You are on page 1of 10

SHELL MODEL OF TURBULENCE

Sugan D Murugan , ICTS


3rd August 2018

1 Turbulence Overview and X 0 0 0


k 2 p̃ = −kj ki ui (k − k )uj (k ) (8)
1.1 Nature of Turbulence k0

A flow is said to be turbulent in nature if it has certain 8 and ??combined to give (fi = 0)
characteristics, such as
 
1. Flow has to be dissipative in nature, because of viscous d X
+ νk 2 ũi (k) = −ikj Pil (k) ũl (k 0 )ũj (k − k 0 ) (9)
shear stress dt 0
k
2. Flow must have non-zero 3D vorticity
where
3. Flow with a relatively high reynolds number  
ki kj
4. Diffusive in nature Pij (k) = δij − 2 (10)
k
Since there is no formal theory of turbulence yet, the phe-
nomenological models describe the characteristics of turbu- 1.4 Statistical Description of Turbulent Flows
lence. In a stationary turbulent flow, the velocity of the fluid u(x, t)
at a particular position and time is a random variable. [3]
1.2 Navier Stokes Equation With its PDF as f(v, t) , we define the mean velocity2 at a
point as
∂t ui + uj ∂j ui = −∂i p + ν∂2 ui + fi (1)
˚
where ui (x, t) is the velocity field,ν is the kinematic vis- hu(t)i = u0 f(u0 , t) dV (11)
cosity p = Pρ , where P is the pressure and ρ is the density.
fi represent external force.1
all

Assuming the velocity field is divergence less, we get By defining û(x, t) = u(x,t) − hu(x,t)i , we define the Stan-
dard moments
∂i ui = 0 (2)
hûn i
Now taking 1and multiplying with ui reduces to µ n = D E n2 (12)
û2
∂2 p = −(∂i uj )(∂j ui ) (3)
A general second order 2 point covariance defined as

1.3 Spectral Representation Γij = ui (x, t)uj (x + r, t0 )




(13)
˚
ik.x
u(x, t) = ũ(k, t)e dk (4) But for a statistically homogeneous and isotropic field,
the relevent corelation functions at one time, depends only
˚ on the magnitude of r and not x . Therefore the structure
ũ(k, t) = u(x, t)e−ik.x dx
1
(5) functions and corelation functions are defined as follows
(2π)3
In spectral space eq 1 the NSE becomes

Rij (r, t) = ui (x, t)uj (x + r, t) (14)

dũi X
= −iki p̃ − νk 2 ũj + f̃i − ikj ũi (k 0 )ũj (k − k 0 ) (6)
dt

Dij (r, t) = (ui (x + r, t) − ui (x, t)) uj (x + r, t) − (uj (x, t)



k
(15)
whereas 2 and 3 becomes Now the corresponding spectral space corelations func-
tions can be defined as
ki ũi = 0 (7)
2 various averaging like time, spatial, ensemble could also be consid-
1 Einstein summation convention is used throughout this report. ered when symmetries of turbulence are known.
3
Although these ranges doesn’t exhibit any exact boundaries,
R̃ij (k, t) = u∗i (k, t) uj (k, t) (16) rough values has been established for these ranges. These


distinction can be made either with real scales or spectral
X scales (wavenumbers or frequencies).
Rij (r, t) = R̃ij (k, t)eik.r (17) Starting from the scales in which the boundaries of the
k physical system to be investigated, i.e length scale L and
˚ its mean velocity U . This gives a timescale at the largest
Rij (r, t) = Φij (k, t) eik.r dk (18) scale
all k L
τL = (26)
X U
Φij (k̄, t) = δ(k − k̄) R̃ij (k,t) (19)
A typical stationary turbulent system majorly does three
k
phenomena namely production of energy, transfer of energy
Spectral kinetic energy and dissipation rates can be de- and dissipation of energy. Since the turbulent behaviour of
fined as the system is complex enough to describe directly, it is con-
venient to introduce different lengthscales to describe each
K() =
1
hui (x, t)ui (x,t)i (20) phenomenon seperately.
2 Starting from the largest length scale L which is the phys-
Defining spectral kinetic energy as ical flow scale of the system itself. In the concept of richard-
son view of energy cascade, turbulence is considered to be
1 composed of eddies4 of different sizes. Let l0 be the size
Ẽ(k, t) = R̃ii (k, t) (21) of the largest eddies that are comparable to L. Energy is
2
pumped into the system at these length scales. The eddies
then using 16 and 17 of size l0 are generally anisotropic in nature, the behaviour
X and evolution of these eddies greatly depend on the bound-
K(t) = Ẽ(k, t) (22) ary conditions of the system. In order to see the isotropic
k nature of turbulence, the concerned eddy sizes need to be
In isotropic turbulences, it is easier to work with average smaller than l0 . Define a length scale lEI ≈ l60 , which de-
spectral energy defined on just the magnitude of wavenum- notes a upper cut off on eddy lengths below which energy
ber k is not pumped into the system. The lengthscales between
lEI < l < l0 , is generally regarded as the energy-containing
˚ range.
1
Ẽav (k, t) = Φii (k, t)δ(|k| − k) dk (23) For the smallest eddies the length scale relevent is the
2
all k kolmogorov length scale η , given by . It is at these scales
at which dissipation of energy is dominant which is evident
Defining turbulent dissipation rate as
from dimensionless NSE and the Reynolds no constructed

with these kolmogorov length scales is unity. Although the
ε(t) = −ν ui ∂2 ui (24) main concern is with the eddies of size much larger than
these kolmogorov scales but also smaller than eddies in the
defining spectral dissipation rate as
energy containing range, for this purpose a length scale lDI is
X constructed with lDI ≈ 60η . The lengthscales between η <
ε(t) = 2νk2 Ẽ(k, t) (25) l < lDI 5 is called dissipation range.
k Apart from energy containing range and dissipation range,
the majority of the eddies lie in the so called inertial range
1.5 Various length scales in Turbulence which ranges from lDI < l < lEI . It is this eddies that trans-
fer energy from large eddies to smaller eddies and finally
Using only E and ν , the kolmogorov scales formed are η = dissipate in the dissipation range. Many experiments laws
−1
ν 4 ε − 4 , uη = ε 4 ν 4 , τη = ν 2 ε 2 respectively.
3 1 1 1 1
has been established in this inertial range which shall be
Consider a system with homogeneous and isotropic tur- discussed in .
bulence which is stationary in its statistical properties, then
the steady state of such system is characterized by three
main macroscopic phenomenon. 1.6 Kolmogorov Hypothesis
The Kolmogorov’s similarity hypothesis provides a useful
1. Energy containing range insight into nature of turbulence at relevant scales. The
chosen parameters in any turbulent system are the char-
2. Inertial Range
acteristics of fluid, dissipation rate, length scale of system
3. Dissipative Range 4 Eddyis the swirling of a fluid localized to a length scale.
5 DI
and EI stands for demarcation line between Dissipation, Inertial
3 It is easy to prove Rij (k, k 0 , t) = Rij (k,t) δk,k 0 and Energy, Inertial respectively
and its boundary conditions. At sufficiently high Reynolds We have already discussed the fact that the autocorelation
number (which??), A.N Kolmogorov presented the following function of rDij in 15 is an isotropic function of r[3]. The only
rr
hypothesis second order tensor formed from the vector r are δij & ri 2j .
The eddies of size l  l0 (as described in 1.5) are statisti- Therefore, we can write Dij can be written as
cally isotropic and statistics of such small scale motions are
universal in nature and are determined by ε and ν only. ri rj
Further if the length scales are particularly well within the Dij (r, t) = DN (r, t)δij + [DL (r, t) − DN (r, t)] (32)
r2
inertial range, then the statistics depends uniquely on ε and
not ν. The two-third law verifies the longitudinal structure func-
Having exploited the length scales at two ends of spectrum tion DL (r, t) scales with r for r belong in the certain range
namely η and l0 , define the velocity scale u(l) to be typical called inertial range.
velocity associated with length scales ' l. Define the time 2
scale τ(l) to be eddy turnover time associated with l given DL (λr, t) = λ 3 DL (r, t) (33)
by τ(l) ' u(l)
l
. Then the velocity scales and time scales for Similarly energy spectrum Ẽ(k, t) shows power law be-
the eddy lying in inertial range can be formed from scaling haviour with k for a certain range of k values
arguments and above hypothesis. For η  l  l0
−5
Ẽ(λk, t) = λ 3 Ẽ(k, t) (34)
u(λl) = λh u(l) (27)
1 1
These experimental results can be related to Wiener-
and from dimensional arguments that [u] = [ε] 3 [l] 3 , which Khinchin formula for the isotropic incompressible case
implies h = 13 . Similar arguments for time scale (since
2 −1 ˆ∞
[τ] = [l] 3 [ε] 3 ) Ẽ(k) =
1
krD(r)sin(kr)dr (35)
π
2
τ(λl) = λ 3 τ(l) (28) 0

which verifies the scaling relations 33and 34


As a consequence, the velocity scale and time scale of
eddies decrease as size of the eddies decreases. Now consid-
ering the view of energy cascade along these length scales, 1.8 Cascade of Energy in Turbulence
if the rate of transfer of energy from eddies larger than l to From spectral NSE , taking a product with ũi and taking
eddies smaller than l is defined as T (l), then T (l) is expected average leads to the following
to be of order
dẼ(k, t)
u(l)2 = −2νk2 Ẽ(k, t) + T̃(k,t) (36)
T (l) ≈ (29) dt
τ(l)
where Ẽ(k, t)is given by 21 ,and T̃(k, t) is the energy trans-
Then with 27 and 28 it follows that fer rate and is given by

( )
T (λl)= T (l) = ε (30) X

ũj (k)ũk (k 0 )ũl (k − k 0 )



T̃(k, t) = kl Pjk (k)R i (37)
Therefore the rate of transfer of energy from larger eddies k0
to smaller eddies is independent of l for l lying in inertial
this is a direct overview of energy cascade in fourier space.
range.
From 22 and 25, by summing over all k for 36 and for
Similarly, the typical timescale for viscous diffusion to at- isotropic turbulence dK(t) = −ε(t), therefore
tenuate excitation on a scale of ' l is given by dt
X
2 T̃(k, t) = 0 (38)
l
τ diff (l) ' (31) k
ν
It is evident using 29,30 and 31 that the length scales l at 1.9 Intermittency
which the diffusion time scale and eddy turnover timescale
−1
are of same order is η ' ν 4 ε 4 which is the kolmogorov dis- A central assumption of the K41 theory is the self-similarity
3

sipation scale. This validates the assumption that viscosity of the random velocity field at inertial range scales. But for
becomes important at length scales l in dissipation range i.e higher order structure functions the exponent is not a linear
η  l  l0 . function of the order.
The notion of intermittency may be quantified in the fol-
lowing sense, if the high pass filtered signal of the random
1.7 Experimental laws of Turbulence function u(t) is defined as follows
There are some experimental laws that are obeyed in every ˚
turbulent flows. Amongst it the two-thirds law reveals the u(t) = d3 ω eiωt ûω (39)
scaling nature of structure function. R 3
˚
u>
Ω (t) = d3 ω eiωt ûω (40)   32  2
t1 t1 tn 0 kn0 kN 2(N+2(N−no )−1)
|ω|>Ω ' ' 'λ 3 (45)
tN tn 0 tN k1 kno
Then the random function u(t)is intermittent at small
where no is the shell no after which the viscous range starts,
scales if the flatness
for N = 22, no = 12 the global stiffness from eqn 45 is around
D 4 E ' 107 . Showing that the system is very stiff.
u>
Ω (t) In intermittent statistics, to numerically see the rare event
(41)
the simulation time has to be significantly larger.
F(Ω) =
h(u>
Ω (t)) i
2 2

grows without bound with the filter frequency Ω. It is


trivial that gaussian distribution and self similar signals are
2.2 Shell model overview
not intermittent as F (Ω) is independent of Ω. A turbulent system transfers a flux of energy from small
A sample of turbulent signal as shown in , subject to high- wave numbers to large wave numbers in spectral space. For
pass filtering with Ω comparable to Kolmogorov scale. It is a system having extreme length scales l0 and η as discussed
thus a characteristic of the dissipation range and does not in ....
imply violations of the self-simliar K41 theory of the inertial Then the effective no of degrees of freedom
range.  3
Experimental results suggest that structure functions of l0 9
dof
' ∝ Re 4
order p follow power-laws in the inertial range
N
η
The main idea of shell model is to divide the spectral space
Sp (l) = h(δu (l))p i ∝ lζp (42) into concentric spheres, and radii of nth sphere given by
where the ζp ’s are called exponents of structure func-
kn = k0 λn (46)
tions.
This makes the spectral space discrete, and spectral veloc-
ities un to averaged velocities over the shell. The dynamical
2 Shell model system will have the form

2.1 Turbulence as a Dynamical system u̇n = F(u1 , u2 ..., uN ) (47)


Given a system obeying PDE’s like Navier Stokes equation, The main criterial for building up the evolution equations
it is not always possible to find an analytic solution. These so as to mimic the characteristics of turbulent motion are
PDE’s can be converted into simplified ODE and then stud- the following
ied as a dynamical system.
The possibility of a dynamical system approach allows 1. The only linear term for un is given by −νkn2 un .
one to capture the physical mechanism in turbulence. From
2. The non-linear interaction terms for un are quadratic
the behaviour of dynamical systems in their phase space the
combinations of the form kn un0 un00 , where n0 6= n00 6= n
traditional statistical terms can be studied.
but close to n.
Shell model, simplified caricatures of equations of fluid
mechanics in wave-vector representation. Although the non- 3. InPthe absence of forcing and damping, the energy
linear interactions are with the neighbouring wave numbers, ∗
n un un has to be conserved.
1
2
it exhibits anomalous scaling. The advantage being less no
of variables, which allows for relatively easy numerical com- In the shell model, the velocity un will be defined com-
putations. plex.The most general shell model which satisfying the above
Aspects of numerical integration: The main difficulty in mentioned criteria is given by
integrating a shell model arises from the stiffness of the sys-
tem. In the inertial range, the natural time scale of the nth u̇n = iΦ∗n (u) − Dn un + fn (48)
shell scales as The particular shell model called GOY model is defined
as
1 1
tn ' ' 2 (43)
kn un kn3
u̇n = i(kn un+1 un+2 − ξkn−1 un−1 un+1 + (ξ − 1)kn−2 un−1 un−2 )∗ − νkn2 un + fn
Whereas, within viscous range, the natural time scale as (49)
described in .... In this form it is transparant that the model is defined
by two free parameters ξ and λ together with dimensional
(44) quantities k0 , ν, fn and initial conditions un (0). The common
1
tn =
νkn2 choice of parameters for 3D GOY model is (ξ, λ) = ( 12 , 2) and
The global stiffness of the system is given by for 2D GOY model is (ξ, λ) = ( 54 , 2) , explained below.
The GOY shell model can be modelled to have conserved
quantities in the case of inviscid type (without viscosity and Hn = kn |un |2 (61)
forcing). Lets consider a generic GOY model
Zn = kn2 |un |2 (62)

u̇n = i(a kn un+1 un+2 +b kn−1 un+1 un−1 + c kn−2 un−1 un−2 ) − ν + Therefore,
fn kn2 un
the corresponding α2 has to be 1 for the invari-
(50) ant to be dimensionally same as helicity and α has to be
by rescaling time, we can set a = 1, then in the inviscid 2 for the invariant to be dimensionally same as enstrophy.
2

case we have in general Therefore for λ = 2 case, the ξ takes value 12 for the model to
have second invariant as Helicity, and so a 3D GOY model
∗ and similarly ξ takes value 54 for the model to have second
u̇n = i(kn un+1 un+2 + b kn−1 un+1 un−1 + c kn−2 un−1 un−2 ) (51)
invariant as Enstrophy, and so a 2D GOY model.
the invariant quantities are defined in general as The rate of change of energy of shells is given by

(un u̇∗n + u∗n u̇n )


1
(63)
N
X
knα |un |2
Ėn =
I(α) = (52) 2
n=1 using 48,
for İ(α) = 0, using 51
Ėn = =(Φn un ) − Dn |un |2 + Fn (64)

N
X Where Fn = fn u∗n + fn∗ un . Therefore the total rate of change
k0α+1 uofn−1energy
un−2 unof
) the model is given by
n n−1 n−2
İ(α) = z IM(z un+1 un+2 un + b z
n α α un+1 un−1 un + c z α

n=1
(53) N
X N
X
where z = λα , therefore for İ = 0,by relabeling indices Ė = Ėn = − (Dn |un |2 − Fn ) (65)
n=1 n=1

defining viscous dissipation rate as


N
X
k0α+1 =(
n−1
İ(α) = z α un−1 un un+1 (z n−1 n
+ bz + cz n+1
) (54) N
X
n=1 ε= 2Dn En (66)
n=1
Since in both 3D and 2D turbulences, energy is an invari-
ant, therefore the model by default should have α = 0 as
energy conservation. The simplified equation becomes 2.3 3D GOY MODEL
For the 3D GOY model , as discussed in ..... the parameters
1 + b z + c z2 = 0 (55) are chosen as follows
Since one root corresponds to z1 = 1or(α = 0) , which
1
k0 = 16
implies 1 + b + c = 0,let b = −ξ and c = ξ − 1. Then the λ=2
other root corresponding to the second invariant is given by ξ = 0.5
ν = 1x10−7
z2 =
1
(56) The inviscid invariants in these model are Total Energy
ξ −1 and Helicity given by
N  n
|un |2
X N N
I2 =k0α
1 X X
|un |2 (57) , H= (−1)n kn |un |2 (67)
ξ −1 E=
n=1 n=1
2
n=1
 
1 The model is built with N =22 shells, and the dynamical
α = logλ (58)
ξ −1 system equations correspoding to each shell is given by
From actual turbulence, the 3D model has Energy and
Helicity as conserved quantities, whereas 2D model has En- Φ1 = k1 u2 u3 (68)
ergy and Enstrophy as conserved quantitied. Helicity and
Enstrophy are defined originally as 1
Φ2 = k2 u3 u4 − k1 u1 u3 (69)
˚ 2
H= ω i ui d 3 x (59)
1 1
ΦN−1 =− kN−2 uN−2 uN − kN−3 uN−2 uN−3 (70)
˚ 2 2
Z= ω 2 d3 x (60)
1
ΦN = − kN−2 uN−1 uN−2 (71)
where ω is vorticity defined as ωi = εijk ∂j uk . 2
The Spectral helicity and enstrophy density goes as for all other n,
3. Each shell has a natural time scale. If the shell is in in-
ertial range then the time scale is defined tn ' kn1un ' 12
1 1 kn3
Φn = kn un+1 un+2 − kn−1 un−1 un+1 − kn−2 un−2 un−1 (72) whereas if the shell is in viscous/dissipation range then
2 2 −1
the time scale is defined as tn ' (Dn )−1 = νkn2 .As
where Φn is the transfer function, which plays the role shell no increases the time scale decreases in both cases,
of non linear terms in NSE as discussed in .... This Φn is at the boundary between inertial range and viscous
responsible for energy cascade in the model. The model with range the time scales match.
this notation is as follows
4. Therefore for numerical computation, this smallest
2 −1

u̇n = iΦ∗n − Dn un + fn (73) scale tN ' νkN is taken as time increment for solv-
ing the model.
where Dn = ν kn2 , fn =fδn,n0 .
The model is numerically solved for the set of un (t) as a 5. If forced at shell no n0 ,then there exists forcing length
function of time. scale lfor ' kn−1
0
.
First, the unforced case (f = 0) is taken and studied, with
This distribution evolves in time to produce a characteristic
the initial condition of velocities are taken as follows
of decaying turbulence, i.e the energy decreases with time
1 as per eqn .... and shown in 1.
un = u0 kn2 for n = 1, 2 (74) Fitting the graph to an exponential decay of energy, the
curve fit gives
kn2
un = u0 kn2 e− 2 for n = 3, 4, ...N
1
(75) Ė = Eo e−λt (76)
The idea behind this distribution is to kickstart the sys- with final set of parameters, Eo = 0.581882 ± 0.02091
tem with the energy majorly concentrated in the initial and λ = 0.0154321 ± 0.0007438.The fit curve shown in red
shells, which means in real space the typical length scales of in 1
velocity distribution is comparable to integral length scales As time progresses, the energy which is mainly concen-
as discussed in .....The velocity of higher shells rapidly decay trated in the smaller shells now cascades to higher shells. In
to zero. the final shells, the dissipation coefficient Dn =νkn2 becomes
The total simulation time is ' 300 units. Time step dominant and it dissipates the energy there.
' 3x10−5 units. Using runga-kutta 4th order method to kick- There is no steady state in this condition since there is no
start the simulation and them Adam Bashforth 4th order forcing, yet the energy vs wavenumber spectrum emerges
as time passes as per Kolmogorov prediction (E(k) ' ε 3 k− 3 )
2 5
method to simulate the results.
(not shown in figure).
0.6 Starting with initial velocity distribution as given in 74
and 75, and allowed to decay for a reasonable time till the
0.5
higher shells get a non-zero velocities. The forcing is calcu-
lated in the following manner
0.4 N
X N
X
−2νkn2 En + < fno u∗no

(77)
Total Energy

Ė = 0 = Ėn =
0.3
n=1 n=1

ε̄
fno = (78)
0.2 < (uno )
Now, with forcing at fourth shell calculated using 78 , with
0.1 a magnitude of f4 = 2.8x (1 + i) x10−4 ., the initial conditions
of velocity is final conditions in the unforced case. The sys-
0
tem of equations are solved until a stationary distribution
0 50 100 150 200 250 300 is achieved. Once steady state is achieved, the time average
Time
of dissipation rate is constant, let
Figure 1: Showing total energy vs time, featuring a expo- ˆ T
nential decay fit to data ε̄ = ε(t)g(t)dt (79)
0

where Tis a timescale in which distribution in stationary


Regarding the eqn 73, the time and length scales involved
and g(t) is the PDF of the distribution.
are as follows
As per K41 theory, with ε̄ and ν , the Kolmogorov scales
1. The integral length scale is taken as L ' k1−1 . can be calculated in the model as follows
 3  14
−1
2. The dissipation length scale is around ldiss ' kn−1 . 1. Kolmogorov Length scale η = νε̄ .
p=1
0 p=2
p=3
p=4
p=5
-10 p=6

-20
St.Function <up>

-30 Shell no 10
0.04
-40
0.035

-50 0.03

-60 0.025

0.02

Shell velocity
-70
0.015
-80
-5 0 5 10 15 20 0.01
log(k)/log(2)
0.005

Figure 2: Spectral energy vs Wave number (log-log plot) 0

-0.005
Shell no 3
0.07 -0.01
0 500 1000 1500 2000 2500 3000
Time
0.06

0.05
Figure 4: Shell 10 velocity vs Time
0.04

0.03
Shell velocity

0.02

0.01

-0.01

-0.02

-0.03
0 500 1000 1500 2000 2500 3000
Time

Figure 3: Shell velocity vs Time Shell no 16


0.016

0.014
ν
 21
2. Kolmogorov Time scale τ = ε̄ .
0.012

The stationary distribution of shell velocities 0.01


(ūn = un − hun i) is shown in 3,4, and 5. As discussed
Shell velocity

0.008
in ... the time scale of the shell velocities decrease as shell
no increases. 0.006
The equivalent of the Navier stokes structure functions
0.004
defined as in .... in shell models is simply given by
0.002

Sp (kn ) = h|un | i p
(80) 0

From dimensional arguments, a simple non-anamalous -0.002


0 500 1000 1500 2000 2500 3000
scaling in the inertial range of turbulence is given by Time

− 3p
Sp (kn ) ≈ kn (81) Figure 5: Shell 16 velocity vs Time

As discussed in the ..., the structure function does not


scale properly for all orders. Define exponent function ζ(p)
as follows

Sp (kn ) ≈ knζ(p) (82)


3 0.0001

9x10-5
2.5
8x10-5

7x10-5
2
6x10-5

Energy flux
zeta

1.5 5x10-5

4x10-5
1
3x10-5

2x10-5
0.5
1x10-5

0 0
0 1 2 3 4 5 6 7 -5 0 5 10 15 20
p log (k)

Figure 6: Anamolous exponent functions vs p. Figure 7: Flux of energy through shell

Order p ζ GOY (p) ∆ζ GOY (p) ζ ESS (p)6 Shell no 2 Shell no 14


1.6x10-11 1.6x10-5
1 -0.48087 0.05166 0.35±0.03

Energy Dissipation Rate

Energy Dissipation Rate


1.4x10-11 1.4x10-5
2 -0.79960 0.06319 0.7±0.03 1.2x10-11 1.2x10-5
1x10-11 1x10-5
3 -1.05042 0.06514 1 8x10-12 8x10-6
4 1.33363 0.07928 1.28±0.03 6x10-12 6x10-6
4x10-12 4x10-6
5 -1.51189 0.09501 1.55±0.05 2x10-12 2x10-6
0 0
6 -1.70527 0.1015 1.77±0.05 500 1000 1500 2000 2500 500 1000 1500 2000 2500
Time Time

Table 1: Exponent function ζ(p)vs.p Total


0.0025
Energy Dissipation Rate

0.002

From the simulation for p = 2, 3, 4, 5, 6 the exponent func- 0.0015

tions are fit to curve and is shown in .... 0.001

Considering only the inertial range (shells from 2 to 19) 0.0005


are taken and their moments of spectral velocities (2) are fit 0
500 1000 1500 2000 2500
for scaling according to 82 and given in table below Time
The exponents calculated from direct simulation of Navier
stokes equation are almost indistinguishable from the ones Figure 8: Energy Dissipation rate ε(t)
found in shell model simulation. This shows the remarkable
resemblance between two equations.
The characteristics of turbulence studied from Navier where ε is given by 79.
stokes equation is proved to be chaotic in nature, similarly The flux as shown in 7, the flux of the inertial shells are
the shell model is proven to be chaotic and is discussed in positive, implying that energy is flowing into it. Thus the
section... by calculating the Lyapunov exponents for the energy is being cascaded forward towards the higher shells
shell model. making dissipating effect dominant in them. The other in-
From Kolmogorov prediction, there exists an inertial variant Helicity, keeps its flux alternating between shells.
range in which there is a scaling law. The energy of the Also the dissipation rate of individual shells, showing high
shell depends only on energy cascade rate and wavenumber intermittency in larger shells, as shown in 8.
in a power law fashion. Therefore from pure dimensional
grounds, the relation
−5
2.4 2D GOY MODEL
2
En ∼ ε kn3 3
(83)
For the 2D GOY model , as discussed in ..... the parameters
The power law spectrum produced in the inertial shells are chosen as follows
of the GOY model does not tell us about the direction in 1
k0 = 16
which the cascade of energy is happening. The flux of energy λ=2
through any given shell is ξ = 1.25
PN ν = 1x10−7
=(Φn un )
Πen
n =
m=n
(84) The model is built with N =32 shells, and the dynamical
ε̄ system equations correspoding to each shell is same as that
6 Extended Self Similarity for 3D GOY model
9

8
u̇n = iΦ∗n − Dn un + fn (85)
7

6
The only difference in the equation is that, the dissipative
coefficient Dn is now modified as 5

zeta
4

Dn = νkn2 + µkn−2 (86) 3

As discussed in ... the second term in Dn , removes energy 1


from smaller shells to aid the role of surface friction that is
0
present in 2D case. The graph of Dn vs n is shown in fig...., 0 1 2 3 4 5 6 7
p
it is seen clearly that Dn is significant only in the initial and
final shells.
Figure 10: Anamolous exponent functions vs p.
The no of shells in the model is increased, to encapitulate
two inertial ranges k− 3 & k−3 .
5
The anomalous exponents calculated for the moments are
From .... in 2D the second invariant is enstrophy, which given in the table below
induces the spectrum E(k) ' k−3 . The forward cascade of en-
strophy carries a constant enstrophy dissipation rate which Order p ζ GOY (p) ∆ζ GOY (p)
determines the statistics in that range, that is how the 1 -1.1229 0.0236
k−3 spectrum is produced. 2 -2.2340 0.0477
All the time and length scales involved in 2D model are 3 -2.9181 0.0653
same as that for 3D model. At first the simulation is done 4 -4.4410 0.0979
with unforced (f = 0) and with no friction (µ = 0)case, with 5 -5.5410 0.1237
similar initial conditions as in 3D model. After sufficient 6 -6.6381 0.1496
iterations, when the higher shells have obtained non-zero
energy, forcing is applied and stationary distribution is ob- Table 2: 2D Exponent function ζ(p)vs.p
tained.

0
As discussed, in 2D enstrophy is cascaded directly and
p=1
p=2 energy is cascaded inversely. Shown in 11 of average flux of
p=3
p=4
p=5
energy and enstrophy of shells.
-100 p=6

500000
0
-200
-500000
St.Function <up>

-1x106
Energy flux

-1.5x106
-300 -2x106
-2.5x106
-3x106
-3.5x106
-400
-4x106
-5 0 5 10 15 20 25
log(k)/log(2)

-500 2.5

2
Enstrophy flux

1.5
-600
-5 0 5 10 15 20 25 30 1
log(k)/log(2)
0.5

0
Figure 9: Moments of Shell velocites vs Wave number (log-
-0.5
log plot) -5 0 5 10 15 20 25
log(k)/log(2)

Figure 11: Flux of energy and enstrophy through shell


Now with forcing applied based on enstrophy flux, the
moments of shell velocities have been shown in 9. Notice difference in sign for energy flux in 3D and 2D
From the simulation for p = 2, 3, 4, 5, 6 the exponent func- denoting the change in direction of cascade. And enstrophy
tions are fit to curve and is shown in .... cascades forward as discussed.
Finally, the dissipation of enstrophy and enstrophy is [12] Qiaoing Chen, Sylyl Chen, Gregory L.Eyink (2003).
shown in The joint cascade of energy and helicity in three-
dimensional turbulence. Physics of fluids, Vol 15 No 2
Shell 4 Shell 15
0.06 0.1
0.09
0.05 0.08
0.04 0.07
Enstrophy

Enstrophy
0.06
0.03 0.05
0.04
0.02 0.03
0.01 0.02
0.01
0 0
100 200 300 400 500 600 700 100 200 300 400 500 600 700
Time Time

Shell 15
0.0006
Enstrophy Dissipation rate

0.0005
0.0004
0.0003
0.0002
0.0001
0
100 200 300 400 500 600 700
Time

Figure 12: Enstrophy in 2D GOY

References
[1] G.K. Batchelor (2000). An Introduction to Fluid Dy-
namics, Cambridge University press.

[2] Stephen B.Pope (2010). Turbulent flows, Cambridge


university press.

[3] Peter D.Ditlevsen (2011). Turbulence and Shell Models,


Cambridge University press.

[4] Uriel Frisch (2005). The Legacy of A.N. Kolmogorov,


Cambridge University press.

[5] Bohr, Jensen (1998). Dynamical Systems approach to


turbulence, Cambridge University press.

[6] Victor S. L’vov, Evgenii Podivilov, Anna Pomyalov,


Itamar Procaccia, Damien Vandembroucq (1998). Im-
proved shell model of Turbulence, Physical Review E,
Vol 58, No 2.

[7] Leo Kadanoff, Detlef Lohse, Jane Wang, Roberto Benzi


(1994). Scaling and Dissipation in GOY shell model,
arXiv:chao-dyn/9409001

[8] Luca Biferale, Shell Models of Energy Cascade in Tur-


bulence. Annu. Rev. Fluid mech.2003. 35:441-68

[9] R. Benzi, S Ciliberto, R. Tripiccione, C. Baudet, F.


Massaioli, S.Succi (1993). Extended self-similarity in
turbulent flows. Physical review E, Vol 48, No 1.

[10] P.D.Ditlevsen (1998), Cascades of energy and helic-


ity in the GOY shell model of turbulence. arXiv:chao-
dyn/9811009v1.

[11] H.K.Moffatt, A.Tsinober. Helicity in Laminar and Tur-


bulent flow. Annu. Rev. Fluid Mech. 1992. 24:281-312

You might also like