HAL Id: jpa-00247359
https://hal.archives-ouvertes.fr/jpa-00247359
Submitted on 1 Jan 1997
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
A Renormalization Group Study of Three-Dimensional Turbulence
Philippe Brax
To cite this version:
Philippe Brax. A Renormalization Group Study of Three-Dimensional Turbulence. Journal de
Physique I, EDP Sciences, 1997, 7 (6), pp.759-765. �10.1051/jp1:1997190�. �jpa-00247359�
A Renormalization Group Study of Three-Dimensional Turbulence
Ph.
Brax(*)
DAMTP, University
ofCambridge,
SilverStreet, Cambridge CB39EW,
UK(Received
24January
1997, revised 12 March 1997,accepted
4April 1997)
PACS.47.27.Ak F1tndarnentals PACS.ll.10.Gh Renormalization
PACS.05.40.+j
Fluctuationphenomena,
random processes, and Brownian motionAbstract. We
study
the three-dimensional Navier-Stokes equation with a random Gaussianforce
acting
onlarge wavelengths.
Our work has beeninspired by Polyakov's analysis
ofsteady
states of two-dimensional turbulence. We
investigate
the time evolution of theprobability
law ofthe
velocity potential. Assuming
that thisprobability
law isinitially
definedby
a statistical field theory in the basin of attraction of a renormalization fixedpoint,
we show that its time evolution is obtainedby averaging
over small scale features of thevelocity
potential. Theprobability
law ofthe
velocity potential
converges to the fixedpoint
in thelong
timeregime.
At the fixedpoint,
thescaling
dimension of thevelocity
potential is determined to be-4/3.
Wegive
conditions for the existence of such a fixed point of the renormalization groupdescribing
thelong
time behaviour of thevelocity potential.
At this fixed point, the energy spectrum of three-dimensional turbulencecoincides with a
Kolmogorov
spectrum.Turbulence is one of the crucial
puzzles
of TheoreticalPhysics. Many
methods have been used to tackle such a difficultproblem [Ii.
Oneapproach studying
the role of scale invariance haslong
beenpursued.
A few years agoPolyakov suggested using
conformal field theories tostudy
turbulence in two dimensions [2].Polyakov's
idea is that the scale invariantregime
ofturbulence can be described
by
a conformal fieldtheory.
In threedimensions, perturbative
renormalization transformations in momentum space have been used to derive an infrared fixedpoint describing
thelong
distancephysics
of the Navier-Stokesequation [3-6]. Similarly,
probabilistic
solutions of the Navier-Stokesequation
with noforcing
term have beenprovided by
self-similarprobability
laws for thevelocity
field[7,8].
We shall useproperties
of the Navier-Stokes
equation
underscaling
transformations in order tostudy
the time evolution of solutions.In
particular,
ouranalysis
will combinenon-perturbative
statements about the renormalization group and results about 3d conformal field theoriesapplied
to the Navier-Stokesequation.
The motion of a fluid is described
by
themacroscopic velocity
field. We shall suppose that themacroscopic velocity
field ua isregularized
at small distancesby
a cut-off a. This cut-off is related to theviscosity by
~2
U "
II)
(*)
On leave of absence fromSPhT-Saclay, CEA,
91191 Gif sur Yvette, France(e-mail: braxflamoco.saday.cea.fr)
©
Lesiditions
dePhysique
1997760 JOURNAL DE
PHYSIQUE
I N°6where T is a time
characterizing
the energydecay
when astirring
force is absent. The cut-offa
plays
a crucial role as it sets the smallest size below which the Navier-Stokesequation
does notprovide
a sensibledescription
of fluid motions. It is also relatedii)
to theviscosity
and the fact that the Navier-Stokesequation
isdissipative.
Forincompressible fluids,
thevelocity
field isdivergenceless:
T7 va = 0.
(2)
This
implies
that thevelocity
field can be writtenva = curl ~fia
(3)
where ~fia is the
velocity potential.
We will be interested in thevorticity
wa = curl va.
(4)
The
vorticity
satisfies thevorticity
Navier-Stokesequation
where AI =
curl
F
isthe
orticityforcingterm
and F is the elocityforcing term.
that the cut-off
a
regulates the productof
fields appearing in the
non-linear terms of (5). In
order to discard
boundary
effects, we
shall
onlybe
interested
in the
case
where theup all space. Furthermore,
we
shall require the orticityto
decrease at infinity. The initialconditions and the external tirring are chosen to be
regularized
randomfields where
the ultraviolet ut-off is given by the scale
a. In
particular, the velocity forcing
term
willto
be Gaussian witha white
noisetemporal
dependence and anTaking into
ccountthe ncompressibilitycondition,
2-point
function isFa,«ik, ~d)Fa,Pik', ~d'ii = WOSL kiibaP - ))bik
+
'ibi~d +~d'i
16)in
ourier space. The constant Wo determines themplitude of
the
forcing while
itsk-dependence
is
defined
by
SL(k).
We
suppose that SL isa
positive function of
width L~~, I.e. (k)
=
here SL has idthL~~
and iscentered
around zero, italso normalized
by
J d~ks(k)=
I. The idthis
supposed tobe
arbitrarily all. Noticethat
the spectrum (~)
of
the stirring orce F is ~SL(k) hichis
L=~ta
(7)
where ~ is a
large
numberdepending
on theforcing
term. Notice that we do notspecify
theshape
of the function S apart from therequirement
that it is a smooth normalized function centered around zero. Thespectrum k~SL(k)
ispeaked
aroundL~~ implying
that the randomstirring only
acts on verylarge
scales. Inparticular
this entails that the2-point
function(6)
defines ageneric
Gaussianstirring
force whose action is confined tolarge
scales.For a
given
initialconfiguration
and agiven
realization of thestirring force,
solutions of(5)
will solve the initial valueCauchy problem. Starting
from an initialprobability
lawdPo (ifi),
onecan
soive (5)
for agiven
realization of thestirring
force. This defines the conditionalprobability
law
dP(~fia(x,t)[Fa ix, t)) expressing
theprobability
law of ~fi~(x,t)
for agiven
realization of thestirring
forceF~.
Thisprobability
law is such that the Navier-Stokesequation
is valid(~
Practically, S(k)
can be taken to be x~I
exp
-k~,
in that casek~SL (k)
has a maximum at k= L~~
in all correlation functions
(ju~~ (xi, t).. u~ujx~, tjj
=f(u~~ (xi, tj.. (-(v~ v)u~~
+ iu~~vjv~
+ u/~u~~ +M~j.. u~~ix~, tjj (8)
for a
given
realization of thestirring
forceFa.
Theseequations
are theHopf equations.
The conditionalprobability
lawdP(~fia(x,t)[Fa ix, t))
is random when seen as a functional of thestirring
force. Afteraveraging
over theforcing
termrealizations,
onegets
theprobability
law of thevelocity potential
at time tdl'iffiaiX>t))
"
/dl'(ffia(~>t)ifa(X>t)16~ d/L(Fa(~>t)1 (9)
where
d~(Fa ix, t))
is the Gaussianprobability
law of thestirring
force at time t. The mainobject
of thepresent
work is to characterize thelong
time behaviour ofdP(~fia).
We shall suppose that the conditional
probability
at time t is definedby
a statistical fieldtheory
dP(~fia ix, t) [Fa (x, t))
= d~fi exp
-St
r
(ifia, Fa) (10)
Z '
where the
partition
function Z is a normalizationfactor,
d~fi stands for the functionalintegral symbol
andSt,r(ifia, Fa)
is an effective action at time tdepending
on theregularized
fields.By
dimensionalanalysis,
the action is ahomogeneous
function oft/T.
Afterintegrating
overthe
stirring force,
one obtains the effective actiondescribing
theprobability
law of thevelocity potential
at time tdl'ilfia)
"jdlfiexp-St,riffia)
~XP~St,r(ffia)
"lexp-St,rilfia>Fa)d/Lifai~>t)). ill)
We shall suppose that the
velocity potential
is a field withscaling
dimensiond~.
This dimension is for instance well-defined if the effective fieldtheory ill)
is in the basin of attraction of afixed
point
of the renormalization group. In that case thescaling
dimension of~fi becomes its conformal dimension. In this
communication,
we shall show that the time evolution of solution of(9) starting
from the initial conditionsii I)
is determinedby
renormalization transformations.The evolution of solutions of
(9)
isgiven by
ilal~~>~~tj
#
~~~~§bA-lal~>t) eXp-sATt,rlial~~>~~t))
#
/ d§beXp-st,A-TrllfiA-lal~,t)). l12)
ii,aj
This is the
precise
statement that the time evolution of solutions follows renormalization tra-jectories.
The firstequation implies
that thevelocity potential
lies on the renormalizationtrajectory
of a field of dimensiond~,
I.e. the solutionj~(~x, ~~t)
of the Navier-Stokes equa- tion at time~~t
issimply
the renormalized field~fia(~x, t)
of~fi>-ia lx, t).
The secondequation
entails that the effective action at time~~t
is obtained afterintegrating
over the fluctuations in the range)~~a, al.
This renormalization transformation isperformed
in momentum spaceintegrating
over modes in[a~~, ~a~~].
Let usemphasize
the roleplayed by
the cut-off a. The cut-off a is necessary to have a well-defined renormalizationprocedure (12).
The timerescaling exponent
T and thescaling
dimension of~fi are
uniquely
determinedT=(, d~=-( l131
762 JOURNAL DE
PHYSIQUE
I N°6Notice that the
scaling
dimension of w is dw =2/3 guaranteeing
that thevorticity
decreases atinfinity.
Let us now prove these results.
They depend crucially
onproperties
of thestirring
forceFa.
Notice that the
spectrum
of thestirring
forceSL(k)
haslength
dimension -3.Moreover,
oneobtains
SL( ~)
=
~~S>-i~(k). (14)
~
Using (6),
one can deduce thefollowing scaling
relation(Fa()> )), Fa(~, )))
"
~~~~(FA-la(k>L°)> FA-la(k'>L°')). lis)
As the
stirring
force isGaussian,
this relation between2-point
functionsyields
anequality Fa(~x, ~~t)
=
~~~/~F>-ia(x, t), (16)
I.e. these two random variables have the same
probability
law. In otherwords,
one has theequality
/ /i~~~~fai~X> ~~t))d/Lifai~~> ~~t))
"
/ /ifAlai~>t))d/LifA~lai~>t)) (171
for any functional
f.
This fact will allow us to relate theprobability
law of the stream function at two different times whentaking
the average of(19).
Let us rewrite the Navier-Stokesequation
afterrescaling
a ~~~~a
and T ~~~~T
~°lt~~
+
iv>-ia v)Ld>-ia
=
iLd>ia v)v>-ia
+U~~~/~Ld>ia
+MAia. i18)
At time
t,
the initial condition(10)
definedby St,fi(ifi>-i lx, t),F>-i~(x,t))
is a solution of(18)
where the range of theintegration
over modes for any correlation function is now[0, ~a~~].
Using (12)
and(16)
one can substitute~~~~fia(~x, ~~t)
and~f+~M~(~x, ~~t)
for~fi>-ia lx, t)
and
M>-i~(x,t)
in each correlation function. Notice that the fields in each correlation func- tion have now modesonly
in the range[0, a~~].
Theintegration
over the modes in the range[a~~, ~a~~] only
affects the Boltzmannweight
expSt,>-T~ (~fi>-i~ ix, t), F>-1~ ix, t)).
Perform-ing
theintegration
of the Boltzmannweight
over these modesgives
the renormalized Boltzmannweight
~Xp
-SlTt,T~'~a~~f, ~~t), ~~ Fa~~~t, ~f)) / dlfi~Xp -St,fi ~'~l~la~~>
~~>~l~~a~~>
~~~' ~~~~jj ~j
A
The
probability
law of thevelocity potential (12)
is obtained afteraveraging
over the Gaussianstirring
force andusing (17).
Theresulting equations
for the correlation functionsonly
involve the fields~fia(~x, ~~t)
andMa (~x, ~~t).
Theseequations correspond
to~)ii~j
+~~+~~~iva b>z)Lda
=~~+~~~(Lda b>ziva
+UbizLda
+~~i~~fifa 120)
when inserted in correlation functions. The derivatives are taken with respect to
~~t
and ~x.We now
require
that(20)
coincides with(5)
at time~~t
and coordinates ~x. This is achieved ifd~
= -1-~, d~
= T 2.
(21)
2
One can then deduce
(13).
This proves that(12, 13) specify
the time evolution of theprobability
law of thevelocity potential.
The evolution
equations (12) prescribe
the timedependence
of thevelocity potential
and itsprobability
law. This result is valid for anygiven
initial conditions. One would like tostudy
thelong
time behaviour of such solutions. One can view(12)
asdefining
adynamical system
in the space of
probability
laws. One candistinguish
two situations. If theprobability
law(12)
does not converge to a limit as time goes toinfinity (~
~cc),
thelong
time behaviour of solutions of the Navier-Stokesequations
is difficult toanalyse.
In view of the numerousevidence in favour of an universal
regime
of three-dimensionalturbulence,
one can assume that theprobability
law(12)
converges as ~ ~ cc. Such a limit is a fixedpoint
of therenormalization group. At the fixed
point
thescaling
dimension-4/3
of thevelocity potential
becomes its conformal dimension(fixed points
of the renormalization group are conformal theories[9]).
Let us derive criteria for the existence of a fixedpoint.
The field~fij
converges to the renormalized field~fi of dimension
-4/3.
Thisimplies
that the renormalizedvelocity potential ~~~~fia(~x, ~~t)
becomes aindependent,
I.e. thedependence
on the cut-off a of the solution of the Navier-Stokesequation
~fia(~x, ~~t) disappears
in thelong
timeregime.
Hence if the evolution(12)
admits a fixedpoint
thelong
time behaviour of the solutions of the Navier-Stokes
equation
isindependent
of the cut-off a. This is thesign
of an universal behaviour.Using a£~fij ix, t)
=
-~fi~fij ix, t)
~ 0 as ~ ~ cc one can derive renormalization groupequations
for then-point
functionsrz~(x~, t)
=
(vu (xi, t)..
vu(xz~,t))
of thevelocity
fieldwhere to is a
time origin. These
equations are valid upon convergence(12) to a fixed point. They
provide
on-trivial tests for the
We
shall
now derive further conditionsfor the xistence
of such a
fixed point. The fixed pointatisfies (18)
in
thelimit
when ~
goes
toinfinity.
Using (13), an seethat the
influence
of
theviscosity ecomes negligible in the
long
time regimeSn(k)
=
(~)~S(~)) (23)
converges to zero as ~ ~ cc. This
implies
that the2-point
correlation ofF>-ia
goes to zero and therefore the Gaussianstirring
force vanishes in thelong
timeregime.
Inparticular
thisimplies
that the fixedpoint
is not influencedby
theprecise shape
S of thestirring
force. It is alsoindependent
of the infrared cut-off Ldefining
thetypical
size of the action of thestirring
force. We have therefore shown that the fixed
point
is neitherdependent
on the ultra-violetnor the infra-red cut-offs
[10].
Assuming
that the fixedpoint represents steady states,
the time derivative of each correlation function vanishes as well. We therefore find that fixedpoints
are characterizedby
thevanishing
of the non-linear terms in the limit ~ ~ cc. The non-linear terms can be
readily
evaluated when the cut-off is rescaled to zerousing properties
of short distanceexpansions.
Products of fields becomesingular
when the cut-off is removed. For fields calledquasi-primary
fieldsill],
the result of theproduct
can beexpanded
in a power series ina/~.
Avanishing
limit is obtained if theleading
term has apositive exponent.
We shall suppose that thevelocity potential
becomesa
quasi-primary
field at the fixedpoint.
In that case, the nonlinear terms read~afl7~7((VA-la ~)VA-la)fl
~())~~
~~~~1~2,a,A-Ia
+(24)
764 JOURNAL DE
PHYSIQUE
I N°6where
d2
is the conformal dimension of theleading pseuso-vector
~fi2,a in theexpansion
of the non-linear terms. We can therefore conclude that the fixedpoint
is characterizedby
theinequality
d2
> 4 +2d~. (25)
This
generalizes
a similarinequality
obtainedby Polyakov
in two dimensions. As a consequence, notice that~fi cannot be identified with ~fi2 as
(25)
is not satisfied in that case. Thisimplies
that
(lfiPlfiulfip)
" o>(26)
the three
point
function of thevelocity potential
vanishesidentically
at the fixedpoint.
This
equation
is an exact closureequation.
It followsdirectly
from the existence of a fixedpoint.
One does not need to know the field~fi2 in order to find
d2.
The dimensiond2
can be determined from a renormalization group
equation involving
the correlation functionrz~,i (x~, t)
=
(£op,b,((va.T7)va) p(xi, t)va(x2, t).. va(xz~+i, t))
of nvelocity
fields and one inser- tion of the non-linear termrim,iii))~/~xi, t)
+ ~~ ~~)~
~~~rn,i(1))~/~xi,t)
= 0.127)
In conclusion the existence of a fixed
point
and the dimensions of the fields~fi and
~fi2 are char- acterized
by
renormalization groupequations. Conversely
any conformal fieldtheory satisfying (25)
such that thevelocity potential
is aquasi-primary
field of dimension-4/3
describes thelong
timeregime
of 3d turbulence.It is
extremely
difficult to constructexplicitly
three-dimensional conformal field theoriesdescribing
an infrared fixedpoint
of the renormalization group. Provided that(12)
convergesto a fixed
point
we can calculate the2-point
function of thevelocity potential
and then the energyspectrum
in thelong
timeregime.
The energyspectrum
iseasily
related to the Fouriertransform of the two
point
functionEa(k)
=
4~k~ / d~x(va (x)va (0))
exp -2~ikx.(28)
In the
long
timeregime,
the energyspectrum
behaves asEa(k)
+~
k~~/~ (29)
This
spectrum
is valid for k < Ila.
The exponent isgiven by 2d~
+1 =-5/3 (~).
Thespectrum
coincides with aKolmogorov
spectrum[12]
The solutions of the Navier-Stokes
equation (12)
have some salient features. Let us no- tice that the evolution of theprobability
law of thevelocity potential (12)
makesexplicit
thebreaking
of time reversal invariancealready
present in the Navier-Stokesequation. Indeed,
the
probability
law at time~~t
is obtained from theprobability
law at time tby averaging
over the small scale features of the
velocity potential.
These small scale features are there-fore lost in the process of evolution. In
particular, knowing
theprobability
law at time~~t
does not enable one to deduce the
probability
law at time t. Another consequence is that(~) Our results are
compatible
with theperturbative
calculations of references[3-5].
In this case, thestirring
force isspecified by (Fa,~ (k, w)Fa,p(k', w'))
=
fiXa-i (k)(&~p fit)&(k
+k')&(w
+uJ')
(k2 +g~2j where x~-i
(k)
isequal
to one for k <a~~
andzero otherwise. The mass term m = L~~ is small but
guarantees
that there are no infrareddivergences
when k ~ 0. The sameanalysis
as in the text shows thatd~
=
~fand
the spectrum behaves likek@
in the range k < a~~ for 0 < y < 5.
the renormalization
trajectories
converge to anon-unitary
conformal fieldtheory
asd~
< 0.This result has also been obtained in 2d
[2].
It reinforcesPolyakov's
idea that turbulence is aflux state and not an
equilibrium
state describedby
anunitary theory. Moreover,
at the fixedpoint
the energy spectrum is determined to be aKolmogorov
spectrum. This resulthighly depends
on the fact that thestirring
force is Gaussian and acts onlarge wavelengths.
Anotherfeature of our
analysis
is that we have derived aspectrum compatible
with aKolmogorov spectrum
without any cascadehypothesis [12].
Moreover our results gobeyond
the usual scale invarianceanalysis ill.
Our renormalizationtechnique
allows us to derive the time evolution of thevelocity
field when theviscosity
is non zero. This is not sufficient to characterize thevelocity
field. It isonly
when thescaling
dimension oftfie velocity
field and the time evolution(12)
of theprobability
law of thevelocity potential
are known that aprobabilistic
solution of the Navier-Stokesequation
isspecified.
Apossible application
would be to use(12)
in order tocompute the time evolution of the
probability
law of thevelocity potential numerically.
In thatcase one could
certainly
use renormalizationtechniques
drawn from lattice fieldtheory.
An- otherinteresting
test would be to use(22)
and(27)
in order tonumerically
test the convergence of(12)
to a fixedpoint. Finally,
let us note that our results can beapplied
in twodimensions,
in
particular (25)
isreplaced by Polyakov's
conditionill (~)
and the energyspectrum
is stillgiven by
aKolmogorov spectrum.
Assuming
that theprobability
law of thevelocity potential
is a statistical fieldtheory,
we have shown that its time evolutioncorresponds
to renormalizationtrajectories.
We have assumed that these renormalizationtrajectories
lead to a fixedpoint.
The fixedpoint
isnon-unitary
and characterizedby
renormalization groupequations.
In that case we have shown that aKolmogorov spectrum
is obtained in thelong
timeregime.
The existence of such a fixedpoint
of the renormalization groupgoverning
thelong
timeregime
of turbulence seemsphysically
clear in view of the numerous
experimental
as well ascomputational
evidence in favour of aKolmogorov spectrum.
It is achallenging problem
to construct such a fieldtheory.
Acknowledgments
am
grateful
to W.Eholzer,
M.Gaberdiel,
H.K. Moflatt and R. Peschanski for useful discus- sions andsuggestions.
References
ill
FrischU.,
Turbulence(Cambridge University Press).
[2]
Polyakov A-M-,
N~cl.Phys.
8396(1993)
397.[3] Forster
D.,
Nelson D-R- andStephen M.J., Phys.
Rev. Lett. 36(1976)
867.[4] Forster D. and Nelson
D.M., Phys.
Rev. A16(1977)
732.[5] Dedominicis C. and Martin
P.C., Phys.
Rev. A19(1979)
419.[6] Yakhot V. and
Orszag S., Phys.
Rev. Lett. 57(1986)
1722.[7] Foias C. and Temam
R.,
Comm~n. Math.Phys.
90(1983)
181.[8] Foias C. and Temam
R.,
India. Univ. Math. J. 29(1980)
913.[9] Belavin
A.A., Polyakov
A-M- and ZamolodchikovA.B.,
N~cl.Phys.
B 241(1984)
333.[10] Vergassola M., Phys.
Rev. E53(1996)
3021.ill]
FerraraS.,
Grillo A.F. and GattoR.,
AnnalsPhys.
76(1973)
161.[12] Kolmogorov A.N.,
C.R. Acad. Sci. URSS30(1941)
301.(~) The stream function ~l satisfies
tfi>-ia(~)tfi~-ia(~)
+~