Turbulence: Course 2

Anand Utsav Kapoor

International Master in Turbulence

Lille/Poitiers, France

A demonstration of the Kolmogorovs 4/5th law and the equivalent law for a passive scalar.

Submitted to: Prof. Thomas Gomez

May 21, 2017

1
 Abstract

When the Navier-Stokes equation is converted into an evolution equation for the velocity
correlation function,the resulting expression is the Karman-Howarth (KH) equation,which
is a very important equation in the theory of isotropic turbulence. It provides the start-
ing point for nearly all dynamical theories and one of the immediate consequences of the
Karman-Howarth equation is Kolmogorovs celebrated four-fifths law.
The first section of this report contains assumptions and definitions of key concepts, the
second section contains a demonstration of the KH equation , the third, of the 4/5th law and
finally, the fourth section contains the development of a law similar to the 4/5th law for a
passive scalar.

2
Contents

1 Key terms and assumptions 5

2 The Karman Howarth equation 7

3 Kolmogorovs 4/5th law 8

4 Kolmogorovs ideas extended to passive scalars dynamics: Yagloms 4/3rd law 9

3
List of Figures

1 Definition of the longitudinal and lateral velocity correlation functions. . . . . 6

2 The shape of the longitudinal and lateral velocity correlation functions. . . . . . 6
3 Schematic shape of K(r) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

4
1 Key terms and assumptions

1. Homogeneous, isotropic turbulence is considered , which implies that there is no mean

shear and hence no injection of energy.
2. Ensemble averages are used instead of time averaging due to (1) and since we need to
differentiate between the mean and the turbulent components , therefore, u itself is the
turbulent component.
3. The second order velocity correlation tensor is defined as:

Qij (r, x, t) = ui (x)uj (x + r) (1)

and for homogeneous turbulence, statistical quantities are independent of x and hence,

Qij (r, t) = ui (x)uj (x + r) (2)

If the time dependence is understood, a more concise notation can be used :

Qij (r) = ui uj (3)

The following hold true:

Qij (r) = Qji (r)

(4)

= (5)
xi ri

= (6)
xj rj

Qij Qij
= =0 (7)
ri rj

4. Longitudinal and lateral velocity correlation functions

Qxx (rex ) = u2 f (r) (8)

Qyy (rex ) = u2 g(r) (9)

u being the typical velocity of the large eddies.

The functions f and g are called the longitudinal and lateral velocity correlation functions
(or coefficients). They are dimensionless and satisfy f (0) = g(0) = 1. They typically
look like as shown in fig. 1

5
 Figure 1: Definition of the longitudinal and lateral velocity correlation functions.

Figure 2: The shape of the longitudinal and lateral velocity correlation functions.

Another particular form of Qij is R(r) and is given by:

1 1
R(r) = Qii = u.u (10)
2 2
5. Second-order longitudinal structure function

[v(r)]2 = [ux (x + rex ) ux (x)]2 (11)
also,
[v(r)]2 = 2u2 (1 f ) (12)

For isotropic turbulence, the longitudinal correlation function and the lateral structure
function are related as:

1
g = f + rf (r) (13)
2
6. Third order velocity correlation function
It is given by:

Sijk (r) = ui (x)uj (x)uk (x + r) (14)

A special case is:

6

u3 K(r) = ux 2 (x)ux (x + rex ) (15)
The function K(r) is known as the longitudinal triple correlation function.

The third-order longitudinal structure function, [v]3 is related to K(r) through:

[v]3 = 6u3 K(r) (16)

Figure 3: Schematic shape of K(r)

7. General form of the correlation tensors in isotropic turbulence

u2 2
Qij = [(r f ) ij f ri rj ] (17)
2r
[ K rK 2K + rK K ]
Sijk = u 3
ri rj rk + (ri jk + rj ik ) rk ij (18)
2r3 4r 2r

2 The Karman Howarth equation

We can convert
the Navier-Stokes equation into an evolution equation for the velocity correlation

function ui uj , as shown below:

ui ui uk (p/)
= + x 2 ui (19)
t xk xi

uj uj uk (p /)
= + 2x uj (20)
t xk xj

On multiplying (19) by uj and (20) by ui , adding and taking an ensemble average , we get:

uj uk ui uk 1 p p
ui uj = ui + u j ui + u j + ui 2
uj + uj x
2
u i
t xk xk xj xi x

(21)

7
(21) can be simplified by considering the following:
1. The operations of taking averages and differentiation commute.
2.

= (22)
xj rj

3. ui is independent of x and uj is independent of x.

and using the definition of Sijk from equation 14, a more simplified form is obtained:

Qij
= [Sikj + Sjki ] + 22 Qij (23)
t rk

The pressure terms have been dropped as the assumption of isotropic turbulence and mass con-

tinuity (eqn. 7) demand that ui p = 0
Substitution of Qij and Sijk from eqns. 17 and 18 yield :

2 4
[u r f (r, t)] = u3 [r4 K(r)] + 2u2 [r4 f (r)] (24)
t r r
This is the Karman Howarth equation. The problem with this equation is that we cannot predict
the evolution of f without knowing the rate of change of K(r). , which depends on the fourth-
order correlations, i.e, we have a closure problem.

3 Kolmogorovs 4/5th law

This law, which is an extremely important result in turbulence, states that [v]3 is equal to
(4/5)r in the inertial sub-range.

Now, by using eqns 12 and 16 and writing the KH equation in terms of [v(r)]2 and [v(r)]3
, we have:

2 r4 [ r4 ] [ 4 ]
r4 [v(r)]2 = [v(r)]3 r [v(r)]2 (25)
3 2 t r 6 r r
Now, in order to simplify eqn. 25, we need to use an estimate of the second order structure
2
function in the inertial sub range, and we know that it is [v(r)] = 2/3 r2/3 or less and
based on this , the second term in eqn. 25 , is at most of the order of r4 (r/l)2/3 and assuming
that r l, we can neglect the second term and hence eqn. 25 reduces to

4
[v(r)]3 = r + 6 [v(r)]2 (26)
5 r
In the inertial sub-range, viscous effects are unimportant and so eqn.26 reduces to

4
[v(r)]3 = r (27)
5

8
This is Kolmogorovs famous four-fifths law.

4 Kolmogorovs ideas extended to passive scalars dynamics: Yagloms 4/3rd law

In such a case we might be interested in the spatial structure of the concentration field and the
time required to achieve near perfect mixing.
The passive scalars concentration C is governed by the advection-diffusion equation

C
+ u.C = 2 C (28)
t
Just like the last section, it is supposed that u and C have zero mean and that their distributions
are statistically homogeneous and isotropic.
A convenient measure of the non-uniformity of the contaminant is provided by the variance of
C, defined as C 2 . We can get an expression for the rate of change of the variance from the
advection-diffusion equation. Multiplying eqn 28 by C, we get:

[ 1 2] [1 ]
C + . C 2u = .[CC] (C)2 (29)
t 2 2
Taking the ensemble average of eqn. 29 makes the second and third terms ( ones with diver-
gences) disappear due to homogeneity.
The following equation is obtained:

d 1 2
C = (C)2 (30)
dt 2

Therefore, the fluctuations in contaminant are destroyed by diffusion at a rate proportional

to

c = (C)2 (31)

We can define c as the characteristic length-scale of the most rapid spatial fluctuations in C.
This is the analogue of the Kolmogorov microscale for u
Also, consider the following :

max = max[, c ] r min[l, lc ] = lmin (32)

This defines the inertial-convective subrange.The name derives from the fact that the restriction
r guarantees the dominance of inertia over viscous forces, while the requirement that
r c ensuresthat the convection of C is much greater
diffusion. Since max[, c ] r ,
than
we would expect that neither nor will influence [C]2 in the inertial-convective subrange.

On the other hand, the restriction r min[l, lc ] = lmin suggests that [C]2 depends on
the large scales only to the extent that they determine the flux of energy and scalar variance

9
from large to small scale, that is, they determine and c . Therefore, in a reasoning close to
Kolmogorovs, we might expect that, in the inertial-convective subrange,

[C]2 = f (, c , r) (33)

In order to find a law equivalent to the 4/5th law, we need to determine the governing equation
for the two point correlations (x)C(x + r) = CC and therefore, we have:

C
= .[uC] + 2 C (34)
t

C 2
= .[u C ] + C (35)
t
where a prime indicates a quantity evaluated at x = x + r .
On multiplying the first by C , and the second by C, adding the two and averaging, we ob-
tain



C C = C (ui C) + C (ui C ) + C2x C + C x 2 C (36)
t xi xi

Since the operations of averaging and differentiation commute, C is independent of x and C

independent of x , and x i and x operating on averages may be replaced by - r i and r i ,
i
respectively. The equation for CC now becomes

C C = (ui ui )C C + 2r 2 CC (37)
t ri

The isotropic tensor (ui ui )C C can be written as



(ui ui )C C = A(r)ri (38)

where A(r) is given by

rA(r) = u|| CC (39)


here, u|| is the component of u u parallel to r.

therefore, we have

1 d
(ui ui )C C = rA (r) + 3A = 2 (r3 A)

(40)
ri r dr

Using eqns 35 and 38, we have:

10
 2
[r C C] = [r2 u|| C C ] + 2 r2 C C (41)
t r r r

If is any scalar , then u = 0
This was done in section 2 for the pressure term.
Therefore , we have

u|| (C)2 = 2 u|| CC (42)

Therefore eqn. 39 can be written as

1 2[ 1
]
C C = 2 r u|| (C ) + 2 C C
2
(43)
t r r 2 r
When
working
in the inertial-convective subrange , It is convenient to use the structure function
C ) rather than CC
2

They are related by

C 2 ) = 2 C 2 2 CC (44)

In the universal equilibrium range, r l, eqn. 43 can be written as

2 1 2
C C = C C 2 C = 2c (45)
t t 2 t

The approximation is valid as the time derivative of C 2 is much less than that of C 2
Thus, in the equilibrium range (r l) eqn. 43 integrates to

4
u|| (C)2 2 (c)2 = c r (46)
r 3
In the inertial-convective subrange, where diffusion is negligible, this simplifies to

4
u|| (C)2 = c r (47)
3
This is the Yagloms Law for a passive scalar.

11
References

[1] P. A. Davidson. Turbulence: an introduction for scientists and engineers. Oxford University
Press, Oxford, UK ; New York, 2004.
[2] Uriel Frisch and Andre Nikolaevich Kolmogorov. Turbulence: The Legacy of A. N. Kol-
mogorov. Cambridge University Press, November 1995.
[3] G.K. Batchelor. Theory of Homogeneous Turbulence. Cambridge University Press, May
1982.

12