Roddam Narasimha
Z
|u(t)| dt as T .
0
2
the lag):
3
R
So R(0) < u2 >= () d.
Thus () is the contribution per unit frequency interval at to
(twice) the disturbance kinetic energy. () is called the power
spectral density or just the power spectrum. (Different notations and
normalizations are however in use.)
As we have seen, u(t) does not possess a classical Fourier series
or transform representation. It can however be written
Z
u(t) = eit dZ(),
4
Figure 2:
Spectra of derivatives
we see Z
u
= ieit dZ(),
t
5
and
2 Z Z
u it
0 0
i t 0
= ie dZ() i e dZ( )
t
Z
= 2 eit () d.
6
derivatives, and in homogeneous isotropic turbulence each is pro-
portional to (u1 /x1 )2 .
Secondly the mean square vorticity is also a sum of spatial (cross)
derivatives of the velocity. We can therefore say
< u2 >
a1 2
< 2 >,
2 < u2 >
< > a2 .
2
In homogeneous isotropic turbulence a1 = , a2 = .
So, the Taylor microscale provides the length scale that, with the
rms fluctuation, gives us the velocity derivatives in turbulent flow.
7
principle says that the large eddy motion will be asymptotically in-
dependent of . On the other hand, some viscous dissipation is
inevitable. The stirring process must therefore balance the energy
input which must be of order U 3 /L with the viscous dissipa-
1/2
tion, of order U 2 /2 . It follows that LRe1/2 , Re = ReL .
Recalling the scaling of the Blasius boundary layer thickness,
therefore is a scale determined by a (laminar) boundary layer type
process. So L, and the two scales we introduced in the auto-
correlation curve are widely separated at high ReL . That is, R( )
falls through a narrow and sharp parabola from its value of unity at
= 0, to a slow decline to zero in the tail.
But this is not enough. The super-mixing of turbulence demands
that it be broken down (crushed into dust like a lump of coal, said
Kolmogorov) into scales at which viscosity is dominant, i.e. local
Re is O(1). Kolmogorov argued that such fine scales must be deter-
mined entirely by and , with no separate memory of U and L or
except through . Loss of memory of large-scale initial conditions
also implies isotropy in the Kolmogorov scales. By dimensional
analysis these scales are found to be
u = ()1/4 ,
l = 3/4 1/4 ,
so u l / is indeed 1. We note that
l 3/4 3/2
ReL Re .
l
Motion at Kolmogorov scales is a kind of random laminar flow- the
former because of the large scale shaking, the latter because of the
dominant influence of visocity.
Important! The Kolmogorov vorticity is of the same order as the
Taylor vorticity:
3 1/2
u 1 U U
= 1/2 1/2 1/2 .
l L
8
That is, there is no order-of-magnitude change in vorticity as we go
from the microscale to the K-scale. This suggests that the process
involved is akin to plane extensional strain of a vortex sheet (accom-
panied of course by normal squashing, to preserve continuity) a
process similar to what happens in relaminarization of a boundary
or shear layer under large favourable pressure gradients.
This phenomenon we have just described is called the cascade
process the process by which energy cascades down from the
large scales U , L to the K scales u , l .
9
putting U 3 /L. As ReL , we therefore have
l L
.
u U
CHANNEL FLOW
10
(velocities, stresses) = 0,
x
p
= const in x.
x
The momentum equation along x is
p 2u
(M o)x : + 2 + = 0,
x
| {z } y y
2
|{z}
U U2
h h
p
h = w , U2
x
p U2
so .
x h
Outer Solution
Uh
As Rh = :
2u U U2 U
2 2 =
y h h hU2
U2 U 2
=
h U h U
U2 1
=
h c f Rh
Our limit process is: cf UU 0,
Uh
.
In laminar flow: cf Rh = constant,
but in turbulent flow: cf Rh .
11
So the outer equation is
p
+ = 0,
x y
U2 y
= 1 ,
h h
to satisfy the boundary condition (y = h) = 0.
2
Inner limit of as y 0, = Uh , =
and does not satisfy boundary conditions at y = 0 that = 0, so
different inner limit is required.
Inner Solution
p 2 u
(M ox ) : + 2 + = 0.
x y y
Let inner scales be v , l h.
Orders of magnitude:
U2 v
+ 2 + U2 l 0.
h l |{z}
pressure gradient term
|{z}
U2 /l
u+ = f (y+ ) + h.o.t
or: u = U f (y+ ) + o (U /U ) ,
= U2 g (y+ ) + o (U /U )2
Outer Expansion
Outer boundary condition on u is
u (y = h) = Uc , centre-line velocity.
u = Uc + h.o.t,
= Uc + (Re) U1 , say
| {z }
gauge function
Now Millikan and Kolmogorov say inner and outer solutions must
match.
Let outer co-ordinate variable be
y
Y .
h
Then
Uc + U1 (Y ) U f (y+ )
| {z } | {z }
as Y 0 as y+
Divide through by Uc :
U 0 U1 0
put = U , = U1 , drop prime.
Uc Uc
Matching condition gives
1 + U1 (Y 0) U f (y+ ) ,
13
As U 0 as Re ,
f (y+ ) as y+ ,
to produce product of O(1).
As y+ = Y h+ , we have:
1 + U1 (Y 0) U f (Y h+ ) ,
U 1
U1 (Y 0) f (Y h+ ) .
Let
U1 (Y 0) = G(Y ),
f (Y h+ ) = F (Y h+ ),
0 U 0 1
G (Y ) = F (Y h+ ) h+ , getting rid of constant, .
0 U
= F (Y ) , say, = h+ , = .
To solve this equation, differentiate in :
0 00
0 = F (Y ) + F (Y ) Y
0 00
F (z) + zF (z) = 0, z = Y .
14
Solution:
00
F 0
0 = z 1 , ln F (z) = - ln z + const,
F
0 1
F (z) = + const, kis constant (independent of Re)
kz
1
F (z) = ln z + const.
k
1
= ln(y+ ) + const.
k
Everything else now follows as usual, including defect law and skin
friction law.
15