Shell models as phenomenological models of turbulence

The Seventh Israeli Applied and Computational Mathematics Mini-Workshop. Weizmann Institute of Science, June 14, 2007

Boris Levant (Weizmann Institute of Science)

Joint work with R. Benzi (Universita di Roma), P. Constantin (University of Chicago), I. Procaccia (Weizmann Institute of Science), and E. S. Titi (University of California

Irvine and Weizmann Institute of Science)

Plan of the talk

Introducing the shell models Existence and uniqueness of solutions Finite dimensionality of the long-time

dynamics Anomalous scaling of the structure functions

Introduction

Shell models are phenomenological model of turbulence retaining certain features of the original Navier-Stokes equations.

Shell models serve as a very convenient ground for testing new ideas.

They are used to study energy cascade mechanism, anomalous scaling, energy dissipation in the zero viscosity limit and other phenomena of turbulence.

Navier-Stokes equations All information about turbulence is contained in

the dynamics of the Navier-Stokes equations

In the Fourier space variables it takes the form

,)( fpuuut

u

.0 u

,)()()()()( 2][ kkk

klmlm

k ftikptuktlutuidt

tdu

.0 kuk

The phenomenology of turbulence Let , where is a characteristic length. For small viscosity , there exist two scales

Kolmogorov , and viscous s.t. Inertial range – the dynamics is

governed by the Euler ( ) equation. Dissipation range – energy from the

inertial modes is absorbed and dissipated Viscous range – the dynamics is

governed by the linear Stokes equation

10

Lk L

k 'k

kkk 0

'kkk

kk '

0

Kolmogorov’s hypothesis

The central hypothesis of the Kolmogorov’s theory of homogeneous turbulence states that in the inertial range, there is no interchange of energy between the shell and the shell if the shells and are separated by at least ``an order of magnitude’’.

One usually considers .

''' kkk 21 kkk ]'','[ kk

],[ 21 kk

1''2 kk

A drastic modification of the NSE The turbulent field in each octave of wave

numbers is replaced by a very few representative variables.

The time evolution is governed by an infinite system of coupled ODEs with quadratic nonlinearities.

Each shell interacts with only few neighbors.

122 nn

n k

Different models

The most studied model today is Gledzer-Okhitani-Yamada (GOY) model.

Sabra model – a modification of the GOY model introduced by V. L’vov, etc.

Other examples include the dyadic model, Obukhov model, Bell-Nelkin model etc.

Sabra shell model of turbulence The equation describe the evolution of

complex Fourier-like components , of the velocity field .

with the boundary conditions .

The scalar wave numbers satisfy

, 2

1 2

21*

11*

12 )2( nnnnnnnnnnn fukuuuuuuik

dt

du

nu ,...3,2,1n,...),,( 321 uuuu

001 uu

.2 0n

n kk

Quadratic invariants

The inviscid ( ) and unforced ( ) model has two quadratic invariants.

1. The energy

2. Second quadratic invariant

0 ,...2,1,0 nfn

.2

1

1

2

n

nu

.1

1

2

1 2

1n

n

n

uW

The ``dimension’’ of the shell model The ``3-D’’ regime . Forward energy

cascade – from large to small scales.

W is associated with a helicity. The ``2-D’’ regime . The energy flux

is backward – from the small to large scales.

W is associated with an enstrophy. The value stands for the ``critical

dimension’’. It represents a point where the flux of the energy changes its direction.

10

21

1

Typical spectrum in the 3-D regime

Existence and uniqueness of solutions

• P. Constantin, B. Levant, E. S. Titi, “Analytic study of the shell model of turbulence”, Physica D, 219 (2006), 120-141.• P. Constantin, B. Levant, E. S. Titi, “A note on the regularity of inviscid shell models of turbulence”, Phys. Rev. E, 75 (1) (2007).

Preliminaries – sequence spaces Define a space to be a space of square

summable infinite sequences over , equipped with an inner product and norm

Denote a sequence analog of the Sobolev spaces

with an inner product and norm

2lH

1

2221 :,...),(u:

nn

dnd ukuuV

,),( *

1n

nnvuvu

,),( *

1

2n

nn

dnd vukvu

.u1

222

n

nd

nduk

.u1

22

n

nu

C

Abstract formulation of the problem We write a Sabra shell model equation in a

functional form for

The linear operator is The bilinear operator is defined as

where

Huu ,...),(u 21

f,u)B(u,uu A

,...),(u 2222

21 ukukA

v),...),(u,v),(u,(v)B(u, 21 BB

).2( 21*

11*

12 2

1 v)(u,

nnnnnnnn uvvuvuikB

.u)0u( 0

Solutions of the viscous model – The viscous ( ) shell model has a unique global

weak and strong solutions for any .

Moreover, for , the solution of the viscous shell model has an exponentially (in ) decaying spectrum

when the forcing applied to the finite number of modes.

nk

0

0

nckn Ceu 2

P. Constantin, B. Levant, E. S. Titi, “Analytic study of the shell model of turbulence”, Physica D, 219 (2006), 120-141.

H0u

Hf,u0

Weak solutions of the inviscid model For the inviscid ( ) shell model

has a global weak solution with finite energy

for any . The solution is not necessarily unique.

0

P. Constantin, B. Levant, E. S. Titi, “A note on the regularity of inviscid shell models of turbulence”, Phys. Rev. E, 75 (1) (2007).

)],,0([)u( HTLt

Hf,u0

T0

Weak solutions – uniqueness

The solution is unique up to time T if

The solution conserves the energy as long as . In other words, if

The last statement is an analog of Onsager conjecture for the solutions of Euler equation.

,...),,( 321 uuuu

.)(sup2

0

2 dssuk n

T

nn

.)(sup1

23/2

0

nnn

Tttuk

,...),,( 321 uuuu

3/1u V

P. Constantin, B. Levant, E. S. Titi, “A note on the regularity of inviscid shell models of turbulence”, Phys. Rev. E, 75 (1) (2007).

Solutions of the inviscid model

For there exists T > 0, such that the inviscid ( ) shell model has a unique solution

In the 2-D parameter regime there exists a such that the norm of the solution is conserved. Using this and the Beale-Kato-Majda type criterion for the blow-up of solutions of the shell model, we show that in this 2-D regime the solution exists globally in time.

0

4/51

P. Constantin, B. Levant, E. S. Titi, “A note on the regularity of inviscid shell models of turbulence”, Phys. Rev. E, 75 (1) (2007).

10 f,u V

).],,0([)u( 11 VTCt

1d dV

Looking for the blow-up

The goal is to show that the norm of the initially smooth strong solution becomes infinite in finite time for some initial data.

This will allow to address the problem of viscosity anomaly. Namely, that the mean rate of the energy dissipation in the 3-D flow

is bounded away from zero when .

1V

2

1u

0

Dyadic model of turbulence

For the following inviscid dyadic shell model

one can show that for any smooth initial data

the norm of the solution becomes infinite in finite time.

This was proved in the series of papers by N. Pavlovich, N. Katz, S. Friedlander, A. Cheskidov, and others.

, 12

11 nnnnnn uukuk

dt

du

3/1V

Damped inviscid equation

Consider the inviscid equation with damping

for some . For any which are supported on the

finite number of modes, the solution of the damped equation exists globally in time for any .

f,u)B(u,uu

0.u)0u( 0

f,u0

0

Finite dimensionality of the long-time dynamics

• P. Constantin, B. Levant, E. S. Titi, “Analytic study of the shell model of turbulence”, Physica D, 219 (2006), 120-141.• P. Constantin, B. Levant, E. S. Titi, “Sharp lower bounds for the dimension of the global attractor of the Sabra shell model of turbulence”, J. Stat. Phys., 127 (2007), 1173-1192.

Degrees of freedom of turbulent flow

Classical theory of turbulence asserts that turbulent flow has a finite number of degrees of freedom. In the dimension d = 2,3

For d=2 it was shown that the fractal dimension of the global attractor of NSE satisfies

.d

l

LN

.log1)(

3/12

l

L

l

LcAd f

Finite dimensionality of the attractor The shell model has a finite-dimensional

global attractor. The fractal and Hausdorff dimensions of the

global attractor satisfy

Moreover, we get an estimate in terms of the generalized Grashoff number

.log)()( 2 Cl

LAdAd fH

.log2

1)()( 2 CGAdAd fH

31

2

f

kG

Attractor dimension in 2-D

In the 2-D parameter regime there exists a such that the norm of the solution is conserved.

Assume that the forcing is applied to the finite number of modes for . Then the fractal and Hausdorff dimensions of the global attractor satisfy

12/1

2 )1(log d dV

,...),,(f 321 fff,0nf

enn

.log1

1)()( 2 cnG

dAdAd efH

Around the critical dimension – 2-D

Note that as .

Therefore, the number of degrees of freedom of the model tends to as we approach the ``critical dimension’’ .

1d

1en

Inertial manifold

An inertial manifold is a finite dimensional Lipschitz, globally invariant manifold which attracts all solutions of the equation in the exponential rate. Consequently, it contains the global attractor.

The concept was introduced by Foias, Sell and Temam in 1988.

The existence of an inertial manifold for the Navier-Stokes equations is an open problem.

Dimension of the inertial manifold

Let the forcing satisfy for . Then the shell model has an inertial manifold of dimension

This bound matches the upper bound for the fractal dimension of the global attractor.

The estimate takes into account the structure of the forcing – if the equation is forced only at the high modes, the attractor is small.

,...),(f 21 ff ,0nf bnn

.log2 cnGN b

Reduction of the long-time dynamics

For such an denote a projection of onto the first modes, and . There exists a function whose graph is an inertial manifold.

The long-time dynamics of the model can be exactly reduced to the finite system of ODEs

for .

NP

NN PIQ HQHP NN :

HPNp

f,(p))p(p),B(ppp NPA

NN

H

How big the attractor can be? The bounds obtained until now predict that

the global attractor is finite-dimensional for any force. But are those bounds tight?

In the 2-D regime of parameters for the forcing concentrated on the first mode the stationary solution is globally stable.

Our goal is to construct the forcing for which the upper bound for the dimension of the global attractor are realized.

21 ...)0,0,(f f

...0,0,/u 21kf

The general procedure

The global attractor contains all the steady solutions together with their unstable manifolds.

The plan is – construct a specific forcing, find a corresponding stationary solution and estimate the dimension of its unstable manifold.

This method has been used by Meshalkin-Sinai, Babin-Vishik, and Liu to estimate the lower bound for the dimension of the global attractor for the 2-D NSE.

Single mode stationary solution The natural candidate – forcing concentrated

on the single mode and the corresponding stationary solution

This is an analog of the Kolmogorov flow for the NSE, used by Babin-Vishik and others.

However, in our case, because of the locality of the nonlinear interactions, the dimension of the unstable manifold is at most 3.

,...),0,0,...,0(f 1 NN k ,...).0,0,...,0(u 3

NN k

N

Stability of a single mode solution Bifurcation diagram of the single mode

stationary solution vs. .

3-D 2-D

4Nk

Construction of the large attractor The conclusion – for any and for small

enough viscosity, there exists such that is stable for all and unstable for all .

To build a large attractor, we consider the following lacunary forcing and the corresponding stationary solution

1

,ff1

5

n

n .uu1

5

n

n

NNu

NN NN

Lower bound for the dimension The solution has a large unstable

manifold. Counting its dimension we conclude that the Sabra shell model at has a large global attractor of dimension satisfying

Therefore, the upper-bounds for the fractal dimension of the global attractor are sharp.

The constant depends only on and tends to as .

u

).(log)dim( 2 cGCA

1

)(c 1

Anomalous scaling of the structure functions

• R. Benzi, B. Levant, I. Procaccia, E. S. Titi, “Statistical properties of nonlinear shell models of turbulence from linear advection models: rigorous results”, Nonlinearity, 20 (2007), 1431-1441.

Structure functions

The n-th order structure function of the velocity field is defined as

where denotes the ensemble or time average. Assuming that the turbulence is homogeneous

and isotropic, one concludes that the structure functions depend only on .

),v( tx

,r

r),v(),v(r)(r,

n

n txtxxS

rr

Kolmogorov scaling

Under various assumptions on the flow, and in particular, assuming that the mean energy dissipation rate is bounded away from zero when viscosity tends to 0, Kolmogorov derived the 4/5 law

Applying dimensional arguments he conjectured

2v

.5

4)(3 rrS

.)( 3/nn rrS

Anomalous scaling

Recent experiments, both numerical and laboratory, predict that the structure functions are indeed universal and for each there exist ``scaling exponents'‘ , such that for large Reynolds number

Moreover, , as predicted by the 4/5 law, but the rest of the exponents are anomalous, different from the prediction n/3.

2nn

.)( nrrSn

13

Application of the shell model Shell models of turbulence serve a useful purpose

in studying the statistical properties of turbulent fields due to their relative ease of simulation.

In particular, shell models allowed accurate direct numerical calculation of the scaling exponents of their associated structure functions, including convincing evidence for their universality.

In contrast, simulations of the Navier-Stokes equations much harder, and one still does not know whether these equations in 3-D are well posed.

Structure functions of the shell model We define the structure functions

For sufficiently small viscosity, and a large forcing, there exists an ``inertial range’’ of -s for which the structure functions follow a universal power-law behavior

All the exponents are anomalous except for the .

,)(2

2 nn ukS ,)(2

2

p

nnp ukS ...Im)( *113 nnnn uuukS

nk

.)( n

nnn kkS

13 3/nn

Linear problem

In the recent years a major breakthrough has been made in understanding the mechanism of anomalous scaling in the linear models of passive scalar advection.

The linear shell model reads

where is a solution of the nonlinear problem.

f,w)B(u,ww Au

Connection to the nonlinear case Let be real, and consider the system

Observe, that for any the two equations exchange roles under the change .

This leads to the assumption that if the scaling exponents of the two field exist they must be the same for any .

Angheluta, Benzi, Biferale, Procaccia, Toschi (2006), Phys. Rev. Lett. 87.

f,u)B(w,u)B(u,uu A

.gw)B(w,w)B(u,ww A0

uw

0

Numerical evidence

The ``compensated’’ sixth order structure function for different values of 74.1

6 )( nn kkS

Rigorous result

For and the solution of the coupled system exists globally in time.

For any , the solutions converge uniformly, as to – the corresponding solutions of the system with

H),0w(),,0u( )),,0([, HLgf

0T0 )0,(),0,( twtu

0,0)0,u(),u(suplim

2

00

tt

Tt

.0)0,w(),w(suplim2

00

tt

Tt

Conclusions

If the scaling exponents of the fields are equal for any they will be the same for .

For the is a solution of the nonlinear equation, while is a solution of the linear equation advected by .

This result is valid for the large but finite time interval.

wu,0

00 u

wu

Summary

Shell models are useful in studying different aspects of the real world turbulence, by being much easier to compute than the original NSE.

Further analytic study of the models may shed light on the long standing conjectures in the phenomenological theory of turbulence.