Essentials of Turbulence

ME 724
Spring 2019

Module 04

Dr. Abhilash J. Chandy

February 12, 2019

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 1 / 41

Part 1: The scales of turbulent motion

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 2 / 41
 Introduction

Turbulent motions range in size from the width of the flow δ to much smaller scales, which
become progressively smaller as the Reynolds number increases.
The idea of the energy cascade (introduced by Richardson (1922)) is that the kinetic energy
enters the turbulence (through the production mechanism) at the largest scales of motion.
This energy is then transferred (by inviscid processes) to smaller and smaller scales until, at
the smallest scales, the energy is dissipated by viscous action.
Kolmogorov (1941) added to and quantified this picture by identifying the smallest scale of
turbulence to be those that now bear his name.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 3 / 41

Energy cascade

Turbulent flow at high-Re with characteristics U and length scale L and Re = U L/ν.
According to Richardson’s view of the energy cascade, turbulence is considered to be
composed of eddies of different sizes. Eddies of size l have a characteristic velocity u(l) and
timescale τ (l) ≡ l/u(l)
An eddy is conceived to be a turbulent motion, localized within a region of size l, that is at
least moderately coherent over the region.
The eddies in the largest size range are characterized by the length scale l0 , which is
comparable to L and their characteristics velocity u0 ≡ u(l0 ) is on the order of the r.m.s
turbulence intensity u0 ≡ ( 23 k)1/2 , which is comparable to U . The Re of these eddies
Re0 ≡ u0 l0 /ν is therefore large (i.e. comprable to Re) so the direct effects of viscosity are
negligibly small.
Richardson’s notion is that the large eddies are unstable and breakup, transferring their
energy to somewhat smaller eddies, which undergo a similar breakup process and transfer
their energy to even smaller eddies - the energy cascade.
The energy cascade continues until the Re is sufficiently small that the eddy motion is
stable, and molecular viscosity is effective in dissipating the kinetic energy.
The rate of dissipation,  is determined by the transfer of energy from the largest of eddies
(the first process in the sequence) and scales according to u20 /τ0 = u30 /l0 . This is consistent
with experimental observations in free shear flows where  is independent of ν.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 4 / 41
 Kolmogorov hypotheses - 1

What is the size of the smallest eddies that are responsible for dissipating the energy? As l
decreases, do the characteristic velocity and times scales u(l) and τ (l) increase, decrease or
remain the same?
Kolmogorov formulated three hypotheses based on the theory that the velocity and
timescales u(l) and τ (l) decrease as l decreases.
(1) In general, the large eddies are anisotropic and are effected by the BC of the flow;
Kolmogorov argued that the directional biases of the large scales are lost in the chaotic
scale-reduction process, by which energy is transferred to successively smaller and smaller
eddies.
Kolmogorov’s hypothesis of local isotropy: At sufficiently high-Re, the small-scale
turbulent motions (l  l0 ) are statistically isotropic.
Introducing a length scale lEI (with lEI ≈ 16 l0 , say) as the demarcation between the
anisotropic large eddies (l > lEI ) and the isotropic small eddies (l < lEI )
Kolmogorov argued that by the same reasoning as presented above, the geometry of the
large eddies (determined by the mean flow and the boundary conditions) is also lost. This
means the statistics of the small-scale motions are in a sense universal - similar in every
high-Re turbulent flow.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 5 / 41

Kolmogorov hypotheses - 2

(2) On what parameters, does this statistically universal state depend? The two dominant
processes are the transfer of energy to successively smaller scales (rate at which the small
scales receive energy from the large scales, TEI ), and the viscous dissipation (kinematic
viscosity, ν).
In fact the dissipation rate  is determined by the energy transfer rate TEI implying  ≈ TEI
The answer to the above question lies in the second hypothesis
Kolmogorov’s first similarity hypothesis: In every turbulent flow at sufficiently high-Re,
the statistics of the small-scale motions (l < lEI ) have a universal form that is uniquely
determined by ν and .
The Kolmogorov scales are defined as

η ≡ (ν 3 /)1/4 uη ≡ (ν)1/4 τη ≡ (ν/)1/2

Two identities can be derived from the above relations: (1) Reη = ηuη /ν = 1, which is
consistent with the notion that cascade proceeds to smaller and smaller scales until the Re is
small enough for dissipation to be effective. (2)  = ν(uη /η)2 = ν/τη2 showing that
(uη /η) = 1/τη provides a consistent characterization of the velocity gradients of the
dissipative eddies - Both these identities reveal that Kolmogorov scales characterize the very
smallest dissipative eddies.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 6 / 41
 Kolmogorov hypotheses - 3

Consider a point x0 in a high-Re turbulent flow at time, t0 . In terms of the Kolmogorov

scales at (x0 , t0 ), non-dimensional coordinates are defined by y ≡ (x − x0 )/η, and the
non-dimensional velocity-difference field is defined by w(y) ≡ [U (x, t0 ) − U (x0 , t0 )]/uη .
According the hypotheses stated above, when the w(y) is examined on a not-too-large a
scale (specifically |y| < lEI /η), it is statistically isotropic and statistically identical at all
points (x0 , t0 ) in all high-Re flows.
On the small scales, all high-Re turbulent velocity fields are statistically similar; i.e. they are
statistically identical when they are scaled by the Kolmogorov scales.
The ratios of the smallest to largest scales are readily determined from the definitions of the
Kolmogorov scales and from the scaling  ∼ u30 /l0 . So

η/l0 ∼ Re−3/4 uη /u0 ∼ Re−1/4 τη /τ0 ∼ Re−1/2

So obviously, the velocity scales and timescales of the smallest eddies are small compared
with those of the largest eddies; Also, the ratio η/l0 decreases with increasing Re.
(3) As a consequence, at sufficiently high-Re, there is a range of scales l that are very small
compared with l0 and yet very large compared with η, i.e. l0  l  η. Since eddies are
much bigger than the dissipative eddies, it may be supposed that that their Re is large, and
hence their motion is little affected by viscosity.
Kolmogorov’s second similarity hypothesis: In every turbulent flow at sufficiently high-Re,
the statistics of the motions of scale l in the range l0  l  η have a universal form that
is uniquely determined by  and independent of ν.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 7 / 41

Kolmogorov hypotheses - 4
!
!
!
!
Energy7containing!
Universal!equilibrium!range! range!

Dissipation!range! Inertial!subrange!

η! lDI ! lEI! l0! L!

Eddy!sizes!l!at!very!high7Re!showing!various!length!scales!

It is convenient to introduce a length scale lDI (with lDI = 60η, say) so that the range in
the above hypothesis canDissipation,!
be writtenε!
lEI > l > lDI . This length scale lDI splits the
Production,!P!
universal equilibrium range (l < lEI ) into 2 subranges: the inertial subrange (lEI > l > lDI )
and the dissipation range (l < lDI ). T(l)!
Transfer!of!energy!to!
The bulk of the energy is contained insuccessively!smaller!scales!
the larger eddies in the size range
lEI = 61 l0 < l < 6l0 , which is therefore called the energy-containing range.
η! l ! DI l ! l! L! EI 0
Given an eddy size l (in the inertia
Dissipation!
subrange), characteristic
Inertial!subrange!
velocity
Energy7containing!
scales and timescales for
the eddy are those formed range!
from  and l. range!

u(l) = (l)1/3 = uη (l/η)1/3 ∼ u0 (l/l0 )1/3

Energy!cascade!at!very!high7Re!!

τ (l) = (l2 /)1/3 = τη (l/η)2/3 ∼ τ0 (l/l0 )2/3

A consequence of the second similarity hypothesis is that the velocity scales and timescales
u(l) and τ (l) decrease as l decreases.
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 8 / 41
 Eddy!sizes!l!at!very!high7Re!showing!various!length!scales!

Kolmogorov hypotheses - 5
Dissipation,!ε! Production,!P!

T(l)!
Transfer!of!energy!to!
successively!smaller!scales!

η! lDI! lEI! l0! L!

Dissipation! Inertial!subrange! Energy7containing!
range! range!

Energy!cascade!at!very!high7Re!!

In the conception of the energy cascade, T (l) is the rate at which energy is transferred from
eddies larger than l to those smaller than l. if this transfer process is accomplished primarily
by eddies of size comparable to l, then T (l) can be expected to be of the order u(l)2 /τ (l).
Also from the above equations u(l)2 /τ (l) = , which means T (l) = .
So
TEI ≡ T (lEI ) = T (l) = TDI ≡ T (lDI ) = 
for lEI > l > lDI , i.e. the rate of energy transfer from the large scales, TEI , determines the
constant rate of energy transfer through the inertial subrange, T (l); hence the rate at which
energy leaves the inertial subrange and enters the dissipation range, TDI ; hence the
dissipation range .

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 9 / 41

CHAPTER 6: THE SCALES OF TURBULENT MOTION

Turbulent Flows
The energy spectrum - 1
Stephen B. Pope
Cambridge University Press, 2000
c
!Stephen B. Pope 2000 How is tke distributed among the eddies of
different sizes?
8
10
Motions of length scale l correspond to
E11(!1)/("#5)1/4

7
10
6 wavenumber κ = 2π/l and that the energy in
10
10
5 the wavenumber range (κa , κb ) is
4
10 Z κb
3
10 k(κa ,κb ) = E(κ)dκ
2
10 U&F 1969, wake 23 κa
1 U&F 1969, wake 308
10 CBC 1971, grid turb. 72

10
0 Champagne 1970, hom. shear 130
S&M 1965, BL401 Also the dissipation rate,  in the range (κa , κb )
Laufer 1954, pipe 170

10
-1 Tielman 1967, BL 282
K&V 1966, grid turb. 540
is Z κb
-2 K&A 1991, channel 53
10
-3
CAHI 1991, return channel 3180
Grant 1962, tidal channel 2000
(κa ,κb ) = 2νκ2 E(κ)dκ
10 Gibson 1963, round jet 780
C&F 1974, BL 850
κa
-4
10 Tielman 1967, BL 23

10
-5
CBC 1971, grid turb. 37
S&V 1994, BL 600 From Kolmogorov’s first similarity hypothesis in
S&V 1994, BL 1500

10
-6
-6 -5 -4 -3 -2 -1 0 1
the universal equilibrium range
10 10 10 10 10 10 10 10
(κ > κEI ≡ 2π/lEI ), the spectrum is a
!1$
universal function of  and ν.
Figure 6.14: Measurements of one-dimensional longitudinal velocity From the second hypothesis, it follows that in
spectra (symbols), and model spectra (Eq. 6.246) for Rλ =
30, 70, 130, 300, 600 and 1500 (lines). The experimental data are
the inertial range (κEI < κ < κDI ≡ 2π/lDI ),
taken from Saddoughi and Veeravalli (1994) where references to the spectrum is
the different experiments are given. For each experiment, the final
number in the key is the value of Rλ. E(κ) = C2/3 κ−5/3

where C is a universal constant.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 10 / 41
 The energy spectrum - 2

Consider a general power-law spectrum E(κ) = Aκ−p , where A and p are constants.
The energy contained in the wave numbers greater than κ is
Z ∞
A
k(κ,∞) ≡ E(κ0 )dκ0 = κ−(p−1)
κ p−1

for p > 1, while the integral diverges for p ≤ 1. Similarly the dissipation in wave numbers
less than κ is Z κ
2νA 3−p
(0,κ) ≡ 2νκ02 E(κ0 )dκ0 = κ
0 3−p
for p < 3, while the integral diverges for p ≥ 3. Thus p = 53 , corresponding to the
Kolmogorov spectrum, is around the middle of the range (1,3) for which the integrals k(κ,∞)
and (0,κ) converge.
5
With p = 3
, k(κ,∞) ∼ κ−2/3 and (0,κ) ∼ κ4/3
Although the Kolmogorov −5/3 spectrum applies only to inertial range, the observations
made are consistent with the notion that the bulk of the energy is in the large scales (l > lEI
or κ < 2π/lEI ) and the bulk of the dissipation is in the small scales (l < lDI or κ > 2π/lDI )

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 11 / 41

Restatement of the Kolmogorov hypotheses

(in terms of N -point distribution in the 4D (x,t) space

Consider a simple domain G within the turbulent flow, and let x(0) , x(1) , ..., x(N ) be a
specified set of points within G. New coordinates and velocity differences are defined by

y ≡ x − x(0) , v(y) ≡ U (x, t) − U (x(0) , t)

and the joint PDF of v at the N points y (0) , y (1) , ..., y (N ) is denoted by fN .
Local homogeneity definition: The turbulence is locally homogeneous in the domain G. If
for every fixed N and y (n) (n = 1, 2, ..., N ), the N -point PDF fN is independent of x(0)
and U (x(0) , t).
Local isotropy definition: The turbulence is locally isotropic in the domain G is it is locally
homogeneous and if the addition the PDF fN is invariant with respect to rotations and
reflections of the coordinate axes.
Local isotropy hypothesis: In any turbulent flow with a sufficiently large Re = UνL , the
turbulence is to a good approximation locally isotropic if the domain G is sufficiently small
(i.e. |y (n)  L for all n) and is not near the boundary of the flow or it other singularities.
First similarity hypothesis: For locally isotropic turbulence, the N -point PDF fN is
uniquely determined by viscosity ν and the dissipation rate .
Second similarity hypothesis: If the moduli of the vectors y (m) and of their differences
y (m) − y (n) (m 6= n) are large compared with the Kolmogorov scale, η, then the N -point
PDF fN is uniquely determined by  and does not depend on ν.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 12 / 41
 Fourier modes

The Karman-Howarth equation is something that is derived from the NS equations and is a
governing equation for the longitudinal autocorrelation function, f (r, t). This equation while
fully describing the dynamics of the two-point velocity correlation does not provide a very
clear picture of the energy cascade.
For this purpose it is helpful to examine the NS equations in wavenumber space. So the
velocity field as the 3D Fourier series is given by

eiκ·x û(κ, t)
X
u(x, t) =
κ

Using the above definition, the NS equations will result in the equations governing the
evolution of the Fourier modes.
Finally the balance equation for the kinetic energy at wavenumber κ can be derived as

Ê(κ, t) = uˆ∗ (κ, t) · û(κ, t)



DNS of homogeneous turbulence are usually performed in wavenumber space (due to zero
truncation errors)

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 13 / 41

Fourier-series representation - 1
An implication of the Fourier series is that the turbulent velocity field is periodic.
Consider a cube 0 ≤ xi ≤ L in physical space, where the length of the side L is large
compared with the turbulent integral scale L11 .
The velocity field is supposed to be periodic, i.e.
u(x + N L, t) = u(x, t)
for all integer vectors N .
In the x1 direction, the Fourier modes are of the complex form
eiκ0 n1 x1 = cos(κ0 n1 x1 ) + i sin(κ0 n1 x1 )
for +ve and -ve integers n1 and where κ0 is the lowest wavenumber κ0 ≡ 2π/L
By defining the wavenumber vector κ = κ0 n = κ0 (e1 n1 + e2 n2 + e3 n3 ) and the Fourier
mode can be written as eiκ·x = eiκ0 n1 x1 eiκ0 n2 x2 eiκ0 n3 x3 . This Fourier mode can be
interpreted in terms of the magnitude κ ≡ |κ| and direction e ≡ κ/κ
Given two wavenumber vectors κ and κ0 , we define
if κ = κ0

1
δκ, κ0 =
0 if κ 6= κ0

Second we denote h...iL as the volume average over the cube 0 ≤ xi ≤ L. The
orthonormality property is then
0
D E
eiκ·x e−iκ ·x = δκ, κ0
L

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 14 / 41
 Fourier-series representation - 2
For a periodic function g(x) (e.g. a component of velocity at any given time) its Fourier
series is
eiκ·x ĝ(κ)
X
g(x) =
κ
where the sum is over the infinite number of discrete wavenumbers κ = κ0 n and ĝ(κ) is the
complex Fourier coefficient at wavenumber κ. Since g(x) is real, ĝ(κ) satisfies conjugate
symmetry,
ĝ(κ) = ĝ ∗ (−κ)
where the asterisk denotes the complex conjugate.
Given g(x), the Fourier coefficients can be determined from the orthogonality condition
* +
0 0
D E
g(x)e−iκ ·x ĝ(κ)eiκ·x e−iκ ·x
X
=
L
κ L
X
= ĝ(κ)δκ, κ0 = ĝ(κ0 )
κ
So it is convenient to define the operator Fκ { } by
Fκ {g(x)} = g(x)e−iκ·x L


Z LZ LZ L
1
= 3 g(x)e−iκ·x dx1 dx2 dx3
L 0 0 0
= ĝ(κ)

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 15 / 41

Fourier-series representation - 3
Thus the operator Fκ { } determines the coefficient of the Fourier mode
One of the principal reasons for invoking the Fourier representation is the form taken by
derivatives. So
   
∂g(x) ∂g(x) −iκ·x
Fκ = e
∂xj ∂xj L
 
∂ −iκ·x
= −g(x) e
∂xj L
= iκj g(x)e−iκ·x L



= iκj ĝ(κ)
Differentiation w.r.t xj in physical space corresponds to multiplication by iκj in wavenumber
space.
The Fourier series of the turbulent velocity field is

eiκ·x û(κ, t)
X
u(x, t) =
κ
where the Fourier coefficients of velocity are
ûj (κ, t) = Fκ {uj (x, t)}
The Fourier modes eiκ·x are non-random and fixed in time. This means that the Fourier
coefficients û(κ, t) are time-dependent and random (Since the velocity field u(x, t) is
random). Since the mean hu(x, t)i is zero, the means hû(κ, t)i are zero. For each κ,
û(κ, t) is a complex vector that satisfies conjugate symmetry, i.e. û(κ, t) = û∗ (−κ, t)
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 16 / 41
 The evolution of Fourier modes: Projection tensor
!
κ2" In wavenumber space, the divergence of
velocity is
! 
∂uj

Fκ = iκj ûj = iκ · û
! ∂xj
! κ!
so that the continuity equation ∇ · u = 0
indicates that û is normal to κ: κ · û = 0
κ1"
Lets examine in more detail the orientation of a vector (such as û) relative to the
k
wavenumber κ. Any vector Ĝ can be decomposed into a component Ĝ that is parallel to
⊥ k ⊥
κ, and a component Ĝ , that is normal to κ, i.e. Ĝ = Ĝ + Ĝ . With e = κ/κ being the
unit vector in the direction of κ, we have
k k κj κk
Ĝ = e(e · Ĝ) = κ(κ · Ĝ)/κ2 or Ĝ = Ĝ
κ2
⊥ k
Hence from Ĝ = Ĝ − Ĝ
⊥
Ĝ = Ĝ − κ(κ · Ĝ)/κ2 or Ĝ⊥ = Pjk Ĝ
where the projection tensor Pjk (κ) is
κj κk
Pjk ≡ δjk −
κ2
The principal observation is that the projection tensor Pjk (κ) determines Ĝ⊥ = Pjk Ĝ to be
the projection of Ĝ onto the plane normal to κ.
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 17 / 41

The evolution of Fourier modes: Navier-Stokes - 1

The equation for the evolution of the velocity vector in wavenumber space û(κ, t) is
obtained by applying the operator F{ } to the NS equations:
∂uj ∂(uj uk ) ∂ 2 uj 1 ∂p
+ =ν −
∂t ∂xk ∂xk ∂xk ρ ∂xj
The time derivative is simply  
∂uj dûj
= Fκ
∂t dt
For the viscous term, the rule for the Fourier transform of the derivative can be used to derive
∂ 2 uj
 
Fκ ν = −νκ2 ûj
∂xk ∂xk
With p̂(κ, t) ≡ Fκ {p(x, t)/ρ} being the Fourier coefficient of the dynamic pressure, p/ρ, the
pressure-gradient term is  
1 ∂p
Fκ = −iκj p̂
ρ ∂xj
For now the nonlinear convection term is written as
 
∂(uj uk )
Fκ = Ĝj (κ, t)
∂xk
So the NS equations with these definitions are now
dûj
+ νκ2 ûj = −iκj p̂ − Ĝj
dt
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 18 / 41
 The evolution of Fourier modes: Navier-Stokes - 2
When the last equation shown previously is multiplied by iκj (or taking the divergence), the
L.H.S vanishes on account of continuity (iκj ûj = 0) resulting in
κ2 p̂ = iκj Ĝj
The above equation is in wavenumber space the Fourier transform of the Poisson equation
for pressure (obtained from the NS equations)
  
∂ ∂(uj uk )
Fκ −∇2 p = Fκ

∂xj ∂xk
By solving the above equation for p̂, we obtain for the pressure term
κj κk k
−iκj p̂ = Ĝk = Ĝj
κ2
k
i.e. the pressure term exactly balances −Ĝ , the component of −Ĝ in the direction of κ.
What remains, then on the RHS of the NS equation (last equation in previous slide) is
⊥
−Ĝ , the component of −Ĝ perpendicular to κ:
dûj  κj κk 
+ νκ2 ûj = − δjk − Ĝk
dt κ2
= −Pjk Ĝk = −Ĝ⊥j
The viscous term has a simple effect. Consider for example the final period of decay of
isotropic turbulence in which the Re is so low that convection is negligible relative to the
effects of viscosity. Then from a specified initial condition, û(κ, 0), with neglect of the term
⊥
Ĝ has the solution 2
û(κ, t) = û(κ, 0)e−νκ t

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 19 / 41

The evolution of Fourier modes: Navier-Stokes - 3

Thus in the final period of decay, each Fourier coefficient evolves independently of all other
modes, decaying exponentially with time at the rate νκ2 . High-wavenumber modes decay
more rapidly than do low-wavenumber modes.
Expressed in terms of û(κ), the nonlinear convective term is
 
∂(uj uk )
Ĝj (κ, t) ≡ Fκ = iκk Fκ {uj uk }
∂xk
( ! !)
0 00
û (κ0 )eiκ ·x û (κ00 )eiκ ·x
X X
= iκ F k κ j k
κ0 κ00
i(κ0 +κ00 )·x −iκ·x
XX D E
= iκk ûj (κ0 )ûk (κ00 ) e e
L
κ0 κ00
XX X
= iκk ûj (κ0 )ûk (κ00 )δκ,κ0 +κ00 = iκk ûj (κ0 )ûk (κ − κ0 )
κ0 κ00 κ0
Substituting this relationship into the NS equations results in the final form of the evolution
equation for û(κ, t):
dûj X
+ νκ2 ûj = −iκl Pjk (κ) ûj (κ0 )ûl (κ − κ0 )
dt
κ0
The LHS involves û only at κ. In contrast, the RHS involves û at κ and κ00 such that
κ0 + κ00 = κ; and in fact the contributions of κ0 = κ and κ00 = κ are zero. Thus in
wavenumber space the convection term is nonlinear and non-local, involving the interaction
of wavenumber triads, κ, κ0 , and κ00 such that κ0 + κ00 = κ
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 20 / 41
 The kinetic energy of Fourier modes
So for the periodic case being considered the above equation is the NS equations in
wavenumber space. It is a deterministic set of ODEs for the Fourier coefficients û(κ, t). In
order to describe the turbulence statistically, one of the features we look at is the kinetic
energy.
Lets assume the mean velocity is zero as before, i.e. hU (x, t)i is zero everywhere implying
the Fourier coefficients of hU (x, t)i, i.e. hû(κ, t)i are also zero.
The next simplest statistic is the covariance of the two Fourier coefficients,
hûi (κ0 , t)ûj (κ, t)i. These coefficients are uncorrelated unless κ0 + κ = 0 or κ0 = −κ. Thus
all the covariance information is contained in
R̂ij (κ, t) = hûi (−κ, t)ûj (κ, t)i
= hû∗i (κ, t)ûj (κ, t)i
Of particular interest is the kinetic energy of the Fourier mode, defined as
1 1
Ê(κ, t) = R̂ii (κ, t) = hû∗i (κ, t)ûi (κ, t)i
2 2
and the turbulent kinetic energy (tke) is
1 X1 X
k(t) = hui ui i = R̂ii (κ, t) = Ê(κ, t)
2 κ 2 κ
The dissipation rate (t) is also related to Ê(κ, t) by
X
(t) = 2νκ2 Ê(κ, t)
κ
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 21 / 41

Velocity spectrum tensor

In homogeneous turbulence, the two-point velocity correlation and the velocity-spectrum
tensor form a Fourier-transform pair:
Z∞
ZZ Z∞
ZZ
1 1
Φij (κ) = Rij (r)e−iκr Rij (r) = Φij (κ)eiκr
2π 3 2π 3
−∞ −∞

Here κ = {κ1 , κ2 , κ3 } is the continuous wavenumber vector.

The velocity spectrum tensor Φij (κ) is a useful quantity to consider because it represents
the Reynolds-stress density in wavenumber space, i.e. Φij (κ) is the contribution (per unit
volume in wavenumber space) from the Fourier mode eiκ·x to the Reynolds stress hui uj i.
So setting r = 0 for the covariance we obtain
Z∞
ZZ
Rij (0) = hui uj i = Φij (κ)dκ
−∞

(Note that Φij has dimensions of (velocity)2 /(wavenumber)3 .

The information contained in Φij (κ) can be considered in 3 parts: (1) The subscripts i and
j give the directions of the velocity in physical space (2) The wavenumber direction, κ/|κ|
gives the direction in physical space of the Fourier mode and (3) The wavenumber’s
magnitude determines the lengthscale of the mode, i.e. l = 2π/|κ|.
And the dissipation rate is
Z∞
ZZ
= 2νκ2 Φii (κ)dκ
−∞

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 22 / 41
 Energy spectrum function
Being a second-order tensor function of a vector Φij (κ) contains a great deal of information.
A simpler though less complete description is provided by the energy-spectrum function
E(κ), which is a scalar function of a scalar.
The energy-spectrum function is obtained from Φij (κ) by removing all directional
information about the velocities by considering (half) the trace, i.e. 21 Φii (κ). The
information about the direction of the Fourier modes is removed by integrating over all
wavenumbers κ of magnitude |κ| = κ.
To express this mathematically, we denote by S(κ) the sphere in wavenumber space,
centered at Hthe origin and with a radius κ; and integration over the surface of this sphere is
denoted by ( )dS(κ). This ie the energy spectrum is defined as
I
1
E(κ) = Φii (κ)S(κ)
2
The properties of E(κ) follow straightforwardly from those of Φij (κ). E(κ) is real,
non-negative and for negative κ, it is undefined or zero. So integration of E(κ) over all κ is
the same as integration of 12 Φii over all κ. Thus
Z ∞ Z ∞
k= E(κ)dκ  = 2νκ2 E(κ)dκ
0 0

E(κ)dκ is the contribution to k from all wavenumbers κ in the infinitesimal shell

κ ≤ |κ| < κ + dκ in wavenumber space.
In general, Φij (κ) contains more information than does E(κ); but in isotropic turbulence,
Φij (κ) is completely determined by E(κ).
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 23 / 41

One-dimensional spectra
In isotropic turbulence, the 1D spectra, E11 (κ1 ) are related to the longitudinal
autocorrelation function and are determined by the energy-spectrum function, E(κ).
Z∞
κ2
 
E(κ)
E11 (κ1 ) = 1 − 12 dκ
κ κ
κ1

The above formula can be inverted to obtain E(κ) (for isotropic turbulence)
 
1 d 1 dE11 (κ)
E(κ) = κ3
2 dκ κ dκ
With κ1 = 0, the longitudinal integral scale is given by
πE11 (0)
L11 =
2 u21



The transverse 1D spectra can be determined from the longitudinal counterpart:

 
1 dE11 (κ1 )
E22 (κ1 ) = E11 (κ1 ) − κ1
2 dκ1
In examining the Kolmogorov hypotheses, the 1D spectra is of a power-law form
E11 (κ1 ) = C1 Aκ−p 1

where C1 is a constant and A is a normalization factor (e.g. A = u21 L1−p


11 ). So it follows
CHAPTER 6: THE SCALES OF TURBULENT MOTION
that E(κ) = CAκ−p with C = 21 p(2 + p)C1 and
Turbulent Flows
Stephen B. Pope 1
E22 (κ1 ) = C10 Aκ−p
1 where
Cambridge University Press, 2000
c
!Stephen B. Pope 2000
C10 = (1 + p)C1 .
2
Dr. Abhilash J. Chandy Essentials
1.0 of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 24 / 41
 CHAPTER 6: THE SCALES OF TURBULENT MOTION

Turbulent Flows
Stephen B. Pope
Comparison of spectra
Cambridge University Press, 2000
c
!Stephen B. Pope 2000

In the center of the wavenumber range, all

E ( κ ), E 11 (κ 1 ), E 22 (κ 1 )

the spectra exhibit power-law behavior with

p = 53 . So consistent with the equations
slope —5/3 above, E11 , E22 and E are in ratios
1 : 34 : 55
18
.
At high wavenumber, the spectra decay
more rapidly than a power of κ.
At low-wavenumber, E(κ) tends to zero as
κ2 . In contrast, the 1D spectra are
maximum at zero wavenumber. This again
illustrates the fact that the 1D spectra
contain contributions from wavenumbers κ
100 101 102 103 104 greater than κ1 .
κ, κ1 At low wavenumber the 1D spectra E11
Figure 6.11: Comparison of spectra in isotropic turbulence at R λ = 500:
and E22 are in the ratio 2 : 1 - consistent
solid line, E(κ); dashed line, E11(κ1); dot-dashed line, E22(κ1). with the ratios of the integral length scales
From the model spectrum, Eq. (6.246). (Arbitrary units.)
L11 and L22 .

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 25 / 41

4
Kolmogorov spectra - 1
According to the Kolmogorov’s hypotheses in any turbulent flow at sufficiently high-Re, the
high-wavenumber portion of the velocity spectra adopts particular universal forms.
For any turbulent flow at sufficiently high-Re the Kolmogorov hypotheses apply to the
velocity field on small length-scales, specifically in the universal equilibrium range defined by
l < lEI , which corresponds to κ > κEI ≡ 2π/lEI
According to the hypothesis of local isotropy, velocity statistics pertaining to the universal
equilibrium range are isotropic and this applies to κ > κEI .
According to the first similarity hypothesis, velocity statistics pertaining to the universal
equilibrium range have a universal form that is uniquely determined by  and ν. Hence for
κ > κEI , E(κ) is a universal function of κ,  and ν.
So using  and ν to nondimensionalize κ and E(κ), we get

E(κ) = (ν 5 )1/4 ϕ(κη)

= u2η η ϕ(κη)
where ϕ(κη) is a universal non-dimensional function - the Kolmogorov spectrum function.
Alternatively if  and κ are used to non-dimensionalize E(κ), the relation is

E(κ) = 2/3 κ−5/3 Ψ(κη)

and these universal functions are related by
Ψ(κη) = (κη)5/3 ϕ(κη)
2πη
for κ > κEI , which corresponds to κη > lEI

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 26 / 41
 Kolmogorov spectra - 2
!
!
!
!
!

5#&'/6783#2)$#$#/! "#$%&'()*!&+,$*$-'$,.!')#/&!
')#/&!

4#&'2$)*!(,-')#/&! 0$(($1)2$3#!')#/&!

#(# #'&# #%&# #"#

!!"$(# !!"$'&# !!"$%&# !!""#

9)%&#,.-&'(!:3#!)!*3/)'$2;.$8!(8)*&<!)2!%&'6!;$/;!=&6#3*>(!#,.-&'!(;3?$#/!2;&!%)'$3,(!')#/&(!

The second similarity hypothesis applies to scales in the inertial subrange, i.e. η  l  l0 or
lDI < l < lEI or in wavenumber range it is κEI < κ < κDI or in terms of κη,
2πη 2πη
1  κη  η/l0 or κDI η = > κη > = κEI η
lDI lEI
In the inertial subrange, according to the second similarity hypothesis, E(κ) has a universal
form uniquely determined by , independent of ν. In the equation above for E(κ), Ψ(κη)
becomes independent of its argument and hence tends to a constants, C. So the second
similarity hypothesis predicts that in the inertial subrange, the energy-spectrum function is
E(κ) = C2/3 κ−5/3
This is the famous Kolmogorov − 35 spectrum, and C is a universal Kolmogorov constant.
Experimental data support the value C = 1.5. Consequently the 1D spectra are given by
−5/3 −5/3
E11 (κ1 ) = C1 2/3 κ1 E22 (κ1 ) = C10 2/3 κ1
where C1 = 18/55 C and C10 = 4/3 C1 .
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 27 / 41

CHAPTER 6: THE SCALES OF TURBULENT MOTION

Turbulent Flows A model spectrum - 1

Stephen B. Pope
Cambridge University Press, 2000
c
!Stephen B. Pope 2000

Lets introduce a model spectrum that can be used for

105 comparison.
4 slope 2 slope -5/3
10
3
E(!) 102 E(κ) = C2/3 κ−5/3 fL (κL)fη (κη)
2 10
" u" 101
0 where fL and fη are specified nondimensional
10
10
-1 functions. The function fL determines the shape of
10 -2 the energy-containing range, and tends to unity for
-3
10 -5 -4 -3 -2 -1 0
large (κL). Similarly, fη determines the shape of the
10 10 10 10 10 10 dissipation range, and it tends to unity for small κη.
!"
In the inertial subrange, both fL and fη are
Figure 6.13: Model spectrum (Eq. 2.246) for R λ = 500 normalized by

the Kolmogorov scales.

essentially unity, so the Kolmogorov − 35 spectrum
with constant C is recovered.
The specification of fL is
 5/3+p0
κL
fL (κL) =
[(κL)2 + cL ]1/2

where p0 is taken to be 2, and cL is a positive constant. Clearly fL tends to unity for large
κL, while the exponent 53 + p0 leads to E(κ) varying as κp0 = κ2 for small κL. (With the
alternative choice p0 = 4, the spectrum becomes the von Karman spectrum, which has
E(κ) ∼ κ4 for small κ.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 28 / 41
 A model spectrum - 2

The specification of fη is

fη (κη) = exp{−β{[(κη)4 + c4η ]1/4 − cη }}

where β and cη are positive constants. For cη = 0, this reduces to

fη (κη) = exp(−βκη)

Because the velocity field u(x) is infinitely differentiable, it follows that for large κ, the
energy-spectrum function decays more rapidly than any power of κ → Hence the exponential
decay. Several experiments support the exponential form with β = 5.2. However the simple
exponential departs from unity too rapidly for small κη, and the value of β is contrained to
be β ≈ 2.1. You can avoid this by using the first equation for fη (κη).
For specified value of k, , and ν, the model spectrum is determined by the above equations
with C = 1.5 and β = 5.2.
Alternatively, the non-dimensional spectrum is uniquely determined by a specified value of
Rλ . The constants cL and cη are determined the requirements that E(κ) and 2νκ2 E(κ)
integrate to k and , respectively: at high-Re their value are cL ≈ 6.78 and cη ≈ 0.40.
The figure in the previous slide is a log-log plot of the model spectrum for Rλ = 500. The
power laws E(κ) ∼ κ2 at low wavenumber and E(κ) ∼ κ−5/3 in the inertial subrange are
evident.

CHAPTER 6: THE SCALES OF TURBULENT MOTION

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 29 / 41
Turbulent Flows
Stephen B. Pope
Cambridge University Press, 2000 Dissipation Spectra - 1
c
!Stephen B. Pope 2000

8
10
E11(!1)/("#5)1/4

7
10
6
10
E11 (κ1 ) plotted with Kolmogorov scaling
5
10
10
4 As is the case with E(κ), the Kolmogorov
10
3 hypotheses imply that the scaled spectrum
10
2 ϕ11 = E11 (κ1 )/(ν 5 )1/4 is a universal
U&F 1969, wake 23

10
1 U&F 1969, wake 308 function of κ1 η at sufficiently high Re and
CBC 1971, grid turb. 72

10
0 Champagne 1970, hom. shear 130
S&M 1965, BL401
for κ1 > κEI
Laufer 1954, pipe 170

10
-1 Tielman 1967, BL 282
K&V 1966, grid turb. 540
For κ1 η > 0.1 all the data lie on a single
10
-2 K&A 1991, channel 53
CAHI 1991, return channel 3180
curve. The high-Re data exhibit power-law
10
-3 Grant 1962, tidal channel 2000
Gibson 1963, round jet 780 behavior for κ1 η < 0.1, the extend of power
C&F 1974, BL 850
10
-4 Tielman 1967, BL 23 law region generally increasing with Rλ .
CBC 1971, grid turb. 37
-5 S&V 1994, BL 600
10 S&V 1994, BL 1500 So the data are consistent with
-6
10
10
-6
10
-5
10
-4
10
-3
10
-2
10
-1 0
10 10
1 E11 (κ1 )/(ν 5 )1/4 being a universal
!1$ function of κ1 η for κ1 η > 0.1 with
departures from the universal behavior
Figure 6.14: Measurements of one-dimensional longitudinal velocity
arising from the energy-containing range
spectra (symbols), and model spectra (Eq. 6.246) for Rλ = κ1 η < 0.1 and the model spectra represent
30, 70, 130, 300, 600 and 1500 (lines). The experimental data are the data quite accurately as well.
taken from Saddoughi and Veeravalli (1994) where references to
the different experiments are given. For each experiment, the final
number in the key is the value of Rλ.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 30 / 41
 Turbulent Flows
Stephen B. Pope
Cambridge University Press, 2000 Dissipation Spectra - 2
c
!Stephen B. Pope 2000

1 E11("1) 100
Compensated 1D spectra (i.e.
5/3
−2/3 κ1 E11 (κ1 )) with Kolmogorov
scaling are shown on a linear-log plot, which
!-2/3"5/3

emphasizes the dissipation range.

For κ1 η > 0.1, there is close agreement
10-1 between measurements in grid turbulence
(Rλ = 60) and in a turbulent boundary
layer (Rλ = 600), again supporting the
universality of the high-wavenumber
spectra.
The straight-line behavior evident in this
10-2 plot for κ1 η > 0.3 corresponds to
exponential decay of the spectrum at the
0.0 0.2 0.4 0.6 0.8 1.0 highest wavenumbers. Furthermore, the
"1# model spectrum represents the data
accurately.
Figure 6.15: Compensated one-dimensional velocity spectra. Measure-
Also shown are 1D spectra deduced from
ments of Comte-Bellot and Corrsin (1971) in grid turbulence at
two alternative models for fη (κη), which do
Rλ ≈ 60 (triangles), and of Saddoughi and Veeravalli (1994) in a
turbulent boundary layer at Rλ ≈ 600 (circles). Solid line, model not match the data as well as the model
spectrum Eq. (6.246) for Rλ = 600; dashed line, exponential spec- spectrum does.
trum Eq. (6.253); dot-dashed line, Pao’s spectrum Eq. (6.254).
CHAPTER 6: THE SCALES OF TURBULENT MOTION
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 31 / 41
Turbulent Flows
Stephen B. Pope
Cambridge University Press, 2000
c
!Stephen B. Pope 2000
Dissipation Spectra - 3
2.0
1.8
7
D(!) 1.6
3 1.4
u" 1.2 Dissipation spectrum D(κ) = 2νκ2 E(κ)
1.0 and the cumulative dissipation
#(0,!) 0.8 (0,κ) ≡ 0κ D(κ0 )dκ0 .
R
# 0.6
0.4 Peak of the dissipation spectrum occurs at
0.2 κη ≈ 0.26 corresponding to l/η ≈ 24, while
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 the centroid (where (0,κ) = 21 ) occurs at
!" κη ≈ 0.34 (l/η ≈ 18).
50 20 10 5
l/" Thus the motions responsible for the bulk
of the dissipation (0.1 < κη < 0.75 or
Dissipation spectrum (solid line) and cumulative dissipa-
Figure 6.16: 60 > l/η > 8) are considerably larger than
tion (dashed line) corresponding to the model spectrum Eq. (6.246)
for Rλ = 600 : ! = 2π/κ is the wavelength corresponding to the Kolmogorov scale.
wavenumber κ.
The Kolmogorov hypotheses imply that the
characteristic size of the dissipative motions
scale with η, not that it be equal to η. So
Table: Charateristic wavenumbers and lengthscales of the dissipation everything is consistent with the previous
spectrum (based on the model spectrum) observation
Defining wavenumbers
8 κη l/η So the conclusion is that the demarcation
Peak of dissipation spectrum 0.26 24 lengthscale between inertial and dissipative
(0,κ) = 0.1 0.10 63 ranges to be lDI = 60η.
(0,κ) = 0.5 0.34 18
(0,κ) = 0.9 0.73 8.6

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 32 / 41
 Turbulent Flows
Stephen B. Pope
Cambridge University Press, 2000 The inertial subrange
c
!Stephen B. Pope 2000

(a)
0.8

j 1 E11(j 1)
0.6 C1
The second Kolmogorov hypothesis predicts
a − 53 spectrum in the inertial subrange.
–2/3 5/3
0.4

e 0.2
With regard to the compensated 1D spectra
5/3
0.0 –5
(−2/3 κ1 E11 (κ1 )), Kolmogorov
10 10–4 10–3 10–2 10–1 100
j 1g
hypotheses predicts that this quantity
(b)
0.8 adopts the constant value C1 in the inertial
j 1 E22(j 1)

C1!

0.6 range.
–2/3 5/3

0.4
Model spectrum shows good agreement,
e

0.2
which is expected since the former is
0.0 –5
10 10–4 10–3
j 1g
10–2 10–1 100 constructed to get good agreement in the
(c)
0.8
inertial subrange.
j 1 E33(j 1)

C1!

0.6 Figure also provides evidence of local

–2/3 5/3

0.4 isotropy in this distinctly anisotropic

turbulent flow: for κ1 η > 2 × 10−3 , E22
e

0.2

0.0 –5
and E33 are very similar and as predicted
10 10–4 10–3 10–2 10–1 100
j 1g
by local isotropy, the plateau value of their
Compensated one-dimensional spectra measured in a tur-
Figure 6.17: compensated spectra is 34 that of E11 .
bulent boundary layer at Rλ ≈ 1, 450. Solid lines, experimental A measured value of E22 (κ1 ) differed from
data Saddoughi and Veeravalli (1994); dashed lines, model spec-
the calculated value of E11 (κ1 ) (with the
tra from Eq. (6.246); long dashed lines, C1 and C1! corresponding
to Kolmogorov inertial-range spectra. (For E11, E22 and E33 the assumption of isotropy) by no more than
model spectra are for Rλ = 1, 450, 690 and 910, respectively, cor- 10%.
responding to the measured values of "u21#, "u22# and "u23#.)
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 33 / 41

CHAPTER 6: THE SCALES OF TURBULENT MOTION

The energy-containing range - 1
Turbulent Flows
Stephen B. Pope Unlike the universal equilibrium range, the
Cambridge University Press, 2000
9
c
!Stephen B. Pope 2000 energy-containing range depends on the
particular flow
Also, the 1D spectra provides little direct
0.25 information because E11 (κ1 ) contains
0.20 contributions from all wavenumbers of
E(!) magnitude greater than κ1 . E(κ) is really
kL11 0.15 the most informative quantity and for
isotropic turbulence, this can be obtained
0.10 experimentally by differentiating the 1D
0.05 spectra.
Appropriate scales for normalization are the
0.00
0 1 2 3 4 5 6 7 8 9 10 tke, k, and the longitudinal integral scale
!L11 L11 . Note in isotropic turbulence,
Z ∞ Z ∞
Energy spectrum function in isotropic turbulence nor-
Figure 6.18:
E(κ) 4
E(κ)dκ = k dκ = kL11
malized by k and L11. Symbols, grid-turbulence experiments 0 0 κ 3π
of Comte-Bellot and Corrsin (1971): !, Rλ = 71; !, Rλ =
65; ", Rλ = 61. Lines, model spectrum, Eq. (6.246): solid, p0 = 2, The shape of the spectrum (from the first
Rλ = 60; dashed, p0 = 2, Rλ = 1, 000; dot-dash p0 = 4, Rλ = 60. figure) does not vary strongly with the Re.
For the model spectrum with p0 = 4 a
slightly different shape is obtained (10%
difference).
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 34 / 41
 CHAPTER 6: THE SCALES OF TURBULENT MOTION

Turbulent The energy-containing range - 2

Flows
Stephen B. Pope
Cambridge University Press, 2000
c
!Stephen B. Pope 2000 For the model spectrum, the cumulative
kinetic energy
1.0 Z κ
k(0,k) = E(κ0 )dκ0
0
k(0,") 0.8 R!=1000
is plotted against l/L11 = 2π/(κL11
k 0.6 The centroid of the spectrum is at
κL11 ≈ 4 (l/L11 ≈ 1 12 ) and 80% of the
0.4 energy is contained in motions of
lengthscale 16 L11 < l < 6L11 .
0.2 So the energy-containing motions are
R!=30 characterized by lengthscales ranging from
l0 = 6L11 to lEI = 61 L11 .
0.0
0 1 2 3 4 5 6 7 8 9 10
l/L11
Table: Charateristic wavenumbers and lengthscales of the energy spectrum
(based on the model spectrum)
10 4 2 1
"L11
Defining wavenumbers κL l/L
CHAPTER 6: THE SCALES OF TURBULENT
Peak of energy spectrum
MOTION
11
1.3
11
5.0
k(0,κ) = 0.1k 1.0 6.1
Cumulative turbulent kinetic energy k(0,κ) against
Figure 6.19:

Turbulent Flows
wavenumber κ and wavelength " = 2π/κ for the model spectrum.
k(0,κ) = 0.5k
k(0,κ) = 0.8k
3.9
15
1.6
0.42
k(0,κ) = 0.9k 38 0.16
Stephen B. Pope
Dr. Abhilash J. Chandy Cambridge
Essentials University
of Turbulence Press,
ME 724 Spring 20192000
Module 04 February 12, 2019 35 / 41

c
!Stephen B. Pope 2000
Effects of Reynolds number

100 106
5
10-1 (a) 10
4 1000 (b)
E(!)10-2 11 E(!) 10
103
kL1110-3 2 102 300
"u" 101
-4 100
10 100
10-5 10-1 R#=30
-2
10
10-6 R#=30 100 300 10
-3
1000
10-7 -1 10-4
10 100 101 102 103 10-4 10-3 10-2 10-1 100
!L11 !"

Figure 6.20: Model spectrum for different Reynolds numbers, scaled by

(a) k and L11 (b) Kolmogorov scales.

For the energy-containing ranges of the spectra (0.1 < κL11 < 10) are very similar, whereas
with increasing Rλ , the extent of the −5/3 region increases, and the dissipation range
(where the spectrum rolls off) moves to higher values of κL11 .
For the same spectra normalized by the Kolmogorov scales, the dissipation ranges (κη > 0.1
say) are very similar, while the energy containing ranges moves to lower values of κη as Rλ
increases.
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 36 / 41
The spectral view of the energy cascade: Energy-containing motions

The bulk of the tke is contained in motions of lengthscale l, comparable to the integral
lengthscale, L11 (6L11 > l > 61 L11 = lEI , say), whose characteristics velocity is of the
order of k1/2 .
Since their size is comparable to the flow dimensions L, these large-scale motions can be
strongly influenced by the geometry of the flow.
Their timescale, L11 /k1/2 is large compared with the mean-flow timescale, so that they are
significantly affected by the flow’s history.
So in contrast to the universal equilibrium range, the energy-containing motions do not have
a universal form brought about by statistical equilibrium.
All of the anisotropy is confined to the energy-containing motions and consequently so also is
all of the production of turbulence; On the other hand, the viscous dissipation is negligible.
During the initial steps in the cascade, energy is removed by inviscid processes and
transferred to smaller scales (l < lEI ) at a rate TEI , which scales as k3/2 /L11 . This
transfer process depends on the non-universal energy-containing motions, and consequently
the non-dimensional ratio TEI /(k3/2 /L11 ) is not universal.

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 37 / 41

Energy-spectrum balance - 1

For homogeneous turbulence, the energy-spectrum balance is quantified by the balance

equation for E(κ, t):

∂ ∂
E(κ, t) = Pκ (κ, t) − Tκ (κ, t) − 2νκ2 E(κ, t)
∂t ∂κ
The RHS represents production, spectral transfer, and dissipation.
The production spectrum, Pκ is given by the product of the mean velocity gradients
∂ hUi i /∂xj and an anisotropic part of the spectrum tensor. The contribution
R to the
production from the wavenumber range (κa , κb ) is denoted by P( κa , κb ) = κκab Pκ dκ.
In the second term, Tκ (κ) is the spectral energy transfer rate: It is the net rate at which
energy is transferred from modes of lower wavenumber than κ to those with wave numbers
higher than κ. TheRrate of gain of energy in the wavenumber range (κa , κb ) due to this
spectral transfer is κκab − ∂κ
∂
Tκ (κ)dκ = Tκ (κa ) − Tκ (κb ).
Since Tκ vanishes at zero and infinite wavenumber, the transfer term makes no contribution
to the balance of the tke, k.
The final term is the dissipation spectrum D(κ, t) = 2νκ2 E(κ, t).

Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 38 / 41
 Stephen B. Pope
Cambridge University Press, 2000
c
!Stephen B. Pope 2000
Energy-spectrum balance - 2
(a) E(κ) D(κ)
In the inertial subrange, spectral transfer is
the only significant process so that (when it
0
κEI κDI κ is integrated from κEI to κDI ) the
(b) Pκ - ∂Tκ
equation yields (for Tκ (κDI ) ≈ TDI )
∂κ
0
κEI κDI κ 0 ≈ TEI − TDI
- ∂E - ∂Tκ -D
∂t ∂κ
Whereas in the dissipation range, spectral
ε
(c) Tκ
transfer balances dissipation so that (when
it is integrated from κDI to ∞) the
0
equation yields
κEI κDI κ

Figure 6.28: For homogeneous turbulence at very high Reynolds num-

0 ≈ TDI − 
ber, sketches of (a) the energy and dissipation spectra (b) the
contributions to the balance equation for E(κ, t) (Eq. 6.284), and When they are added together, the last 3
(c) the spectral energy transfer rate. equations give the turbulent kinetic energy
In the energy containing range, all the terms equation dk/dt = P − .
are significant except for dissipation. Say
So in the energy-containing range TEI
k(0,κEI ) ≈ k, (0,κEI ) ≈ 0, P(0,κEI ) ≈ P
depends in a non-universal way on several
and Tκ (κEI ) ≈ TEI , when it is integrated
factors. However this transfer rate
over the energy-containing range (0, κEI ),
establishes an inertial subrange of universal
the energy-spectrum balance yields
character with Tκ (κ) = TEI ; and finally the
dk high wavenumber part of the spectrum
≈P −T EI MOTION
CHAPTER 6: THE SCALES OF TURBULENT dissipates the energy at the same rate as
dt
Turbulent Flows that at which it receives it.
Dr. Abhilash J. Chandy
Stephen B. Pope Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 39 / 41
Cambridge University Press, 2000
c
!Stephen B. Pope 2000
20
Limitations - 1
1.7
A limitation of the Kolmogorov hypotheses
1.6 is that they apply only to high-Re flows and
that a criterion for sufficiently high-Re is
1.5
p
not provided. Many laboratory or practical
1.4 flows have reasonably high-Re (10000 or
Reλ = 150) and also the motions on the
1.3
dissipative scales are found to be
1.2 anisotropic.
102 103 104
R! On closely looking at the inertial range
Spectrum power-law exponent p (E(κ) ∼ κ−p) as a func-
Figure 6.29: spectra, the Kolmogorov −5/3 spectrum is
tion of Reynolds number in grid turbulence: symbols, experimental approached slowly as the Re increases.
data of Mydlarski and Warhaft (1998); dashed line, p = 53 ; solid
−3
From experiments on grid turbulence at
line, empirical curve p = 53 − 8Rλ 4 .
high-Re (Reλ = 50 − 500), the inertial
The notions of energy cascade, vortex range spectrum is indeed a power law
stretching and the Kolmogorov hypotheses E(κ) ∼ κ−p , but that exponent depends on
provide an invaluable conceptual Reλ .
framework. However both conceptually and
So p does approach 5/3 at very large Reλ ,
empirically, there are some shortcomings.
but at Reλ = 200, p ≈ 1.5
But it should also be appreciated that these
shortcomings have a minor impact on the All the experimental data considered so far
study and modeling of turbulent flows pertain to II order velocity statistics (i.e.
because the small scales contain little statistics that are quadratic in velocity).
energy (and less anisotropy) and so have These are the most important quantities
little direct effect on the flow. since they determine the kinetic energy and
the Reynolds stresses.
Dr. Abhilash J. Chandy 21 Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 40 / 41
 c
!Stephen B. Pope 2000

Limitations - 2
102

K In contrast to the Re effects discussed

previously, here k does not appear to reach
101 an asymptote, but instead the data are
consistent with an indefinite increase,
3/8
possibly as K ∼ Reλ .
100 1
10 102 103 104
The discrepancies between the Kolmogorov
R! predictions and the experimental values of
Measurements (symbols) compiled by Van Atta and An-
Figure 6.30: the higher-order moments Mn are
tonia (1980) of the velocity-derivative kurtosis as a function of
3 attributed to the internal intermittency.
Reynolds number. The solid line is K ∼ Rλ8 .

The simplest examples of higher-order It is again emphasized that in the context

statistics are the normalized of the mean velocity field and the Reynolds
velocity-derivative moments stresses in turbulent flows, these issues are
 ,*  +n/2 of minor significance. The turbulence
∂u1 n ∂u1 2

Mn = energy and anisotropy are predominantly
∂x1 ∂x1 contained in the large-scale motions.
Internal intermittency on the other hand,
For n = 3 and n = 4 these are the concerns rare events that are manifest only
velocity-derivative skewness S and the in the higher-order statistics of small-scale
kurtosis K (For a Gaussian random quantities. In the context of energy cascade
variable, S is 0 and K is 4). and turbulent flows. of much more
According to the Kolmogorov hypotheses, importance than internal intermittency is
22
for each n, Mn is a universal constant. the question of what determines the rate of
However it is found that S and K are not energy transfer TEI from the
constant and increase with Re. energy-containing scales.
Dr. Abhilash J. Chandy Essentials of Turbulence ME 724 Spring 2019 Module 04 February 12, 2019 41 / 41