect to a probability
measure ma fa.il to hold in a ne Ii ible fraction of cases and thus one
cannot conclude from (5.1) that vortex lines stretch) even in isotropic flow.
. ,.. . .
The rate at which lines stretch is not estimated in this argument, and
" ,
decreases (in agreement with the "bathtub" picture).
e me J
. .e use 0 e er a
tions, except for the constraint of incompressibility. It is independent of the
direction 0 time: one st' gets stret ing' (Pt w is rep ac Y -t w. t
i~ the randomness of (Pt that is essential. At t = 0 fJxo is known, at a later
time fJxt. is less well known, and as a result IfJxt I, appropriately averaged,
. . . ~p
--
. t
~ ~ r
a lower bound on Z in terms of line length or line complexity, 3
rl c~ c~ •
AssumIng d1 ~l;. -) > U, we coutO use ~l;.) as an entropy tnot tne en-
, tropy), i.e., a monotonically non-decreasin~ function of time that is time
invariant only when the flow is statistically time invariant (= stationary).
~nl'.h "
..... " l"n" 1~ hp n~pn . ..1 nf thp trllP
..'"
• •• ;n "
space dimensions does not ~ave the relatively simple expressions that result
.c..~_ ... 1. ~ •• 1: •• _& ... 1._ .L ,. . , U _ _ :l4- __ : __
. . . v...... u....... .. V.l. u......." 'V 'V" "~A •
- -",-,.,- -.--.--. e
depends on the direction of in V2 • n is the winding number of V2 around
1· 1 - 02
if there are n unknotted filaments VI, V2 , ••• ,Vn , the quantities ri =]; u·
ds = L:j eli;""; are invariants, where Gi is the center-line of the j-th filament
and the elij are integers. The quantity ri""i '(no summation) is also an
. .. .
into smaller loops, and the same conclusion holds. In the presence of viscos-
1 y, owever sm ,vor ex amen scan reconnec ; near y coun erro a mg
vortices flatten and merge. The transformation of Figure 5.2 is then possi-
ble. The a roximate conservation of helicit in sli htl viscous flow uts
constraints on possible reconnection geometries.
If vortex filaments stretch the must also fold. 9 The reasons can be
seen on an example: suppose a vortex loop as in Figure 1.1 is expanded
.
arms is reduced from p to pi va, lei. in.the arms increases
. to aiel, and the
.
, . , ,
Ix - x/I -+ a/x - x/I, and thus Er(au, at) _aEI(U, l). Plck a = u: , then 1
1 , U = u- r u, ,or 1 u, = u u ,were u = q IS a
function of a single argument q whose precise form depends on eo.
Clearly,
d~~q) > 0 (lengthe~ng the tube with u fixed increases the energy). For q
m, is propor ion q pu i g w s w vor ic n 0
each other roughly doubles the velocity field and quadruples the energy).
For q large, E(q) q is an increasing function of q adding up the velocity
fields of two Ion vortices ut on to of each other does more than add u
their energies, but maybe no~ by much). An asymptotic analysis that we
. .
Consider now a closed vortex filament V of unit circulation and a small,
approximately circular cross-section of small but positive area. Suppose V
an e approxima e y covere y circ cy in e ,0
equal lengths l and of radii Ui, i = 1, ... ,N. The energy of the velocity
field associated with this filament is
98 5. Vortex Stretching
E - ..!-
811"
JJ
dx dx' e(x) . e(x')
Ix - x'i
,eex) .eex')
l?f - x'i
(5.2)
Let ti be a vector tying alongtlte axis of 1"" originating at the eenter «-4,
e
and pointing in the ·direction of in li. If Ii and I j are far from each other,
1 ti' tj
approximated by N' > N cylinders of length.e. The sum L:i Eii associated
. . . .
The double sum has now O(N ) entries, and has decreased. The typical
.. .. .
products ti ·tj must decrease,o Le., the filament must fold, or else the energy
of the vortex increases.
ge , rex , ,
can increase, and thus vortex stretching can act as an energy sink for
turbulent flow. The stretching and folding transfer energy to smaller sc es.
5.4. Fractalization and Capacity 99
In equation (5.2), the double sum is the "interaction energy" and the
su ii
.
was omitted when the two-dimensional vortex Hamiltonian was derived
from the energy of two-dimensional flow because it was constant, albeit
possibly infinite. Here it cannot be omitted without discussion because it
can var . The division of the ener into self-ener and interaction ener
is arbitrary, as it depends on the choice of cylinder length l; the sxnaller i,
. .
Suppose there is a natural length l in the problem (for example, suppose
. .
cross-sections of the tubes are comparable; the energy of the vortex can be
wnt en as
p. - t.
i j=ti
As vortex filaments stretch and fold, their axes converge to fractal sets.
It is immaterial for our purposes here whether this happens in a finite time,
so that the fractal limits are actually achieved, or in an infinite time, so
.
only the axes of the filaments. What happens to the cross-sections will be
Consider a bounded subset C of the plane, and consider all the vorticity
c Ions suc a _, x = ,supp I.e., ways 0
distributing a unit of vorticity among the points of C). For each consider e,
the energy of the resulting flow, including the self-energy of any small
ieces of . Given C is it ossible to make the ener in a finite vol
containing C finite? If the answer is yes, C is said to have positive capacity,
•. . . • • 10
,
A set consisting of a single point has zero capacity because a point vortex
has an infinite self-energy. A set consisting of a finite number of points also
as zero capacl y. se a as nl e area as POSI lve capac!. us a
that has positive Hausdorff dimension has positive capacity. The problem
. . . ~,
itive scalar capacity if it can support a unit charge with a finite electrical
. .
SI
5.5. Intermittency 101
its vorticity field, i.e., the transformation in Figure 5.3 is not allowed. If
any po 10n 0 e amen 18 8 re e so a i: COl apses on 0 a curve,
that curve carries vorticity. The associated eneft';Y is infinite. If the en-
ergy is conserved in the stretching, this kind of co; <;;.pse is forbidden. Thus
stretchin must be accom anied b a foldin thht revents the a ear-
ance of smooth vortex filaments of zero cross-section. Note that smooth
D = log 4/ log 3 > 1. The length of the resulting curve is infinite (earh
• • • • l' •
;;.:J
5.5. Intermittency
Consider a vortex filament that intersects a box of finite volume, and
ppose e amen s re c es an e ox moves so a; icon inues a e
crossed by the filament. The portion of the filament in the box lengthens
as it folds, or else the forbidden collapse to a curve will occur. One may
wonder whether folds uniformly spread in the box are sufficient to conserve
11 B. Mandelbrot, 1975, loco cit.
102 5. Vortex Stretching
energy, or whether the filament must fold into ever tighter folds, in which
. g a .n
...u .....u.JL1o.......
ing. Suppose some specific scale has been reached. Smaller scales will be
ce i a or i n 0 e e . in amen is s re
Q. To simplify matters, assume a is constant along
of the tube. H a -.- 1 the stretching has stopped. If a > 1, the next
stretchin will add another factor of a and the filament will fractalize as
in the examples of the preceding section. The radius (j of the filament will
decrease by a factor 1/.;a when the filament stretches by a .
.
j:¢i ij
for E· E ii , the self-energy. Compare E (j a, the interaction energy
of the part stretched by a factor a, with El = E1(q), its energy before
hi - . 2 ••
A. Chorin, 1988b.
-
5.5. Intermittency 103
~
~
~
r7
T
"-,.
_..-
--
T
FIGURE 5.5. A lattice vortex.
",.
.LV
. 1. ... 1. •
" ..UO ~ c
. ,.
~
. a.
.£.1. .__ -I.' . .1 ..
lattice, with bond length h, and a "vortex filament" that coincides with
a connectea sequence or Donas In tnat lattice. Toe quotatIon marKS WIll
be dropped in the future. As before, there is no claim that such a vortex
filament can be generated by the equations of motionj the lattice is a useful
device for estimating the effects of stretching and folding. to the extent
that they do not depend on the precise form of the equations of motion;
....;~""'"' .1 ....".,.'"
t'
.......
~" ...... ~h" ..
.......'"' . - -" .". ,..t
..
....
.c.
....
..........1
v.......
......
. --
n-o.
.L ne VVl 1l~A.,
•
i:LIlU
..
keeping track of possible singularities in the energy integral.
. , , ,.
line
•
. "
. \ lIue
II
VVl110A. leg~
....
} UIUlSL ue
... -
interaction energy on the whole crude lattice, as a result of the usual scaling
,
J.'
""', ,
•
,.,
~ Ut;
. " Vl
. ., ,. , , ,- ..
1
4 uy "
\11.1.':: VV.l u~A 11.1 \lUt;.l \J.! 0 JIill
is larger than 1/2· of the number of bonds occupied by the vortex in the
unrenned. whole and the selt energy ot eacn bonet nas grownj thus Its self-
energy (J.L' N)new > ~(J.L' N)old. The self-energy increases faster than the
interaction energy. H one refines the lattice repeatedly, the self-energ.:y will
increase without bound, and there is no way to balance it out by bending
I.~.x on t.hp. .
- In t,hi~
thp. Vl', 1"4+;",,. 80 a..q to .J its' ~.
E,
argument, changes in the shape of vortex cores and in how they fit into the
, t, 1- ,1
ua.V'V Uvu Uvvu
J. J. •
"u.uv
,So
.
The growth of the self-energy in the process of refining the description 1~ _
13 A. Chorin, 1986.
, .....
'U.
-.
r.n::t\;ll tllt ~1., ~:1ro.
5.5. Intermittency 105
at each refinement one keeps only a fraction {3 of the (1/8)th of box rather
-3 . .
available volume is 2D ' -3 .
n a emp as een m e 0 e ermine numenca y. e ra 10 0
the energy in a box on one scale to the energy in a box on anot~er scale is
determined by the Kolmogorov spectrum and will be estimated in Chapter
7' it re uires an elaborate anal sis. However at each level of refinement
one halves all the scales and one makes a step to the right in the graph of
.
microscope. The ratio Z(k)1 E(k) = k 2 is multiplied by 4. At each step
. 2 .
, ,
depend on D' and on hi (J, set the ratio of the new ratio to the old ratio
o t ese quantItIes equa to ,rna e an assumptIOn a ou (J an so ve
for D'. The problem is that one does not know how to chop off a finite
piece of the lattice without affecting the outcome. The surface of the box
• 2 • . i n e n vor i
dies out only as L -1, thus no truncation at a finite distance makes real
. . ,
and 3. Most people seem to believe D is not far from 3, and thus D' is not
ar rom . 0 e a even i measure 0 s .. .
e
may be 0 and thus supp may still be very sparse.
If the vorticity stretches into small volumes, one presumably obtains
106 5. Vortex Stretching
figuration near the point on the filament that is being considered. One can
.
5.4)
= I t.
o
It is plausiblethat'b(tt,w), b(t2,W), tl of: t2, are independent. By the central
• • lit.
varies wildly. One way to check this conclusion is to observe that if one
LUC'~"''''O'''''''~OI ,
leJ is large, one obtains a collection of unconnected tubular pieces. The
filament portions in between the highlighted portions are wide and have a
relatively small lei.
The 10 normalit of is com atible with the conclusions in the re-
ceding section. If, at each level of refinement, the vortices are stretched
• . -3.
,
stop stretching, one obtains 8 pieces of vortex, each of length (1- d), where .
.nl. . . n
2Dc - 3 < 1; one can imagine that as the vortex centerlines stretch, the
cr ss- ec ions . pre on 0 n ig r' u n c a
cross-sections with eventually unbounded eccentricity. This process is often
called "sheetificationn • Such cross-sections could then organize themselves
into coherent objects, like the fingers in two dimensions do, in a local
u' n •
One may conclude that the geometry of vortex filaments is, not unex-
pecte y, extreme y comp ca e; e r
in eneral interpreted in a highly generalized sense. H one views vortex
filaments as "coherent structures", their "coherence in t ee ImenslOns,
unlike that of vortex structures in two dimensions, is incomplete. Of course,
. eri en s smooth out
the small-scale structure, and thus look more coherent than is possible in a
se - 1
representation of what happens to a vortex tube has been given recently
(Figure 5.7 .'
~-'----""~ = constant
. ' 0
, I
This chapter contains an assortment of facts and tools needed in the anal-
i=l j=l
l r- __
G depends only on a finite number of spins; spins very far away do not
.
non-analytic behavior at Pc cannot happen. If ( = OOt m is described by an
infinite sum, and infinite sums of analytic functions can behave in various
odd ways. It fol~ows in particular that the non-analytic behavior of ~ near
T. can be observed in full detail onl when N = 00.
For T near Tc , the system can have large fluctuations, i.e. t there is a non-
. . .
. r non-trivially from their mean (E): if ( is large, the
•
can be seen from Tschebysheff's theorem, the departure of their sum from
t e mean may be ess t an we oun e. n ition, lone IVl es·t e
lane into re ions where l > 0 and their com lements where f. < 0, one sees
near Tc islands within islands of positive and negative spins, on all scales
down to the scale of the lattice, distributed in a self-similar way when the
onent
Consider again a cubic lattice in a d-dimensional space, and a connected
sequence 0 on s, WIt no SIte separating more t an two on s in t e
sequence (Figure 6.2). Such a sequence is a "self-avoiding walk", or SAW for
short. (The lattice vortices constructed in Section 5.3 were self-avoiding.)
, . ,
that can be obtained from each other by rigid rotation or translation as
being identical, there is, on a square lattice in two space dimensions, one
, , ,
but fewer than 33 = 27 four-step SAW's, since visits to earlier sites are
or 1 en.
6.2. Polymers and the Flory Exponent 117
• .
h
~
..- .,"
...
/ T...
i/
-.I
FIGURE 6.2. A self-avoiding walK.
~
A . 11 . ~
VL -...
1
- to'
1 . ~ '1'
v
,.. ..
J. Y -~"ICJ:I
rot A TT",
~~ YV ~
.
~
.l\.UUW 11 ~ a.
1
"}JULY u.u:a " ,
because of its uses in theoretical chemistry. Let TN be the ~nd-to-end
straight-line distance between the beginning and the end of a SAW; we
clrom that
(6.1) TN
- 21
= {TN} "J NJM for large N ,
.,
"" u ..v~ '<;i fA'
.
l~
1 ~,
"UIC J.' IUI:1 -r •
',..T
\!'lUl>1C
1
"lftL"
. . 1 .
WIC It: Ul:SIUl:) 1I111C It::''''t::l fL
1
UVlJJ..l
.1.
for chemical potential and for Flory exponent. This usage is common, and
the mearung of p, should be apparent from the context:} TIle caTcuTatIOn
of J.L is related to problems in spin statistics,2 but we shall not exploit this
relation here. It should be clear that p, > p, = 4is the Brownian walk !;
1••" (~
...n • ::un, A.nn t.hp. • nr' AAlf- • .. • .. • {'
the number of folds, straighten the walk and increase p,. p, is a critical
__ ...3 :_ .. ,. . _1 __ =_ L 1.. _ l:_:L I\r --
-~
,
.1
Q.l............ ~ ~ .... -., VJ..l"J .loU OJ.l"IJ .u.uu'" ~ ~ ~,
logrN .
J1. = J~nn Inu
.... l\T
Flory gave an argumentS from which one can deduce p, = 3/(d + 2),
where d is the dimension of the space. Flory's argument makes, as far
as one can see no sense whatsoever" it is a strikinu case of rhrht answers
derived by wrong arguments. For d = 1, the only SAW is the line itself and
.
,..
1 "[;I" ...
. . . . . . . ........
.J
IN
t') ......
..... ........
'... _1 :-
J ................., "'u ,...
3 . . . _...l:......,,11
4' ........ '-&.-..u. ....
.... . . .11u
-0/ •
1/JL ~ 1.70; more crudely, D 1.66 = 5/3 if JL = 3/5, the Flory value.
f'oJ
phisms of the cubic lattice, Le., the set of transformations that map the
. .
anal matrix with integer entries. An orthogonal matrix has orthonormal
columns, and thus has only one non-zero entry in each column and on each
line, the non-zero entry being +1 or -1. There are 48 such matrices in
three-dimensional s ace. "
Pick a SAW of N steps. Pick one of its ends and label it 0 I the "floating'
n •
,
,site to be picked; rotate the piece of the SAW between 0 and P by one
,
. the new SAW i8 self 8JleiQiaS1 accept ij;;.jfjtj§ pQt§§lf-a;~r()i4iIlg, id~ntifl
the new con guration wit teo con guration. is otatlOn e nes
the new SAW in the Markov chain of SAW's. It is eas to check that the
sequence is ergodic (any SAW has a chance to be folded into any other
SAW) and satisfies the detailed balance condition for polymers (SAW'S
nent f.l changes value. We shall never have to worry about this situation in
. . .
are vectors. Pick a lattice site on the polymer labeled I, and let 1"1 be the
vector along the polymer issuing from I (as in Section 5.4). Consider the
sum
Sr= L rI·rJ
iI-J!$r
I.e., consi er a v J ,
add +1 if r I and r J are parallel, -1 if they are anti-parallel, 0 otherwise.
Sr satisfies the relation
and spectrum, and then the energy spectrum. Indeed, following the steps
. . . .. . ,
we find that there are fV positive co~tributions t_o Sr between rand
r D- 1
e
Note that when supp is topologically one-dimensional, i.e., smoothly
e orma e In 0 a curve, 'Y an im supp 0 no ways grow 0-
gether, unlike what happens in the scalar case and unlike the general trend
outlined with caveats in Chapter 3. Indeed, compare b = 2" = 0 for
9 A. Chorin and J. Akao, 1991.
R. Zeitak, 1991.
6.4. Percolation 121
"-"
1\,-
.. 1
~
"~ l
I'
-/
if
~
~V
/'
FIGURE 6.3. Estimation of D.
Brownian walks, D = 5/3, I 0.37 for polymers. The inequality D < 1 f'V
can support a finite vorticity as defined in Section 5.4 so that Ju 2dx < +00,
~
{;(j,l1 l:tl:::iU "rc
,.. ..... ~ <;~/~x, W),lIIHUt
L lOr eal:Il w a vOHIcny nelU <;
. .1M . . "y ..
translation invariant and has (u2 ) finite. This assertion is plausible, and
does not contradict the claim in Section 5.4: It was claImed there that gIven
£ > 0, one can find a set of dimension 1 + E: that has positive vector capac-
itv. not that every set of dimension 1 + £ has Dositive vector caoacitv. One
conclusion from this argument about jj is that one cannot have any part
.
,,; "" ........ ~ 4- hn
..... ,,; +'1-. ....
....... ..
, ....... UJ'
1-...... ~
....
1 . \11 ... _ _ l' . _1
- ,
~
'r~ ~, -rr ~
"J: '-' -"
...... T"O>
0.'*, rerCOlatlOn
.
,
T'" 1 . Lt.
lI"U~LI.L
11'.li:) l'. •
l.l:t
~ ". .11 .:vv 1. ~
,...
.
cn::1I" ~--r
~ .- -,
-"V "UC VoL U.lC v
" " ,
sider the squares ("plaquettes") defined by its bonds (Figure 6.4). Go from
.p aque e 0 p aque e, an In e one, cover e p aque e WI a 1
with probability P, or leave it white with probability 1 - P, with what is
done in one plaquette being independent of what had been done in the pre-
vious ones. Black la uettes that have a common bond in their boundar
are neighbors; black plaquettes that touch each other at one point are not
.' .
'other through black ones that are neighbors form a "percolation cluster".
. .
.problem, in which the sites or the lattice are either occupied (probabil-
.. ..
assumed to be connected, and the question is whether there exists an infi-
nl e connee e c us er. on perco a Ion IS 1 eren : assume e on In
the square lattice are either conducting (probability p) or not [proba.bility
(1-p)]. Is there an infinite conducting cluster? Yes, ifp > Pc; no, up < Pc,
with = 1 2 for the s uare lattice. A bond ercolation roblem can be
reduced to a site percolation problem on the "dual" lattice, constructed as
,
line if the corresponding bonds touch. A bond on a square lattice touches
. , . . .
sites. The lattice dual to a square lattice is drawn in Figure 6.5. 13
On this dual lattice Pc = 1/2; note that by adding links we have made it
easier to have an infinite connected cluster and reduced Pc. The plaquette
v r i
coordinates (i, j) in the middle of each plaquette [thus the lattice sites are
at (i±l,j±l)]. If (i+j) is even, blackplaquettesat (i,j),(i+l,j+l), and
at (i,j) and (i - l,j' - 1) (to the northeast and southwest) are viewed as
nnected while black 1 uettes to the northwest and southeast are not.
If (i + j) is odd, the conv~rse holds: (i, j) is connected to the northwest
,
more elaborate structure that will be discussed in the context of vortex
dynamics.
Consider SAW's in the plane, open i.e., wit dangling end points or
closed Le. formin closed 100 s that divide the lane into an interior'
and an exterior). Assume that the bonds in the SAW attract each other,
. .. .
only. The easiest way to incorporate this attraction into the statistics is
o eae
. . . ,
the number of close encounters in the SAW; this probability replaces the
equal probabilities of t e preceding non-interactmg po ymer pro em. e
fractal dimension of the resultin 01 mers is independent of whether the
are closed or open.
Tc such that for T > Tc the effect is negligible; for T < Tc the effect is
ca as r p i - p o y . . n
2, the maximum possible for a planar lattice. We now show that at T = Tc
there is an intermediate state, whose statistics are t ose 0 the u loa
ercolation cluster at the ercolation threshold and in articular D = 7 4.
We shall take for granted that only three states are possible:the unaI-
state. If we find a state with 4/3 < D < 2 it must be the intermediate
. ..
onallattice, relying on "universality" to carry the conclusions over to other
a Ices.
One cannot in general construct a SAW of N steps on a two-dimensional
lattice b sim I walkin at random and avoidin reviousl occu ied sites
has a boundary bond that has already been visited, t~en there may be
ew ~ ~'Oi in a ap an e. .n s y .
n
An N-step walk has then. probability Z-l . 2 , where Z-l = 2- is a N
in universality, and thus the walk generates the intermediate state of the
polymer. That was the claim. The probability of a black hexagon is 1/2,
. i cri ic r ii x .
intermediate polymeric state traces out cluster hulls at the percolation
t eshold.
126 6. Polymers, Percolation, Renormalization
6.6. Renormalization
At its simplest, renormalization is a technique for lumping together vari-
ables, thus reducing their number, in such a way that the statistical be-
. .. .
, , ,
. . .. Perform first the summation over every second spin, say, £2,£4, £S.
D , W
neighbors to the right and left. After summation over £2, £4,£S, ... , one
obtains
z= + ....
Suppose one can find a function F(f3) and a new temperature /3 s.uch that
(6.4)
6.5
(6.4) can be satisfied, plug into it the values of £1, la, obtaining
2cosh2{3 - F(/3)e~
n
~
. T.'II n\ -8 ,
.I." ''-')~
and hence,
The expressions for (E), S, etc., involve log Z, not Z standing alone. Let
4> = N-1log Z. The recursion relation (6.5) reads, in terms of ¢:
Equations (6.6), (6.7), (6.8) are the "renormalization group equations" for
the system. They express log Z for a system of N /2 particles at an inverse
temnerature B in terms of IOl! Z for N narticles at B. The number of
particles has been reduced, and the system has been changed, but the
. . .1 .
u .. ~
........~ +l.....
~ u ...
~
• • ,.,.1
~J ... -~........- ~ _~u --~- _-v.
1-"n............ "fo "h"""""'...... ..1".,,+ ~, .
c C ~ o.
(6.6) expresses the "parameter flow", Le., the way the parameter f3 evolves
(Il)
,1
- , . . . . . .
or equatIons 0.4} cannot De S01vea exactlY, ana one proauces onlY an
~
.
approximate renormalization group.
Each renormalization step halves the system, and thus halves the corre-."
lation lemrth ( 1. two snios have been renlaced bv one (see Section
6.1). At a critical point ( = 00, and therefore Pc = llTc should be a.
l:,. _.1 . .& ~1. ,. . t' . ".,\...:- :_. ~ 'I
.UJ>.\JU .P'vu~u Vol. u ..~v O' -r • ........& g ~'"
the case if things are done right. A critical point is an unstable point of
the renormalization "flow", because if T =f: Tc, ( is finite, and successive
halVlngs will eventually make It small. We shall present a renormalizatlOD
transformation for a system with a critical point in the next section, where
the system is a vortex system in the plane.
128 6. Polymers, Percolation, Renormalization
i j:¢i
where the Xi are the locations of the vortices, ri - ±1, the logarithm
-1
prime on J.L ;have been omitted as irrelevant. We shall have occasion to think
of a vortex; for a sparse system, p. ,...., g~; 211- can be thought of as the
consider a system in which the number of positive r's equals the number
of negative r's and Eri = O. The partition function for this system is
z=
, 1
to write e-{3/-L = K and
Suppose that all pairs of vortices within a distance). of each other make
wish to delete all pairs of vortices with separ~tion between ). and >. + A).
, . .,
the problem so that it has fewer small-scale structures but its statistics
are unc ange. e imi ourse yes 0 e si ua ion were er ar ew
vortices permitted in the given area; p. has to be large. The charge from x
to ). +.6.), removes small scales of motion from the system. The difference
between the previous spin renormalization;and the present renormalization
i n f i; i . he renormalization will
change 11-, and p. controls N. '
XN-l" .
where Dk is V, the domain to which the vortices have been confined, from
which the circles Ix - Xj I < A, j = k + 1, k + 2, , N, have been excised.
. .
We
om
130 6. Polymers, Percolation, Renormalization
sign contribute much more to Z than pairs of the same sign; it is reasonable
t, J t t J
signs are opposite: ri = - r;. This assumption holds better for T < Tc
If the vortex "gas" is sparse, then the probability that there is a second
. i , ; - k
is small. Expand the integrand of the last integral in powers of >..2 fIx; -xkI 2 ,
omIt erms , mtegrate over ,an Integrate over j jot e same or
all the terms in HI, and sum. The result is
k 14k
6.7. The Kosterlitz-Thouless Transition 131
log so as to make all the logs non-negative, and changing A into >.. + ll>",
we n
-2 -2
where r K,\ are the new "renormalized" values of these uantities and
-2-2
Zo = exp(21rK2>..ll>"/1'I). {3r, K>.. are given by
-K>..
(The last term, K>..2 ·2 AoX comes from the variation of the factor >..2 in
-2 2
(6.10)
dX
dy dx
dq
- dq -
132 6. Polymers, Percolation, Renormalization
One can readily see that these equations leave invariant the half-line y ==
1f ,y_, .., " ,y
start near it, they. will eventually get away. This degeneracy is connected
with the indefinite starting value A in the renormalization. Note how-
ever the important fact that as A increases, the "effective" value of (3r2
. . ed r 2 decreases. Since 0 low-en r
•
figurations are favored; given a vortex pair with, large separation, other,
. . . ,
creasing the "effective" interaction of the vortices in the large pair. IT small
palfS are remove ,one as m y re ueing .
~.uooe cQm.patesthis conclusion 'Veith the discussioa ill SootieR 3.1, 8ftee8es
that for T > 0 the ~'large scales" pairs with large separation) and "small
scales" airs with small se arations are not statisticall inde endent: the
former rearrange the latter, the latter attenuate the former.
/ ./
,""'" Jt1
A
V ,
--"'.'.~'-'-. --,,--- _.
..-. • .-
, .• ~
I,
+: I •
•
.. •
•
--"'
.... IHI<; fl R Turn '. .. lnnnQ
v ~ ~
events in different Darts of a lattice are not indenendent. Assume the IOODS
Hvp nn ~_ 1
-..1.' . .1 ,.
nl
1 .
Place, or fail to place, on each square facet (= "plaquette") of a unit
.1-
... u. LJI;;; (l<U.
1
J YVAl1t;;iA
1
J,vv'p
IT:':
\~-A5'-U-.1;;;
~.,.\
V.f/ •
m1.
.LI.I.I;;;At;;i CoW.t;;i
.
0 ....... lJ.J.Pt;;iO
~£
VI.
.1
t;;iJ,v"
mentary loops: in a plane parallel to the (Xl, X2) plane, in a plane parallel
to the tX2, X3) plane, In a plane parallel to the ~Xl, X3) plane, and In each
case, oriented in one of two ways. Two adjoining elementary loops with
the same orientation have a bond that cancels, and the result is a larger,
". • " L~A. loon that
VUJ. out the of the union of 'L
.'I
y
18
macroscopic vortex filaments that trace out the boundaries of percolation
• • • . ;n ~ •
ClUSterS or vortex lOOpS tnat llve on tne races 01 tne CUOlC lattIce. HIS
plausible (but not well verified) that at the critical temperature of the
lattice loop problem there appear vortex loops of the type just described
_1 ..
related to that of polymers. This suggestion is very attractive, both because
. .... ,.
.1 ,. .. 00 ...... ;1 ............ ;+0 • . .
.
.
I" ...
UL lIU" _~"
L_ L1..
lIV u."'- v
L
_ - -"
In particular' since -viscosity
-~nn
at a boundary
, . . - . -;a- . ---, --_.- "-..-'-.-' --
slows down fluid by
.... _-.__...-_.._..- ..._- -, - ,-... - ...... _- ... - " " - ' . ,
creating
• ..,.00
are discussed.
the full range of interactions in a vortical flow. As was shown in the two-
ImenSlon case, a e e emen s 0 v r ici y e in ra r n y.
Here, only the pieces of vorticity that belong to a single filament interact
strongly while different filaments interact weakly, acting as a heat bath
for each other.. The features of the roblem that miti ate this flaw are:
the interaction in three space dimensions has a shorter range than in two
. .. . . . .. .
the number of imposed invariants) shows that imperfect models may give
. , , . .
one to gauge the validity of the conclusions in the non-sparse case will be
escn e ; It rrns t e cone uSlOns.
II. The model fails to represent the extraordinary complexity of vortex
cross-sections. Some of that com lexit will have to be reintroduced at a
later stage, in the calculation of inertial exponents. This initial neglect of
. ,
On the plus side, the model will afford a relatively simple analysis and
. ..
dynamics, in particular the close association of stretching and folding, are
preserve In a natura way.
Note that if T = 00, P(C) is a constant independent of C, and the fila-
ment is a polymer (equal-probability SAW). When ITI = 00, dE = T- 1 = 0
e i see that i de reasin id f
ITI = 00 point, and therefore S has a maximum at the ITI = 00 "poly-
meric" point. The vortex equilibrium of maximum entropy is a polymer.
7.2. Self-Avoiding Filaments of Finite Length 137
N=301
ean energy as a
,
constructing an appropriate Markov chain of filament configurations. The
. v iy , ways, en r r 0
when no ambiguity arises) is an increasing function of the temperature.
T e variation 0 E with T is plotted in Figure 7.1 for convenience, r =
Irl = 1). The values of /3 ~ T-l are arranged so that the temperature
increases as one mov he ri h see ha ter 4 .
The mean energy of the filament incre~es with N (Figure 7.2).
.
2 A.Chorin, 1991.
H. Meirovitc , 1984; A. Chorln and J. Akao, 1991.
138 7. Vortex Equilibria in Three-Dimensional Space
40~
/T=-1
.......
~
/
20~
/
'In ~ - ... _~
~ •
- T=1
. N
JL2(N, T, r) is well defined for r small enough so that the vortex has a small
probability of ending within the sphere II - JI < r. A good estimate of the
.1-.10 r il:: Ii'~V~
~
hv trt\1/2
v
;;: AT· Wp. 4-'
.,
JO[)fP rlennp
~ ~
, ~- ,.....~ 10~rN
f'2,N f'2V V ,.l ) ,
logSN
where rN has just been defined and SN = SrN. IflimN_ooJ.tl,N exists and
eoual u, then u, is an analotme of the Florv exnonent and D, = 1/ u, is
the fractal dimension of the filament. Similarly, if JL2,N ~ [L, the energy
fJ
spectrum is E(k) k- , D = 1/[L. JLl,N, J.t2,N as functions of /3 = liT
f"J
• A'" _ • ... _ . . •
ana or IV are r In J:'.o (.4 ana / .D.
I, tW51
N
• ••
2 1 o
, ,
1/J.Ll,N decreases as T increases (remembering that T·< 0 is "hotter" than
> IS can e un ers 00 m e amewor 0 e wo- ImenSlona
theory of Chapter 4 and of the Kosterlitz-Thouless theory of Chapter 6:
Cut the filament b a lane' at the oints of intersection the direction of
the vortex alternates; the set of intersection points looks like a collection
opposing sign cannot increase and lead to segregation into vortex patches
. , , .
dimensional object. An increase in T can only express itself as an increase
In e separatIOn r etwoon pomt vo ICes 0 opposmg Sign, an ere ore
is an unfolding of the vortex and a decrease in D N • Furthermore, if the
energy of the vortex is fixed, an increase in N must bring a decrease in T
.
it must fold.
,
sake of definiteness, f = 1). Contrast two vortex filaments with the same
finite N but different circulations f l , r 2 , say r 1 > r 2 • The energy integral
emg propor Ion 0 , e 1 s welg s a ac 0 e wo amen s
are z-l exp{ -(3riE), Z-l exp( -,Br~E), where E is the energy that results
from r = 1. These weights are the same as those one would obtain. with
140 7. Vortex Equilibria in Three-Dimensional Space
__J.._ P=-1
.8 ..
------
~
..::~••__...- -...,...._.......--...,j•..--.... p.-o.4
.8 . _ p=o.
~ ."._.
Cl.l\A
- .-'-,.,._,..- ---.
.5 I-
- - _ ~1
.4 ...
N
•
. -- --- --- -_..
I
• • I I I
--
~u IUU --
I~U 'UU - .-
il::~U ~ v;N
only physically relevant case), then the vortex with larger r has a smaller
T'\..1 ,t C1, • 1 Po.t.tt ,1.' 1. 1..
.L/ I ,N allU -r c • .. u y v ... ""''''vO """ >;. ">;'00 , 0 ,ua.o U~!!
• • ••• 1;...... • .. • to •
.
rZ,N
/
T=+ 2.5
IR-.&. A\
~.
3 ...
-_______---=.:---_'""':....._ ...... L.-_....I. '_ T=zoo
(~=O)
• •
filament is smooth, and when the limit is 1/3 the filament is as balled-up
as tne space avallaDle WOUld allow.
Similarly,
(,
·
11m J,l2N = ~I r-II.- ITI=oo
1\i...........
T>O,
. .... . . ..
wnere tne nmn;s J. ana 00 correspona to toe upper ana lower Dounas on
.
iJ = 1/{J.: 0 < iJ < 1 (see Section 6.3). In the same section, we gave the
D = dim supp
'nTA"lrl .1... • n
e= 1, and 1 = 1 [E(k) ~ k-"'( for large k]; for T > 0, we
rl;TV> ~"T\n ~ ~ '" 0 ~nrl f"... I~I
n I'V'\ I"V 1 70
- r r ,. ~".' .. I ' .,
this inte al is the continuum version of the sum used to define the vector-
vector exponent for a polymer. Suppose the tube is thin, has centerline C
,..., r D - Dc +D, and following the usual analysis, E(k) ,..., k-"1, with
+ 0.37 = 1.67 ~ 5 3.
e 0 mogorov exponen as een recovere In t e = 00 case.
Here are some of the things that are unsatisfactory about this argument:
(i) The main problem is that the correction for the presence of a
I!.0n-trivial cross-section is ad hoc. We obtained Dc ~ 1.70 and
i' r i n.
treme variability of a vortex cross-section makes plausible the idea
t at cross-sectIOns can e tac ed on as an afterthought, but the
7.3. The Limit N - 00 and the Kolmogorov Exponent 143
Thus a finite) non-fractal cross-section ~imply chops off the spectrum and
does not roduce a self-similar s ectrum.
A positive aspect of the calculation is the fact that numerical calcula-
6 -.
what one usually obtains at the upper critical dimension, and , = 5/3 is
w a one ge s a er an in e i e ','
E= -
81r
2:2:rI .rJ/II - JI
becomes ineffective.
The easiest way to deal with this limit is to reintroduce a chemical poten-
1
when the filament with finite fixed N was discussed. One can cover the fil-
, ,
of the bonds within a ball to a chemical potential term. The interaction
energy IS now mte, an remaIns nlte as -. 00 W 1 e e r 11 0 t e
balls remain bounded from below. This is a simplified way to dealing with
the non-trivial cross-sections of the vortex filament. The icture we are
presenting is self-consistent if for every finite ball radius, the limit N -+ 00
that result from this vortex equilibrium are not Gaussian; indeed the Biot-
.
nents j1., ji. are kept at their equilibrium value by reconnection. Thus the
. .
7 A. Chorin, 1990.
8 A. Charin and J. Akao 1991.
9 A. Chorin, 1991; A. Bershadski and A. Tsinober, 1991.
7.4. Dynamics of a Vortex Filament: Viscosity and Reconnection 141
Ea:5.0
&10.0
energy spectrum E(k) "J k- 5/ 3 • For T > 0 the slope would be flatter than
-5 3 ,
Consider for a moment closed rather than open vortex loops. If one
rep aces a vor ex amen 'Y e equlva en co ec Ion 0 e emen ary u ~
tke) loops that it spans (see Section 6.8), then the physical vortex is the
boundary of clusters of elementary loops. The shedding of a closed vortex
100 b a iven 100 is a decrease in the size of the enclosed cluster ~ i
picture is consistent with the idea that at ITI = 00 the size of the cluster is
. , " . .
coincide with the hulls of clusters of elementary Buttke vortex loops at a
. , . w r e - i ion in ec i n
6.8; the ITI = 00 state is the intermediate state between the smooth vor-
tices at T < 0 and the very folded vortices at T > o. An analogous identity
between an intermediate state for folding polymers and percolation cluster
reconnections should destroy all the large loops; the constraints imposed by
. ..
n.
vortex loop does not polarize smaller ones (see Sections 6.7 and 6.8), nor
. ." . . .
USee, e.g., U. Frisch and G. Parisi, 1985; note that these authors label the Kol-
mogorov range as not being an equilibrium.
7.5. Relation to the .A Tra.nsition in Superfluids: Denser Suspensions of Vortices 149
empty, each decision being independent of the previous ones. Trace out
e res tlng macroscopIC vor ex oops, W lC are e Dun anes 0 c us-
ters of occupied plaquettes. Note that if an empty pI uette is surrounded
by occupied plaquettes the resulting macroscopic vortex is oriented anti-
. .
the macroscopic ones can be oriented either way.
Jl.
II I
------.. ""- I I I I
~
1 0 -1 -2 fJ
~
III IV
w "tIle pl
. _ _ .. .
. a . . ...nne . line 01
.. ....... •
see Figure 7.8. The fact that the critical states lie along curves is analogous
. III
• 'P I '
n. I
renormalization.
The fractal dimenslOn of the long macroscoplc vortIces remams 7/4 along
both curves. 15 The percolation curves divide the ({3, JL) plane into .four
regions, I, II, III, IV. The area far down where J.L < 0 is presumably un-
, . .1. . T ..1~ to~. ~~l. flow. TT to ,. , ~lv flow~ ITT CR.n
- • .
. ...... 0 ........ '-h"t:>
A .,
'"' A
h .... ,
..,- •~ ...... -
• ~.H"h '"
A
..0 '.-1
• ~ ••-
,.>03 .....
- -• • _ •• - A
'I
a
- A
14 A. Chorin, 1992.
15H. Saleur and B. Duplantier} 1985; A. Chodn, 199~.
16 8. Shenoy, 1989.
152 1. Vortex Equilibria in Three·Dimensional Space
,
dimensional equations that we have not written out. These equat50ns do
,
correct renormalization is clear: ITI = 00, J1- = constant. Indeed, it has
e y een pomte out t at w en = 00, a arge vortex oop neit er
polarizes nor antipolarizes smaller loops, and thus the latter can be removed
without affecting the former.
l Jl
~ • ~ ~
~ '+-/
.1
___ _ ,
.I. 1 . J . .J.
u......a" ... a"v ..e
1.
..cw.o'" ........ - .-
_ ~J. l..
,," .. u ....
_.1.._.1.
", .......u.. ..
"L~ ~l..
..,,, "'..,.
served at the same viscosity but in a very smooth flow.
. . .
~lll) lVlucn or 'tne smaU-SCale VOrtlCal Structures navmg Deen removeu,
.
what remains is a collection of large vortical structures, usually
tubular, that are the raw materials for further modeling on scales
other than inertial scales (Fimre 7.10).
(iv) The effect of vortex removal on small scales is only partially dif-
,. '.
'J
.1.
.n01"ol101
-
• Q1"O n , . . + ' ;1"1+1"0 O!U·.'h
-
' £:Inri
(v) The removal of small vortex loops can greatly affect the appear-
ance of large-scale vortex structures, sharpemng and tightemng the
structures that remain.
.schemes
. proposed elsewhere to the theory presented
I ~ I
•
ue ..
here.
. ~ .~
Ol,;btaUC UCOl,;llUIljU llen~
• I ~ ~
lUloU
The renormal-
~
ltue
~ ..
01
•
non-universal scales above Amax (broader claims have been made on be-
half of other renormalization strategies 19 ). It would seem that the large
scales remain the domain of the inventive modeler and the computational
scientist.
The irony in the theory of the inertial range is that it leads, in practice,
,. , . . Th;o
_.- -;0
ion t'Y'Oof l.,,,A,,,
-~
+.. - -
h .. f n
.~ .-
;f fro. h.n
~-
"
.,
.. ....
hor\t'Y'O • .1
of course a success, because the removal reduces the range of scales that
- ._.-
, • , . . , , • . OUl,;U , . . . 1
~
,
pp. 46-53, (January 1991).
, en an . ang, ores re rno e
on turbulence", Phys. Lett. 147, 297-300 (1990).
P. Bak, C. Tang and K. Wiesenfeld, "Self-organized criticality", Phys.
Rev. A 8 64- 4 1
G. Batchelor, The Theory of Homogeneous Turbulence (Cambridge Uni-
.LIlJCL.H:;; ,
61-66 (1984).
,
dimensions", Math. Compo 39, 1-27 (1982).
ea e an j a, or x 0
in two and three space dimensions", Math. Compo 32, 29-52 (1982).
J. T. Beale and A. Majda, "High-order accurate vortex methods with
ex licit'velocit kernels" J. Com. Ph S. 58 188-208 1985.
J. Bell and D. Marcus, "Vorticity intensification and the transition to
turbulence in the three-dimensional Euler equations", Comm. Math. Phys.
,
G. Benfatto, P. Picco and M. Pulvirenti, "On the invariant measures for
t e two-dimension Euler ow", J. tat. ys. 46, 729-7 2
P. Bernard J. Thomas and R. Handler "Vortex d amics in near wall
turbulence", Proceedings International Conference on Near- Wall Thrbu-
, ,
P. Bernard, J. Thomas and R. Handler, "Vortex dynamics and the pro-
uc ion 0 eyno s s ress, . ui ec anics, P l'
A.G. Bershadski, "Large-scale fractal structure in laboratory turbulence,
astrophysics, and the ocean", Usp. Fiz. Nauk. 160, 189-194 [Sov. Phys.
Usp. 33, 1073-1074 (1990)].
,
properties of turbulent dissipative fields", Phys. Lett. A 165, 37-40 (1992) .
.
uctlOn
Appl. Math.
u e an orin, r u ence ca cu a ions In magne iza Ion
variables", to appear in Proceedings 1992 Israel/U.S. Symposium on Sci-
entific Computing.
R. Caflisch Mathematical Anal sis 0 SIAM Philadel-
phia, 1988).
1-53 (1990).
", ,
A. J. Chorin, "The evolution of a turbulent vortex", Gomm. Math.
ys.
A. J. Chorin, "Thrbulence and vortex stretching on a lattice", Comm.
Pure Appl. Math. 39 (special issue), S47-S65 (1986).
. " .
Motion, Lax 60th birthday volume, A. Chorin and A. Majda, editors
Phys.
, ,
A. J. Chorin, "A vortex model with turbulent and superfluid percola-
lon, . a . ys. ,
A. J. Chorin and J. Akao, "Vortex equilibria in turbulence theory and
quantum analogues", Physica D 52, 403-414 (1991).
A. J. Chorin and P. Bernard, "Discretization of a vortex sheet with an
", . ,
A. J. Chorin and J. Marsden, A Mathematical Introduction to Fluid
ec anics pflnger, ew or,
W. J. Cocke, "Thrbulent hydrodynamic line stretching: Consequences
pf isotropy", Phys. Fluids 12, 2488-2492 (1969) .
..
Bibliography 161
,
mations to the Navier-Stokes equations", Math. Compo 56, 45-60 (1991).
n e y, uan
Press, 1991).
D. Dritschel, UContour surgery", J. Camp. Phys. 77, 240-266 1988a.
D. Dritschel "The re eated filamentation of two-dimensional vorticit
interfaces", J. Fluid Mech. 194,511-547 (1988b).
1992.
,
termittent fully developed turbulence" , J. Fluid Mech. 87, 719-736 (1978).
U. Frisch and G. Parisi, "Fully developed turbulence and intermittency",
in Turbulence and Predictability in Geophysical Fluid Dynamics, M. Ghil
,
I. I. Gikhman and A. V. Skorokhod, Introduction to the Theory of Ran-
om rocesses aun ers, 9 9 .
J. Goodman "The conver ence of random vortex methods" Comm.
Pure Appl. Math. 40, 189-220 (1987).
. .
omp.
C. Green ard and E. Thomann "Sin lar vortex s stems and weak
solutions of the Euler equations", Phys. Fluids 31, 2810-2812 (1988).
. .
o.. H. Hald, "Convergence of vortex methods II", SIAM J. Sc. Stat.
,
O. H. Hald, "Smoothness properties of the Euler flow map", unpublished
manuscript,
O. H. Hald, "Convergence of vortex methods for Euler's e uations III"
SIAM J. Num. Anal. 24,538-582 (1987).
"
ture", J. Fluid Mech. 107, 297-338 (1981).
E. Hopf, "Statistical hydromechanics and functional calculus", J. Rat.
ec. na. ,
T. Y. Hou and J. Lowengrub, "Convergence of a point vortex method for
the 3D Euler equations", Comm. Pure Appl. Math. 43, 965-981 (1990).
M. B. Isichenko "Percolation s ati . i
random media" , Report DOE/ET-53088-528 (Institute for Fusion Studies,
. . , .,
C. Itzykson and J. M. Drouffe, Statistical Field Theory (Cambridge Uni-
versl y ress, am rl ge,
J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1974).
G. Joyce and D. Montgomery, "Negative temperature states for the two-
dimensional guiding center plasma", J. Plasma Physics 10, 107-121 (1973).
. .
,
metriques (Hermann, Paris, 1963).
. ,
. , Iva, ,
"The 4/5 Kolomogorov law and the odd-order moments of velocity differ-
ences in turbulence", unpublished manuscript, 1992.
T. Kambe and T. Takao "Motion of distorted vortex rin s"
Soc. Japan 31, 591-599 (1971).
Bibliography 163
aments
Phys.
,. er. a. DC. ,
D. G. Long, "Convergence of random vortex methods in three dimen-
steM" , iR t*ess, ]J,lath. (A)mp, 1992
H. Lu t Vortex Flows i Nature and Technolo New York
1983), p.123.
, ,
ical behavior of the hard-sphere gas", J. Stat. Phys. 30, 681-698 (1984).
eneveau, ua spec ra an mlxe energy casca e 0 ur u ence In
the wavelet representation", Phys. Rev. Lett. 66, 1450-1453 (1991).
c. Meneveau and K. R. Sreenivasan, "The multifractal spectrum of the
49-76 (1987).
C. Meneveau and K. R. Sreenivasan, "Interface dimension in intermit-
en ur ence, y. eVe ,
J. Miller, "Statistical mechanics of Euler equations in two dimensions",
Phys. Rev. Lett. 65, 2137-2140 (1990).
"D· e . n-dir ti orrelations of oriented 01
Stat. Phys. 63 (1991).
..
"
. . . "
,
simulation and static and dynamic behavior of the plane rotator model",
rog. eore. ys. ,
H. K. Moffatt, "On the degree of knottedness of tangled vortex lines" ,
J. Fluid Meeh. 35, 117-132 (1969).
P. Moin, A Leonard and J. Kim, "Evolution of a curved vortex filament
", ,
D. Montgomery and G. Joyce, "Statistical mechanics of negative tem-
era ure s s, m,
· D. Montgomery, W. Matthaeus, W. Stribling, D. Martinez and S. Oughton,
"Relaxation in two dimensions and the 'Sinh-Poisson' equation", Phys.
l i A 4 3-6 1992.
D. Montgomery and L. Phillips, "Minimum dissipation and maximum
166 Bibliography
," . "
6, 279-287 (1949).
mg...
1 US. U ,
G. Papanicolaou, D. Stroock and S. R. S. Varadhan, "A Martingale
approach to some limit theorems", in Statistical Mechanics and Dynamical
S stems DuketurbllIence conference D.RuelIe editor Duke Universit
Series, vol. 3, 1977.
,
200-223 (1985).
A 2, 792-807 (1990).
E.G. Puckett, "A review of vortex methods", in Incompressible Compu-
a iona ui ec anics, ico ai es an i z
bridge University Press, 1992).
D. Pullin, P. Jacobs, R. Grimshaw and P. Saffman, "Instability and
fHam ntation in finite-am litude waves on vortex 1a ers of finite thickness"
J. Fluid Mech. 209, 359-384 (1989).
"
, ,
R. Robert, "A maximum entropy principle for two-dimensional perfect
.
1 ynm
. ,. ,
P. H. Roberts, "A Hamiltonian theory for weakly interacting vortices" ,
A{athematica 19, 169-179 1972.
S. Roberts "Conver ence of random walk methods" Ph.D. thesis, Uni-
versity of California l Berkeley, 1986.
,
C. Thompson, Classical Equilibrium Statistical Mechanics (Clarendon,
,
S. Widnall, "The structure and dynamics of vortex filaments", Ann.
168 Bibliography
non-
, ,
sions of passively convected scalar gradients in chaotic fluid flows", Phys.
1.£1 S ,
A. Weinrib and S. 'frugman, "A new kinetic walk and ercolation erime-
ters", Phys. Rev. B 31, 2993-2997 (1985).
. .
Polymer Science {World SCientific, Singapore, 1986).
, . . .
(Springer, New York, 1990).
G. Williams, "Vortex ring model of the superfluid lambda transition" ,
Phys. Rev. Lett. 59, 1926-1929 (1987).
" .
dimensions", in Excitations in Two-Dimensional and Three-Dimensional
York, 1991).
mte wave-num er super Ul lty near
the 4He lambda transition" ,Phys. Rev. Lett. 68 2054-2057 1992.
A. Yaglom, An Introduction to the Theory of Stationary Random Func-
C Coordinates, 5
Lianomcal ensemDle, ~.1 \Jorrelatea perCOlaLlon, 1':>~-1.1.1
Cantor set, 59 Correlation function, defined, 31
Capacity Correlation length, 114
IOU
170 Index
rt· 'L'.1
. ~
'
, 11 If
~~ ~ -, ...,....,
Curdling, progressive, 88 Equipartition ensemble, 73
J:!Jrgoaicity, 110
Euler equations, 7, 57, 79
D
....... ,. , '" gauge-invariant form of, 17-
, <:1
18
Detailed balance condition , 116 '}fi
DifferentIatIOn vector, 6
independent, 27
Dirac delta function. 14 T'" '1 '""~
T VCl.lUt:, ':'u
Dissipation range, 52
n" , 'UL~O
Experiments, 25
.. .. .J
DomaiBB
1 1 1 ...
,' ll'
infinite, 7-8 Fields, random, 30-36
perIOdIC, 7 Finite additivity, 40
Flory exponent, 117, 140
li'l,..,.ur
Eddies, 49 inhomogeneous, 98
• • • 1 ,.. I"Il.
.lUV.li:ll.;IU, I, ~
Elementary random measure, 40
Elementarv vortex loons 132-134 Lal-Madras-Sokal, 119
Energy cascade, 55-58 periodic, Fourier representa-
...... "" ~n tion for, 21-24
001 - ..- ,~
~~ ~I ~ £~ •• , ~~
. , oJ.l .l~1 ,
JT~ 1
~ ,UV. .~,
..n ...
.lVoJ
L L
impUlse, ll-iL: Lal-lVlaaras-l:iOKal now, lUI
Impulse density, 20 'x-transition, 134
Independent 'events, 27 in superfluids, 149-152
Inertial range, 52 Laplace operator, 6
T. ~ '-~ r1(un~jn~ 7_R T ~itA~ 11 'l
random field, 44
Kinetic energy, 8 Multifractal vorticity distribution,
tip-fined 11 ,.
U'i
Knotted vortex filaments, 96
v ' 'J 'T.. ,
1 All
... ~ ....
in component form, 38
.·no.
~~~~
~~_4- ~C A"7
p~u '.. , •• , , ......
Fourier form of, 21-22 defined, 44
. . .... . ---
InvlSCla lImIt OI, OJ moving average represenl.adOIl
magnetization form of, 21 of, 44
projection form of, 7 Random flow, 31
random solutions of. 36-39 Random flow field. 32
Newton's law, 6 Random Fourier transform, 39,43
..'TI.T.~· , ~ ~
..... J'. lL....n.l: ... l~ ~O_
<:f'F