LECT URE N OT ES ON T URBULEN CE

B. MUTLU SUMER
Technical University of Denmark
DTU Mekanik, Section for Fluid Mechanics, Coastal & Maritime Engineering
Building 403, 2800 Lyngby, Denmark, bms@mek.dtu.dk

Revised 2013

Contents
1 Basic equations
1.1 Method of averaging and Reynolds decomposition
1.2 Continuity equation . . . . . . . . . . . . . . . . .
1.3 Equations of motion . . . . . . . . . . . . . . . .
1.4 Energy equation . . . . . . . . . . . . . . . . . . .
1.4.1 Energy equation for laminar ows . . . . .
1.4.2 Energy equation for the mean ow . . . .
1.4.3 Energy equation for the uctuating ow .
1.5 References . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

2 Steady boundary layers

2.1 Flow close to a wall . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 Idealized ow in half space y>0 . . . . . . . . . . . .
2.1.2 Mean and turbulence characteristics of ow close to a
smooth wall . . . . . . . . . . . . . . . . . . . . . . .
2.1.3 Mean and turbulence characteristics of ow close to a
rough wall . . . . . . . . . . . . . . . . . . . . . . .
2.2 Flow across the entire section . . . . . . . . . . . . . . . . .
2.3 Turbulence-modelling approach . . . . . . . . . . . . . . . .
2.3.1 Turbulence modelling of ow close to a wall . . . . .
2.3.2 Turbulence modelling of ow across the entire section
2.4 Flow resistance . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Bursting process . . . . . . . . . . . . . . . . . . . . . . . .
2.6 Appendix. Dimensional analysis . . . . . . . . . . . . . . . .
2.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

7
7
8
9
11
11
12
13
15

17
. 17
. 17
. 23
.
.
.
.
.
.
.
.
.

31
43
48
49
57
59
61
66
69

3 Statistical analysis
73
3.1 Probability density function . . . . . . . . . . . . . . . . . . . 73
3

CONTENTS
3.2 Correlation analysis . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Space correlations . . . . . . . . . . . . . . . . . . . .
3.2.2 Time correlations . . . . . . . . . . . . . . . . . . . .
3.3 Spectrum analysis . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1 General considerations . . . . . . . . . . . . . . . . .
3.3.2 Energy balance in wave number space . . . . . . . . .
3.3.3 Kolmogoro s theory. Universal equilibrium range and
inertial subrange . . . . . . . . . . . . . . . . . . . .
3.3.4 One-dimensional spectrum . . . . . . . . . . . . . . .
3.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 Di usion and dispersion

4.1 One-Particle analysis . . . . . . . . . . . . .
4.1.1 Di usion for small times . . . . . . .
4.1.2 Di usion for large times . . . . . . .
4.2 Longitudinal dispersion . . . . . . . . . . . .
4.2.1 Mechanism of longitudinal dispersion
4.2.2 Application of one-particle analysis .
4.3 Calculation of dispersion coe cient . . . . .
4.3.1 Formulation . . . . . . . . . . . . . .
4.3.2 Zeroth moment of concentration . . .
4.3.3 Mean particle velocity . . . . . . . .
4.3.4 Longitudinal dispersion coe cient . .
4.4 Longitudinal dispersion in rivers . . . . . . .
4.5 References . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

5 Wave boundary layers

5.1 Laminar wave boundary layers . . . . . . . . . .
5.1.1 Velocity distribution across the boundary
5.1.2 Flow resistance . . . . . . . . . . . . . .
5.2 Laminar-to-turbulent transition . . . . . . . . .
5.3 Turbulent wave boundary layers . . . . . . . . .
5.3.1 Ensemble averaging . . . . . . . . . . . .
5.3.2 Mean ow velocity . . . . . . . . . . . .
5.3.3 Flow resistance . . . . . . . . . . . . . .
5.3.4 Turbulence quantities . . . . . . . . . . .
5.3.5 E ect of Reynolds number . . . . . . . .
5.3.6 Boundary layer thickness . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

. . . . . . .
layer depth
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .

.
.
.
.
.
.

76
76
81
84
85
87

. 91
. 94
. 98
.
.
.
.
.
.
.
.
.
.
.
.
.

99
99
103
104
113
113
114
118
119
121
123
124
126
129

131
. 132
. 132
. 135
. 136
. 140
. 140
. 141
. 144
. 144
. 144
. 146

CONTENTS

5.3.7 Turbulent wave boundary layers over rough walls . . . 148

5.4 Other aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6 Turbulence modelling
6.1 The closure problem . . . . . . . . . . . .
6.2 Turbulence models . . . . . . . . . . . . .
6.3 Mixing-length model . . . . . . . . . . . .
6.3.1 Turbulent transport of momentum
6.3.2 Mixing length . . . . . . . . . . . .
6.4 k-omega model . . . . .
6.4.1 Model equations
6.4.2 Model constants .
6.4.3 Blending function

. .
. .
. .
F1

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

6.4.4 Eddy viscosity and Blending

6.4.5 Boundary conditions . . . .
6.4.6 Numerical computation . . .
6.4.7 An application example . .
6.5 Large Eddy Simulation (LES) . . .
6.5.1 Model equations . . . . . .
6.5.2 An application example . .
6.6 References . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

155
. 155
. 156
. 157
. 157
. 160

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

169
169
171
171

F2
. .
. .
. .
. .
. .
. .
. .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

172
172
174
175
181
181
183
189

function
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .

CONTENTS

Chapter 1
Basic equations
1.1

Method of averaging and Reynolds decomposition

The mean value of a hydrodynamic quantity, for example, the xi

of the velocity, ui , is dened by
ui =

1
T

component

t0 +T

ui dt

(1.1)

t0

in which t0 is any arbitrary time, and T is the time over which the mean is
taken. Clearly T should be su ciently large to give a reliable mean value.
This is called time averaging. There are also other kinds of averaging, such
as ensemble averaging, or moving averaging. We shall adopt the ensemble

Figure 1.1: Time series of ui .

7

CHAPTER 1. BASIC EQUATIONS

averaging in the case of wave boundary layers, as will be detailed in Chapter

5.
The instantaneous value of the velocity ui may be written (see Fig. 1.1)
ui = ui + ui

(1.2)

in which ui is called the uctuating part (or uctuation) of ui , and ui is its

mean. This is called the Reynolds decomposition. The Reynolds decomposition can be applied to any hydrodynamic quantity.
Let f and g be any such quantities (any velocity component, pressure,
etc.). The following relations are known as the Reynolds conditions (here,
the overbar denotes the time averaging, Eq. 1.1):
f +g
af
a
f
s
fg

= f +g
= af
= a
f
=
s
= fg

(1.3)

f = f
f = 0
fg = fg
fh

= f

h =0

in which a = constant, and s = x1 , x2 , x3 , t.

1.2

Continuity equation

For an incompressible uid (such as water), the continuity equation is (from

the conservation of mass)
ui
=0
(1.4)
xi
where Einsteins summation convention is used, namely there is summation
over the repeated indices:
u1
u2
u3
+
+
=0
x1
x2
x3

(1.5)

1.3. EQUATIONS OF MOTION

in which ui is the i component of the velocity.

Averaging (Eq. 1.1) gives:
ui
ui
=
=0
xi
xi

(1.6)

Subtracting the preceding equation from Eq. 1.4 gives

ui
=0
xi

(1.7)

To sum up, Eq. 1.6 is the continuity equation for the mean velocity and
Eq. 1.7 is the continuity equation for its uctuating part.

1.3

Equations of motion

For a Newtonian uid (such as water) the Navier-Stokes (N.-S.) equation is

(

ui
ui
+ uj
) = gi +
t
xj

ij

(1.8)

xj

in which t is the time, gi is the volume force (such as the gravity force) and
xi is the Cartesian coordinates. The quantity ij is the stress on the uid,
and given as
ui
uj
p ij + (
+
)
(1.9)
ij =
xj
xi
The preceding relation is the constitutive equation for a Newtonian uid.
Here, p is the pressure, and is the viscosity, and ij is the Kronecker delta
dened as
ij
ij

= 1 for i = j,
= 0 for i = j

(1.10)

From the continuity equation (Eq. 1.4), the N.-S. equation becomes
ui
+
( ui uj ) = gi +
t
xj

ij

xj

(1.11)

and averaging each term

ui
+
( ui uj ) = g i +
t
xj

ij

xj

(1.12)

10

CHAPTER 1. BASIC EQUATIONS

Now, the second term on the left hand side of the preceding equation (using
the Reynolds decomposition)

xj

( ui uj ) = uj

ui
+
( ui uj )
xj
xj

(1.13)

Inserting the preceding equation in Eq. 1.12

(

ui
ui
+ uj
) = gi +
(
t
xj
xj

ij

ui uj )

(1.14)

This equation is known as the Reynolds equation.

As seen, the equation of motion for the mean ow in the case of the
turbulent ow (Eq. 1.14) is quite similar to that in the case of the laminar
ow (Eq. 1.8). Comparison of the two equations indicates that, in the case
ui uj . This
of the turbulent ow, there is an additional stress, namely
additional stress is called the Reynolds stress. There are nine such stresses

ui uj =

u1 u1
u2 u1
u3 u1

u1 u2
u2 u2
u3 u2

u1 u3
u2 u3
u3 u3

(1.15)

As seen, these stresses form a symmetrical second order tensor (the Reynolds
stress tensor).
To summarize at this point, there are four equations for the mean ow.
These are
1. the continuity equation (Eq. 1.6) and
2. three equations of motion (Eq. 1.14),
and there are ten unknowns (namely, three components of the velocity,
ui , and the pressure, p, and six components of the Reynolds stress,
ui uj ).
Therefore the system is not closed. This problem is known as the closure
problem of turbulence. We shall return to this problem later.
(In the case of laminar ow, there are four unknowns (namely, three components of the velocity, ui , and the pressure, p) and four equations, namely
the continuity equation (Eq. 1.4) and three N.-S. equation (Eqs. 1.8 and
1.9). Therefore, the system is closed.)

1.4. ENERGY EQUATION

1.4

11

Energy equation

1.4.1

Energy equation for laminar ows

N.-S. equation:
ui
+u
t

ui
= gi
x

ui
x x

1 p
+
xi

(1.16)

Multiplying both sides of the equation by uj

uj

ui
+ uj u
t

ui
= uj gi
x

uj

2
1 p
ui
+ uj
xi
x x

(1.17)

The rst term on the left hand side is

ui
= (ui uj )
t
t

uj
and solving
gives
uj

uj
t

uj
t

ui

(1.18)

from Eq. 1.16, and inserting it into the preceding equation

ui
= (ui uj )
t
t

1 p
+
xj

ui (gj

uj
x x

uj
)
x

(1.19)

Substituting the latter equation in Eq. 1.17, and setting i = j,

p
1
ui ui ) + ui
+ ui u
t 2
xi
(

2
ui
ui
= ui gi + ui
x
x x

(1.20)

Adding the term

ui
u
ui
+
)
x
xi x
on both sides of the equation and also the term
(

ui

u
xi x

on the right hand side of the equation (this term is actually zero due to
continuity), and after some algebra, the following equation is obtained
K
+
[Ku + pu
t
x

ui (

ui
u
+
)] = u g
x
xi

(1.21)

12

CHAPTER 1. BASIC EQUATIONS

in which K is the kinetic energy per unit volume of uid

K=
and

1
1
ui ui = (u2 + v 2 + w2 )
2
2

is the energy dissipation per unit volume of uid and per unit time
= (

1.4.2

(1.22)

ui
u
ui
+
)
x
xi x

(1.23)

Energy equation for the mean ow

The equation of motion for the mean ow (the Reynolds equation, Eq.1.14):
ui
+ u
t

ui
= gi
x

p
+
xj

ui
+
(
x x
x

(1.24)

ui u )

Now, in the preceding subsection, the energy equation, Eq. 1.21, is obtained from the equation of motion, Eq. 1.16. In a similar manner, the energy
equation for the mean ow can be obtained from the equation of motion for
the mean ow (i.e., from Eq. 1.24). The result is
K0
+
[K0 u + ui u ui +p u
t
x

ui (

ui
u
+
)] = u g
x
xi

ui u )

(1.25)
in which K0 is the kinetic energy per unit volume of uid for the mean ow
K0 =

1
1
ui ui = (u2 + v2 + w2 )
2
2

(1.26)

and 0 is the energy dissipation per unit volume of uid and per unit time,
again, for the mean ow
0

= (

ui
u
ui
+
)
x
xi x

(1.27)

Now, note the last term on the right hand side of Eq. 1.25. We shall return
to this term in the next subsection.

ui
x

1.4. ENERGY EQUATION

1.4.3

13

Energy equation for the uctuating ow

Averaging Eq. 1.21 gives

1
1
ui ui ) +
[ ui ui u + pu
t 2
x 2
(

ui (

ui
u
+
)] = u g
x
xi

(1.28)

Furthermore, Eq. 1.25:

1
1
ui ui ) +
[ ui u ui + ui u ui + p u
t 2
x 2
(

= u g

ui (

ui u )

ui
u
+
)] =
x
xi

ui
x

(1.29)

Subtracting Eq. 1.29 from Eq. 1.28 gives

Kt
1
[Kt u + ui ui u +p u
+
t
x
2

ui (

ui
u
+
)] = u g
x
xi

t +(

ui u )

(1.30)
in which Kt is the kinetic energy per unit volume of uid for the uctuating
ow (i.e., the kinetic energy of turbulence)
Kt =

1
1
ui ui = (u 2 + v 2 + w 2 )
2
2

(1.31)

and t is the viscous dissipation of the turbulent energy per unit volume of
uid and per unit time
t

= (

ui
u
u
+
) i
x
xi x

(1.32)

Eq. 1.25 is the energy equation for the mean ow, and Eq.1.30 is that
for turbulence. As seen, the term ( ui u ) xui is common in both equations,
but with opposite signs. This implies that if this term is an energy gain for
turbulence, then it will be an energy loss for the mean ow, or vice versa.
Now, we shall show that this term is an energy gain for turbulence.
Integrating Eq.1.30 over a volume V gives
t

Kt dV +
V

F n dS =
S

u g dV
V

t
V

dV +

(
V

ui u )

ui
dV
x
(1.33)

ui
x

14

CHAPTER 1. BASIC EQUATIONS

in which
F = Kt u +

1
u uu +pu
2 i i

ui (

ui
u
+
)
x
xi

(1.34)

and S is the surface encircling the volume V. Here, the volume integral in
the second term on the left hand side of Eq. 1.33 is converted to a surface
integral, using the Green-Gauss theorem. The rst term on the right hand
side of Eq. 1.33 drops when the volume force is taken as the gravity force,
since g = 0. The second term on the left hand side of Eq. 1.33, on the other
hand, is the energy ux at the surface S. Therefore, it is not a source term.
Now, consider a situation where there is no inux of turbulent energy. In
this case, the maintenance of turbulence (i.e., t V Kt dV
0 in Eq. 1.33) is
ui
only possible when V ( ui u ) x dV in Eq. 1.33 is positive, meaning that
the term ( ui u ) xui in the turbulence energy equation (Eq. 1.30) should
be positive. In other words, this term is an energy gain for turbulence, and
therefore it is an energy loss for the mean ow. This is an important result,
because it implies that turbulence extracts its energy from the mean ow, and
the rate at which the energy is extracted from the mean ow is ( ui u ) xui .
An important implication of the above result is that turbulence is generated only when there exists a velocity gradient in the ow xui . When
there is no velocity gradient, no turbulence will be generated. In this case,
any eld of turbulence introduced into the ow will be dissipated (the decay of turbulence). See Monin and Yaglom (1973, pp. 373-388) for further
discussion.
Example 1 Find the turbulence energy generation and the total energy dissipation (per unit volume of uid and per unit time) for a turbulent boundary
ow in an open channel (Fig. 1.2).
The energy extracted from the mean ow is (
ow, this will be
u
( uv)
y

ui u )

ui
x

. For the present

This is the energy extracted from the mean ow per unit time and per unit
volume of uid (Fig. 1.2 c). As seen from the gure, the energy generation
for turbulence is zero at the free surface while it increases tremendously as
the wall is approached.

1.5. REFERENCES

15

Figure 1.2: (a) Shear stress, (b) velocity and (c) turbulent energy extracted from
the mean ow.

From Eq. 1.25, the total energy dissipation, on the other hand, is
0

in which
be

+(

ui u )

ui
x

is given in Eq. 1.27. The total energy dissipation will therefore

u u
u
)
+( uv)
y y
y
From the preceding relation, it is seen that the energy dissipation is very
large near the wall, and it decreases with the distance from the wall, similar
to the variation of the turbulent energy generation (Fig. 1.2 c).
(

1.5

References

1. Monin, A.S. and Yaglom, A.M. (1973). Statistical Fluid Mechanics:

Mechanics of Turbulence, vol. 1, MIT Press. Cambridge, Mass.

16

CHAPTER 1. BASIC EQUATIONS

Chapter 2
Steady boundary layers
In this chapter, we will study steady, turbulent boundary layers over a wall,
in a channel, or in a pipe. (Examples of turbulent boundary layers over a
wall may be: ow over a at plate, ow over the seabed, the atmospheric
boundary-layer ow over the earth surface, etc.).
We will rst concentrate on the ow close to a wall (a general analysis),
and then we will study the ow close to a smooth wall, and subsequently
that close to a rough wall. (The latter will include the ow near the bed
of a channel, and that near the wall of a pipe, etc.). Finally, we will turn
our attention to ows in channels/pipes, considering the entire channel/pipe
cross section.

2.1
2.1.1

Flow close to a wall

Idealized ow in half space y>0

First, consider the following idealized case.

The ow is steady, and takes place in the half space y > 0 (Fig. 2.1),
and there is no pressure gradient in the horizontal direction, i.e., no pressure
gradient in the x direction.
Shear stress.
The basic equations are the continuity equation and the two components
of the Reynolds equation. The continuity equation is automatically satised. The y component of the Reynolds equation gives an equation for the
17

18

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.1: Shear stress in the idealized half space y > 0.

Reynolds stress
v 2 . The x component of the Reynolds equation (Eq.
1.14), on the other hand, reads
(

u
u
u
+ u + v ) = gx
t
x
y

2
2
p
u
u
+ ( 2 + 2)+ (
x
x
y
x

u 2) +

uv)

(2.1)
Since ut = 0, v = 0, g x = 0, xp = 0, u is a function of only y, and the ow is
uniform in the x direction (i.e., ux = 0), then the Reynolds equation will
be
d2 u
d
( uv)=0
+
(2.2)
dy 2 dy
or,
d du
d
(
+ ( u v )) =
=0
(2.3)
dy dy
dy
in which is the total shear stress:
=

du
+(
dy

uv)

(2.4)

which is composed of two parts, namely the viscous part (the rst term), and
the part induced by turbulence (the Reynolds stress, the second term).
Integrating 2.3 gives
= c;

c = constant

(2.5)

The boundary condition at the wall:

y=0:

(2.6)

2.1. FLOW CLOSE TO A WALL

in which

19

is the wall shear stress. From Eqs. 2.4 and 2.6, one obtains
(=

du
+(
dy

u v )) =

(2.7)

This is an important result. It shows that the shear stress is constant

over the entire depth, and equal to the wall shear stress, 0 .
Mean velocity.
Our objective is to determine the velocity u as a function of y. To this end,
we resort to dimensional analysis (see Appendix at the end of this chapter
for a brief account of the dimensional analysis).
Now, the mean properties of the ow (such as u) depend on the following
quantities
,
(2.8)
0 , y,
Here we consider a smooth wall, therefore no parameter representing the wall
roughness is included.
The ow depends on 0 because the shear stress determines the ow
velocity; the larger the shear stress, the larger the ow velocity. (Also, it must
be stressed that the shear stress in this idealized case remains unchanged
across the depth, as shown in the preceding paragraphs).
Likewise, the ow depends on the distance from the wall, y, because the
closer to the wall, the smaller the velocity (just at the wall, obviously the
velocity will be zero due to the no-slip condition).
Clearly, the ow should also depend on the uid properties, namely ,
and .
Hence
u = F ( 0 , y, , )
(2.9)
or
(u,

0,

y,

, )=0

From dimensional analysis, the preceding dimensional function

converted to a nondimensional function
(
yUf

in which Uuf , and

velocity dened by

u yUf
,
)=0
Uf

(2.10)
can be

(2.11)

are two nondimensional quantities, Uf is the friction

Uf =

(2.12)

20

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.2: Flow in an open channel.

and

is the kinematic viscosity

Solving

u
Uf

(2.13)

u
= f (y + )
Uf

(2.14)

from Eq. 2.11

in which
y+ =

yUf

(2.15)

Eq. 2.14 is known as the law of the wall. The quantities Uf and
the inner ow parameters, or wall parameters.

are called

Example 2 Boundary layer in an open channel. Flow close to the bed of

the channel.
Now, consider a real-life ow, namely the ow in an open channel (Fig.
2.2). The ow in the channel is driven by gravity, with the volume force
gx = gS in which S is the slope of the channel. The x component of the
Reynolds equation, Eq. 2.3, will in the present case read
d du
(
+(
dy dy
in which

u v )) =

d
=
dy

gx =

gS =

(2.16)

is the specic weight of the uid. Integrating gives

=

Sy + c;

c: constant

(2.17)

2.1. FLOW CLOSE TO A WALL

21

The boundary condition at the free surface:

y=h:

=0

(2.18)

which gives c = Sh where h is the depth. Then the shear stress

y
)
h

= Sh(1

(2.19)

The wall shear stress (the bed shear stress) will then be
0

Hence, the shear stress

=

(2.20)

= Sh
y
)
h

0 (1

(2.21)

(see sketch in Fig. 2.2).

Our main concern in this section is the ow close to the wall. Close to
the wall, i.e., y
h, the wall shear stress
=

0 (1

y
)
h

(2.22)

i.e., near the wall, the shear stress can be assumed to be constant and equal
to the wall shear stress,
0 . This thin layer of uid is called the constant
stress layer.
As seen, the ow in the constant stress layer is similar to the idealized
ow considered in the preceding section; in both ows, the shear stress is
constant across the depth. The immediate implication of this result is that,
the velocity distribution close to the wall in the present case (i.e., in the
constant stress layer) is given by the law of the wall (Eq. 2.14):
u
= f (y + )
Uf

(2.23)

Example 3 Boundary layer in a pipe. Flow close to the wall of the pipe.
Now, consider a fully developed turbulent boundary layer in a pipe (Fig.
2.3).
Similar to the previous example, the x component of the Reynolds equation in the present case:
1
p
(r ) =
(2.24)
r r
x

22

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.3: Flow in a pipe.

Here, the gravity term is included in the pressure gradient term for convenience. The ow is driven by the pressure gradient xp .
Integrating Eq.2.24 gives
r =

p r2
+c
x2

(2.25)

in which c is a constant.
The boundary condition at the center line of the pipe
r=0:

(2.26)

=0

which gives c = 0. The boundary condition at the wall of the pipe

D
:
2
From Eqs.2.25 and 2.27, one gets
r=

=2

r
or
D

(2.27)

0 (1

2y
)
D

(2.28)

in which y = D2 r is the distance from the pipe wall.

D
Again, similar to the previous example, for y
, i.e., close to the wall,
2
the shear stress can be assumed to be constant and approximately equal
to the wall shear stress
0 , the constant stress layer. Similar to the
boundary layer in an open channel, in this thin layer of uid near the wall
of the pipe, the velocity is given by the law of the wall
u
= f (y + )
Uf

(2.29)

2.1. FLOW CLOSE TO A WALL

2.1.2

23

Mean and turbulence characteristics of ow close

to a smooth wall

In this subsection, mean and turbulence characteristics of the boundary layer

ow close to a wall will be studied in greater detail. As seen in the previous examples, the shear stress near the wall is approximately constant, and
can be taken equal to the wall shear stress (the constant stress layer). This
subsection is concerned with the ow in the constant stress layer. The analysis regarding the mean velocity will be undertaken for two extreme cases:
namely, for small values of y + , and for large values of y + .

Figure 2.4: Shear stress distribution in open channel. Grass (1971).

Mean velocity for small values of y + . Viscous sublayer.
As seen in the preceding paragraphs, the total shear stress is composed
of two parts, the viscous part and the turbulent part (Reynolds stress) (Eq.
2.4). Fig. 2.4 shows the contribution of each part to the total shear stress
as a function of the distance from the wall. The gure indicates that, very
near the wall, the turbulent part is small compared with the viscous part,
du
( uv)
. Hence
dy
=

du
+(
dy

uv)

du
dy

(2.30)

24

CHAPTER 2. STEADY BOUNDARY LAYERS

or
0

du
dy

(2.31)

y+c

(2.32)

Integrating gives
u=

Uf2

The no-slip condition at the wall:

(2.33)

u=0

y=0:

This gives c = 0. Therefore, the mean velocity distribution will be

u=

Uf2

(2.34)

or in terms of the law of the wall (Eq.2.14 and Eq.2.15)

u
= y+
Uf

(2.35)

The velocity varies linearly with the distance from the wall. This analytical
expression agrees well with the experiments for y + < 5. Beyond y + = 5, the
data begins to deviate from the analytical expression (Fig. 2.5). The layer in
which the analytical expression agrees with the experimental data is called
the viscous sublayer:
v

=5

Uf

, or alternatively

+
v

=5

(2.36)

It must be mentioned that, although this very thin layer of uid is called
viscous sublayer, turbulence from the main body of the ow does penetrate
into this layer. This can be seen in the velocity signal in the viscous sublayer,
and also in the wall shear stress signal in the form of turbulent uctuations.
To get a feel of how thin this layer is, for example, for Uf = O(1 cm/s),
2
O(1) = O(0.05 cm) = O(0.5 mm).
v = 5 Uf = 5 10
Mean velocity for large values of y + . Logarithmic layer.
Fig. 2.4 shows that, for large values of y + , the viscous part of the total
shear stress is small compared with the turbulent part, du
( u v ). (It
dy

2.1. FLOW CLOSE TO A WALL

25

should be noted that, although y + is large, yet y is still small compared with
h so that
0 , namely y is in the constant stress layer).
Now, the shear stress generates the velocity variation over the depth; i.e.,
the cause-and-e ect relationship between the shear stress and the velocity
variation du
is such that generates du
.
dy
dy
Since the shear stress for large values of y + is practically uninuenced
by the viscosity, the velocity variation du
should also be uninuenced by the
dy
viscosity; therefore, dropping the viscosity , from Eq. 2.9,
du
d
= F ( 0 , y, )
dy
dy

(2.37)

or

du
, 0 , y, ) = 0
(2.38)
dy
From dimensional analysis, this dimensional equation is converted to the
following nondimensional equation:
(

du 1
)=0
dy Uf

(2.39)

This equation has a single variable, namely

du 1
dy Uf
y

Solving this variable from Eq. 2.39 will give

du 1
=A
dy Uf

(2.40)

in which A is a constant (the solution).

Integrating Eq. 2.40 gives
u = AUf ln y + B1

(2.41)

or in terms of the law of the wall (Eq.2.14 and Eq.2.15):

u = Uf (A ln y + + B)

(2.42)

in which B is another constant. This expression is known as the logarithmic

law.

26

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.5: Velocity distribution. Taken from Monin and Yaglom (1973).
1. Experiments done in pipe ows, in channel ows, and in boundary layer
ows over walls, etc. have all conrmed the logarithmic law with
A = 2.5,

B = 5.1

(2.43)

Constant A is also written in the literature in the form of 1/ , in

which is called von Karman constant, = 0.4. Apparently, A is a
universal constant; no matter what the category of the wall is (smooth,
rough, transitional), this constant always assumes the same value, i.e.,
A = 2.5, as will be demonstrated later in the section.
2. Fig. 2.5 compares the logarithmic law with the experimental data.
It is seen that the logarithmic law agrees quite well with the data in
the range 30 < y + < 500. (It may be noted, however, that the upper

2.1. FLOW CLOSE TO A WALL

27

bound of the latter interval is related to the ow-depth (boundary-layer

thickness) e ect; Obviously, this bound is determined by the condition
hUf
y
h, or alternatively y +
. Hence, depending on the Reynolds
hUf
, the numerical value of the upper bound may change).
number
3. The layer in which the logarithmic law is satised is called the logarithmic layer.
4. The lower bound of the logarithmic layer is often taken as y + = 70 in
the literature. The layer of uid which lies between the viscous sublayer
and the logarithmic layer is called the bu er layer, 5 < y + < 70.
Clearly, in this layer, both the viscous e ects and the turbulence e ects
are equally important.
The above analysis implies that, for the velocity distribution to satisfy
the logarithmic law, the following two conditions have to be met:
1. y should be in the constant stress layer,
y

0,

and for this

in which h is the ow depth in the case of the open channel ow, or

it is the radius in the case of the pipe ow, or the boundary-layer
thickness in the case of the boundary layer over a wall. (The above
condition can be taken as y < 0.1h. However, the data given in Monin
& Yaglom (1971, p. 288-289) implies that it can be relaxed even to y <
(0.2 0.3)h);
2. y should be su ciently large so that the variation of the velocity,
is independent of the viscosity. For this, y + should be
y + > 30

du
,
dy

70

These conditions are to be observed when tting the logarithmic law

(Eq. 2.42) to a measured velocity distribution.
Turbulence quantities.
By turbulence quantities, we mean the Reynolds stresses, namely
ui uj =

u2
vu
wu

uv
v2
wv

uw
vw
w2

(2.44)

28

CHAPTER 2. STEADY BOUNDARY LAYERS

For the present ow,

u w should be zero because there is no variation
of the velocity in the z direction (see Fig. 2.6). Also,
v w should
be zero because there is no mean ow in the y, z plane (no secondary
currents). Furthermore, the Reynolds stress tensor is symmetric with respect
to the diagonal. Therefore, there are only four independent Reynolds stresses.
These are the three diagonal components, and one o -diagonal component,
namely,
u v , or, for convenience, dropping - in the diagonal components,
and in the o -diagonal components:
u 2 , v 2 , w 2 , and

uv

(2.45)

Now, close to the wall, these quantities should depend on the same independent quantities as in the case of the mean velocity (Eq. 2.9), namely
0,

y,

(2.46)

From dimensional analysis, we get the following nondimensional relations

u 2 = Uf f1 (y + ),

v 2 = Uf f2 (y + ),

w 2 = Uf f3 (y + ),

u v = Uf2 f4 (y + )
(2.47)
The functions f1 , ..f4 are universal functions, and their explicit forms are to
be determined from experiments.
For large values of y + , we know that the variation of the velocity with y,
i.e., du
, is independent of the viscosity (see the argument in conjunction with
dy
, and
Eq. 2.37). Since the turbulence is generated by the velocity gradient du
dy
du
since dy is independent of the viscosity, then the turbulence should also be
independent of the viscosity. Therefore, for large values of y + , the viscosity

Figure 2.6: Zero components of the Reynolds stresses.

2.1. FLOW CLOSE TO A WALL

29

Figure 2.7: Turbulence quantities. Taken from Monin and Yaglom (1973).
should drop out in the expressions in Eq. 2.47, meaning that the turbulence
quantities in Eq. 2.47 should tend to constant values:
u2
= f1 (y + )
Uf

A1 ,

v2
= f2 (y + )
Uf

A2 ,

w2
= f3 (y + )
Uf
uv
= f4 (y + )
Uf2

(2.48)

A3 ,
A4

Fig. 2.7 shows the experimental results. As expected, the universal func-

30

CHAPTER 2. STEADY BOUNDARY LAYERS

tions f1 , ..f4 approach zero, as y + goes to zero, and also, in conformity with
Eqs. 2.48, they tend to constant values as y + takes very large values. From
the gure, the latter constant values are
2.3, A2

A1

0.9, A3

1.7, and A4

(2.49)

Example 4 Turbulent energy budget

From Eq. 1.30, the turbulent energy equation for a steady boundary-layer
ow
F
ui
( ui u )
+ t=0
(2.50)
x
x
in which t is given in Eq. 1.32, and F
F = Kt u +

1
u uu +pu
2 i i

ui (

ui
u
+
)
x
xi

(2.51)

The second term in Eq. 2.50 represents the turbulence production, and
the third term the viscous dissipation of turbulent energy.
The rst term, on the other hand, represents the ux of turbulent energy;
it consists of four parts:
1. Kt u represents the transport of turbulent energy by convection;
2.

1
2

ui ui u represents the transport due to turbulent di usion;

3. p u represents the transport due to pressure uctuations; and

4.

ui (

ui
x

u
xi

) represents the transport due to viscous dissipation.

For the present boundary-layer ow, the turbulent energy budget, Eq.
2.50, reduces to
u
F
+ t=0
( uv)
(2.52)
x
y
It can easily be seen that F in the present case consists of only items 2, 3
and 4 above.
Fig. 2.8 shows the contributions to the turbulent energy budget of the
previously mentioned e ects plotted versus the distance from the wall y + . It
is seen that most of the energy production and energy dissipation takes place
near the wall in the bu er layer, namely 5 < y + < 70.

2.1. FLOW CLOSE TO A WALL

31

Figure 2.8: Turbulent energy budget. Laufer (1954).

2.1.3

Mean and turbulence characteristics of ow close

to a rough wall

Mean velocity.
Consider that the wall is now covered with roughness elements (Fig. 2.9).
Let the height of the roughness elements be k. Assume that k is su ciently
large (larger than the thickness of the viscous sublayer) so that it inuences
the ow.
In this case, there is one additional parameter to describe the ow, namely
the roughness height, k. Therefore, Eq. 2.9 will read
u = F ( 0 , y,

, , k)

(2.53)

Now, the total shear stress is given as in the case of the smooth wall (Eq.

32

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.9: Rough wall.

2.4), namely
=

du
+(
dy

(2.54)

uv)

However, the turbulent part of the total shear stress,

present case, consist of two parts:
1. one is induced by the mean velocity gradient,
in the case of the smooth wall); and

du
,
dy

u v , will in the

(in the same way as

2. the other is caused by the vortex shedding from the roughness elements.
(Observations show that lee-wake vortices are shed from the roughness
elements into the main body of the ow in a continuous manner, Sumer
et al., 2001).
in
Clearly, for large distances from the wall, (1) the viscous part du
dy
the total shear stress (Eq. 2.54) is negligible, as discussed in the previous
section, and (2) the vortex-shedding-induced part of
u v is also negligible,
because the vortices shed into the ow cannot penetrate to large distances
due to their limited life time. This means that, for large distances from the
wall, the shear stress is practically independent of the viscosity and the wall
roughness.
Now, as discussed in the previous section, in the cause-and-e ect relationship between the shear stress and the velocity variation, the shear stress
is the cause and the velocity variation du
is the e ect. Since the shear stress
dy
for large distances from the wall is independent of the viscosity and the wall

2.1. FLOW CLOSE TO A WALL

33

roughness, then the velocity variation du

should also be independent of these
dy
e ects. Hence, dropping out these quantities (i.e., and k) in Eq. 2.53,
d
du
= F ( 0 , y, )
dy
dy

(2.55)

This is precisely the same equation as in the case of the smooth wall (Eq.
2.37). Therefore, the result obtained in conjunction with Eq. 2.37 is directly
applicable (Eq. 2.41), i.e.,
u = AUf ln y + B1

(2.56)

Here, A is the same constant as that in the case of the smooth wall, A =
2.5. Our analysis implies that this constant has the same value, irrespective
of the category of the wall (smooth, or rough). Therefore, A must be a
universal constant.
In the present context, the preceding expression may be written as
u = AUf ln(

y
)
y0

(2.57)

in which y0 is called the roughness length (not to be confused with the roughness height k). Eq. 2.53 suggests that this new quantity should be a function
of , k, 0 and ,
y0 = y0 ( , k, 0 , )
(2.58)
and from dimensional analysis,
y0 = k g(

kUf

(2.59)

kU

The explicit form of the function g( f ) can be found from experiments.

Nikuradse carried out such experiments (see Schlichting, 1979); He coated
the walls of circular pipes with sand grains of given size. The sand grains
were glued on the wall in a densely packed manner. The height of Nikuradses
sand roughness will be denoted by ks .
Changing k by ks , Eq. 2.59 will be
y0 = ks g(

ks Uf

(2.60)

According to Nikuradses experiments, there are three categories of wall:

34

CHAPTER 2. STEADY BOUNDARY LAYERS

k U

1. Hydraulically smooth wall when s f < 5. (This is when the roughness

height is smaller than the thickness of the viscous sublayer, Eq. 2.36.
Clearly, in this case, the roughness elements are covered by the viscous
sublayer, and therefore not directly exposed to the main body of the
ow. Hence, the wall, although it is rough, acts as a smooth wall);
k U

2. Completely rough wall when s f > 70. (In this case, the viscous sublayer is completely destructed, and the roughness elements are completely exposed to the main body of the ow); and
3. Transitional wall when 5 <

ks Uf

< 70.

Now, in the case of the hydraulically smooth wall, the function g takes
the following form
ks Uf
a
g(
) = ks Uf
(2.61)
This is because ks will drop out only when g is given in the form depicted
in the preceding equation. Here a is a constant. Inserting the preceding
expression into Eq. 2.60
a
y0 = ks ks Uf
and from Eq. 2.57
u = AUf ln(

y
y
1 yUf
)
) = AUf ln( a ) = AUf ln(
y0
a
Uf

Comparing the preceding equation with Eqs. 2.42 and 2.43 gives the constant
a as 1/9.
In the case of the completely rough wall, on the other hand, the function g is given as
ks Uf
)=b
(2.62)
g(
This is because the viscosity does not play any role, and therefore should
drop out of formulation; and this is possible only when g is given as in the
preceding form. Here, b is another constant. Nikuradses experiments show
that this constant is
ks Uf
1
g(
)=b=
(2.63)
30

2.1. FLOW CLOSE TO A WALL

35

Figure 2.10: Velocity distribution. Rough wall. Grass (1971).

Substituting this into Eq. 2.60, the roughness length will be found
y0 =

ks
30

(2.64)

and from Eq. 2.57

u = AUf ln(

30y
)
ks

(2.65)

This is the logarithmic law for ow over a completely rough wall. Fig. 2.10
compares the above expression with the experimental data (Grass, 1971, Fig.
4). As seen, the logarithmic law begins to deviate from the measured velocity
prole for y 0.2ks . This is not unexpected, because it has been seen that,
for the velocity prole to satisfy the logarithmic law, y should be su ciently
large, so large that the variation of the velocity du
is independent of the
dy
wall roughness; otherwise, the velocity distribution will clearly not satisfy
the logarithmic law, as revealed by Fig. 2.10.
In the case of the transitional wall, the function g is given in Fig. 2.11.
The data can be represented by the following empirical expression:
g(ks+ ) =

1/9
1
+
exp[ 140(ks+ + 6)
+
ks
30

1.7

36

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.11: Variation of function g(ks+ ). Data: Nikuradse (1933), reproduced

1/9
1
from Monin and Yaglom (1973). Solid curve: g(ks+ ) = k+ + 30
exp[ 140(ks+ +
s
6) 1.7 ]
Nikuradses equivalent sand roughness.
The wall roughness is inuenced by various factors such as the height of
the roughness elements, the shape of the roughness elements, and the packing
pattern. Therefore, it will di er from that in Nikuradses experiments. To be
able to compare the results obtained with di erent wall roughnesses, we need
a common base line. The concept Nikuradses equivalent sand roughness
facilitates this.
The idea is to nd Nikuradses equivalent sand roughness for any wall
roughness. Nikuradses equivalent sand roughness, ks , is dened such that
the velocity distribution across the depth is given by Eq. 2.65. In practice,
it is easy to nd ks ; the measured velocity distribution is plotted on a semilogarithmic diagram, and the y intercept of the line which represents the
ks
logarithmic law (Eq. 2.65) will be 30
.
The values of Nikuradses equivalent sand roughness are reported in the
literature for various wall roughnesses, and they are generally in the range
ks =(2 4)k (Kamphuis, 1974, and the review in the paper by Bayazit, 1983).
It may be noted that the upper bound of the previously mentioned range can

2.1. FLOW CLOSE TO A WALL

37

sometimes reach values as much as 5k and even more, depending on the shape
of the roughness elements, the packing pattern as well as the roughnessheight-to-ow-depth ratio (the latter is for large values of this ratio such as
> O(0.3)) (Bayazit, 1983). In particular, ks =2k for walls covered by sand,
pebbles, or stones the size in the range 0.54-46 mm (Kamphuis, 1974); and
ks =(3 6)k for the ground covered with ordinary grass or agricultural crops
(Monin and Yaglom, 1973). Bayazit (1983) reports the following values for
the equivalent sand roughness for various boundaries:
Author
Roughness
ks
Leopold et al. (1964)
Limerinos (1970)
Bayazit (1976)
Charlton et al. (1978)
Hey (1979)
Thompson and Campbell (1979)
Gladki (1979)
Denker (1980)
Bray (1980)
Gri ths (1981)

Gravel
Gravel
Closely packed spheres
Gravel
Gravel
Gravel
Gravel
Closely-packed cylinders
Gravel
Gravel

3.5D84
3.5D84
2.5D
3.5D90
3.5D84
4.5D90
2.5D80
2D
3.5D84 (3.1D90 )
5D50

Figure 2.12: Theoretical wall.

Finally, ks =2.5k for a river bed where the sediment grains are in motion
(k being the grain size). This is when the bed remains plane (i.e., no bed
forms are developed) (Engelund and Hansen, 1967). This corresponds to a
weak sediment transport, and before bed ripples emerge. When the bed is

38

CHAPTER 2. STEADY BOUNDARY LAYERS

covered with ripples, ks is found to be ks =(2 3)k with 2-D ripples (Fredse et
al., 2000), k being the ripple height. In the case of 3-D ripples, the previous
relation can, to a rst approximation, be used. For very high velocities,
ripples are washed away and the bed becomes plane again; This sedimenttransport regime is called the sheet-ow regime (Sumer et al., 1996). In this
case, the ks value can be calculated from the following empirical relations
(Sumer et al., 1996):
1. No-suspension sediment transport (w/Uf > 0.8
ks
= 2 + 0.6
k

2.5

0<

2. Suspension sediment transport (w/Uf < 0.8

ks
1
w
= 4.5 + exp 0.6( )4
k
8
Uf

1):
2

(2.66)

1):
2

2.5

(2.67)

in which k is the sediment grain size, w the fall velocity of sediment

grains and the Shields parameter dened by
=

Uf2
g(s

1)k

(2.68)

in which s is the specic gravity of sediment grains and g the acceleration due to gravity.
Theoretical wall.
In the case of the rough wall, the exact location of the wall is not well
dened. Does it lie at the top of the roughness elements, or does it lie at
the base bottom, or does it lie somewhere between? This issue brings in the
concept of the theoretical wall.
Formally, the theoretical wall is dened as the location from which the y
distances in Eq. 2.65 are measured (Fig. 2.12).
Now, for convenience, change the coordinate to y , the distance from the
base wall. Now suppose that the theoretical bed lies at the distance y1 from
the wall (Fig. 2.12), then the relationship between y and y :
y=y

y1

(2.69)

2.1. FLOW CLOSE TO A WALL

Figure 2.13:

39

Semilog plots of the velocity distribution for three di erent

theoretical-wall locations.

40

CHAPTER 2. STEADY BOUNDARY LAYERS

Hence the logarithmic law (Eq. 2.65) will read

u = AUf ln(

y1 )

30(y
ks

(2.70)

For a given measured velocity prole u(y ), and taking A = 2.5, the
quantities Uf , ks and y1 (i.e., the location of the theoretical wall) can be
determined from Eq. 2.70. The procedure may possibly be best described
by reference to the following example:
In the laboratory, time-averaged velocity proles are measured with a
Laser Doppler Anemometer over a stone-covered bed in an open channel
(the stones the size of k = 3.8 cm, and the ow depth h = 40 cm from
the top of the stones). The measurements are made at four, equally spaced,
vertical sections between the crests of two neigbouring stones. Then, the
space-averaged velocity prole is obtained from the measured, four velocity distributions. Subsequently, the following procedure is adopted to get
the location of the theoretical wall (namely, y1 , or alternatively y, see the
denition sketch in Fig. 2.13), Uf and ks .
1. Plot u (here, u denotes the time- and space-averaged velocity) in a semilog graph for various values of y1 (or alternatively for various values of
y). (The measured velocity proles are plotted for eight values of
y, namely y/k = 0; 0.1; 0.15; 0.2; 0.25; 0.3; 0.35; 0.4. Only three
proles are displayed here, in Fig. 2.13, to keep the gure relatively
simple).
2. Identify the straight line portion of each curve in Fig. 2.13.
3. For this, look at the interval 0.2ks y 0.1h to start with. (The latter
interval can be taken as 0.2ks y (0.2 0.3)h with the upper bound
relaxed if necessary). This is the interval where the logarithmic layer is
supposed to lie. Here the upper boundary 0.1h (or (0.2 0.3)h in the
relaxed case), ensures that the y levels lie in the constant stress layer,
while the lower boundary, 0.2ks , ensures that the variation of u with
respect to y is not inuenced by the boundary roughness, two conditions
necessary for the velocity distribution to satisfy the logarithmic law.
4. Identify the case where the thickness of the logarithmic layer (where the
velocity is represented with a straight line) is largest. The previously

2.1. FLOW CLOSE TO A WALL

41

mentioned eight velocity proles, when analyzed, reveals the following

result:
y
k

0
0.1
0.15
0.2
0.25
0.3
0.35
0.4

Logarithmic layer

Thickness of the
logarithmic layer

1.5 < y(cm) < 5

1.5 < y(cm) < 5
1.5 < y(cm) < 5
1.5 < y(cm) < 6
1.5 < y(cm) < 5
2.5 < y(cm) < 5.5
2.5 < y(cm) < 5
2.5 < y(cm) < 5

3.5
3.5
3.5
4.5
3.5
3.0
2.5
2.5

cm
cm
cm
cm
cm
cm
cm
cm

5. As seen from the preceding table, the case where y/k = 0.2 gives
the thickest logarithmic layer. Therefore, adopt this location as the
location of the theoretical wall (Fig. 2.13 b).
6. The straight line portion of this velocity prole (corresponding to y/k =
0.2, Fig. 2.13 b) has a slope equal to AUf , Eq. 2.70. From this information, nd Uf .
7. Extend the straight line portion of the velocity prole (corresponding
ks
to y/k = 0.2) to nd its y-intercept; this is equal to 30
(Fig. 2.13 b).
Returning to the theoretical wall, research shows that the origin of the
distance y(= y y1 ), namely the level of the theoretical bed, lies (0.15-0.35)k
below the top of the roughness elements (Bayazit, 1976, and 1983).
Turbulence quantities.
The turbulence quantities given in Eq. 2.45 in the case of the rough
wall should depend on the same independent quantities as in the case of
the smooth wall (Eq. 2.46) except that (1) will drop out (no inuence of
viscosity) and (2) ks will be added to the list of the independent quantities:
0,

y, , ks

(2.71)

From dimensional analysis, we get the following nondimensional relations

y
y
y
y
u 2 = Uf g1 ( ), v 2 = Uf g2 ( ), w 2 = Uf g3 ( ), u v = Uf2 g4 ( )
ks
ks
ks
ks
(2.72)

42

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.14: Turbulence quantities. Rough wall. Averaging over both time and
space. Sumer et al. (2001).

Fig. 2.14 shows data regarding u 2 , v 2 , and u v . The data are plotted in a format slightly di erent from that in the above equation, namely,
the actual roughness height is used to normalize y, rather than Nikuradses equivalent sand roughness. Notice the large di erence in the roughness
Reynolds numbers of the two sets of data. Despite this, the good agreement
between these two data may imply that the functions g1, .., g4 are universal.
Fig. 2.15 compares the turbulence quantities obtained in another study;
in these tests, everything else is maintained unchanged, but the wall roughness is changed, to see the inuence of the wall roughness on the end results.
From the gure, the following two points are worth emphasizing:

1. The inuence of the wall roughness disappears with the distance from
the wall, as expected.
2. The distribution of turbulence across the depth is more uniform in the
case of the rough wall than in the case of the smooth wall.

2.2. FLOW ACROSS THE ENTIRE SECTION

43

Figure 2.15: Turbulence quantities. Grass (1971).

2.2

Flow across the entire section

In the preceding section, we have been concerned with the ow close to a

wall (a wall of an open channel, that of a pipe, etc.). In this section we will
turn our attention to the ow across the entire section of the channel/pipe.
Clearly, in this case, there will be an additional parameter, namely the ow
depth in the case of the channel-ow (or the pipe radius in the case of the
pipe ow, or the boundary-layer thickness in the case of the boundary-layer
ow).
Mean ow.
In the analysis developed for ows close to a wall, the mean ow velocity
has been written as a function of four quantities, 0 , y, , (Eq. 2.9). When
we consider the ow across the entire section, however, two changes will take
place:
1. There will be one additional parameter, namely the ow depth (or
the pipe radius, or the boundary-layer thickness) a ecting the ow, as
mentioned in the preceding paragraph (Fig. 2.2); and

44

CHAPTER 2. STEADY BOUNDARY LAYERS

2. The shear stress

(Fig. 2.2),
=

y
)
h

0 (1

will replace the wall shear stress

(2.73)

Hence, the velocity is now dependent of the following ve parameters

u = F ( , y,

, , h)

(2.74)

However, from Eqs. 2.73 and 2.74, the preceding equation may be written as
u = F ( 0 , y,

, , h)

(2.75)

From dimensional analysis, the following nondimensional function may

be obtained
Uf h y
u = Uf (
, )
(2.76)
h
Now, we will follow the same line of thought as in the preceding sections;
at large distances from the wall is
namely, the variation of the velocity du
dy
independent of the viscosity. Dropping out the viscosity in Eq. 2.75,
du
d
= F ( 0 , y, , h)
dy
dy
or

du
, 0 , y, , h) = 0
dy
and from dimensional analysis one obtains
1(

du Uf
=
dy
h

1(

y
)
h

Integrating gives
u = Uf

1(

y
y
) d( ) + c
h
h

and using the boundary condition

u = U0

y=h:
one obtains

U0 = Uf

1(
y=h

y
y
) d( )
h
h

(2.77)

(2.78)

2.2. FLOW ACROSS THE ENTIRE SECTION

45

Figure 2.16: Flow regions in a turbulent boundary layer.

or denoting the integral by another function, f1 ( hy ),
u

y
U0 = Uf f1 ( )
h

(2.79)

in which U0 is the velocity at the free surface in the case of the open-channel
ow, and the center line velocity in the case of the pipe ow. The preceding
equation is known as the velocity-defect law. This velocity distribution is
satised in the outer-ow region, Fig. 2.16.
The explicit form of the function f1 ( hy ) can be found in the following way.
As sketched in Fig. 2.16, the outer region and the logarithmic layer is
expected to overlap over a certain y interval. Therefore, in the overlapping
region, the two velocity distributions, namely the logarithmic law (Eq. 2.42)
and the velocity-defect law (Eq. 2.79), should be identical; therefore from
Eqs. 2.42 and 2.79,
AUf ln(

yUf

) + BUf

y
U0 + Uf f1 ( )
h

(2.80)

and solving the function f1

y
y
f1 ( ) = A ln( ) + B
h
h

(2.81)

Therefore, the velocity-defect law

u

y
U0 = Uf (A ln( ) + B )
h

(2.82)

46

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.17: Velocity distribution over the entire section of the pipe.
1. A is the same universal constant as in the analysis for the near-wall
ow, i.e., A = 2.5.
2. B , on the other hand, is found from experiments as B = 0.8 for pipe
ows (Monin and Yaglom, 1971). However, if Eq. 2.82 is assumed to
be applicable for the entire cross section, then B has to be B = 0.
3. Experiments show that the velocity-defect law begins to deviate from
the measured velocity proles (not very signicantly, however) for hy >
0.2 0.3 (Fig. 2.17).
4. When the general form of the equation for the velocity is considered
(Eq. 2.76), it will be seen that U0 is a function of the Reynolds number.
From Eq. 2.80,
U0 = AUf ln(

hUf

) + Uf (B

B)

(2.83)

5. The velocity-defect law is also given in terms of the cross-sectional

average velocity. In the case of the open-channel ow
1
V =
h

u(y) dy
y=0

and
V = U0

Uf (A

B)

(2.84)

2.2. FLOW ACROSS THE ENTIRE SECTION

47

Figure 2.18: Streamwise turbulence intensity distributions. Compiled by Wei and

Willmarth (1989).
In this case, the velocity-defect law (from Eqs. 2.82 and 2.84) will read
as
y
u V = AUf (ln( ) + 1)
(2.85)
h
Note that, from the preceding equation, the velocity u is equal to the
mean ow velocity, u = V, when y = 0.37h.
6. Finally, it can readily be seen that the velocity-defect law in its form in
Eq. 2.82 is equally valid for the case of the rough wall as well. However,
the velocity U0 in this case will be di erent from that in Eq. 2.83:
U0 = AUf ln(

30h
)
ks

Uf B

(2.86)

48

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.19: Transverse turbulence intensity distributions. Compiled by Wei and

Willmarth (1989).

Turbulence quantities.
The functional forms of the turbulence quantities depicted in Eq. 2.47
hU
will include one additional parameter in the present case, namely f , or
alternatively hV , the Reynolds number.
Figs. 2.18 and 2.19 illustrates how this latter parameter inuences the
turbulence quantities u 2 and v 2 (Wei and Willmarth, 1989). As seen,
near the wall, the e ect of Re disappears, as expected (see Section 2.1.2).
However, away from the wall, the Reynolds number signicantly inuences
2

the turbulence quantities. As Re increases, Uuf , and Uvf increase. This

is basically due to the increase in the ow depth when Re increases; indeed,
the ow depth in terms of the wall parameters is like h+
Re p , p > 0 in
hU
which h+ (= f ). (As Re increases, the turbulence generated near the wall
will be di used to larger and larger h+ , as revealed by Figs. 2.18 and 2.19).

2.3

Turbulence-modelling approach

The preceding sections give a considerable insight into the process of turbulent boundary layers in channels/pipes, and over walls. However, the method
used for the analysis does not enable us to do a parametric study, a study

2.3. TURBULENCE-MODELLING APPROACH

49

where the inuence of various parameters on the boundary layer process can
be studied in a systematic manner. This can be achieved with the aid of
turbulence modelling. A detailed account of turbulence modelling is given in
Chapter 6. We restrict ourselves here only with the so-called mixing-length
model and use it to address the problem of determining the mean velocity
distribution as function of the distance from the wall.

2.3.1

Turbulence modelling of ow close to a wall

Smooth wall.
Our objective is to obtain the velocity distribution, u, near the wall (in
the constant-stress layer). The x component of the Reynolds equation leads
to the following equation (Eq. 2.7):
(=

du
+(
dy

u v )) =

(2.87)

u v . To close the system,

in which there are two unknowns, namely u, and
we need a second equation. Drawing an analogy to Newtons friction law
=

du
dy

the turbulence-induced shear stress can be written as

du
uv = t
dy

(2.88)

(2.89)

in which t is called turbulence viscosity. This is the simplest turbulence

model, and was suggested by Boussinesq in 1877.
The second step is to determine the turbulence viscosity in the above
equation. Making an analogy, this time, to the kinetic theory of gases, t is
written as
(2.90)
m Vt
t =
in which m is what is called the mixing length (interpreted as the length
of penetration of uid particles, or the length over which uid particles/parcels keep their coherence), and Vt is the turbulence velocity (a charv 2 ). The
acteristic velocity of the turbulent motion, for example, Vt
hypothesis leading to Eq. 2.90 is known as Prandtls mixing-length hypothesis. The preceding equation is obtained in a di erent way in Chapter 6,
Section 6.3.2.

50

CHAPTER 2. STEADY BOUNDARY LAYERS

There are two quantities to be ascribed in the preceding equation to

determine the turbulence viscosity: m and Vt . According to Prandtls second
hypothesis, these two quantities are related in the following way (see Chapter
6, Section 6.3.2):
du
(2.91)
Vt = m
dy
Clearly, it is expected that (1) the larger the mixing length, the larger the
turbulence velocity; and also (2) the larger the velocity gradient du
, the
dy
larger the turbulence generation (Section 1.4.3), and therefore the larger the
turbulence velocity, revealing Eq. 2.91.
From Eqs. 2.90 and 2.91, the turbulence viscosity
t

2
m

du
dy

(2.92)

The third step is to determine the mixing length in the preceding equation. van Driest (1956) gives the mixing length as follows
m

= y[1

exp(

y+
)]
Ad

(2.93)

in which is the von Karman constant (=0.4), and Ad is called the damping
coe cient, and taken as 25. As seen,
m

y, for large values of y +

(2.94)

while
0, as y +

(2.95)

The latter implies that there will be no turbulence mixing as the wall is
approached.
Eqs. 2.89, 2.92 and 2.93 are the model equations (three equations); so,
there are four equations altogether (along with the Reynolds equation, Eq.
2.87), and four unknowns, namely, u,
u v , t , and m (the latter already
given by Eq. 2.93). Hence, the system is closed. Therefore the unknowns can
be determined. Let us focus on u.
From Eqs. 2.87, 2.89 and 2.92, one gets
2 du 2
)
m(

dy

du
=
dy

(2.96)

2.3. TURBULENCE-MODELLING APPROACH

51

Figure 2.20: Velocity distribution. Nezu and Rodi (1986).

Solving du
from the preceding equation, and then substituting Eq. 2.93 into
dy
the solution, and subsequently integrating gives
y+

dy +

u = 2Uf
0

1+ 1+4

2 y +2 [1

exp(

y+ 2
)]
Ad

1/2

(2.97)

The velocity distribution described by this expression is known as the van

Driest velocity prole, Fig. 2.20.
1. As seen from the gure, for small values of y + , the van Driest prole
goes to the linear velocity distribution in the viscous sublayer (Eq.
2.35),
u
= y+
(2.98)
Uf
while, for large values of y + , it goes to the logarithmic velocity distribution in the logarithmic layer (Eq. 2.42),
u = Uf (A ln y + + B)

(2.99)

2. Furthermore, the van Driest prole agrees remarkably well with the
measured velocity prole including the bu er layer (5 < y + < 70).

52

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.21: van Driest velocity proles for di erent values of roughness.
Regarding the other two unknowns,
u v and t , inserting Eqs. 2.93
and 2.97 in Eq.2.92, t is obtained, and from this and Eqs. 2.97 and 2.89,
u v is obtained.
Transitional and rough walls.
In this case, the mixing length may be given as (Cebeci and Chang, 1978)
m

= (y +

y)[1

exp(

(y + + y + )
)]
Ad

(2.100)

1. y in the above equation is the distance from the wall; it is measured

from the level where the velocity u is zero. This level lies almost at the
top of the roughness elements. (y here should not be confused with the
distance measured from the theoretical wall. As will be pointed out
later (Item 3 below), apparently y + y is the distance measured from
the theoretical wall).
2.

y is the so-called coordinate displacement, or the coordinate shift (see

Example 5 below), and given by Cebeci and Chang (1978)
y + = 0.9[ ks+

ks+ exp(

ks+
)];
6

5 < ks+ < 2000

(2.101)

2.3. TURBULENCE-MODELLING APPROACH

53

in which ks+ is the roughness Reynolds number (ks being Nikuradses

equivalent sand roughness):
ks+ =

ks Uf

(2.102)

3. Apparently, the distance y + y is the distance measured from the

theoretical wall, as will be shown in Example 5.
Similar to the case of the smooth wall, u can be obtained from the model
equations, with m given in Eq. 2.100:
y+

dy +

u = 2Uf
0

1+ 1+4

2 (y +

y + )2 [1

exp(

(y+ + y + ) 2
)]
Ad

1/2

(2.103)

Fig. 2.21 displays the velocity proles obtained from Eq. 2.103 including the
smooth-wall prole. In Fig. 2.21, a represents the linear prole in the viscous
sublayer (Eq. 2.98), b the logarithmic prole in the case of the hydraulically
smooth wall (Eq. 2.99), and c the logarithmic prole in the case of the
completely rough wall (Eq. 2.65), namely, by putting z = y + y,
30z
)
ks
z being measured from the theoretical wall.

(2.104)

u = AUf ln(

Example 5 Rottas (1962) theory about the coordinate shift

y.

The velocity distribution for large y + values is written in the following

form:
1
u = Uf [ ln y + + c(ks+ )]
(2.105)
in which y is measured from the level where the velocity u = 0 (not to be
confused with y measured from the theoretical wall). The function c(ks+ ) in
the above equation is determined from Nikuradses experiments (Fig. 2.22).
Note that 1 < ks+ < 1000 (see the gure).
For the smooth-wall case, the preceding equation reduces to
u = Uf [

ln y + + c(0)]

(2.106)

Now, in the case of the smooth wall, we know that there are three distinct
regions, namely,

54

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.22: Function c(ks+ ) in Rottas (1962) theory.

1. y + < 5 where the momentum transfer across the depth is due to viscosity (Region 1, Fig. 2.23);
2. 5 < y + < 70 where this transfer is due to both viscosity and turbulence
(Region 2, Fig. 2.23); and
3. y + > 70 where it is practically due to turbulence alone (Region 3, Fig.
2.23).
In the case of the transitional/rough wall, Region 2 and Region 3 are still
present (Fig. 2.24 a). However, Region 1 is practically not existent. Now, we
imagine a ctitious region (Region 1) just beneath Region 2, where a ctitious
momentum transfer takes place. Let us assume that this momentum transfer
is due to viscosity. Let the thickness of this layer be y. So, with the addition
of this ctitious layer, the ow over the rough wall will be identical to that
over the smooth wall. Therefore, we can implement the smooth-wall velocity
distribution (Eq. 2.106) for the present case (with the wall coordinate y+ y,

2.3. TURBULENCE-MODELLING APPROACH

55

Figure 2.23: Three regions in the ow over a smooth wall.

Fig. 2.24 b):
u + U ( y) = Uf [

ln(y + +

y + ) + c(0)]

(2.107)

in which u is the actual velocity and U ( y) is the ctitious velocity that is

to be added to u, to get the ctitious smooth-wall velocity (see the gure).
Considering large values of y + , therefore y + + y + y + , the above equation
becomes
1
u + U ( y) Uf [ ln(y + ) + c(0)]
(2.108)
The ctitious velocity U ( y) can be calculated from, for example, Eq. 2.97,
replacing y + in Eq. 2.97 by y + ( y + cannot be calculated from Eq. 2.106
because the latter equation is valid only for large values of y + ). Therefore,
from Eqs. 2.108 and 2.97, the actual velocity is obtained as
u = Uf [

ln(y + ) + c(0)]

U ( y+)

(2.109)

On the other hand, this velocity is given independently in Eq. 2.105. So,
setting these two equations equal,
Uf [

ln(y + ) + c(0)]

U ( y+)

one gets
c(ks+ )

c(0) +

Uf [

ln y + + c(ks+ )]

U ( y+)
=0
Uf

(2.110)

(2.111)

56

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.24: Flow over a rough wall in Rottas (1962) theory.

Given the function c(ks+ ) (Fig. 2.22), and U ( y + ) from Eq. 2.97,
y+

dy +

U ( y ) = 2Uf
0

1+ 1+4

y +2 [1

exp(

y+ 2
)]
Ad

1/2

the coordinate shift y + can be calculated from Eq. 2.111. Rotta (1962)
carried out this calculation. Fig. 2.25 shows the result.
It may be noted that the variation depicted in Fig. 2.25 agrees fairly well
with the expression given by Cebeci and Chang (1978) in Eq. 2.101.
As a nal remark, the coordinate shift normalized by the roughness
height, ky , can be found from Eq. 2.101 (where ks may be taken as 2k).
The result of this exercise is plotted in Fig. 2.26. As seen, the ky values
exhibited in the gure agree rather well with the experimental observations
that the level of the theoretical bed lies (0.15-0.35)k below the top of the
roughness elements, given in conjunction with the theoretical bed in Section

2.3. TURBULENCE-MODELLING APPROACH

57

Figure 2.25: Coordinate shift in Rottas (1962) theory.

2.1.3. This implies that the coordinate shift y may be interpreted as the
distance of the theoretical wall from the top of the roughness elements.

2.3.2

Turbulence modelling of ow across the entire

section

The velocity distribution given in Eq. 2.103 is obviously valid near the wall.
When applied for the entire depth, the velocities predicted from this equation are apparently smaller than the measured velocities. This di erence is
corrected by the so-called Coles wake function (see Coleman and Alanso,
1983)
y+
y+
( ) ( + ) = ( ) 2 sin2 ( + )
(2.112)
h
2h
in which is a wake strength coe cient, h is the ow-depth/pipe-radius, or
the boundary-layer thickness.
When this equation is used together with Eq. 2.103, it does not produce
equal to zero at the center line of a pipe, or at
a prole with a derivative du
dy
the free-stream edge of the boundary layer over a wall (clearly, in the case of

58

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.26: Coordinate shift

(1978). (ks is taken as 2k ).

y from Rotta (1962) and Cebeci and Chang

pipe ows and boundary-layer ows, this derivative has to be zero). To avoid
this drawback, Coles wake function has been corrected. The nal form of
the velocity expression (including the corrected wake function) is as follows
(Coleman and Alanso, 1983):
u
Uf

y+

2 dy +

=
0

1+ 1+4

y+ 2
+( + ) (1
h

2 (y +

y + )2 [1

y+ 2
y+
2
) + ( ) ( + ) [3
h+
h

(2.113)
1
exp(

(y + + y+ ) 2
)]
Ad

y+
2 +]
h

Regarding the values, this parameter basically depends on two factors:

(1) the driving pressure gradient, and (2) the degree of turbulence in the outer
region. A convergent ow (favourable pressure gradient) tends to produce low
values of , while a divergent ow (adverse pressure gradient) produces high
values of . Likewise, if the turbulence in the outer region is intense, will
have low values, while, the turbulence in the outer region is weak,
will
have large values (Coleman and Alanso, 1983). The latter authors report a
range of from 0 to 0.55.

2.4. FLOW RESISTANCE

2.4

59

Flow resistance

The question here is that, given the mean ow velocity V, how we can calculate the friction velocity Uf , i.e.,
Uf =

(2.114)

This issue may be best described by reference to a circular pipe ow.

For a smooth circular pipe, the ow velocity is given by Eq. 2.42 with
coe cients given in Eq. 2.43. Schlichting (1979, p. 603) takes it in the
following, slightly di erent form:
u = Uf (2.5 ln y + + 5.5)

(2.115)

Assuming that this velocity distribution is, to a rst approximation, valid

across the entire section, and therefore putting y = R, the pipe radius, will
give:
RUf
) + 5.5)
(2.116)
U0 = Uf (2.5 ln(
On the other hand, we have obtained in the preceding paragraphs a relationship between the mean ow velocity and the maximum velocity (Eq. 2.84),
which reads
V = U0 (A B )Uf
(2.117)
Now, Schlichting (1979, p.609) takes this relation in the following form
V = U0

(2.118)

3.75Uf

Combining Eq. 2.116 with Eq. 2.118, we obtain the following equation
(Schlichting, 1979, p. 610)
1 V
VD
Uf
= 2.035 log(
8 )
V
8 Uf

0.91

(2.119)

in which D is the pipe diameter. (Note that log in the above equation should
not be confused with the natural logarithm). The coe cients can be adjusted
slightly so that the preceding equation agrees with the experimental data.
This latter exercise gives the following relation
1 V
VD
Uf
= 2.0 log(
8 )
V
8 Uf

0.8

(2.120)

60

CHAPTER 2. STEADY BOUNDARY LAYERS

This is Prandtls universal law of friction for smooth pipes (Schlichting, 1979,
p.611).
In the case of a rough circular pipe, we can do the same exercise, but this
time with the velocity distribution given by Eq. 2.65
u = AUf ln(

30y
)
ks

(2.121)

30R
)
ks

(2.122)

Putting y = R in the above equation

U0 = AUf ln(

and combining Eqs. 2.122 and 2.118, we get (Schlichting, 1979, p. 621)
Uf
=
V

1
8(2 log kRs + 1.68)

(2.123)

A comparison with Nikuradses experimental results shows that closer agreement can be obtained, if the constant 1.68 is replaced by 1.74 (Schlichting,
1979, p.621):
Uf
1
=
(2.124)
V
8(2 log kRs + 1.74)
In the case of a transitional-wall-category circular pipe, Colebrook and
White obtains the following resistance relation (Schlichting, 1979, p.621)
1 V
= 1.74
8 Uf

2 log(

18.7
ks
)
+ VD
R ( )( 8 Uf )
V

(2.125)

Given the mean-ow velocity, the friction velocity can easily be calculated
from Eqs. 2.120, 2.124 or 2.125.
In the case of a non-circular pipe ow or an open ow, the preceding
resistance relations can be used provided that the pipe diameter D should
be replaced by 4rh in which rh is the hydraulic radius, equal to the crosssectional area divided by the wetted perimeter (Schlichting, 1979, p. 622).
For example, in the case of an open channel with a rough bed, the resistance
relation can be found as
V
14.8rh
= 5.75 log(
)
Uf
ks

(2.126)

2.5. BURSTING PROCESS

or

61

V
14.8rh
= 2.5 ln(
)
Uf
ks

(2.127)

It may be noted that the ratio V /Uf may be traditionally written in terms
of the friction coe cient, f, as
Uf =

f
V
2

(2.128)

Figure 2.27: Snapshot of ow near the wall.

2.5

Bursting process

Experimental work conducted in 1960s and 70s has shown that the nature
of the ow pattern near the wall in a turbulent boundary layer is repetitive;
the ow near the wall occurs in the form of a quasi-cyclic process, called the
bursting process. Reviews of the subject can be found in Laufer (1975), Hinze
(1975, pp. 659-668), Cantwell (1981), Grass (1983) and Nezu and Nakagawa
(1993).

62

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.28: Flow separation induced by adverse pressure gradient.

The description of the bursting process in the following paragraphs is
mainly based on the work of O en and Kline (1973, 1975).
Fig. 2.27 gives a snapshot of the ow picture near the wall. As seen, the
ow consists of a transverse alternation of low-speed (A) and high-speed (B)
regions. These are called low-speed and high-speed streaks. The low-speed
and high-speed streaks form locally and temporarily.
Similar to the separation of a boundary layer subject to an adverse pressure gradient (such as, for example, the boundary layer separation over a
cylinder, see Fig. 2.28 a), a low-speed wall streak is subject to separation
due to a local and temporary adverse pressure gradient xp , Fig. 2.28 b. This
causes this coherent, low-speed uid to entrain into the main body of the
ow. This is called the ejection event. See Fig. 2.29, Frames 1 and 2.
The uid ejected into the ow grows in size, as it is convected downstream,
while keeping its coherent entity. This continues for some period of time (Fig.
2.29, Frame 3), and nally, at some point (t = T ), the previously mentioned
coherent structure breaks up (Fig. 2.29, Frame 4). At this point, some uid
from the coherent structure returns back to the near-wall region (Fig. 2.29,
Frame 5). This uid impinges the wall (in the form of a high-speed wall
streak, called the in-rush event), and spreads out sideways (Fig. 2.29, Frame
6). Two-such neighbouring high-speed streaks eventually merge (Fig. 2.29,
Frame 6), and retard the ow there, and, as a result, a new low-speed wall
streak is formed. The time period spent from the formation of the low-speed
streak at time t = 0 (Fig. 2.29, Frame 1) to the formation of the next one
at time t = T1 (Fig. 2.29, Frame 6) is called the bursting period. After the
formation of this new low-speed streak, the same chain of events occurs, and
the process continues in a repetitive manner.
Fig. 2.30 displays two snapshots (viewed from the side), illustrating the

2.5. BURSTING PROCESS

Figure 2.29: Sequence of ow pictures over one bursting period.

63

64

CHAPTER 2. STEADY BOUNDARY LAYERS

Figure 2.30: (a): ejection; and (b): in-rush events. Grass (1971).
ejection event (Fig. 2.30, a), and the in-rush event (Fig. 2.30 b).
The bursting process has been studied in the case of rough walls, and it
was found that practically the same kind of quasi-cyclic process takes place
(Grass, 1971, and Grass et al., 1991).
The rest of this section is concerned with quantitative information on the
bursting process.
1. The mean spacing of the low-speed wall streaks (Fig. 2.29, Frame 1) is
+

Uf

100

(2.129)

(see for example, Lee et al., 1974)

2. Fluid ejections originate in
5 < y + < 50

(2.130)

(Nychas et al., 1973). Corino and Brodkey (1969), on the other hand,
report that lower half of the viscous sublayer y + < 2.5 is essentially

2.5. BURSTING PROCESS

65

passive and the rest (2.5 < y + < 5) active, being inuenced by the
quasi-cyclic, bursting events occurring in 5 < y + < 70.
3. The x and z extents of ejected low-speed uid structures involve
dimensions of the order of
20 to 40 of x+ , and 15 to 20 of z +

(2.131)

4. Ejected uid elements reach y + s of

y + = 80

100

(2.132)

(Nychas et al., 1973). Praturi and Brodkey (1978) report that, in some
rare cases, the ejected uid elements travel up to 300 y + ).
5. The mean streamwise distance from the onset of lift-up of a low-speed
wall streak to the break-up of any sign of coherency is about
1300 in x+

(2.133)

(O en and Kline, 1973).

6. The mean time between the onset of ejection and break-up (the life
time of a burst) is
T U0
= 2.3 for the smooth boundary data, and
h
1.3 for the rough boundary data

(2.134)

of the study by Jackson (1976). As seen, T scales with the outer ow

parameters, namely U0 , the free-stream velocity (or the velocity at the
free surface in the case of the open channel ow), and h is the boundarylayer thickness/the ow depth.
7. Finally, the mean bursting period
T1 U0
=5
h

(2.135)

(Jackson, 1976). Note that this, too, scales with the outer ow parameters.

66

CHAPTER 2. STEADY BOUNDARY LAYERS

There have been several implications of this newly discovered aspect of

turbulent boundary-layer ows. One of them is the process of sediment
suspension in rivers and in the marine environment. The work conducted in
1970s and 80s indicated that the bursting process is the key mechanism for
sediment suspension from/near the bed (see, for example, Sumer and Oguz,
1978, Sumer and Deigaard, 1981, Grass, 1983, and Sumer, 1986).

2.6

Appendix. Dimensional analysis

We may write the unit of any hydrodynamic quantity in the following form
(2.136)

[A] = L T K

in which [A] is the unit of A, and L, T, and K are the fundamental units,
namely, L is the length, T the time, and K is the force. , and are
positive or negative integers, or they may be zero.
Now, Buckinghams Pi Theorem states the following.
A dimensional function
f (A1 , A2 , ...., An ) = 0

(2.137)

can always be converted to the following nondimensional function

F ( 1,

2 , ...... n r )

(2.138)

=0

in which Ai (i = 1, .....n) are the dimensional quantities, and r is the number

of the fundamental units (normally r = 3, i.e., the length, the time and the
force). The quantities i (i = 1, ..., n r) are the nondimensional quantities
and calculated from
1
1

n r

= Ax1 1 Ay21 Az31 A4

= Ax1 2 Ay22 Az32 A4
..........
= Ax1 n r Ay2n r Az3n r A4

Note that the above combinations of Ai (for each

that

i)

(2.139)

should be selected so

1. The unit of any one of Ai should not be a combination of the units of

the others; and

2.6. APPENDIX. DIMENSIONAL ANALYSIS

67

2. Each and every fundamental unit should be included in at least one of

the Ai s.
The advantage of the dimensional analysis is that, rst of all, it translates the dimensional function into a nondimensional function in a systematic way, secondly and more importantly, in doing so, the number of the
quantities included in the dimensional function is reduced by r (normally
r = 3).
Example 6 Law of the wall.
In Section 2.1.1, the ow velocity is written as a function of the following
nondimensional quantities (Eq. 2.9)
u = F ( 0 , y,

(2.140)

, )

or
f ( 0,

(2.141)

, , u, y) = 0

As seen, n = 5, and therefore the number of nondimensional quantities will

be n r = 5 3 = 2. Hence
F ( 1,

2)

(2.142)

=0

in which the nondimensional quantities

and

x1 y1
0

z1

(2.143)

x2 y2
0

z2

(2.144)

and
The left and right hand sides of Eq. 2.143 should have the same units.
Writing the units of the quantities on both sides of the equation according
to Eq. 2.136,
L0 T 0 K 0 = (KL 2 )x1 (KL 2 T )y1 (KL 4 T 2 )z1 (LT

(2.145)

T y1 +2z1 1 K x1 +y1 +z1

(2.146)

or
L0 T 0 K 0 = L

2x1 2y1 4z1 +1

68

CHAPTER 2. STEADY BOUNDARY LAYERS

which requires
(2.147)

2y1 4z1 + 1 = 0,
y1 + 2z1 1 = 0,
x1 + y1 + z1 = 0

2x1

This gives

1
1
, y1 = 0, and z1 =
2
2
Hence the rst nondimensional quantity
x1 =

1/2 0 1/2
0

u=

u
u
=
1/2
( )
Uf
0

(2.148)

(2.149)

Regarding the second nondimensional quantity, from Eq. 2.144

L0 T 0 K 0 = (KL 2 )x2 (KL 2 T )y2 (KL 4 T 2 )z2 L

(2.150)

and
2x2

(2.151)

2y2

4z2 + 1 = 0,
y2 + 2z2 = 0,
x2 + y2 + z2 = 0

and

1
x2 = , y2 = 1, and
2
Then, the second nondimensional quantity
2

1/2
0

1 1/2

y=

z2 =

y( 0 )1/2

1
2

yUf

(2.152)

(2.153)

Therefore, the nondimensional function will be

F(

u yUf
)=0
,
Uf

(2.154)

or solving the velocity

yUf
u
)
= f(
Uf
This relation is known as the law of the wall.

(2.155)

2.7. REFERENCES

2.7

69

References

1. Bayazit, M. (1976): Free surface ow in a channel of large relative

roughness, J. Hyd. Res., 14, no. 2, 1976, p. 115.
2. Bayazit, M. (1983): Flow structure and sediment transport mechanics
in steep channels. In: Mechanics of Sediment Transport, Proc. of
Euromech 156, Istanbul, 12-14. July 1982, A.A. Balkema (Rotterdam),
p.197.
3. Cantwell, B.J. (1981): Organized motion in turbulent ow. Ann.
Rev. Fluid Mech., vol 13, p.457.
4. Cebeci, T. and Chang, K.C. (1978): Calculation of incompressible
rough-wall boundary-layer ows. AIAA Journal, vol. 16, No. 7,
p.730.
5. Coleman, N.L. and Alanso, C.V. (1983): Two-Dimensional channel
ows over rough surfaces. J. Hydraul. Eng., Proc. ASCE, vol 109,
No. 2, p.175.
6. Corino, E.R. and Brodkey, R.S. (1969): A visual investigation of the
wall region in turbulent ow. J. Fluid Mech., vol. 37, p.1.
7. Engelund, F. and Hansen, E. (1967): A Monograph on Sediment Transport in Alluvial Streams, Teknisk Forlag, Copenhagen.
8. Fredse, J. Andersen, K.H. and Sumer, B.M. (2000): Wave plus current over a ripple-covered bed. Coastal Engineering.
9. Grass, A.J. (1971): Structural features of turbulent ow over smooth
and rough boundaries. J. Fluid Mech., vol. 50, p.233.
10. Grass, A.J. (1983): The inuence of boundary layer turbulence on the
mechanics of sediment transport. In: Mechanics of Sediment Transport, Proc. of Euromech 156, Istanbul, 12-14. July 1982, A.A. Balkema
(Rotterdam).
11. Grass, A.J., Stuart, R.J. and Mansour-Tehrani, M. (1991): Vortical structures and coherent motion in turbulent ow over smooth and
rough boundaries, Phil. Trans. R. Soc. London A, vol. 336, p. 35.

70

CHAPTER 2. STEADY BOUNDARY LAYERS

12. Hinze, J.O. (1975): Turbulence. McGraw-Hill.

13. Jackson, R.G. (1976): Sedimentological and uid-dynamic implications of the turbulent bursting phenomenon in geopyhsical ows. J.
Fluid Mech., vol. 77, p. 531.
14. Kamphuis, J. W. (1974): Determination of sand roughness for xed
beds. J. Hydraulic Res., Vol. 12, No.2, p.193.
15. Laufer, J. (1954): The Structure of Turbulence in Fully Developed Pipe
Flow. NACA, Report No. 1174.
16. Laufer, J. (1975): New trends in experimental turbulence research.
Ann. Review of Fluid Mech., vol. 7, p.307.
17. Lee, M.K., Eckelman, L.D., Hanratty, T.J. (1974): Identication of
turbulent wall eddies through the phase relation of the components of
the uctuating velocity gradient. J. Fluid Mech., vol. 66, p. 17.
18. Monin, A. S. and Yaglom, A. M. (1973): Statistical Fluid Mechanics:
Mechanics of Turbulence. MIT Press, Cambridge, Mass.
19. Nezu, I. and Rodi, W. (1986): Open-Channel ow measurements with
a laser Doppler anemometry. ASCE J. Hyd. Eng., vol. 112, No. 5,
335-355.
20. Nezu, I. and Nakagawa, H. (1993): Turbulence in Open-Channel Flow.
IAHR Monograph, A.A. Balkema (Roterdam).
21. Nychas, S.G., Hershey, H.C. and Brodkey, R.S. (1973): A visual study
of turbulent shear ow. J. Fluid Mech., vol. 61, p. 513.
22. O en, G.R. and Kline, S.J. (1973): Experiments on the velocity characteristics of Bursts and on the Interactions Between the Inner and
Outer Regions of a Turbulent Boundary Layer. Dept. of Mech. Engrg.,
Stanford University, Report MD-31.
23. O en, G.R. and Kline, S.J. (1975): A proposed model of the bursting
process in turbulent boundary layers. J. Fluid Mech.., vol. 70, p. 209.

2.7. REFERENCES

71

24. Praturi, A.K. and Brodkey, R.S. (1978): A stereoscopic visual study
of coherent structures in turbulent shear ow. J. Fluid Mech., vol. 89,
p.251.
25. Rotta, J.C. (1962). Turbulent boundary layers in incompressible ow.
Progress in Aerospace Science, vol. 2, p.1.
26. Schlichting, H. (1979). Boundary-Layer Theory. McGraw-Hill.
27. Sumer, B.M. (1986): Recent Developments on the Mechanics of sediment suspension, General-Lecture Paper, in the book: Euromech 192:
Transport of suspended solids in open channels, A.A. Balkema Publishers, Rotterdam, 1986.
28. Sumer, B.M., Cokgor, S. and Fredse, J. (2001): Suction removal of
sediment from between armour blocks. Journal of Hydraulic Engineering, ASCE, vol. 127, No. 4, pp. 293-306, 2001.
29. Sumer, B.M. and Deigaard, R. (1981): Particle motions near the bottom in turbulent ow in an open channel - Part 2. Journal of Fluid
Mechanics, Vol. 109, p. 311.
30. Sumer, B.M. and Oguz, B. (1978): Particle motions near the bottom
in turbulent ow in an open channel. Journal of Fluid Mechanics, Vol.
86, p. 109, 1978.
31. Sumer, B.M., Kozakiewicz, A., Fredse, J. and Deigaard, R. (1996):
Velocity and concentration proles in the sheet ow layer of movable
bed. Journal of Hydraulic Engineering, ASCE, vol. 122, No. 10,
549-558.
32. van Driest, E.R. (1956): On turbulent ow near a wall. J. Aeronautical Sciences, vol. 23, p.1007.
33. Wei, T. and Willmarth, W.W. (1989): Reynolds-number e ects on
the structure of a turbulent channel ow. J. Fluid Mech., vol. 204, p.
57.

72

CHAPTER 2. STEADY BOUNDARY LAYERS

Chapter 3
Statistical analysis
As has been seen in the previous chapters, hydrodynamic quantities such
as the velocity components appear to be random variables. Although recent research has shown that the near-wall turbulent-boundary-layer process
is repetitive in nature (Chapter 3, Bursting Process), unless special pattern recognition techniques are used to detect these repetitive, quasi-cyclic
events, the turbulence signals even near the wall appear to be largely random. Therefore, it is imperative to use statistical techniques to study the
turbulence.
Three topics will be discussed in the present chapter. First we will briey
study the probability density function of the turbulent velocity, then we will
concentrate on the correlation and subsequently spectral analysis of turbulence. In the entire analysis, it will be assumed that the turbulence is isotropic
unless otherwise stated.

3.1

Probability density function

Consider, for example, the x component of the velocity, u (Fig. 3.1 a). The
velocity u (or its uctuating component u ) is a random variable. Thus, u
(or u ) must have a probability density function (p.d.f.) such that
p(u ) du = Pr[u < U < u + du ]

(3.1)

and
p(u ) du = 1
73

(3.2)

74

CHAPTER 3. STATISTICAL ANALYSIS

Figure 3.1: Time series and p.d.f. of velocity.

in which p(u ) is the p.d.f. of the uctuating velocity u , and Pr[u < U <
u + du ] is the probability of occurrence of the uctuating velocity between
u and u + du .
Experiments show that p(u ) can be represented by the Gaussian (normal)
distribution
1
u2
p(u ) =
)
(3.3)
exp(
2 2u
u 2
in which

is the standard deviation of the velocity of u :

u

(u

u)2 =

u2

(3.4)

The signicance of u is that the probability of occurrence of u outside the

range, for example, 2 u < u < +2 u is only 4.6%, while that outside the
range 3 u < u < +3 u is even much smaller, namely 0.3% (if the p.d.f. is
represented by the Gaussian distribution).
It may be noted that the p.d.f. of the uctuating velocity components
may be di erent from the Gaussian distribution, with non-zero skewness,
particularly near the wall (see the measurements made very close to the wall,
reported in Eckelmann, 1974). This may be linked to the bursting process
(Chapter 3).
Fig. 3.2 shows the p.d.f.s (in the form of frequency histograms) for the
two components of the turbulent velocities, u and v, near the wall for three
di erent boundaries, namely the smooth, transitional and rough walls. Note
the non-zero skewness in all distributions.
The statistical moments (or central moments) also give some information
on the statistical properties of the velocities and other quantities:

3.1. PROBABILITY DENSITY FUNCTION

75

Figure 3.2: Frequency histograms of u and v velocity uctuations. Grass (1971).

1. The rst statistical moment is the mean
u=

(3.5)

u p(u) du

2. The second central moment gives the variance

2
u

(u

u)2 p(u) du =

u 2 p(u ) du

(3.6)

u)3 p(u) du =

u 3 p(u ) du

(3.7)

3. The third central moment,

u3 =

(u

76

CHAPTER 3. STATISTICAL ANALYSIS

gives the skewness factor :
Su =

u3
(

(3.8)

2)2
u

4. The fourth central moment,

u4 =

(u

u)4 p(u) du =

u 4 p(u ) du

(3.9)

gives the atness factor:

Fu =

3.2

u4
( 2u )2

Correlation analysis

There are two kinds of correlations: (1) Space correlations, and (2) time
correlations. We will rst focus on the space correlations.

3.2.1

Space correlations

Denitions.
Fig. 3.3 gives the denition sketch. Consider the i component of the
uctuating velocity at Point A, (ui )A , and j component of the uctuating velocity at Point B, (uj )B . Suppose that these velocities are measured
simultaneously.
Two-point double correlation is dened by
(Oij )AB = (ui )A (uj )B

(3.10)

in which the overbar indicates the time averaging (Eq. 1.1). There are
other correlations as well: three-point, four-point,..., and triple, quadruple,...correlations.
Two most important two-point double correlations (Fig. 3.4) are
u (x) u (x + r)

(3.11)

v (x) v (x + r)

(3.12)

and

3.2. CORRELATION ANALYSIS

77

Figure 3.3: Correlation measurement.

in which r is the separation between the two points.
For convenience, these correlations are normalized by u 2 and v 2 , respectively:
u (x) u (x + r)
(3.13)
f (r) =
u2
v (x) v (x + r)
g(r) =
(3.14)
v2
f (r) and g(r) are called the correlation coe cients.
The physical interpretation of the correlation functions/coe cients is as
follows. The larger the correlation, the better correlated the motions at the
two measurement points, and therefore the more likely that these two points
belong to the same coherent structure. This would obviously give an idea
about the length scale of the coherent structures, as will be detailed later in
the section.
Properties of two-point double correlations.
1. The correlation coe cients are unity at r = 0:
f (0) = g(0) = 1

(3.15)

78

CHAPTER 3. STATISTICAL ANALYSIS

Figure 3.4: Correlation measurement.

(from Eqs. 3.13 and 3.14).
2. There is symmetry between positive and negative separations:
f (r) = f ( r), and g(r) = g( r)

(3.16)

(from the requirement of homogeneity).

3. The correlation coe cients are smaller than or equal to unity:
f (r)

1, and g(r)

(3.17)

This can easily be seen from

(u (x)

u (x + r))2 = u 2 (x) + u 2 (x + r) 2u (x)u (x + r)

= 2(u 2 u (x)u (x + r))

Since the left hand side of the equation

0, hence f (r) 1.

0, then u 2

u (x)u (x + r)

4. The correlation coe cients tend to zero as the separation distance goes
to innity
f (r)
0, and g(r)
0, as r
(3.18)
This is because when r is very large, there will be as many u (x)u (x+r)
products with the negative sign as with the positive sign, therefore,
when averaged, u (x)u (x + r) will be zero. Physically, r is so large
that the two points will not lie in the same coherent structure, therefore, their motions will be uncorrelated, meaning that the correlation
coe cients should be zero for such large r distances.

3.2. CORRELATION ANALYSIS

79

Figure 3.5: Correlation coe cients f and g, and micro scales.

5. While f (r) is always positive, g(r) can take negative values as well. To
show this, we employ
f
1
1
=
(r2 f (r))
rg(r) = rf (r) + r2
2
r
2 r

(3.19)

This is the relationship between f (r) and g(r) in isotropic turbulence,

as will be detailed in the following paragraphs. Integrating gives

1
rg(r)dr = [r2 f (r)]0 = 0
2

(3.20)

provided that r2 f (r) = 0 as r

. (Experiments show that f (r) is
like f (r) e r , then it is easy to see that r2 f (r)
0 as r
.) For
the integral on the left hand side of the equation to be zero, clearly
g(r) must have positive and negative values.
The general forms of the functions of f (r) and g(r) are shown in Fig. 3.5.
Finally, we note the following. There is a special class of turbulence,
called isotropic turbulence where turbulence characteristics such as u 2 are
independent of the rotation of the axes. This implies that u 2 = v 2 = w 2 .
Turbulence in the core region in a wind tunnel (where the wall e ects
are practically nonexistent) may be considered to be isotropic. In such a
turbulence eld, the correlations coe cients are related with the following
equation
r
g(r) = f (r) +
f (r)
(3.21)
2 r
(see Hinze, 1959, p. 150).

80

CHAPTER 3. STATISTICAL ANALYSIS

Scales of turbulence.
Micro scales. Expanding f (r) as a Taylor series, and for small values
of r, gives
1 2f
f (r) = 1 + ( 2 )r=0 r2 + ...
(3.22)
2 r
Note that the terms with rn (n being an odd number, n = 1, 3, 5, ..) will drop
because f (r) = f ( r).
Now, the formal denition of the micro time scale is
1

2
f

1 2f
(
)r=0
2 r2

(3.23)

in which f is the micro length scale of turbulence. Inserting the preceding

equation in Eq. 3.22, one gets
f (r) = 1

r2

(3.24)

2
f

As seen from Fig. 3.5, this is a parabola t to the correlation coe cient for
small values of r , and f is the x intercept of the tted parabola.
Likewise,
r2
(3.25)
g(r) = 1
2
g

in which g is another micro length scale.

These length scales can also be expressed in terms of the uctuatingvelocity gradients:
1
u 2
1
)
(
(3.26)
2 =
2u 2 r
f
and
1
2
g

1
v 2
)
(
2
r
2v

(3.27)

(see Raudkivi and Callander, 1975, p. 167).

f and g are not independent of each other. In isotropic turbulence,
from Eq. 3.21,
2

g(r))r=0 = 2(
2

r2

f (r))r=0

(3.28)

3.2. CORRELATION ANALYSIS

81

and inserting Eqs. 3.24 and 3.25,

f

(3.29)

Macro scales. The macro scales of turbulence are dened by

f

f (r)dr

(3.30)

g(r)dr

(3.31)

and
g

=
0

This length scale is indicated in Fig. 3.6 where the area of the rectangle
1 is equal to the area under the correlation curve.

Figure 3.6: Denition of macro scale.

or g can be interpreted as the length scales of coherent structures
(or the so-called eddies, the rotating entities) of turbulence.
Similar to the micro scales, the macro scales are also related to each other.
In isotropic turbulence, integrating Eq. 3.21 gives
f

g(r)dr =
0

1
f (r)dr + [rf (r) |0
2

1
2

f (r)dr

(3.32)

which gives
f

3.2.2

=2

(3.33)

Time correlations

There are two kinds of time correlations: Eulerian time correlations, and
Lagrangian time correlations.

82

CHAPTER 3. STATISTICAL ANALYSIS

The Eulerian time correlation is dened by
RE (t) =

u ( )u ( + t)
u2

(3.34)

This is for the x component of the velocity. Likewise, other Eulerian correlation coe cient can also be dened for the y and z components of the
velocity.
Similar to space correlations, one has
RE (0) = 1

(3.35)

RE (t)

(3.36)

0, as t

(3.37)

and
RE (t)

(Similar relations for the other components of the velocity).

The micro and macro time scales can also be dened, similar to the space
correlations.
The micro time scale is dened by
1
2
E

1
(
2

RE (t)
)t=0
t2

(3.38)

(cf. Eq.3.23), or
RE (t) = 1

t2
2
E

(3.39)

(cf. Eq. 3.24).

The macro time scale, on the other hand, is dened by
TE =

RE (t)dt

(3.40)

The macro time scale can be interpreted as the time during which a coherent
structure (or an eddy) passes the measurement point.
We expect that the macro time scale and the macro length scale are related. To nd this relation, G.I. Taylor made the so-called frozen-turbulence
approximation. Consider the coherent structure at time t in Fig. 3.7. The
size of this coherent structure is f , the macro length scale of turbulence.

3.2. CORRELATION ANALYSIS

83

Figure 3.7: Taylors frozen turbulence approximation.

Consider the measurement point A. Now, with the frozen-turbulence approximation, the size of the coherent structure will remain unchanged, as it
is convected downstream. The time for the coherent structure to pass Point
A is TE (by virtue of the physical interpretation in the previous paragraph).
Therefore, the distance that the coherent structure travels can, to a rst
approximation, be written as uTE in which u is the mean velocity at the
measurement point. From the gure, this distance is equal to the size of the
coherent structure, hence
f

= uTE

(3.41)

This relationship enables the length scale to be calculated from the knowledge
of the time scale. (The latter is easier to measure than the former).
The Lagrangian time correlation is dened in the same way as in the case
of the Eulerian time correlation. However, the velocity is now measured by
tracking the uid particle, as the particle travels. Therefore, the Eulerian
time series (which is the basis for the Eulerian time correlation) is to be
replaced by the time series from the Lagrangian particle tracking. Then, all
the equations given for the Eulerian correlation (Eqs. 3.34 - 3.41) are equally
valid for the Lagrangian correlation.

84

CHAPTER 3. STATISTICAL ANALYSIS

So, the Lagrangian correlation coe cient (cf. Eq. 3.34)
RL (t) =

V ( )V ( + t)
V 2

(3.42)

in which V is the Lagrangian uctuating velocity, and, for later use, the
Lagrangian macro time scale
TL =

RL (t) dt

(3.43)

cf., Eq. 3.40).

It may be noted that, in the laboratory, tracking of particles can be
achieved by means of an ordinary camera (or, two special cameras used in
the eld of photogrammetry for 3-D tracking), combined with a stroboscope
(Sumer and Oguz, 1978, and Sumer and Deigaard, 1981), or using video
techniques.
Of particular interest is the physical interpretation for the Lagrangian
macro time scale. The Lagrangian macro time scale may be interpreted as
the time during which a coherent structure keeps its identity. Also, the
length scale to be obtained by application of Eq. 3.41 may be interpreted as
the distance traveled by the coherent structure.

3.3

Spectrum analysis

In turbulent ows, there is a continuous, broad range of length/time scales.

For example, smoke coming out of a smoke stack has ow structures with
varying sizes; Larger eddies (coherent structures) contain smaller eddies;
these smaller eddies contain even smaller eddies, and so on.
The correlation analysis cannot address the issue of the previously mentioned continuous, broad range of length/time scales. (It gives only two
discrete length/time scales (the micro and the macro scales), and no more
than that).
This continuous, broad range of scales can be studied only by use of the
spectrum analysis. We will study this in this section.
The analysis will be developed for isotropic turbulence for convenience.
There are two kinds of spectrum analyses. One involves the space correlations, and the other involves time correlations. The spectrum function in
the rst case is resolved in the wave number space, while that in the second

3.3. SPECTRUM ANALYSIS

85

Figure 3.8: Denition sketch.

case is resolved in the frequency space. We will consider in this section the
former spectrum analysis. However, the analysis in the latter is basically the
same, provided that the wave number is to be replaced by the frequency.

3.3.1

General considerations

The Fourier transform of two-point double correlation Qij (see Eq. 3.10 for
the denition of Qij ) is
ij (k)

1
(2 )3

Qij (r) e

ikr

dr

(3.44)

in which r is the separation vector (Fig. 3.8), dr =dr1 dr2 dr3 , k is the wave
number vector and i is the imaginary unit, i =
1. The inverse transform,
on the other hand, is
Qij (r) =

ij (k)

eikr dk

(3.45)

in which dk = dk1 dk2 dk3 is an element of volume in wave number space.

Now, for r = 0, the preceding equation becomes
Qij (0) =

ij (k)

dk

Qii (0) =

ii (k)

dk

and setting i = j,
(3.46)

86

CHAPTER 3. STATISTICAL ANALYSIS

Figure 3.9: Energy spectrum. Isotropic turbulence.

Here, Qii (0) is two times the mean kinetic energy of turbulence per unit
mass in the i-th direction:
1
Qii (0) = ui ui = 2 [ (u 2 + v 2 + w 2 )]
2

(3.47)

We write Eq. 3.46 in the following form

Qii (0) =

ii (k1, k2 , k3 )

(3.48)

dk1 dk2 dk3

Since the left hand side of the above equation is proportional to the turbulent
energy, then ii (k1, k2 , k3 ) will be like the energy density in the wave number
space. We shall return to this in the following paragraphs.
In the case of isotropic turbulence, there will be spherical symmetry.
Therefore, the spherical coordinates can be used in the wave number space.
Eq. 3.48 in this case will read
Qii (0) =

ii (k)
k=0

ii (k)

=
k=0

dk dA(k)

A(k)

dk

(4 k 2 )

dA(k) =
A(k)

k=0

ii (k)

dk

3.3. SPECTRUM ANALYSIS

87

or
1
Qii (0) =
2

E(k) dk

(3.49)

k=0

in which
E(k) = 2 k2

ii (k)

(3.50)

1. Eq. 3.49 is an important equation. As pointed out in the preceding

paragraphs, the left hand side of the equation is the turbulent energy
(per unit mass), therefore, E(k) will be the energy density in the wave
number space k. The function E(k) is called the energy spectrum function.
2. It may be noted that the spherical symmetry enables us to replace the
three-dimensional spectrum function ii (k1 , k2 , k3 ) (Eq. 3.48) with a
one-dimensional spectrum function, E(k).
3. The relation given in Eq. 3.49 can be envisaged as follows (see the
sketch in Fig. 3.9): The velocity time series of the three components of
the velocity are recorded at the measurement point P. From these time
series, the turbulent energy per unit mass, 12 Qii (0), is calculated, using
Eq. 3.47. Then, according to Eq. 3.49, this energy is decomposed into
innite number of small energies, i.e., E(k) dk, in the wave number
space. Each of these small energies corresponds to a wave number, k,
which itself corresponds to a wave length (or to a turbulence length
scale)
2
=
(3.51)
k
As seen, the objective set at the beginning of this section, namely the
study of the continuous, broad range of length scales of turbulence,
apparently can be achieved by the application of the energy spectrum
function E(k). In the following section, we will have a closer look at
this function.

3.3.2

Energy balance in wave number space

In isotropic turbulence, the previously mentioned correlation function Qij (r)

satises the following equation (Hinze, 1959, p. 159, and Raudkivi and

88

CHAPTER 3. STATISTICAL ANALYSIS

Callander, 1975, p. 163)

2

Qij (r) = Sij (r) + 2

rk rk

Qij (r)

(3.52)

in which
Sij (r) =

rk

(3.53)

(Sik,j + Skj,i )

in which
(Sik,j )AB = (ui )A (uk )A (uj )B

(3.54)

(Si,kj )AB = (ui )A (uk )B (uj )B

(3.55)

and
It should be noted that Eq. 3.52 stems from the N.-S. equation.
The Fourier transform of Eq. 3.52
t

ij (k)

in which
ij (k)

2 k2

ij (k)

1
(2 )3

ij (k)

(3.56)

ikr

(3.57)

Sij (r) e

dr

and the inverse transform reads

Sij (r) =

ij (k)

ikr

dk

(3.58)

ii (k)

(3.59)

Now, putting i = j in Eq. 3.56 gives

ii (k)

ii (k)

2 k2

and multiplying both sides of the latter equation by 2 k 2 , and recalling

E(k) = 2 k2 ii (k) (Eq. 3.50), one gets
t

E(k) = T (k)

2 k2 E(k)

(3.60)

in which
T (k) = 2 k2

ii (k)

(3.61)

3.3. SPECTRUM ANALYSIS

89

Figure 3.10: Energy spectrum and dissipation spectrum. (Note that peaks are
well apart).

This equation (Eq. 3.60) represents the turbulent energy balance in

the wave number space. To study the physical meaning of each term in the
equation, we integrate Eq. 3.60 from k = 0 to :
t

E(k)dk =
0

T (k)dk

k 2 E(k)dk

(3.62)

It can be shown that the rst term on the right hand side of the preceding
equation is zero (Hinze, 1959, p. 178). Hence Eq. 3.62 will be
t

E(k)dk =
0

k2 E(k)dk

(3.63)

Recall that 0 E(k)dk is the turbulent energy per unit mass (Eq. 3.49);
hence, the left hand side of the above equation will represent the rate of
change of the turbulent energy. Now, the integral on the right hand side of
the equation, 0 k2 E(k)dk, is always positive. This means that
t

E(k)dk < 0

(3.64)

From the above analysis, the following deductions can be made:

1. Turbulent energy in isotropic turbulence decreases with time (Eq. 3.64).
This is due to viscous dissipation, governed by the term on the right
hand side of Eq. 3.63.

90

CHAPTER 3. STATISTICAL ANALYSIS

2. As seen from Eq. 3.63, the latter term is weighted by the factor k2 ,
and clearly is more e ective at large wave numbers (i.e., at small size
eddies) than at small wave numbers (i.e., at large size eddies), Fig.
3.10.
3. The fact that the rst term on the right hand side in Eq. 3.62 is zero
implies that it is neither a source term nor a sink term. It can be shown
that this term, namely T (k), is responsible for the transfer of turbulent
energy from small wave numbers to large wave numbers (see Example
7 below).
Summarizing at this point,
1. Most of the turbulent energy is contained in small wave numbers (large
eddies, or large scale motion) (Fig. 3.10);
2. Most of the dissipation occurs at large wave numbers (small eddies, or
small scale motion) (Fig. 3.10); and
3. Between the small wave numbers and large wave numbers, there is a
range of wave numbers responsible for transferring turbulent energy
from small to large wave numbers.

Example 7 Transfer of turbulent energy in the wave number space.

Integrate Eq.3.60 from k = 0 to k:
k

E(k)dk =
0

T (k)dk
0

k 2 E(k)dk

(3.65)

Consider the interval in the wave number space from k = 0 to k (Fig. 3.10).
The rst term on the left hand side of the above equation represents the
change in the turbulent energy in this interval per unit time. The third term
represents the viscous dissipation in this interval per unit time (Fig. 3.11).
k
Experiments show that the sign of the term 0 T (k)dk is the same as
k
that of the viscous dissipation. Therefore, 0 T (k)dk should be an e ux,
rather than an inux. This e ux cannot take place at the k = 0 end of the
considered interval, because k = 0 forms a closed end (there is no negative

3.3. SPECTRUM ANALYSIS

91

Figure 3.11: Transfer of energy in wave-number space.

wave numbers!). Therefore, this e ux must take place at the other end of
the interval, namely at k = k (Fig. 3.11).
So, the picture is as follows. In the interval (0, k) in the wave-number
space, some part of the turbulent energy is dissipated into heat (viscous dissipation), and some part of it is transferred to wave numbers larger than k.
The latter means that there is, on average, a continuous ow of turbulent
energy from small wave numbers to large wave numbers. The term responsible for this is T (k). This term originates from the inertia terms in the N.-S.
equations. Therefore, the previously mentioned transfer of energy is called
the inertial transfer.

3.3.3

Kolmogoro s theory. Universal equilibrium range

and inertial subrange

In the process of turbulent energy transfer from small wave numbers (large
scale motion) to large wave numbers (small-scale motion), a point is reached
beyond which the turbulence will forget the inuence of the initial largescale motion. From this point to the wave numbers where the viscous dissipation occurs, there is a range of wave numbers in which the character of
turbulence will be the same in all ows (in streams, in atmospheric boundary-

92

CHAPTER 3. STATISTICAL ANALYSIS

layer turbulence, etc.). This range is called the universal equilibrium range.
It is called universal because the character of turbulence in this range is independent of the ow environment; and it is called equilibrium because the
energy input is in balance with the energy dissipation, as will be detailed
later in the section. Kolmogoro s theory is concerned with this range.
Kolmogoro s rst hypothesis states that, irrespective of the large-scale
motion, the turbulence in the universal equilibrium range is isotropic. Now,
the energy spectrum function in this range is a function of the following
quantities
E = f ( , , k)
(3.66)
in which
time:

is the dissipation of turbulent energy per unit mass and per unit
= (

ui
u
u
+
) i
x
xi x

(3.67)

(Eq. 1.32).
is involved because it represents the dissipation of turbulent energy.
should be involved because the energy is dissipated into heat through viscosity. The wave number, k, is involved because the turbulent energy process is
primarily dependent on the wave number (Eq. 3.60). (k is taken instead of
k1 , k2 and k3 because of the spherical symmetry; isotropic turbulence!). Note
that E (rather than ii ) is adopted here to represent the energy spectrum
because we assume that turbulence is isotropic.
From dimensional analysis, one gets
E(k) = v2
in which

(3.68)

( k)

is a nondimensional function of k, the quantity v is

1

(3.69)

v = ( )4
and

is
3

=(
2

(3.70)

)4
1

cm 2 4
Here the unit of v is [v] = ( cms cms ( s.cm
) ) = cm
, and that of is
s
1
2
1
cm 3
[ ] = (( s ) cm2 cm 2 ) 4 = cm. The quantity v is called the characteristic
s

( s.cm )

velocity, while is called the characteristic length. Based on the quantities

v and , the following characteristic time may be dened
=

= ( )2

(3.71)

3.3. SPECTRUM ANALYSIS

93

The above characteristic length, , and the characteristic time, , are involved in the large wave number range (small scale motion); they represent
the smallest scales of turbulence, namely the smallest length scale ( known
as the Kolmogoro scale), and the smallest time scale. These are the smallest scales of turbulence because scales smaller than these involve molecular
motion.
The nondimensional equation obtained for the spectrum function (Eq.
3.68) can be developed further, following Kolmogoro s second hypothesis.
The latter states that the peaks of the energy spectrum and that of the
dissipation spectrum are su ciently apart. This is illustrated in Fig. 3.10.
The latter implies that there is a dominating wave number range in the
energy spectrum (small wave numbers, or alternatively, large eddies) which
contain most of the turbulent energy, while there is another dominating wave
number range (large wave number, or alternatively small eddies) which are
associated with the viscous dissipation. According to Kolmogoro s second
hypothesis, these wave number ranges are su ciently apart, meaning that
there must be a wave number range between these two ranges where there is
no signicant energy, and there is no signicant dissipation.
First of all, this range should be a part of the universal equilibrium range,
because it contains no signicant energy. (Recall the denition of the universal equilibrium range at the beginning of this subsection: In the universal
equilibrium range, the energy input is in balance with the energy dissipation;
no signicant energy presence).
Secondly, in this range, the only signicant activity is that the turbulent
energy is transferred from small wave numbers to large wave numbers. This
process is governed by the inertia terms in the N.-S. equation, as described
in Example 7 above. Therefore, this particular range of wave numbers is
called the inertial subrange. (inertial because of the previously mentioned
inertia terms; and subrange because this range is a part of the universal
equilibrium range, as pointed out in the preceding paragraphs).
Now, in the inertial subrange, the energy spectrum should be independent
of the viscosity because no signicant dissipation takes place in this range,
as mentioned previously. This is an important statement. Because, for the
viscosity to drop out in Eq. 3.68, the function ( k) has to be given in the
following form
( k) = Constant (k )

5
3

(3.72)

94

CHAPTER 3. STATISTICAL ANALYSIS

Figure 3.12: General picture of energy spectrum.

Thus, inserting Eq. 3.72 in Eq. 3.68, the spectrum function is found to be
E(k) = Constant

2
3

5
3

(3.73)

This is known as -5/3-law. Experiments conrm the presence of this subrange

where the spectrum is given by the -5/3 law.
Finally, Fig. 3.12 gives a schematic picture of the spectral representation
of the process.

3.3.4

One-dimensional spectrum

E(k) is not an easily measurable quantity. To get around the problem, we

dene the following one-dimensional spectrum
1
2
1
=
2

F (k) =

Q11 (r) e
u 2 f (r) e

ikr

dr

ikr

(3.74)

dr

Since f (r) is an even function of r,

F (k) =

u2

f (r) cos(kr) dr
0

(3.75)

3.3. SPECTRUM ANALYSIS

95

The inverse transform is

u 2 f (r) = 2

F (k) cos(kr) dk

(3.76)

It can be shown that, in isotropic turbulence, the one-dimensional spectrum F (k) is related to the energy spectrum function E(k)
E(k) = k2 F (k)

(3.77)

kF (k)

(see for example Raudkivi and Callander, 1975, p.174).

Incidentally, inserting Eq. 3.73 into Eq. 3.77 gives
2
3

5
3

= k 2 F (k)

kF (k)

(3.78)

k2
C1 + C2
2

(3.79)

This has a solution:

F (k) =

9
55

2
3

5
3

The second and third terms on the right-hand-side of the preceding equation
should drop because F (k) should go to zero for large values of k, and therefore
F (k) = Constant

2
3

5
3

(3.80)

Example 8 The correlation function f (r) is approximated by

f (r) = exp(

Find the one-dimensional and three-dimensional spectrum functions.

The one-dimensional spectrum function from Eq. 3.75
F (k) =

u2

f (r) cos(kr) dr =
0

For k = 0:
k = 0 : F (0) =

u2

u2
(1 + (k f )2 )
f

96

CHAPTER 3. STATISTICAL ANALYSIS

Figure 3.13: 1-D and 3-D energy spectra for the numerical example.
Hence,
F (k) =

F (0)
1 + (k f )2

(see Fig. 3.13 a).

Regarding the three-dimensional energy spectrum function E(k), considering that the turbulence is isotropic, therefore using Eq. 3.77
E(k) = 8F (0)

(k f )4
(1 + (k f )2 )3

(see Fig. 3.13 b for E(k)).

Example 9 One-dimensional spectrum function for a boundary layer in an
open channel.
Fig. 3.14 a shows the one-dimensional spectrum function obtained for
an open channel ow by Raichlen (1967). (The results are normalized as

3.3. SPECTRUM ANALYSIS

97

Figure 3.14: 1-D energy spectra. (a): measured in an open channel. Raichlen
(1967). (b): measured in a jet. Champagne (1978).

98

CHAPTER 3. STATISTICAL ANALYSIS

indicated in the gure. The Reynolds number is dened as Re = 4RV in

which R is the hydraulic radius). Fig. 3.14 b, on the other hand, shows the
one-dimensional spectrum measured in a jet.
It may be noticed that, over a wave number interval, namely 1 < kh
< 6,
2
in Fig. 3.14 a, E(k) is apparently
E(k)

5
3

revealing the previously mentioned Kolmogoro s -5/3 law. The same holds
true for the second example (Fig. 3.14 b) for the range 4 10 2 cm 1 < k1 <
10 cm 1 .

3.4

References

1. Champagne, F.H. (1978): The ne-scale structure of the turbulent

velocity eld. Journal of Fluid Mechanics, Vol. 86, pp. 67-108.
2. Eckelman, H. (1974): The structure of the viscous sublayer and the
adjacent wall region in a turbulent channel ow. J. Fluid Mech., vol.
65, p.439.
3. Hinze, J.O. (1959): Turbulence. McGraw-Hill.
4. Raichlen, F. (1967): Some turbulence measurements in water. J.
Eng. Mech. Proc. ASCE, p.73.
5. Raudkivi, A.J. and Callander, R.A. (1975). Advanced Fluid Mechanics.
Edward Arnold.
6. Sumer, B.M. and Deigaard, R. (1981): Particle motions near the bottom in turbulent ow in an open channel - Part 2. Journal of Fluid
Mechanics, Vol. 109, pp. 311-337.
7. Sumer, B.M. and Oguz, B. (1978): Particle motions near the bottom
in turbulent ow in an open channel. Journal of Fluid Mechanics, Vol.
86, pp. 109-127.

Chapter 4
Di usion and dispersion
Di usion/dispersion is another important property of turbulent ows. Some
examples are: (1) Waste products released into a recipient environment such
as a stream or the ocean undergo rapid di usion; (2) Smoke (or exhaust gases)
released into the atmosphere di uses rapidly when there is strong wind; (3)
Likewise, dredged material (sand, silt, etc.), when dumped into the sea, is
subjected to large di usion.
The objective of this chapter is to describe the di usion and dispersion
processes in turbulent ows. First, we will study the di usion process in a
turbulent ow eld (a general analysis), and then, we will turn our attention to the so-called longitudinal dispersion process, a process caused by the
combined action of the turbulent di usion and the velocity variation in the
ow.

4.1

One-Particle analysis

Release a lump of dispersant into the ow in the form of an instantaneous

point source. The point of release is x = 0, and time t = 0 (Fig. 4.1). This
dispersant will undergo di usion as it is convected downstream, as sketched
in Fig. 4.1.
(Note that we will use the term dispersant to refer to any kind of passive
quantity such as waste products, suspended sediment, heat, etc. Also note
that we assume that there is no interaction between the dispersant and the
ow; the dispersant acts as a passive quantity).
The di usion of the dispersant is caused by the following mechanism.
99

100

CHAPTER 4. DIFFUSION AND DISPERSION

Figure 4.1: Denition sketch.

The instantaneous point source is actually formed by many dispersant
particles. Since the ow is turbulent, each particle will follow a random
path. This will cause the particles to get farther and farther apart as they
travel (Fig. 4.2). The end result will be the continuous spreading (di usion)
of the dispersant, as sketched in Fig. 4.1.
The di usion of a passive quantity can be studied by adopting one of the
following two approaches.
1. Release the dispersant in the form of a lump at x = 0 and at time t = 0;
and follow the cloud of the dispersant as it grows and is convected
downstream; or
2. Release one particle at x = 0 and at time t = 0, and follow its path
x = x(t); Repeat this exercise many times, and determine the statistical
properties of x(t). This approach is called the one-particle analysis.
In the following paragraphs, the latter approach will be adopted.
For simplicity, we shall study the process in one dimension.
The particles are released at x = 0, at time t = 0. They follow random paths x(t). The objective of the analysis is to determine the statistical
properties of x(t). These statistical properties are (Fig. 4.3)
1. The mean value of x, namely x; this will give the mean position of the
dispersing cloud as it is convected downstream.
2. The variance of x, namely (x x)2 ; this will give a measure of the
degree of dispersion of the cloud.

4.1. ONE-PARTICLE ANALYSIS

101

Figure 4.2: Random paths of particles.

3. Other statistical moments; and
4. The probability density function (p.d.f.) of the particle position, namely
p(x). This quantity is proportional to the concentration, as will be
demonstrated later in the chapter.

Figure 4.3: One-dimensional di usion.

The particle velocity (the Lagrangian velocity):
V =

dx
dt

(4.1)

102

CHAPTER 4. DIFFUSION AND DISPERSION

Integrating gives
t

(4.2)

V (t ) dt

x(t) =
0

Now, ensemble averaging gives the mean particle position (the position of
the centroid of the particle cloud):
t

x(t) =

V (t ) dt =

(4.3)

V (t ) dt

Here, the ensemble averaging is

N
n=1

(4.4)

in which is any hydrodynamic quantity, n the value of in one realization, and N the total number of realizations. (In the present context, one
realization is one release of particle).
The variance, on the other hand, can be calculated in the following way.
Let
y(t) = x(t) x(t)
(4.5)
Then, from Eq. 4.2
t

y(t) = x(t)

x(t) =

V (t ) dt
0

V (t ) dt

(4.6)

[V (t )

V (t )] dt

The integrand here is the uctuating part of the particle velocity

V

(4.7)

V (t ) dt

(4.8)

V =V
Hence,
t

y(t) =
0

Di erentiating y 2 (t) with respect to t

d 2
dy
y (t) = 2 y
dt
dt

(4.9)

4.1. ONE-PARTICLE ANALYSIS

103

and from Eq. 4.8

d 2
y (t) = 2 [
dt

(4.10)

V (t ) dt ] V (t)
0
t

V (t ) V (t) dt

= 2
0

The integrand in the preceding equation, V (t ) V (t), recalling the Lagrangian

correlation coe cient (Eq. 3.42):
V (t ) V (t) = V

V (t )V (t)
= RL (t
V 2

Then,
d 2
y (t) = 2 V
dt
To summarize at this point,

t) V

(4.11)

t
2

RL (t

t ) dt

(4.12)

1. The mean particle position is given by Eq. 4.3; and

2. The variance of the particle position is given by Eq. 4.12.
Now, we will continue with the analysis for the following two extreme
cases: (1) For small times (t
TL ); and (2) For large times (t
TL ) in
which TL is the Lagrangian macro time scale dened by
TL =

RL (t) dt

(4.13)

(Eqs. 3.42 and 3.43). Recall that Lagrangian macro time scale, TL , may be
interpreted as the time during which a coherent structure keeps its identity
/ coherence (Section 3.2.2).

4.1.1

Di usion for small times

For small times (t

TL ), the uid particle keeps its identity/coherence,
meaning that the particle velocity will be pretty much the same as that at
the point where the particle starts its travel, namely at x = 0 at time t = 0.
This velocity is the Eulerian velocity at x = 0:
V (t)

V (t = 0)

u(x = 0)

(4.14)

104

CHAPTER 4. DIFFUSION AND DISPERSION

Now, the mean particle position, from Eq. 4.3:

t

x(t) =

V (t ) dt =
0

u(x = 0) dt = u(x = 0)
0

dt = u(x = 0) t
0

(4.15)
Regarding the variance, for small times (t
TL ), rst of all, V 2
u 2 (x = 0), and secondly, RL (t t ) 1. Therefore, Eq. 4.12
d 2
y (t) = 2 u 2 (x = 0) t
dt
and integrating gives
y 2 (t) = u 2 (x = 0) t2

(4.16)

Now, assume that the ow/turbulence is homogenous. Then the statistical properties of velocity should be invariant with respect to the position:
u(x = 0)

u(x) = u = constant

(4.17)

u 2 (x) = u 2 = constant

(4.18)

and
u 2 (x = 0)

Thus, the mean particle position and the variance will be (from Eqs. 4.15,
4.16, 4.17, 4.18)
x(t) = u t
(4.19)
and
y 2 (t) = u 2 t2

(4.20)

The p.d.f. of the particle position, on the other hand, will be the same
as that of the Eulerian velocity provided that u should be replaced with xt .

4.1.2

Di usion for large times

Mean and variance of particle position.

For large times (t
TL ), the analysis will be developed further for the
following special case. Namely, the particle velocity is a stationary random
function (s.r.f.) of time.
(The particle velocity is a s.r.f. of time if the probability density function
of particle velocity is invariant with respect to time, i.e.,
p[V (t)]

p[V (t + )]

(4.21)

4.1. ONE-PARTICLE ANALYSIS

105

meaning that the statistical properties of the particle velocity such as the
mean V and the variance (V V )2 = V 2 are time invariant. Homogeneity
in space is equivalent to being stationary in time).
First, consider the mean particle position. From Eq. 4.3 (taking V (t )
outside the integral because it is constant),
t

x(t) =

(4.22)

V (t ) dt = V t
0

The mean particle velocity should be equal to the Eulerian mean velocity,
u. (The fact that the Lagrangian velocity is a stationary random function
of time requires the homogeneity in the Eulerian sense; therefore, V
u).
Hence,
x(t) = u t
(4.23)
Next, consider the variance of the particle position. Eq. 4.12:
d 2
y (t) = 2 V
dt

t
2

RL (t

(4.24)

t ) dt

Change the notation

=t

(4.25)

then Eq. 4.24 will be

d 2
y (t) = 2 V
dt

t
2

(4.26)

RL ( ) d
=0

Integrating by parts gives

t

y 2 (t) = 2 V 2 (

RL ( ) d ) t
=0

2V

RL ( ) d

(4.27)

=0

For large times (t

TL ), t is so large that the integrals in the preceding
equation converge (provided, of course, that the form of RL ( ) is such that
the integrals do converge). Furthermore, the second term on the right hand
side is small compared with the rst term. Hence, for large times,
t

y 2 (t) = 2 V 2 (

RL ( ) d ) t
=0

(4.28)

106

CHAPTER 4. DIFFUSION AND DISPERSION

The integral for large times can be approximated to

t

RL ( ) d

RL ( ) d = TL

=0

(4.29)

=0

(from Eq.4.13). Inserting the latter equation in Eq. 4.28, and also considering
that V 2 is identically equal to u 2 (due to the fact that V (t) is a stationary
random function of time), the variance is obtained as
y 2 (t) = 2 u 2 TL t

(4.30)

(As will be detailed in the following paragraphs, this is di usion because

y 2 (t) t).
To summarize, the mean particle position and the variance are given by
Eqs. 4.23 and 4.30, respectively.
As seen from Eqs. 4.20 and 4.30, the variance is
y 2 (t)

t2 for small times (t

TL )

(4.31)

y 2 (t)

t for large times (t

TL )

(4.32)

and
This implies that the spreading occurs very fast in the beginning of the
process (y 2 (t) t2 ), and then the rate at which the spreading occurs reaches
an equilibrium stage where y 2 (t) t, as sketched in Fig. 4.4.
P.d.f. of particle position.
The particle displacement (Eq. 4.2)
t

V (t ) dt

x(t) =
0

can be considered as the sum of a series integral over a xed interval, say ,
t

x(t) =

(n+1)

V (t ) dt =
0

V (t ) dt
n

(4.33)

such that each integral has only a partial statistical connection with each
other. For times t
TL , the number of such integrals becomes su ciently
large so that we have a situation like that covered by the central limit theorem.

4.1. ONE-PARTICLE ANALYSIS

107

Figure 4.4: Variance of particle position.

The central limit theorem states that for a series of statistically independent terms, the p.d.f. of the sum tends to the Gaussian distribution (normal
distribution), as the number of terms approaches innity, irrespective of the
probability distributions of each term, which form the sum.
Therefore, for large times t
TL , the p.d.f. of x(t) will be
1

p(x) =
2

exp(
x)2

(x

(x

x)2

2(x

x)2

or, for convenience, changing the notation to y = x

1

p(y) =
2

exp(
y2

(4.34)

x,

y2
)
2 y2

(4.35)

Inserting Eq. 4.30 in the preceding equation,

1

p(y) =
2

2 u 2 TL t

exp(

y2
)
2 (2 u 2 TL t)

(4.36)

We shall return to this equation later in the section.

Now, consider the di usion equation
2
c
c
=D 2
t
y

(4.37)

108

CHAPTER 4. DIFFUSION AND DISPERSION

in which D is the di usion coe cient, y the space coordinate, and t the time.
Consider the initial condition
t=0:

(4.38)

c = (y)

in which (y) is the Dirac delta function, which is dened by

(y) = 0 for y = 0
(y) =
for y = 0
and
(y)dy = 1
The solution to the di usion equation with the above initial condition is
c(y) =

1
2

2Dt

exp(

y2
)
2 (2 D t)

(4.39)

Now, compare the above equation with Eq. 4.36. From this comparison,
the following deductions can be made:
1. The p.d.f. of the particle position for large times (t
di usion equation (cf., 4.37)
2
p
p
=D 2
t
y

TL ) satises the

(4.40)

2. The di usion coe cient of the di usion process is

D = u 2 TL

(4.41)

which can, from Eq. 4.30, also be written as

D = u 2 TL =

1 dy 2 (t)
2 dt

(4.42)

D is called the turbulent di usion coe cient.

The theory summarized in the preceding paragraphs was developed by
G.I. Taylor (1921). It was later extended to the three-dimensional case

4.1. ONE-PARTICLE ANALYSIS

109

by Batchelor (1949). For further reading, consult Batchelor and Townsend

(1956) and Monin and Yaglom (1973).
Flux due to turbulent di usion.
The probability that a particle is present in the interval (y, y + dy)
Pr[y < Y < y + dy] =

m
M

(4.43)

where M is the total number of released particles, and m is the number of

particles occupying the interval (y, y + dy).
This probability can also be written as
(4.44)

Pr[y < Y < y + dy] = p(y) dy

From the preceding two equations, one gets
p(y) =

1 m
M dy

(4.45)

m
, the number of particles per unit length, is the concentration c in
Here, dy
the one-dimensional case. So,

c(y)
M

p(y) =

(4.46)

Now, the probability that a particle is present in the half space (

, y)

is

m1
M
in which m1 is the number of particles occupying the interval (
wise, this probability can be written as
Pr[

< Y < y] =

(4.47)
, y). Like-

Pr[

p(y) dy

< Y < y] =

(4.48)

Then from Eqs. 4.47 and 4.48

y

m1 = M

p(y) dy

(4.49)

Di erentiate the above equation with respect to time

m1
=
t

p(y) dy

(4.50)

110

CHAPTER 4. DIFFUSION AND DISPERSION

Figure 4.5: Flux due to turbulent di usion.

m1
Here
is the rate of change in the number of the particles which occupy
t
the region (
, y), i.e., the decrease in the number of particles in the region
(
, y) per unit time. This means that there is a continuous supply of
m1
particles from the region (
, y) to the region (y, + ) at a rate
, i.e.,
t
m1
the number of particles passing through the section y per unit time is
.
t
The expression for p(y) is given in Eq. 4.35. Substituting this equation in
Eq. 4.50, and integrating and then di erentiating with respect to time gives

y
m1
=M
p(y)
t
2t

(4.51)

Di erentiating Eq. 4.35 with respect to y, on the other hand, gives

p
=
y

y
p(y)
2Dt

(4.52)

From the preceding two equations,

m1
=
t

p
y

MD

or

m1
(Mp)
= D
t
y
From Eq. 4.46, c = Mp. Inserting this in the above equation gives
m1
=
t

c
y

(4.53)

(4.54)

4.1. ONE-PARTICLE ANALYSIS

111

Changing notation
=

m1
t

then Eq. 4.54

=

c
y

(4.55)

The quantity , the number of particles passing through the section y per
unit time (Fig. 4.5), is called the ux ( the ux due to turbulent di usion).
Notice the similarity between the ux due to turbulent di usion and that
due to molecular di usion, namely mol = Dmol yc (Ficks law) in which
Dmol is the molecular di usion coe cient.
Example 10 Conservation of mass in turbulent ows

Figure 4.6: Denition sketch.

Consider the uid element dx dy dz in Fig. 4.6. The inux of mass in
the x direction is
u c dy dz
and
Dx

c
dy dz
x

(due to convection)

(due to turbulent di usion)

(4.56)

(4.57)

(from Eq. 4.55).

The e ux, on the other hand,
[u c +

(u c) dx] dy dz

(due to convection)

(4.58)

112

CHAPTER 4. DIFFUSION AND DISPERSION

and
[ Dx

c
+ ( Dx
x
x

c
) dx] dy dz
x

Thus the net inux in the x

x

direction is

(u c) dx dy dz

Similar expressions for the y

(due to turbulent di usion) (4.59)

and z

(v c) dx dy dz

( Dx

c
) dx dy dz
x

(4.60)

directions:
y

( Dy

c
) dx dy dz
y

(4.61)

c
) dx dy dz
(4.62)
z
z
z
On the other hand, the rate of change of mass in the volume dx dy dz
(w c) dx dy dz

( Dz

(4.63)

(c dx dy dz)

The conservation of mass requires that the rate of change of mass should
be equal to the net inux. Hence, from Eqs. 4.60-4.63, and using the continuity equation (Eq. 1.6), namely
u
v
w
+
+
=0
x
y
z

(4.64)

one gets
c
c
c
c
+u
+v
+w
=
(Dx
t
x
y
z
x

c
) + (Dy
x
y

c
) + (Dz
y
z

c
) (4.65)
z

or, in index notation

c
c
+ ui
=
(Di
t
xi
xi

c
)
xi

(4.66)

and including the molecular di usion e ect as well

c
c
c
+ ui
=
(Di + Dmol )
)
t
xi
xi
xi
However, note that, normally, Dmol

Di .

(4.67)

4.2. LONGITUDINAL DISPERSION

4.2
4.2.1

113

Longitudinal dispersion
Mechanism of longitudinal dispersion

The mechanism of longitudinal dispersion may be best described by reference

to an open-channel ow, Fig. 4.7. Release a lump of dispersant in the form
of a point source at section x = 0 and t = 0. Consider that the released
dispersant is formed by many particles. Now, these particles are subject to
two e ects:
1. The turbulent motion in the y direction; Under this e ect, the particles travel up and down in a random fashion.
2. The streamwise motion in the x direction. The particles take the
velocity u(y), plus a small uctuating velocity u . (However, the latter
is negligible with respect to the former).

Figure 4.7: Longitudinal dispersion.

114

CHAPTER 4. DIFFUSION AND DISPERSION

Particles near the bottom will experience relatively smaller velocities u(y),
and therefore will be retarded (Path 1 in Fig. 4.7 a) while particles near the
free surface will experience relatively larger velocities u(y), and therefore
will be moved large distances (Path 2 in Fig. 4.7 a). This implies that the
particles are prone to disperse farther and farther apart in the streamwise
direction. Thus, the end result will be the progressive dispersion of the
dispersant in the streamwise direction, as sketched in Fig. 4.7 b. This process
is called the longitudinal dispersion.
When compared with the ordinary di usion process studied in the previous section (cf., Fig. 4.1 and 4.2), the key e ect here is that the dispersion is
caused mainly by the velocity variation u(y) across the depth, whereas the
dispersion in the ordinary di usion process is caused by the random paths of
particles; no spatial variation in the mean velocity exists in the latter process.
For the above reason, the present di usion process is sometimes called
the mechanical dispersion.
From the preceding considerations, it can easily be seen that
1. there will be no longitudinal dispersion, when the velocity distribution
across the depth is uniform; and
2. likewise, there will be no longitudinal dispersion when there is no turbulent motion across the depth (i.e., when there is no turbulent di usion
across the depth). (Note, however, that the existing molecular di usion will cause longitudinal dispersion although the degree of dispersion
in this case is several orders of magnitude smaller than the turbulent
longitudinal dispersion, Taylor, 1953, 1954, Elder, 1959).

4.2.2

Application of one-particle analysis

Fig. 4.8 shows the path of a particle. Here, V (t) is the streamwise velocity
of the particle. V (t) is actually the sum of two velocities:
1. the velocity u(y) at the position where the particle is at that particular
time t, and
2. the x component of the uctuating velocity u , which occurs at that
position, and at that particular time, t.

4.2. LONGITUDINAL DISPERSION

115

Figure 4.8: Particle velocity V samples from the velocity prole u(y).
The uctuating velocity is small compared with u(y), and therefore can
be neglected, as already stated in the preceding paragraphs. Hence, the
particle velocity can be taken as V (t) = u(y(t)).
Now, release many particles at time t = 0, and determine their velocities,
V (t), at time t. Let the probability density function (p.d.f.) of these velocities be p(V (t)). Next, do the same exercise, but this time determine their
velocities at time t + . Let the p.d.f. of the latter velocities be p(V (t + )).
Now, if t is su ciently large, then the two p.d.f.s will be identical
p(V (t))

p(V (t + ))

(4.68)

This implies that the velocity V (t) is a stationary random function of time
(Section 4.1.2).
The immediate implication of this is that the theory given in the previous section in conjunction with the ordinary di usion process for large times
(Section 4.1.2) can be applied to the present process because, as in the previous case, the particle velocity is a stationary random function of time. By
application of the one-particle analysis, we reach the following conclusions.
1. The mean particle position (or the position of the centroid of the dispersant) will be at (Fig. 4.7 b)
x(t) = Um t

(4.69)

(from Eq. 4.23). Here, Um is the mean Eulerian velocity, and equal to
the cross-sectional average velocity (Fig. 4.8)
Um =

1
h

u(y) dy
0

(4.70)

116

CHAPTER 4. DIFFUSION AND DISPERSION

2. The variance will be

(x

x)2 = 2 V 2 TL1 t

(4.71)

(from Eq. 4.30). Here TL1 is the Lagrangian time scale dened by
TL1 =

(4.72)

RL1 (t) dt
0

in which RL1 (t) is the Lagrangian correlation coe cient

RL1 (t) =

V ( )V ( + t)
V 2

(4.73)

in which
V =V

(4.74)

Um

Here, the subindex 1 is used to distinguish the present application from

the ordinary di usion.
3. The p.d.f. of the particle position tends to the Gaussian (normal)
distribution for large times (t
TL1 ):
1

p(x) =
2

(x

exp(
x)2

(x

x)2

2(x

x)2

(4.75)

(from Eq. 4.34).

4. The di usion coe cient of the process is
D1 = V 2 TL1 =

1 d(x x)2
2
dt

(4.76)

(from Eq. 4.42). Here, D1 is called the longitudinal dispersion coe cient.
5. The p.d.f. of particle position satises the di usion equation
2
p
p
= D1
t
(x x)2

(4.77)

4.2. LONGITUDINAL DISPERSION

117

(from Eq. 4.40). Here, t is the time derivative when considering the
moving coordinate x x(t). Changing to the xed coordinate x,
t

x
x(t)
=
+
=
+ Um
x t
t
t x
t
x

(4.78)

Then, Eq. 4.77 becomes

2
p
p
p
+ Um
= D1 2
t
x
x

(4.79)

6. From Eqs. 4.79 and 4.46

<c>
<c>
+ Um
= D1
t
x

<c>
x2

(4.80)

in which < c > is the cross-sectional average concentration:

< c >=

1
h

(4.81)

c dy
0

Eq. 4.80 is called the dispersion equation.

The above analysis shows that the longitudinal dispersion process is represented by a one-dimensional di usion equation with constant coe cients
Um and D1 (Eq. 4.80). The process is also governed by the general di usion
equation (Eq. 4.65):
c
c
c
+u
+v
=
(Dx
t
x
y
x

c
) + (Dy
x
y

c
)
y

(4.82)

Clearly, the one-dimensional equation (Eq. 4.80) is much simpler than the
two-dimensional representation (Eq. 4.82). (This is the advantage of the
longitudinal-dispersion concept). Obviously, this is at the expense of the
calculated concentration. In the one-dimensional representation, the crosssectional average concentration is calculated, whereas, in the two-dimensional
representation, the point-concentration is calculated. However, this is not a
problem, because, in longitudinal-dispersion problems, the calculation with
the cross-sectional average concentration is normally good enough (the concentration variations in the longitudinal direction are important, not the
variations over the cross section).

118

CHAPTER 4. DIFFUSION AND DISPERSION

For the application of the previously mentioned one-dimensional dispersion equation, two quantities need to be known: the cross-sectional average
velocity, Um , and the longitudinal dispersion coe cient, D1 .
Um is obtained simply by integrating the velocity distribution over the
cross-section, and dividing it by the cross-sectional area. This is done if
the velocity distribution is known theoretically (for ex., in the form of the
logarithmic law). Otherwise, the velocity distribution over the cross-section
is measured, for example, as in the case of a natural stream, and the mean
velocity calculation is based on the measured velocity.
The longitudinal dispersion coe cient, on the other hand, can be calculated theoretically (see the next section), or it can be calculated numerically,
based on the measured velocity distribution over the cross section, as will be
detailed in Section 4.4.
G.I. Taylor (1953, 1954) introduced the concept of the longitudinal dispersion. In his study, the ow environment was a circular pipe. His theoretical
work and his experiments showed that the longitudinal dispersion coe cient
D1
in a turbulent pipe ow (Taylor, 1954) is aU
= 10 in which a is the radius of
f
the pipe, and Uf is the friction velocity. Taylors work laid the groundwork
for the later research in 1960s and 70s in the area of dispersion in natural
streams, estuaries, etc. and also in industrial ows (such as pipe ows).

4.3

Calculation of dispersion coe cient

In this section, we shall calculate the longitudinal dispersion coe cient for
an open channel ow (Fig. 4.7). We shall do this exercise for heavy particles
(for example, sand, or silt in water). This feature will be represented by the
fall velocity of the particles, w. However, the case of the neutrally-buoyant
particles/dispersant will be captured by setting w equal to zero in the end
solutions.
The analysis presented in the following paragraphs is mainly based on the
work of Sumer (1974). It may be noted that the method used in the analysis
was rst introduced by Aris (1956) and also employed by Sayre (1968).

4.3. CALCULATION OF DISPERSION COEFFICIENT

4.3.1

119

Formulation

The conservation of mass (Fig. 4.7) from Eq. 4.65:

c
c
+ u(y)
t
x

c
c
c
=
( (y) ) + ( (y) )
y
x
x
y
y

(4.83)

in which w is the fall velocity of the particles, and (y) is the turbulent
di usion coe cient. It is assumed that (y) is the same for both x and y
directions.
The boundary conditions:
y = 0 and h:

(y)

c
+ wc = 0
y

(4.84)

implying that there is no net transport across the boundaries.

The initial condition is (for a uniformly distributed plane source)
t = 0:

c(x, y, 0) =

M
(x)
h

(4.85)

in which M is the total number of particles released into the ow.

Furthermore, we write
u(y) = Um (1 + (y)), and (y) = Dm (y)

(4.86)

in which Um and Dm are the cross-sectional average velocity and the crosssectional average di usion coe cient, respectively. We shall return to these
quantities and the nondimensional functions (y) and (y) later in the section.
Now, dene the pth moment of concentration
p

Cp =
in which

Cd

(4.87)

is the normalized streamwise distance dened by

=

Us t
h

(4.88)

in which Us is the mean particle velocity, and C is the normalized concentration dened by
c
(4.89)
C= M
h2

120

CHAPTER 4. DIFFUSION AND DISPERSION

Now, multiply the two sides of the above equations (Eqs. 4.83-4.85) by
, and integrate the equations from
to + :
Cp

)pCp

Cp
1

= p(p

Cp

= 0 and 1:

1) Cp
+

s Cp

Cp

=0

) (4.90)
(4.91)

and
= 0:

Cp ( , 0) = 1 and 0, respectively, for p = 0 and p > 0

(4.92)

Here,
Dm t
y
, = 2
(4.93)
h
h
are the normalized distance from the bottom of the channel and the normalized time, respectively. Also, , s and s are the normalized mean ow
velocity, the normalized mean particle velocity and the normalized mean fall
velocity, respectively:
=

Um h
,
Dm

Us h
,
Dm

wh
Dm

(4.94)

Regarding the ow velocity, u(y), and turbulent di usion coe cient, (y),
distributions across the depth, rst of all, from Eq. 2.85,
u

Um =

y
Uf [ln( ) + 1]
h

(4.95)

(the logarithmic law). The distribution of the turbulent di usion coe cient,
on the other hand, is found from the analogy between the momentum ux
(Eq. 2.89) and the ux due to turbulent di usion
=

dc
dy

uv
du
dy

(4.96)

Taking
uv
Uf2 (1 hy ) (see Eq. 2.21), and using the above analogy
(Eq. 4.96), the di usion coe cient
(y) = Uf y(1

y
)
h

(4.97)

4.3. CALCULATION OF DISPERSION COEFFICIENT

121

and Dm is
Dm =

1
h

(y) dy =
0

1
hUf
6

(4.98)

The functions (y) and (y) in Eqs. 4.86 are then

=

4.3.2

6
2

(1 + ln ), and

= 6 (1

(4.99)

Zeroth moment of concentration

For p = 0, the solution to the Eqs. 4.90-4.92 is found to be as follows

C0 =

) 1
(

sin(

) +

aK (

) F ( K, 1+K; 1

K=1

; ) exp{ 6(K 2 +K) }

(4.100)

in which aK
aK =

(2K + 1) (K + 1
)
3 F2 ( K, K + 1, 1; 1
2 (1
) (K + 1 + )

, 2; 1)

(4.101)

F is a hypergeometric function, 3 F2 is a generalized hypergeometric series,

and is the Gamma function (Gradshteyn and Ryzhik, 1965). The quantity
here is the fall velocity parameter dened by
=

w
Uf

(4.102)

and the solution is found for = 0, < 1.

Fig. 4.9 displays C0 ( , ) for two values of . Note that the physical
meaning of C0 is that C0 is identical to the p.d.f. of the particle position
across the depth. As seen from the gure, the larger the value of , the more
weighted the distributions towards the bottom, as expected.
From Eq. 4.100
C0 ( ) =

sin(

) 1
(

as

(4.103)

Sumer (1974) argues that, for the case of neutrally buoyant particles ( = 0),
1. Indeed the above equation reveals this.
C0 should go to unity, C0

122

CHAPTER 4. DIFFUSION AND DISPERSION

Figure 4.9: C0 ( , ) for two di erent sediment with (a):

= wUf = 0.3. Sumer (1974).

w
Uf

= 0.1, and (b):

4.3. CALCULATION OF DISPERSION COEFFICIENT

4.3.3

123

Mean particle velocity

For this, we dene

1

mp =

(4.104)

Cp d
0

Now, integrate Eqs. 4.90 and 4.92 from

conditions (Eq. 4.91),
dmp
d

= 0 to 1, and using the boundary

s+

)Cp

p(p

1)

Cp

d =0

mp (0) = 1 and 0, respectively, for p = 0 and p > 0

= 0:

(4.105)

(4.106)

Now, the preceding two equations for p = 1

1

dm1
d

)C0 d = 0

(4.107)

=0

(4.108)

and
m1 (0) = 0 at

Note that m1 represents the mean particle displacement relative to the

1
axis, and dm
represents the mean particle velocity relative to this axis. Since
d
1
0 as
this axis moves with the mean particle velocity (Eq. 4.88), then dm
d
time
. Hence, from Eq. 4.107, for large times
1

)C0 ( ,

)d =0

or, from Eq. 4.103,

1
s

C0 ( ,
0

)d =

sin(

) 1
(

) d

(4.109)

This is the mean particle velocity relative to the mean ow velocity. Fig.
4.10 illustrates how the mean particle velocity varies with the fall velocity
parameter. Note that for
= 0, the case of neutrally-buoyant particles,
obviously this velocity must be equal to the mean ow velocity, meaning
that s
must be zero, as revealed by the gure.

124

CHAPTER 4. DIFFUSION AND DISPERSION

Figure 4.10: Mean velocity of heavy particles relative to the mean ow velocity.
Sumer (1974).

4.3.4

Longitudinal dispersion coe cient

The longitudinal dispersion coe cient can be written in the normalized form
as
D1
1 dm2
)
= (
(4.110)
hUf
6 2 d
(see Eqs. 4.87 and 4.104).
From Eq. 4.105,
1 dm2
=6
2 d

(1
0

)C0 d +

(
0

)C1 d

(4.111)

4.3. CALCULATION OF DISPERSION COEFFICIENT

125

in which C1 is found from Eqs. 4.90-4.92, setting p = 1

C1 =

aK fK ( ) exp{ 6(K 2 + K) }

( )+
K=1

dK (

(4.112)

; ) exp{ 6(K 2 + K) }

) F ( K, 1 + K; 1

K=1

in which
= A (

sin(

) 1
(

)(

1
6 (1

d (4.113)

) d

as
.
in which < 1. It may be noted that C1
Now, from Eqs. 4.111, 4.113 and 4.103, the dispersion coe cient is obtained as
D1
hUf

(1
(
0

sin(

[(

36
)

s)

](

) d (4.114)

1
(1

[(

s)

](

) d

in which < 1.
For the case of neutrally buoyant particles/dispersant, putting
gets
1
D1
1
d
=
d
d
hUf
6
36 0
(1
)
0
0

= 0, one
(4.115)

The expression in Eq. 4.115 was rst obtained by Elder (1959). The integral
can be calculated numerically. The result is (taking , the von Karman
constant, 0.41)
D1
= 0.07 + 5.86 = 5.93
(4.116)
hUf
As seen from Eq. 4.115, the dispersion coe cient consists of two terms; the
rst term on the right hand side of the equation, 6 , (the numerical value of
which is 0.07, Eq. 4.116) represents the e ect of the turbulent di usion in
the streamwise direction; it stems from the rst term on the right hand side
of Eq. 4.83. The second term is actually the main portion of the longitudinal

126

CHAPTER 4. DIFFUSION AND DISPERSION

Figure 4.11: Ratio of the longitudinal dispersion coe cient of heavy particles to
that of neutrally-buoyant particles. Sumer (1974).

dispersion coe cient, and induced by the combined action of the turbulent
di usion across the depth y and the variation of the ow velocity in the y
direction.
Returning to the case of heavy particles, the dispersion coe cient is plotted in Fig. 4.11. The gure shows that the dispersion coe cient increases
with increasing , the fall velocity parameter. This can be explained as follows. A heavy particle spends most of the time near the bottom where the
velocity gradient is largest. On the other hand, the larger the velocity gradient, the larger the dispersion coe cient. Therefore, a heavier particle (i.e.,
with larger ) should have a larger dispersion coe cient.

4.4

Longitudinal dispersion in rivers

In the previous section, the longitudinal dispersion coe cient has been calcuD1
lated for an open-channel ow, and found to be hU
6. Field experiments in
f

4.4. LONGITUDINAL DISPERSION IN RIVERS

127

Figure 4.12: (a): The cross-sectional velocity distribution at Renton Junction in

the Green-Duwamish River, Washington. Contours are lines of constant velocity,
in increment of 0.2 ft/s; (b): The transverse prole of the depth-averaged velocity
corresponding to the velocity distribution shown in (a). Taken from Fischer et al.
(1979).
D1
rivers show, however, that hU
takes values much larger than 6. The largest
f
D1
7500 in the Missouri river.
value reported in Fischer et al. (1979) is hU
f
This evidence makes it clear that the value obtained for the case of the open
D1
channel, namely hU
6, cannot apply to real streams. The reason why
f
D1
6 value does not apply was studied by Fischer (1966, 1967). It was
hUf
shown that, in the case of real streams, although the velocity distribution
satises the logarithmic law in the vertical direction, the variation of the
velocity across the river width apparently plays the major role.
A quantitative estimate of the dispersion coe cient in a real stream can be
obtained by neglecting the vertical prole completely, but instead, by taking

128

CHAPTER 4. DIFFUSION AND DISPERSION

the transverse velocity prole, i.e., the velocity prole across the river width.
(Fig. 4.12 displays an example illustrating the transverse velocity variation in
a river, the velocity isolines at a measurement section in the Green-Duwamish
River in Washington, and the corresponding transverse velocity prole).
Fisher obtained the following expression for the longitudinal dispersion
coe cient
D1 =

1
A

(u(y)
0

1
dy
t d(y)

Um ) d(y) dy
0

(u(y)

Um ) d(y) dy

(4.117)
in which A is the cross-sectional area of the river, W width at the surface,
Um mean ow velocity, y the transverse distance measured from one bank
of the river (Fig. 4.12) (not to be confused with the vertical distance used
in the previous section), d(y) the depth (the depth is a function of y, Fig.
4.12), t is the turbulent di usion coe cient in the transverse direction (it
may or may not be a function of y), and u(y) is the depth-averaged velocity
dened by (Fig. 4.12) :
0

1
u(y) =
d(y)

(4.118)

u(y, z) dz
d(y)

Returning to the expression obtained for the case of the open-channel

ow in the previous section, Eq. 4.115, (omitting the rst term on the right
hand side of the equation)
1

D1
=
hUf

36

d
0

(1

d
0

This is actually in terms of dimensional quantities

D1
=
hUf

1
h

(u(y)

Um )dy

1
dy
(y)

(u(y)

Um )dy

or alternatively
D1
=
hUf

1
h2

(u(y)
0

Um )hdy
0

1
dy
(y)h

(u(y)

Um )hdy

(4.119)

As seen, this expression is quite similar to that obtained by Fischer (1966,

1967), Eq. 4.117. The only di erence is that, in the case of the river dispersion, y is the transverse distance, while, in the case of the open-channel
dispersion, y is the vertical distance.

4.5. REFERENCES

129

Fischer et al. (1979) give a numerical example to illustrate how to use

Eq. 4.117 in a eld application.
Finally, it may be noted that real streams may have irregularities such as
bends, sandbars, side pockets, pools, ri es, bridge piers, man-made revetments, and even occasional junked car, or sunken barges. Every irregularity
contributes to dispersion. (Fischer et al., 1979, and Pedersen, 1977). Empirical expressions have been given for the longitudinal dispersion coe cient in
terms of gross ow characteristics of the river (Fischer et al., 1979, p. 136).
These can be used to make rst assessments of the dispersion coe cient.

4.5

References

1. Aris, R. (1956): On the dispersion of a solute in a uid owing through

a tube. Proc. Roy. Soc., vol. A235, p. 67.
2. Batchelor, G.K. (1949): Di usion in a eld of homogenous turbulence. Australian J. Sci. Res., vol. A2, No. 4, p.473.
3. Batchelor, G.K. and Townsend, A.A. (1956):Turbulent di usion. In:
Surveys in Mechanics (ed. G.K. Batchelor and R. M Davies), Cambridge University Press, p.352.
4. Elder, J. (1959): The dispersion of marked uid in turbulent shear
ow. J. Fluid Mech.., vol. 5, p. 544.
5. Fischer, H.B. (1966): Longitudinal Dispersion in Laboratory and Natural Streams. Tech. Report KH-R-12, California Institute of Technology, Pasadena, California.
6. Fischer, H.B. (1967): The mechanics of dispersion in natural streams.
J. Hydraul. Div., Proc. ASCE, vol. 93, p. 187.
7. Fischer, H.B., List, E.J., Koh, R.C.Y., Imberger, J. and Brooks, N.H.
(1979): Mixing in Inland and Coastal Waters. Academic Press, Inc.
8. Monin, A.M. and Yaglom, A.S. (1973): Statistical Fluid Mechanics:
Mechanics of Turbulence, vol. 1, MIT Press. Cambridge, Mass.

130

CHAPTER 4. DIFFUSION AND DISPERSION

9. Pedersen, F.B. (1977): Prediction of Longitudinal Dispersion in Natural Streams. Series Paper No. 14, Technical University of Denmark,
Institute of Hydrodyn. and Hyd. Eng. (ISVA).
10. Sayre, W.W. (1968): Hydraulic Papers, No. 3, Colorado State University, Fort Collins.
11. Sumer, B.M. (1974): Mean velocity and longitudinal dispersion of
heavy particles in turbulent open-channel ow. J. Fluid Mech., vol.
65, p.11.
12. Taylor, G.I. (1921): Di usion by continuous movements. Proc. London Math. Soc., vol. 20, p. 196.
13. Taylor, G.I. (1953): Dispersion of soluble matter in solvent owing
slowly through a tube. Proc. Roy. Soc. London, A, vol. CCXIX,
p.186.
14. Taylor, G.I. (1954): The dispersion of matter in turbulent ow through
a pipe. Proc. Roy. Soc. London, vol. A223, p.446.

Chapter 5
Wave boundary layers
In shallow waters, the orbital motion of water particles under a progressive
wave degenerates into a bottom-parallel, straight line, oscillatory motion at
the seabed (Fig. 5.1). For each half cycle of this oscillatory motion at the
bed, a new time-dependent boundary layer develops. This cyclic boundary
layer is called the wave boundary layer.
The wave boundary layers are of great importance for several processes:
Turbulent mixing of mass and momentum takes place in the wave boundary layers; Likewise, frictional dissipation takes place in the wave boundary
layers; The wave boundary layers are one of the key components of the sediment transport processes; Exchanges of chemicals and organisms between
seabed and the main body of the water take place in the wave boundary layers. Therefore, understanding of wave-boundary layers is essential to develop
predictive tools and capabilities for the previously mentioned processes.
Fredse and Deigaard (1992, Chapters 2 and 3) can be consulted for a
detailed treatment of the subject including the mathematical modelling.
The ow in the wave boundary layers may be in the laminar regime, it
may be in the transitional, or it may be in the turbulent regime.
We will start o with the laminar-regime wave boundary layers, and then
we will focus on the turbulent wave boundary layers. The latter will be
mainly based on the work of Jensen et al. (1989), an extensive experimental work conducted in an oscillating water tunnel, covering both smoothand rough-boundary ows, with very broad range of the Reynolds number
including very high Reynolds numbers.
131

132

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.1: Denition sketch.

5.1
5.1.1

Laminar wave boundary layers

Velocity distribution across the boundary layer
depth

The streamwise component of the N.-S. equation (Eqs. 1.8 and 1.9) in the
boundary layer (Fig. 5.2):
u
u
u
+u
+v
=
t
x
y

1 p 1
+
x
y

(5.1)

First of all, v in the boundary layer is practically zero, therefore, the third
term on the left hand side of the equation drops out. Secondly, the x distances
are large compared to the y distances (the wave length in the laboratory
L = O(2-3 m) while the boundary layer thickness = O(1 mm) O(1 cm);
the same is also true for the turbulent wave boundary layers in the sea,
L = O(100 m) while the boundary layer thickness = O(10 cm)); therefore,
the x dependence in the above equation can be neglected, hence the second
term on the right hand side of the equation also drops out. Therefore, Eq.
5.1:
u
1 p 1
=
+
(5.2)
t
x
y
Now, the boundary layer is actually driven by the pressure gradient xp .
This term can be calculated from the equation of motion outside the boundary layer (in the free-stream region). This equation, by setting = 0 in the

5.1. LAMINAR WAVE BOUNDARY LAYERS

133

Figure 5.2: The boundary-layer thickness, , is small compared with the wave
length, L.
above equation,
U0
=
t

1 p
x

(5.3)

in which U0 is given as
(5.4)

U0 = U0m sin( t)
in which is the angular frequency of the wave ( =
period.
From Eqs. 5.2 and 5.3, one gets
u
U0 1
=
+
t
t
y
in which

is the shear stress

=

2
T

), T being the wave

(5.5)

u
y

(5.6)

Therefore Eq. 5.5

u
= (U0m sin( t)) +
t
t

u
y2

(5.7)

The boundary conditions

y

U0m sin( t)

(5.8)

134

CHAPTER 5. WAVE BOUNDARY LAYERS

and
y = 0:

(5.9)

u=0

The solution to Eq. 5.7 with the preceding boundary conditions is

u(y, t) = U0m sin( t)

U0m exp(

y
1

in which

1
1

=(

)2

) sin( t

(5.10)

(5.11)

representing the thickness of the boundary layer, called the Stokes length.
The time development of the velocity distribution is plotted from Eq.
5.10 in Fig. 5.3. In the gure, is the boundary layer thickness, slightly
di erent from that in Eq. 5.11; namely, is the distance from the bed where
the velocity becomes maximum at the phase value t = 900 ; this distance is
found from the above solution as
=

3 2 1
( )2
4

(5.12)

or, in terms of the Reynolds number, Re,

a

3
2 1
( )2
4 Re

in which
Re =

aU0m

(5.13)

(5.14)

in which a is the amplitude of the orbital motion of water particles at the

bed,
U0m
(5.15)
a=
As seen from Fig. 5.3, the ow near the bed leads over the ow in the
free stream region. (As a result, the near-bed ow reverses earlier).
This is related to the driving pressure gradient. From Eqs. 5.3 and 5.4,
it can readily be seen that xp > 0 after t = 900 in the half period of the
motion. This means that the boundary layer will experience an adverse pressure gradient during this phase of the half period (Fig. 5.4). This pressure
gradient dictates the boundary-layer ow to reverse. However, for the ow
to respond, it will have to dissipate its momentum. Clearly, the near-bed
ow dissipates its momentum earlier than the outer ow, meaning that it will

5.1. LAMINAR WAVE BOUNDARY LAYERS

135

Figure 5.3: Time development of velocity proles in laminar wave boundary layer.
respond to the pressure gradient earlier, therefore will reverse earlier than the
outer ow, as revealed by the velocity proles in Fig. 5.3.
It may be noted that the preceding description may also explain the
overshooting exhibited by the plotted velocity proles in Fig. 5.3.

5.1.2

Flow resistance

The bed shear stress,

0

= (

u
U0m
[sin( t) + cos( t)]
)y=0 =
y
1

(5.16)

or
2 U0m

)]
(5.17)
4
This equation shows that the bed shear stress leads over the free-stream ow
with a phase di erence (cf. Eq. 5.4)
0

[sin( t +

= 450
This also implies that the ow at the bed reverses
free stream.

(5.18)
= 450 earlier than the

136

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.4: Two di erent pressure gradients over one half cycle.
Finally, the friction coe cient in wave boundary layers is dened by
fw =

0m /
2
U0m

(5.19)

or
fw
U0m
(5.20)
2
in which fw is the wave friction coe cient (cf. Eq. 2.128), and Uf m is the
maximum value of the friction velocity dened by
Uf m =

Uf m =

0m

(5.21)

Here 0m is the maximum value of the bed shear stress. From Eqs. 5.16 and
5.20, fw
2
fw =
(5.22)
Re

5.2

Laminar-to-turbulent transition

One way of illustrating the laminar-to-turbulent transition is to plot the

friction factor as a function the Reynolds number. Fig. 5.5 a shows this.

5.2. LAMINAR-TO-TURBULENT TRANSITION

137

The gure also includes, for convenience, the phase di erence (lead) between
the bed shear stress and the free-stream velocity (Fig. 5.5 b) plotted versus
the Reynolds number.

Figure 5.5: (a): Wave friction coe cient, and (b): phase di erence. Jensen et al.
(1989).

1. The transition to turbulence occurs at Re 1.5 105 . The bed shear

stress measurements of Jensen et al. (1989) showed that the bed shear
stress signal is disturbed rst at Re = 1.6 105 . Hydrodynamic instability calculations apparently seem to capture this critical value of Re
(see Sleath, 1984).
2. The data for both the friction coe cient and the phase di erence agrees
well with the theoretical results for the laminar-ow case, depicted in
Eqs. 5.22 and 5.18.

138

CHAPTER 5. WAVE BOUNDARY LAYERS

3. The phase lead decreases with the transition. This can be explained
as follows. In the transitional/turbulent ows, the momentum exchange
between the layers of uid will be enhanced. Therefore, the near-bed
ow will be supplied with the new, momentum-rich uid from the upper layers of the boundary layer, and as a result, it will dissipate its
momentum not as fast as in the case of the laminar ow. Hence, its
response to the adverse pressure gradient will be delayed, leading to
the decrease in the phase lead, as indicated in Fig. 5.5 b. In the fullydeveloped turbulent case (very large Re), the phase lead is only about
50 .

Figure 5.6: Normalized friction coe cient at di erent phase values. Jensen et al.
(1989).

Fig. 5.6 o ers another way of illustrating the transition to turbulence.

Here, the friction coe cient is dened in a way slightly di erent from that
in Eq. 5.19, namely
2 0/
fw = 2
(5.23)
U0m sin( t + 4 )

5.2. LAMINAR-TO-TURBULENT TRANSITION

139

Note that the bed shear stress above is not the maximum value of the shear
stress, but rather the phase-resolved bed shear stress, 0 = f ( t). The advantage of this representation is that, when divided by cos( t 4 ), the bed
shear stress will be independent of the phase t in the case of the laminar
ow (from Eq. 5.17):
2
U0m
=
)
(5.24)
0
1 sin( t +
4
Re 2
but will depend only on Re:
2
(5.25)
fw =
1
Re 2
which represents a straight line when plotted on a log-log paper.
Now, this enables us to identify the ow regime in the following way.
1. Measure the bed shear stress,

= f ( t).

2. Calculate the friction coe cient fw from Eq. 5.23.

3. Plot the latter quantity versus Re on a log-log paper, with
parameter.

t as a

4. If fw versus Re on this plot collapses on the straight line representing

2
fw =
1
Re 2
then the ow will be in the laminar regime; otherwise it will be either
in the transitional regime, or in the turbulent regime.
Jensen et al. (1989) did this exercise. Their results are depicted in Fig.
5.6.
This gure is an important gure. It implies that, even for very large Re,
the ow regime may remain laminar for quite some period of time during the
boundary layer development in the half cycle. For example, for Re = 5 105 ,
the ow remains laminar until t
400 , while for Re = 106 , this gure is
t 200 250 .
Fig. 5.6 also suggests the following. The pattern of the variation of fw
with Re is not the same outside the laminar line; apparently there are two
distinct patterns, implying a change in the ow regime. These two regions are
interpreted by Jensen et al. as the transitional and fully developed turbulent
regimes. From the gure, it can be seen that, even for Re as high as 106 ,
the ow becomes a fully developed turbulent ow only after the phase value
t 600 .

140

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.7: Sample velocity records. Re = 6 106 . U0 was measured at y = 140

mm, and u and v at y = 1 mm. Jensen et al. (1989).

5.3
5.3.1

Turbulent wave boundary layers

Ensemble averaging

Fig. 5.7 displays time series of the horizontal and vertical velocity components of the velocity in the boundary layer (the bottom two time series). The
top time series is the free-stream velocity. This is from an experiment with
Re = 6 106 .
The so-called ensemble averaging is used to calculate the mean values:
u(y, t) =

1
N

u[y,

(t + (j

1)T )]

(5.26)

j=1

Likewise, for example, the root-mean-square (r.m.s.) value of the uctuating

5.3. TURBULENT WAVE BOUNDARY LAYERS

141

Figure 5.8: Mean velocity distributions. Smooth bed. Re = 6106 . Jensen et al.
(1989).

velocity u = u

u 2 (y, t) =

5.3.2

u,

1
2

1
1

[u[y,

(t + (j

1)T )]

u(y, t)]2

(5.27)

j=1

Mean ow velocity

Fig. 5.8 shows the time development of the velocity proles for Re = 6 106 .
The way in which the velocity proles evolve with time is similar to that in
the case of the laminar wave boundary layer (Fig. 5.3). Features such as the
early reversal of the ow near the bed, and the overshooting in the velocity
proles are observed here, too.
The velocity proles in the previous gure are plotted in wall units on
semi-logarithmic plots in Fig. 5.9. The gure also includes the van Driest
prole (Eq. 2.97, the velocity prole obtained for steady boundary layers) as
a reference line in each panel. Note that Uf used in the normalized quantities
yU
y + = f and Uuf is the temporal value (the phase resolved value) of the
friction velocity. Recall that the van Driest prole tends to the logarithmic

142

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.9: Mean velocity distributions in semi-log plot. Smooth bed. Re =

6 106 . Solid curve: van Driest prole. Dashed lines: log law. Jensen et al.
(1989).

distribution for large values of y + (Eqs. 2.97 and 2.99)

u = Uf (A ln y + + B)

(5.28)

Fig. 5.9 shows the following.

1. The oscillatory-ow velocity distributions agree remarkably well with
the steady-ow ones in the range of t = 700 1100 where the ow
experiences almost zero pressure gradient.
2. The logarithmic layer exists already at t = 150 and is maintained
throughout the t range from t 150 up to almost 1600 .
3. At earlier phases, the logarithmic layer does not exist, however. This
is because the boundary layer at these stages is not thick enough to
house the logarithmic layer in it.
4. It may be noted that the velocity proles in the early phases of the
acceleration stage (where the pressure gradient is favourable) resemble quite closely those obtained in favourable pressure-gradient steady
boundary layers, and likewise, those in the later phases of the deceleration stage (where the pressure gradient is adverse) resemble, again,
quite closely the velocity distributions obtained in adverse pressuregradient steady boundary layers (Coles, 1956).

5.3. TURBULENT WAVE BOUNDARY LAYERS

143

Figure 5.10: Mean velocity distributions. Sumer et al. (1987).

Figure 5.11: Shear stress distributions, corresponding to the velocity distributions

in the previous gure. Sumer et al. (1987).

Finally, note the following. Fig. 5.10 shows the semi-log plots of the
velocity distributions obtained in another research (Sumer et al., 1987), while
Fig. 5.11 shows the corresponding shear stress distributions. As seen from
Fig. 5.11, the constant stress layer is displaced to higher elevations in the
later phases (starting from t = 912 ) of the deceleration stage. This means
that the logarithmic layer should also be displaced to these higher elevations.
The velocity distributions may still be represented by the logarithmic law.
However, Uf in such a representation cannot be the wall shear velocity, but
1
a velocity dened by ( ) 2 where is the shear stress occurring at these
elevations rather than at the wall.

144

5.3.3

CHAPTER 5. WAVE BOUNDARY LAYERS

Flow resistance

The friction coe cient, dened by Eq. 5.19 or Eq. 5.20, has been determined from experiments in oscillatory ows and plotted in Fig. 5.5 for both
transitional and fully developed turbulent oscillatory boundary layers.
The variation of the friction coe cient with Re in Fig. 5.5 in the fully
developed turbulent regime (Re
106 ) can be approximated by (Fredse
and Deigaard, 1992, p. 29)
0.035
fw =
(5.29)
Re 0.16

5.3.4

Turbulence quantities

Fig. 5.12 and 5.13 presents the turbulence data in terms of

u2
,
Uf

v2
Uf

w2
,
Uf

uv
Uf2

(5.30)

versus y + . Also plotted in the gures are the corresponding steady boundarylayer distributions (see, for example, Fig. 2.7).
Except at very early phases of the acceleration stage, and also late phases
of the deceleration stage, the oscillatory-ow proles appear to be in reasonable accord with the steady boundary-layer distributions throughout the
phase space. See Jensen et al. (1989) for further details.

5.3.5

E ect of Reynolds number

Fig. 5.14 gives the mean velocity proles for di erent Reynolds number.
The characteristic features of the time development of the velocity proles
shown in the preceding gure are as follows.
1. The velocity prole eventually reaches a state where the logarithmic
layer comes into existence.
2. This logarithmic layer grows in thickness, as one proceeds further in
the phase space t.
3. This continues to be the case until practically the point of reversal near
the bed is reached.

5.3. TURBULENT WAVE BOUNDARY LAYERS

145

Figure 5.12: R.m.s. values of u , v and w . Smooth bed. Re = 6 106 . Jensen

et al. (1989).

146

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.13: Reynolds stresses. Smooth bed. Re = 6 106 . Jensen et al. (1989).
From the gure, it is seen that the higher Re, the earlier the logarithmic
layer comes into existence. This can be explained by reference to Fig. 5.6;
From the latter gure, it is seen that, for the establishment of the logarithmic
layer, the ow has to reach the fully developed turbulent state.
Jensen et al. (1989) also give the e ect of Re on the turbulence quantities.

5.3.6

Boundary layer thickness

Fig. 5.15 presents the normalized boundary-layer thickness a plotted against

Re. Here is the distance from the bed where the velocity becomes maximum
at the phase value t = 900 . The laminar ow solution has already been given
in Eq. 5.12; in terms of Re:

3
2 1
( )2
4 Re

(5.31)

From the gure, the boundary-layer thickness for storm conditions (very
large Re)
a

(5.32)

= O(0.01)

This corresponds to a boundary-layer thickness of

= O(10 cm).

5.3. TURBULENT WAVE BOUNDARY LAYERS

147

Figure 5.14: Mean velocity distributions. Smooth bed. Diamonds: Re = 6 106 .

Jensen et al. (1989). Empty circles: Re = 1.6 106 . Filled circles: Re = 5 105 .
Dashed lines: Re = 5 105 , Spalart and Baldwins (1987) direct N.-S. simulation.
Triangles: Re = 2.8 105 , Hino et al. (1983). From Jensen et al. (1989).

148

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.15: Boundary-layer thickness. Smooth bed. Jensen et al. (1989).

Figure 5.16: Mean velocity u. Circles: Smooth bed. Triangles: Rough bed. Re
= 6 106 . Jensen et al. (1989).

5.3.7

Turbulent wave boundary layers over rough walls

Figs. 5.16-5.19 compare the rough-bed results with the corresponding smoothbed ones in the tests of Jensen et al. (1989). Re = 6 106 and ks+ = 84 in
k U
which ks+ = s f m where ks is Nikuradses equivalent sand roughness. The
parameter kas , on the other hand, is 3700.
From these gures, the following points appear to be noteworthy.
1. The description given in the preceding paragraphs of how the ow and
turbulence evolve during one half-cycle is qualitatively the same for the
rough-bed ows, too.
2. The introduction of the roughness, however, results in a decrease in the

5.3. TURBULENT WAVE BOUNDARY LAYERS

149

Figure 5.17: R.m.s. value of u . Circles: Smooth bed. Triangles: Rough bed. Re
= 6 106 . Jensen et al. (1989).

Figure 5.18: R.m.s. value of v . Circles: Smooth bed. Triangles: Rough bed. Re
= 6 106 . Jensen et al. (1989).

150

CHAPTER 5. WAVE BOUNDARY LAYERS

Figure 5.19: Reynolds stress u v . Circles: Smooth bed. Triangles: Rough bed.
Re = 6 106 . Jensen et al. (1989).
mean ow velocity and an increase in the turbulence quantities.
3. However, this e ect tends to disappear with the distance from the bed;
the mean ow velocity is practically the same for the two cases for
distances ay
0.01. The same is also true for the turbulence quantities, although this time the e ect of roughness disappears at somewhat
larger distances from the bed, namely for ya 0.03. The decrease in the
mean velocity is due to the retarding e ect of the rough boundary while
the increase in the turbulence quantities is due to the enhancement of
the momentum transfer by the presence of the roughness elements..
4. The ow begins to feel the e ect of the roughness only after t reaches
the value of about 150 . This is because the temporal value of the roughness Reynolds number is still far too small for these phases for the
bed to act as a completely rough boundary and, as a result, the ow
quantities for this short interval of time remain practically unchanged,
regardless of the boundary category.
The rough turbulent boundary layers are governed solely by the parameter
This parameter plays a similar role to that of Re in the case of the smooth
turbulent wave boundary layers. Jensen et al. (1989) discussed the e ect of
this parameter on the mean and turbulence ow quantities.
The friction coe cient in the case of the rough bed can be represented
by the following empirical expressions (Fredse and Deigaard, 1992, p. 25
a
.
ks

5.4. OTHER ASPECTS

151

Figure 5.20: Boundary-layer thickness. Rough bed. Jensen et al. (1989).

and 26):

a 1
a
) 4
when
> 50
(5.33)
ks
ks
a 3
a
fw = 0.4( ) 4
when
< 50
(5.34)
ks
ks
The latter expression is due to Kamphuis (1975) (Fredse and Deigaard,
1992, p. 26).
In the case of the transitional-category-bed wave boundary layers, the friction coe cient can be obtained from Kamphuis (1975) empirical diagrams.
Also consult Fredse and Deigaard (1992, p. 33).
Finally, Fig. 5.20 depicts the boundary-layer thickness. The gure shows
that, the boundary-layer thickness for storm conditions (very large kas )
fw = 0.04(

which corresponds to a boundary-layer thickness of

5.4

(5.35)

= O(0.02)
= O(20 cm).

Other aspects

Several other aspects of wave boundary layers have been studied in recent
years. The following is a partial list of these studies undertaken at ISVA.
Fredse, Sumer, Laursen and Pedersen (1993) have studied the nonuniform wave boundary layers, where the non-uniformity is due to a sudden
change in the bed roughness, while Sumer, Laursen and Fredse (1993) have

152

CHAPTER 5. WAVE BOUNDARY LAYERS

studied the non-uniform wave boundary layers, where the non-uniformity is

due to the convergent/divergent geometry of the ow environment.
Lodahl, Sumer and Fredse (1998) have investigated the combined oscillatory boundary layer and current in a circular pipe. The circular pipe
has been selected in favour of a rectangular-section tunnel, to achieve a clean ow environment (to avoid secondary currents, and more importantly
to achieve a 2-D ow environment).
Arnskov, Fredse and Sumer (1993) have studied combined wave and current boundary-layer ow in a basin. In this study, the waves were propagating
at an angle to the current, including the normally incident waves.
Tanaka, Sumer and Lodahl (1998) have studied the e ect of cnoidal waves
on laminar wave boundary layers.
Fredse, Andersen and Sumer (1999) have studied combined wave and
current boundary-layer ows over a ripple-covered bed.
Fredse, Sumer, Kozakiewicz, Chua and Deigaard (2003) have studied
the e ect of externally-generated turbulence on wave boundary layers.

5.5

References

1. Arnskov, M. M., Fredse, J. and Sumer, B. M. (1993): Bed shear stress

measurements over a smooth bed in three-dimensional wave-current
motion. Coastal Engineering, Vol. 20, pp. 277- 316.
2. Coles, D. (1956): The law of the wake in a turbulent boundary layer.
J. Fluid Mech., vol. 1, p. 191.
3. Fredse, J. (1984): Turbulent boundary layer in wave-current motion. J. Hydraul. Eng. ASCE. 110, p. 1103.
4. Fredse, J., Andersen, K. H. and Sumer, B. M. (1999): Wave plus
current over ripple-covered bed, Coastal Engineering, vol. 38, 177221.
5. Fredse, J. and Deigaard, R. (1992). Mechanic of Coastal Sediment
Transport. Advanced Series on Ocean Engineering., Vol. 3, World
Scientic.

5.5. REFERENCES

153

6. Fredse, J., Sumer, B.M., Laursen, T. S.and Pedersen, C. (1993): Experimental investigations of wave boundary layers with a sudden change
in roughness. Journal of Fluid Mechanics, Vol. 252, pp.117-145.
7. Fredse, J., Sumer, B.M., Kozakkiewicz, A., Chua, L. H. C. and Deigaard,
R. (2003): E ect of externally generated turbulence on wave boundary
layer. Coastal Engineering, vol. 49, pp. 155-183.
8. Jensen, B. L., Sumer, B. M. and Fredse, J.: Turbulent oscillatory
boundary layers at high Reynolds numbers. Journal of Fluid Mechanics, Vol. 206, p. 265, 1989.
9. Kamphuis, J. W. (1975): Friction factor under oscillatory waves.
J. Waterways, Port, Coastal and Ocean Eng. Div., ASCE, Vol. 101
(WW2), pp. 135-144.
10. Lodahl, C. R., Sumer, B. M. and Fredse, J. (1998): Turbulent combined oscillatory ow and current in a pipe, J. Fluid Mech., vol. 373,
313-348.
11. Sleath, J. F. A. (1984): Sea Bed Mechanics, John Wiley & Sons.
12. Sumer, B. M., Jensen, B. L. and Fredse, J: Turbulence in oscillatory boundary layers. Advances in Turbulence. Proc. 1st European
Conference, Lyon, France, published by Springer, p. 556, 1987.
13. Sumer, B. M., Laursen, T. S. and Fredse, J. (1993): Wave boundary
layers in a convergent tunnel. Coastal Engineering, Vol. 20, pp. 317342.
14. Tanaka, H., Sumer, B. M. and Lodahl, C. (1998): Theoretical and
experimental investigation on laminar boundary layers under cnoidal
wave motion, Coastal Engineering Journal, vol. 40, No. 1, 81-98.

154

CHAPTER 5. WAVE BOUNDARY LAYERS

Chapter 6
Turbulence modelling
6.1

The closure problem

We have derived the basic equations of turbulent ows in Chapter 1. These

equations are the continuity equation (Eq. 1.6) and the Reynolds equations
(the equations of motion) (Eq. 1.14), respectively:
ui
=0
xi
(
with the quantity

ui
ui
+ uj
) = gi +
(
t
xj
xj
ij ,

(6.1)

ij

ui uj )

(6.2)

the stress on the uid

ij

ij

+ (

ui
uj
+
)
xj
xi

(6.3)

and the Reynolds stress tensor

ui uj =

u1 u1
u2 u1
u3 u1

u1 u2
u2 u2
u3 u2

u1 u3
u2 u3
u3 u3

(6.4)

in which there are, due to symmetry, six independent Reynolds stresses (three
diagonal components and three o -diagonal ones).
As has already been discussed (Chapter 1), there are four equations for
the mean ow, i.e., the continuity equation (Eq. 1.6) and the three Reynolds
155

156

CHAPTER 6. TURBULENCE MODELLING

equations (Eq. 1.14) whereas there are ten unknowns (namely, three components of the velocity, ui , and the pressure, p, and six components of the
Reynolds stress,
ui uj ), hence the system is not closed. This problem is
known as the closure problem of turbulence.
To close the system, we must nd enough number of equations to solve
for the unknowns. We do this through turbulence modelling.
(In the case of laminar ow, there are four unknowns (namely, three
components of the velocity, ui , and the pressure, p), and four equations, i.e.,
the continuity equation (Eq. 1.4) and three N.-S. equation (Eqs. 1.8 and
1.9). Therefore, the system is closed.)

6.2

Turbulence models

There are several kinds of turbulence models the most important of which
are
1. Algebraic models;
2. Turbulence-energy equation models; and
3. Simulation models such as Direct Numerical Simulation (DNS), Large
Eddy Simulation (LES) and Detached Eddy Simulation (DES).
Algebraic models, also known as the zero-equation models, are the simplest of all turbulent models. Here, (1) the Reynolds stress is expressed as
the product of a turbulence viscosity (eddy viscosity) and the mean strain
rate, the so-called Boussinesq eddy-viscosity approximation, and (2) the turbulence viscosity, in turn, is often computed in terms of a mixing length. Of
the algebraic models, the mixing-length model is the simplest and the most
popular one. This model will be described in the next section.
Turbulence-energy equation models are advanced models. Here, too, the
Reynolds stress is expressed as the product of a turbulence viscosity (eddy
viscosity) and the mean strain rate. However, the turbulence viscosity is not
computed in terms of a mixing length, but rather in terms of the turbulence
kinetic energy, k = 12 (ui ui ). There are two kinds of turbulence-energy equation models: (1) One-equation models, and (2) Two-equations models. As an
example, the so-called k
model, a two-equations model, will be described
in the following paragraphs.

6.3. MIXING-LENGTH MODEL

157

Simulation models are most advanced models. These models directly address the turbulence process itself, and therefore they do not involve the afore
mentioned closure problem. In a DNS study, a complete time-dependent solution of the Navier-Stokes equations and the continuity equation is sought.
Therefore the solution furnishes the time series of (1) the three components
of the velocity and (2) the pressure in 3-D space. Up until now, only low
Reynolds number-ows are calculated with the DNS method. For large
Reynolds numbers, the scales of the dissipative part of turbulence motion
are so small that this kind of small scale motion can not be resolved in a
numerical calculation (the number of grid points required to resolve this motion increases approximately with Re3 , Rodi, 1992). However, the upper
limit of the Reynolds number range is pushed to higher and higher values as
more and more computer power becomes available. Rogallo and Moin (1984)
gives a review of the DNS method where the principal/early publications are
cited. Regarding the LES model, here the large eddies (large-scale motions)
are computed while the smallest eddies are modelled. The details of the LES
model along with an application example are given in the chapter.
Wilcoxs (1994) book can be consulted for a detailed account of turbulence
models.

6.3

Mixing-length model

The mixing-length model is an algebraic model. We will describe this model

with reference to the steady boundary-layer ow close to a smooth or rough
wall (i.e., in the constant stress layer, see Section 2.1).

6.3.1

Turbulent transport of momentum

Under this subsection, we will focus on the turbulent transport of momentum.

We need this piece of information to develop the mixing-length theory/model.
The treatment will be quite general, not necessarily limited to the ow close
to a wall.
For this, we refer to Fig. 6.1, and we consider a small uid element (a
uid parcel) adjacent to the unit area indicated in Fig. 6.1, A (Fig. 6.2),
at time t. This uid parcel will be transported to the other side of the area
during the time interval t (B, Fig. 6.2), and this is due to the turbulent
motion. (No such transportation would occur if the ow was in the laminar

158

CHAPTER 6. TURBULENCE MODELLING

Figure 6.1: Denition sketch.

Figure 6.2: Travel of uid element A from one side of the area (dx,dz) considered
in the previous gure to the other side.

regime.)
The mass of the uid transported from one side of the unit area to the
other is
(1 1 dy)
and therefore the momentum transported in the y direction from one side
of the unit area to the other (x component):
x

component of momentum transported: (1 1 dy) u

or, inserting u = u + u
x

component of momentum transported: (1 1 dy) (u + u )

and, similarly, the y component:

y

component of momentum transported: (1 1 dy) v

6.3. MIXING-LENGTH MODEL

159

Now, Newtons second law (for the x direction)

d
[x component of momentum transported]
dt
d
[ (1 1 dy) (u + u )]
=
dt
d(dy)
(u + u )
=
dt

Fx =

in which Fx is the force generated by this momentum transport. Here,

d
(Vertical distance travelled by uid) = v = v , and therefore
dt

d(dy)
dt

Fx = v (u + u )
and by time averaging
Fx = v (u + u ) = v u + u v ) = 0 + u v = u v

(6.5)

Likewise, Newtons second law for the y direction gives

Fy = v v

(6.6)

These forces, Fx and Fy , are exerted on the unit area indicated in Fig. 6.1,
and therefore they are, respectively, the shear and normal stresses. (Notice
that these stresses are induced by turbulence (Fig. 6.3.)
So the above
analysis demonstrates that
1. The turbulent transport of momentum in the y direction generates
stresses in the uid;

Figure 6.3: Forces (stresses) induced by turbulence.

160

CHAPTER 6. TURBULENCE MODELLING

Figure 6.4: Stresses generated by turbulence.

2. The transport corresponding to the x component of the momentum
generates a shear stress, equal to
u v , considering the sign convention adopted in Chapter 1 (the positive shear stress is in the direction
opposite to the x direction); and
3. The transport corresponding to the y component of the momentum
generates a normal stress, equal to
v v (again, considering the positive normal stress is in the direction opposite to the y direction).
Hence the stresses generated by turbulence are
xx

xy

xy

yy

u2
uv

uv
v2

(6.7)

(see Fig. 6.4).

These are the Reynolds stresses introduced already in Chapter 1. We
obtained the Reynolds stresses in Chapter 1 by time averaging the NavierStokes equations. Here, we have obtained the Reynolds stresses in a di erent
way, in a way which will prove useful when developing the mixing-length
model/theory in the next sub-section.

6.3.2

Mixing length

1) Reynolds shear stress in terms of the mixing length

Consider a uid parcel, A, at location I (Fig. 6.5). Suppose that this
uid parcel has just been formed as a new coherent entity at Location I, and

6.3. MIXING-LENGTH MODEL

161

Figure 6.5: A uid parcel travelling from Level I to Level II.

suppose that it starts to move upwards. Now, from Level I, the uid parcel
will travel a distance m , called mixing length, and subsequently mix with
the surrounding uid at Level II. (We expect that m is much the same as
the Lagrangian macro length scale, or the Eulerian macro length scale in the
y direction, see Chapter 3).
With the upward motion of the uid parcel,
1. Momentum will be transported over the distance
Level II;

m,

from Level I to

2. This momentum transport will generate a shear stress, equal to

and

uv;

3. Newtons second law implies that the change in the momentum over
the time from I to II should be equal to the shear stress
u v , i.e.,
u v = Momentum II

Momentum I

(6.8)

The right-hand-side of the preceding equation is the momentum gained

with the travel of the uid parcel from I to II.
With the travel of the uid parcel from I to II, the x component of the
uid velocity will increase from uI to uII , and therefore the gained velocity
will be
du
(6.9)
uII
uI = m
dy
The gained momentum (x component) will then be, Fig. 6.6:
Momentum II

Momentum I = ( q)(

du
)
dy

(6.10)

162

CHAPTER 6. TURBULENCE MODELLING

Figure 6.6: Volume associated with Fluid Parcel A passing through per unit area.
in which q is the volume associated with the uid parcel (Fig. 6.5) passing
through per unit area per unit time (Fig. 6.6). The quantity q is
q = Vt (1 1) = Vt
and therefore the gained momentum (x component) will be
Momentum II

Momentum I = Vt

du
dy

(6.11)

in which Vt is the uid velocity in the vertical direction, an average velocity

representing the vertical velocity of the uid parcel over the travelled distance
m . From Eqs. 6.8 and 6.11:
u v = Momentum II
= Vt

Momentum I =
du
dy

(6.12)
(6.13)

2) Boussinesq eddy-viscosity approximation and Prandtls mixing length

hypothesis
Now, drawing an analogy to Newtons friction law
=

du
dy

(6.14)

6.3. MIXING-LENGTH MODEL

163

the turbulence shear stress can be written as (as was rst suggested by Boussinesq in 1877)
du
uv = t
(6.15)
dy
in which t is called turbulence viscosity (or eddy viscosity). Hence, from
Eqs. 6.13 and 6.15, the turbulence viscosity is
t

m Vt

(6.16)

The hypothesis leading to Eq. 6.16 is known as Prandtls mixing-length

hypothesis.
(It may be noted that, in a recent publication, Nielsen and Teakle, 2004,
develop a theory of turbulent di usion of momentum and suspended particles,
a nite-mixing-length theory, in which the turbulence viscosity is expressed
as t = m Vt (1 + O( m /L)2 ). Here L is the scale of the velocity distribution
u(y). It may also be noted that di erent di usivities exhibited by di erent
species, like density, heat and salinity, in turbulent ows may be qualitatively explained in terms of Nielsen and Teakles theory. See also Absi, 2005,
in relation to the latter publication.)
To sum up at this point: (1) The Reynolds shear stress is expressed as the
product of a turbulence viscosity (eddy viscosity), t , and the mean strain
rate, du
(Eq. 6.15), the so-called Boussinesq eddy-viscosity approximation;
dy
and (2) The turbulence (eddy) viscosity, in turn, is computed in terms of the
mixing length (Eq. 6.16).
3) Prandtls second hypothesis
The question is, however: What is Vt ? To assess and therefore model this,
we consider the control volume in Fig. 6.7. The motion associated with the
uid parcel A in Fig. 6.5 is from the left bottom corner of the control volume
to the right top corner, and involves three kinds of velocities:
1. The vertical velocity Vt at the instant I (It is important to notice that
there is no vertical velocity at the instant II because the uid parcel
mixes with the surrounding uid at II; Recall the denition of the
mixing length m .);
2. The horizontal velocity at the instant I; and

164

CHAPTER 6. TURBULENCE MODELLING

Figure 6.7: Control volume.

3. The horizontal velocity at the instant II.
The amount of uid entering the control volume in the vertical direction
(associated with this motion) per unit time is then:
Vt dx

(6.17)

while the amount of uid entering the control volume in the horizontal direction per unit time at the instant I:
uI

(6.18)

and that leaving the control volume in the horizontal direction per unit time
at the instant II:
uII m
(6.19)
Continuity requires:
Vt dx + uI

= uII

and therefore
Vt = (uII
and since O(

m)

uI )

dx

= O(dx) (small), then

Vt = uII

uI

or from Eq. 6.9:

Vt =

du
dy

6.3. MIXING-LENGTH MODEL

165

We write the term du

in terms of its absolute value to ensure that the amount
dy
of uid passing through the unit area, i.e., q(= v) is always positive (Fig.
6.5):
du
(6.20)
Vt = m
dy
This is known as Prandtls second hypothesis.
With the latter hypothesis (Eq. 6.20), the turbulence (eddy) viscosity
from Eq. 6.16 is expressed as
t

2
m

du
dy

(6.21)

4) van Driest mixing length

The fourth step is to determine the mixing length in the preceding equation. Until now the analysis is given for the general case, and therefore it can
be implemented for any distance y from the wall. However, from now on,
we will restrict our attention to the region close to the wall, the layer called
constant stress layer. (See Chapter 2). So, y is so small that we are in the
constant stress layer,
0 , and for this
y

in which h is the ow depth in the case of an open channel ow, or it is the

radius in the case of a pipe ow, or the boundary-layer thickness in the case
of a boundary layer over a wall. (See discussion in Section 2.1.2);
On the other hand, y should be su ciently large, so large that the transport of momentum is due to turbulence. (Viscosity is neglected in the transport of momentum; This is for the case of the smooth wall. The rough-wall
case will be treated later.) We require that the transport of momentum is
due to turbulence because the previously described mixing length (Section
1.3.2) is essentially based on the turbulent transport of momentum. This
last condition requires that y + should be
y + > 30
in which y + =
2).

yUf

70

, Uf being the wall shear stress (friction) velocity (Chapter

166

CHAPTER 6. TURBULENCE MODELLING

Now, we should expect that the mixing length (for y values in the range
y
h and y + > 30 70) increases with increasing distance from the wall,
and the simplest form of this dependence is
m

(6.22)

= y

in which
is a constant. From the solution for the velocity (Eq. 6.28),
turns out to be the von Karman constant ( = A1 = 0.4) introduced in
Chapter 2. Eq. 6.22 is valid for the ranges 30 70 < y + , and y
h.
van Driest (1956) revised Eq. 6.22 so as to accommodate the inuence of
viscosity in the transport of momentum. This led to
m

= y[1

exp(

y+
)]
Ad

(6.23)

in which Ad is called the damping coe cient, and taken as 25, and this
equation for the mixing length is valid for the range 0 y
h. From Eq.
6.23, it is seen that
y, for large values of y +

(6.24)

while
0, as y +

(6.25)

The latter implies that there will be no turbulent transport of momentum

as the wall is approached. Note that, for rough-wall ows, the mixing-length
will inevitably change. We shall consider the transitional and rough-wall
situations later.
5) Summary of the model equations and solution
Eqs. 6.15, 6.21 and 6.23 are the model equations:
uv =

= y[1

2
m

du
dy

du
dy

exp(

y+
)]
Ad

6.3. MIXING-LENGTH MODEL

167

So, for the problem we consider, i.e., the steady boundary-layer ow close
to a smooth wall (in the constant stress layer, see Section 2.1), there are three
model equations (the preceding equations), and one Reynolds equation, Eq.
2.87, i.e., altogether four equations, and four unknowns, namely, u,
uv,
t , and m . Hence, the system is closed. Therefore the unknowns can be
determined. Of particular interest is the velocity, u. The solution for the
latter quantity is obtained in Section 2.3.1 (Eq. 2.97):
y+

dy +

u = 2Uf
0

1+ 1+4

2 y +2 [1

exp(

y+ 2
)]
Ad

1/2

(6.26)

(the van Driest velocity distribution, Fig. 2.20). This equation is valid in
the constant stress layer, namely for the range 0 y
h.)
+
For small values of y , the van Driest prole reduces to the linear velocity
distribution in the viscous sublayer (Eq. 2.20),
u
= y+
Uf

(6.27)

while, for large values of y + , it reduces to the logarithmic velocity distribution

in the logarithmic layer (Eq. 2.20),
u = Uf (A ln y + + B)

(6.28)

It is seen from Fig. 2.20 that the van Driest prole agrees remarkably well
with the experiment.
6) Mixing length in the case of transitional and rough walls and solution
In this case, the mixing length may be given as (Cebeci and Chang, 1978)
m

= (y +

y)[1

exp(

(y + + y + )
)]
Ad

(6.29)

y in the above equation is measured from the level where the velocity u is
zero. This level lies almost at the top of the roughness elements, as discussed
in Section 2.3.1 (y here should not be confused with the distance measured
from the theoretical wall. y+ y is the distance measured from the theoretical
wall. See the detailed discussion in Section 2.3.1).

168

CHAPTER 6. TURBULENCE MODELLING

y is the so-called coordinate displacement, or the coordinate shift (see

Example 5 in Section 2.3.1), and given by Cebeci and Chang (1978)
y + = 0.9[ ks+

ks+ exp(

ks+
)];
6

5 < ks+ < 2000

(6.30)

in which ks+ is the roughness Reynolds number (ks being Nikuradses equivalent sand roughness):
ks Uf
ks+ =
(6.31)
Here the distance from the wall y should be in the interval 0 y
h where
the upper boundary ensures that the y levels lie in the constant stress layer.
The solution for u for the set of the model equations along with the
Reynolds equation is obtained in Section 2.3.1 (Eq. 2.103):
y+

dy +

u = 2Uf
0

1+ 1+4

2 (y +

y + )2 [1

exp(

(y+ + y + ) 2
)]
Ad

1/2

(6.32)

Fig. 2.21 displays the velocity proles obtained from the preceding equation
including the smooth-wall prole. In Fig. 2.21, a represents the linear prole
in the viscous sublayer (Eq. 6.27), b the logarithmic prole (Eq. 6.28) in the
case of the hydraulically smooth wall, and c the logarithmic prole in the
case of the completely rough wall (Eq. 2.65), namely, by putting z = y + y,
u = AUf ln(

30z
)
ks

(6.33)

z being measured from the theoretical wall.

7) Turbulence quantities
The turbulence quantities, namely the Reynolds stresses,
u 2,
v 2,
2
w and u v , also are unknowns. The Reynolds shear stress,
u v , can
du
be determined, using the model equation u v = t dy , once the velocity u is
computed. The mixing-length model is designed only to be able to handle
the velocity u, (and from the latter, the Reynolds stress
u v ). It does not
o er solutions for the other three unknowns, the turbulence quantities
u 2,
2
2
v and
w .

6.4. K-OMEGA MODEL

6.4

169

k-omega model

This model is a turbulence-energy equation model. As mentioned in the preceding paragraphs, there are two kinds of turbulence-energy equation model,
one-equation model and two-equations model. The k
model is a twoequations model.
There are three versions of the k
model in the literature: (1) the original k
model, which is due to Wilcox (1994), (2) the k
, BSL (baseline)
model, and (3) the k
, SST (shear-stress transport) model. The latter two
models were developed by Menter (1993) to improve Wilcoxs original model
so that an even higher sensitivity could be obtained for adverse-pressuregradient ows. In this section, we will describe the k
, SST model.
Incidentally, the following note is worth mentioning. Menter (1993) made
an extensive comparison between (1) the classic k
model; (2) the original
k
model; (3) the k
, BSL model; and (4) the k
, SST model for
various well documented ows. The tested ows were, among others, two
kinds of adverse pressure gradient ow (one having a very strong adverse
pressure gradient, so strong that separation occurs); the backward-facingstep ow; and the ow past a NACA 4412 airfoil at an angle of attack near
maximum lift condition. The latter two ows also have substantial adverse
pressure gradient e ects. The main conclusion from this inter-comparison
exercise was that the k
, SST model gave the most accurate results while
the k
model did not yield as accurate results as the other three for the
tested adverse-pressure-gradient ow cases.

6.4.1

Model equations

There are two basic equations in the k

model: one equation for k, and
the other equation for in which k is the turbulent kinetic energy (Eq. 1.31
or Eq. 6.36) and the specic dissipation of turbulent kinetic energy (Eq.
6.37). (We will use the symbol k for turbulent energy in favour of Kt , to be
consistent with the usual notation used in the literature in conjunction with
turbulence modelling.)
The equation for k:
k
+
t

uj k
xj

xj

( +

k t)

k
=
xj

ij

ui
xj

(6.34)

(It should be noted that this equation is in a form slightly di erent from that

170

CHAPTER 6. TURBULENCE MODELLING

given in Chapter 1 (Eq. 1.30). The derivation of this equation is given in

Wilcox (1994, pp. 74-76). This reference also includes a detailed discussion
on the k
model.)
Now, the second equation, the equation for :
t

uj
xj

( +

xj

+2 (1

t)

F1 )

xj

ij
t

ui
xj

k
xj xj

(6.35)

Above, k is:
1
k = ui ui
2

(6.36)

and :
=
in which

(6.37)

is the dissipation of turbulent kinetic energy

ui ui
xk xk

=
In the preceding equations,
ij

ij

(6.38)

is the Reynolds stresses (=

ui uj =

2
k
3

ui
uj
+
xj
xi

ij

ui uj ),
(6.39)

ui is the uctuating components of the velocity, ij is the Kronecker delta

( ij = 1 for i = j and ij = 0 for i = j), and t is kinematic eddy viscosity,
is one of the model closure constants.
t = t / , and
As seen from Eq. 6.39, this model, too, assumes the Boussinesq eddyviscosity approximation where the Reynolds stress is related to the rate of
mean strain. Note that we need the second term on the right-hand-side of
Eq. 6.39 so that, when contracting (that is, when i is set equal to j), Eq.
6.39 gives k = 12 ui ui , compatible with Eq. 6.36 (Wilcox, 1994, p. 75).
Now, k in Eq. 6.34 and , and in Eq. 6.35 are given as
k

= F1

k1

+ (1

F1 )

k2

(6.40)

= F1

+ (1

F1 )

(6.41)

6.4. K-OMEGA MODEL

171
= F1

+ (1

F1 )

(6.42)

= F1

+ (1

F1 )

(6.43)

in which k1 , k2 ,
1,
2,
1, 2,
1 and
2 , the model constants, are
combined as a function of the distance z to the nearest wall with the socalled blending function , F1 .

6.4.2

Model constants

The model constants for inner (wall) region are:

2
1
1
)
k1
1
1
1 (=
0.0750 0.09

0.567

and for the outer region (

2
2
2 (=
0.0828 0.09

6.4.3

0.85

0.5

in the tables is the von Karman constant):

2
2
)
k2
2

0.463

1.0

0.856

Blending function F1

The blending function F1 is

F1 = tanh arg14

(6.44)

and

arg1 = min max

k 500
;
0.09 z z 2

4 k 2
(CDk )z 2

(6.45)

with
(CDk ) = max 2

k
; 10
xj xj

20

(6.46)

172

CHAPTER 6. TURBULENCE MODELLING

Figure 6.8: Computation domain.

6.4.4
In the k

Eddy viscosity and Blending function F2

, SST model, the eddy viscosity is calculated from
t

0.3k
max (0.3 ; F2 )

(6.47)

where is the absolute value of the vorticity. The function F2 in Eq. 6.47
is another blending function, dened by
F2 = tanh arg22

(6.48)

in which
arg2 = max 2

k 500
;
0.09 z z 2

(6.49)

Both F1 and F2 take the value unity at the wall and gradually decrease and
approach to zero as the distance from the wall is increased.

6.4.5

Boundary conditions

The boundaries of the computational domain (Fig. 6.8) are (1) Inlet, (2)
Outlet, (3) Symmetry Boundaries and (4) Walls (e.g., the bed in the example

6.4. K-OMEGA MODEL

173

given in Fig. 6.8 and the surface of a rigid body present in the computational
domain, if any).
1) Inlet and outlet boundary conditions
At the inlet, zero transverse, and vertical velocities are specied. At
the outlet, zero-gradient conditions ( / n = 0), Neumann conditions, are
applied for all quantities.
2) Symmetry boundaries
At the symmetry boundaries , Neumann conditions are applied for k and
and for the three components of the velocity (u, v and w). This is (1) at
the sides and (2) at the top surface of the computational domain where a
lid is placed in the case when the model cannot handle the free surface,
Fig. 6.8).
3) Walls (e.g., the bed and/or the surface of a rigid body)
1. Zero velocity is specied for u , v and w.
2. Zero turbulent energy is specied for k when the wall is smooth, while
the Neumann condition is applied when the wall is rough and transitional.
Regarding the latter condition, experiments reveal that the Neumann condition is satised ( k/ n = 0) at the wall for rough boundaries, n being the
direction normal to the wall (Nezu, 1977, Sumer et al., 2001 and 2003). A
validation exercise for the case of a steady boundary layer in an open channel ow revealed a much better agreement with the experiments (Nezu, 1977,
and Sumer et al., 2001) when the boundary condition was taken in the form
of the Neumann condition (Roulund, 2000, p. 25).
3. The Dirichlet condition is applied for , namely
= Sr

Uf2

at the wall

(6.50)

where Uf =
o / is the friction velocity based on the wall shear stress o .
The quantity Sr is a tuning parameter, and it is used to account for the bed
roughness:
Sr =

3
40
+
ks
100
0.85
(ks+ )

, ks+ < 20.2

, ks+ > 20.2

(6.51)

174

CHAPTER 6. TURBULENCE MODELLING

where ks+ = (ks Uf )/ is the wall roughness in wall units, and ks is Nikuradses
equivalent sand roughness. Wilcox (1994) was the rst to introduce the
boundary condition in Eq. 6.50. The tuning parameter Sr was given by
Wilcox (1994) as in Eq. 6.51, but with constants slightly di erent from
those in Eq. 6.51 due to the special implementation of the velocity boundary
conditions. With the present constants (Eq. 6.51), the friction velocity
determined for a wide range of velocities for the case of the undisturbed ow
was found to agree rather well with that obtained from the classic Colebrook
and White equation (the di erence being less than 6%, depending on the
wall roughness, Roulund, 2000, p. 23).
4. For the implementation of the preceding boundary condition, the wall
shear stress, 0 (= Uf2 ), is needed. Traditionally, 0 is calculated from 0 =
u
( + t )( xuji + xji )W all . However, it may also be calculated from the van
Driest (1956) velocity prole (Eq. 6.32):
u
=2
Uf

y+
0

dy +
1+

1+4

(y + +

y + )2

exp

y+ +

y+

(6.52)

This improves the stability of the model and allows a coarser mesh resolution
at the bed, as was revealed by the work of Roulund et al. (2005). In the
implementation of Eq. 6.52, u is taken as the tangential velocity in the cell
adjacent to the wall.
Similar implementation of the bed shear stress in the wall boundary conditions is reported in Srensen (1995) for the k
model.
As a nal remark in conjunction with the previous description, it may be
emphasized that the k
model is a well tested model and has been used
successfully in various contents in computational uid dynamics (Wilcox,
1994, Menter, 1993). The only change made in the present implementation
is the values of the constants in Eq. 6.51 which are slightly di erent from
those originally given by Wilcox (1994) (item 3 above), plus the way in which
the wall shear stress is calculated is through the van Driest equation (item 4
above).

6.4.6

Numerical computation

To sum up, the set of equations which needs to be solved consists of

1. The continuity equation, Eq. 6.1;

6.4. K-OMEGA MODEL

175

2. The Reynolds equations, Eqs. 6.2-6.4; and

3. The model equations, Eqs. 6.34-6.35.
With the solution, the three velocity components, the pressure, the turbulent kinetic energy and the specic dissipation of turbulent kinetic energy
are calculated.
Unlike the mixing-length model, this set of equations is too complex for
one to seek analytical solutions. Instead, the equations are solved numerically. To this end, one resorts to a CFD (Computational-Fluid-Dynamic)
code coupled with a turbulence model (e.g., the k
model in the present
case) which is incorporated into the code. CFD codes may also o er various
turbulence-modelling options. In these codes, the momentum, continuity
and turbulence-model equations are typically transformed into curvilinear
coordinates, and they may be linearized and decoupled. The equations are
discretized, using a computational mesh (e.g., the latter can be based on
a multi-block structure with each block containing N 3 computational cells
where N is the number of cells in each spatial direction.)
The k
model was implemented in a Ph.D. study to calculate ow
around a vertical cylinder (a bridge pier or a pile) in relation to numerical
modelling of scour around such structures (Roulund, 2000, and Roulund et
al., 2005). More information about the model, boundary conditions, numerical computation, etc. with regard to the previously mentioned application
can be found in the latter publications.

6.4.7

An application example

Under this subsection, we will present the results of a small study where the
k
model was employed.
The CFD code used in this study, EllipSys3D, is a three-dimensional
general purpose ow solver (an incompressible general purpose Navier-Stokes
solver). It is a multiblock nite-volume numerical model that solves the
incompressible Reynolds equations (Eqs. 6.2-6.2). The k
, SST model is
used as the turbulence closure. EllipSys3D has been developed at the Ris
National Laboratory, Denmark and at the Technical University of Denmark,
Department of Fluid Mechanics. The code o ers a variety of turbulencemodel options, including the k
, SST model. The basic principles of the
model are described in Michelsen (1992) and Srensen (1995).

176

CHAPTER 6. TURBULENCE MODELLING

The problem studied in this example is a steady boundary layer ow in an

open channel with a smooth, or alternatively, a rough bed. The ow domain
is the same as that depicted in Fig. 6.8. As the model does not have the
free-surface facility, a lid is placed at the top surface of the computational
domain. The boundary conditions are exactly the same as described above
under the heading Boundary conditions. The study is a part of the Ph.D.
project of Dixen (2007).
Figs. 6.9 and 6.10 compare the velocity distributions obtained in this
study with the experimental data, in linear and semi-logarithmic plots, respectively. (The Reynolds number is Re(= hV ) = 1.9 104 , V being the
depth-average velocity.) The agreement between the model prediction and
the experiment is very good. (It may be noted that the experimental data
agree very well with the van Driest velocity prole, not shown here for reasons
of space.)
Figs. 6.11 and 6.12 present the kinetic energy of turbulence, k, obtained
by the model, Fig. 6.11 in y/h, and 6.12 in y + (the latter gure gives a closeup picture near the bed). The gures also present data from experiments.
k = 12 (u 2 + v 2 + w 2 ) is obtained in the experiments by measuring the three
components of the velocity. The model prediction is not as good as that
regarding the mean velocity. It underpredicts the turbulent energy by about
30% near the bed. Wilcox (1994, pp. 132-135) also reports similar behaviour
for channel- and pipe-ow applications.
Next we give the results obtained for a rough bed. Nikuradses equivalent
k U
sand roughness, ks+ (= s f ), is 200. The Reynolds number, Re(= hV ), is
1.4 104 . Similar to the smooth-bed case, Figs. 6.13 and 6.14 compare
the velocity distributions with the experimental data, in linear and semilogarithmic plots, respectively. The agreement between the model results
and the experiment is generally good although the model underpredicts the
velocity by slightly less than 20% near the free surface (at the lid in the
computational domain). The model prediction captures the velocity quite
well in the logarithmic region and in the bu er zone.
Finally, Figs. 6.15 and 6.16 present the kinetic energy of turbulence,
yU
k, obtained by the model, Fig. 6.15 in y + (= f ), and 6.16 in kys . The
agreement between the model prediction and the experiment is quite good.

6.4. K-OMEGA MODEL

Figure 6.9: Velocity prole. Smooth wall.

Figure 6.10: Velocity prole. Smooth wall. Semi-log plot.

177

178

CHAPTER 6. TURBULENCE MODELLING

Figure 6.11: Turbulence kinetic energy versus distance from the wall. Smooth
wall.

Figure 6.12: Turbulence kinetik energy versus distance from the wall, in terms of
inner ow parameters. Smooth wall.

6.4. K-OMEGA MODEL

Figure 6.13: Velocity prole. Rough wall.

Figure 6.14: Velocity prole. Rough wall. Semi-log plot.

179

180

CHAPTER 6. TURBULENCE MODELLING

Figure 6.15: Turbulence kinetic energy in terms of inner ow parameters. Rough

wall.

Figure 6.16: Turbulence kinetic energy in terms of inner ow parameters where

the length scale is taken as ks. Rough wall.

6.5. LARGE EDDY SIMULATION (LES)

6.5
6.5.1

181

Large Eddy Simulation (LES)

Model equations

The principal idea in the method of LES is to average the N.-S. equations over
a small volume of space. (Recall that the N.-S. equations are averaged over
time to obtain the Reynolds equations, i.e., Eqs. 6.1 and 6.2). Normally, the
size of the averaging volume is chosen such that the small-scale component
of turbulence is, for convenience, not resolved. The aforementioned spaceaveraging gives the following set of equations: (1) the continuity equation,
ui
=0
xi

(6.53)

and (2) the equation of motion

ui
ui
+ uj
=
t
xj

1 p
+
xi

xj

ui
uj
+
)
xj
xi

ui uj
xj

(6.54)

in which ui = the space-averaged (the resolved) component of the velocity;

p = the space-averaged (the resolved) component of the pressure; and ui =
the uctuating component of the velocity over the averaging volume (called
the subgrid-scale component of the velocity, to di erentiate it from the term
uctuation in the conventional Reynolds decomposition), and ui uj = the
subgrid-scale stresses, analogous to the Reynolds stresses in the case of the
Reynolds decomposition.
The subgrid-scale stresses (SGS) are written (by analogy to the Boussinesq approximation)
ui
uj
ui uj = t (
+
)
(6.55)
xj
xi
in which t = the turbulence viscosity associated with the unresolved component of turbulence. (Note that the subgrid-scale stresses in the above
equation are the o -diagonal part of the subgrid-scale stress tensor. The diagonal part is automatically included in the pressure term.) Hence, Eq. 6.54
will be
ui
ui
+ uj
=
t
xj

1 p
( +
+
xi
xj

t )(

ui
uj
+
)
xj
xi

(6.56)

(It may be noted that the turbulence viscosity t here is di erent from that
in the k
model dened in the previous section. As emphasized in the

182

CHAPTER 6. TURBULENCE MODELLING

present denition, Eq. 6.55, t is associated with the unresolved small-scale

turbulence averaged over the averaging volume, while t dened in conjunction with the k
model is associated with the turbulence over the
entire spectrum).
On dimensional grounds, the turbulence viscosity is written as
t

= Cs2

(6.57)

in which is a length scale and Cs is a nondimensional constant. (By making

an analogy to Prandtls mixing-length hypothesis, it may be seen that s must
be a quantity like u/ y). We shall return to this point later.
Clearly, the resolved quantities, ui (xi , t), p(xi , t) and t (xi , t) in Eqs. 6.556.57 should be dependent on the size of the averaging volume. The smaller
the size of this volume, the better the resolution (both in space and time),
and therefore the more information on space and time variations of ui (xi , t),
p(xi , t) and t (xi , t), implying that these quantities are dependent on the
size of the averaging volume. This feature in LES is handled through t as
follows.
As seen in Eq. 6.57, t involves a length scale, . This quantity is taken
as the length scale representing the size of the averaging volume
= ( x1 x2 x3 )1/3

(6.58)

in which x1 , x2 , x3 are the grid sizes in the x1 , x2 , x3 directions,

respectively. (The averaging volume being considered to be x1 x2 x3 ).
Furthermore, the quantity s in Eq. 6.57 is taken as
s = (2Sij Sij )1/2

(6.59)

1 ui
uj
Sij = (
+
)
2 xj
xi

(6.60)

in which Sij

It is easy to see that s can also be written in the following form

s= (

ui
uj ui
+
)
xj
xi xj

1/2

(6.61)

The above model (Eqs. 6.53-6.59) is known as the Smagorinsky (1963)

model for LES. As seen, only one closure coe cient is required, namely

6.5. LARGE EDDY SIMULATION (LES)

183

Cs , the Smagorinsky constant. The value of the Smagorinsky constant is

determined from experiments.
To sum up, the solution of the above set of equations, namely Eqs. 6.536.55, and Eqs. 6.57-6.60, basically gives the time series of the hydrodynamic
quantities (the velocity components and the pressure) in space, similar to
the time series obtained in a laboratory experiment using, for example, the
Laser Doppler Anemometer (LDA), or the Particle Image Velocimetry (PIV)
techniques. Once we obtain this raw data, then we process the data in
terms of statistical quantities such as the mean, the variance, other statistical
moments, and even correlations and spectra, in exactly the same fashion as
we process data from experiments. (It should be added that, clearly, the time
series obtained through the LES experiments are resolved with a resolution
equal to the grid size selected in the simulation, just as the experimentally
obtained time series are resolved with a resolution equal to the size of the
measurement volume of the measuring instrument, e.g. to the size of the
measurement volume of the LDA probe.)
A detailed account of turbulence models for LES has been given in a
review paper by Meneveau and Katz (2000).

6.5.2

An application example

In this subsection, we will present the results of a study where the LES model
was used to simulate the wave boundary layer in a U-shaped oscillating tunnel
(Fig. 6.17); The study is essentially a part of a Ph.D. work conducted at the
Technical University of Denmark, Lohmann et al. (2006). The CFD code
employed in the study, called NS3, a 3-D numerical ow solver, was made
available by DHI Water and Environment.
The Smagorinski subgrid-scale model was used in the calculations. Regarding the value chosen for the Smagorinski constant,Cs , experience shows
that Cs equals 0.2 in isotropic turbulence and 0.1 in channel ow. In Lohmann
et al.s (2006) study, the latter value was found to be most appropriate, based
on comparison with experimental data. The van Driest damping function
y+
f = 1 exp( A
) (see Eq. 6.23) is generally used as a standard part of
d
the Smagorinski subgrid-scale formation; it was also used in the study of
Lohmann et al. (2006) (see p. 3 of the latter publication).
Note that the free-stream velocity varies according to
U0 = U0m sin( t)

184

CHAPTER 6. TURBULENCE MODELLING

Figure 6.17: U-shaped oscillating tunnel.

in which
is the angular frequency of the motion,
= 2T , T being the
period. (See Chapter 5 for a detailed discussion of wave boundary layers.)
Fig. 6.18 presents the raw time series of the three components of the
resolved velocity at three di erent values of the distance y from the bed. As
seen from Fig. 6.18, the time series is, qualitatively, no di erent from the
ones obtained experimentally using LDA (c.f. Fig. 5.7 in Chapter 5).
Ensemble averaging is used to calculate the mean value of the resolved
velocity, e.g.,
N
1
u(y, t) =
u[y, (t + (j 1)T )]
(6.62)
N j=1
and the root-mean-square (r.m.s.) value of the subgrid-scale component of
the velocity, e.g., u = u u,
u 2 (y, t) =

1
N

1
2

[u[y,

(t + (j

1)T )]

u(y, t)]2

(6.63)

j=1

Lohmann et al. (2006) found that, for the chosen phase values, averaging

6.5. LARGE EDDY SIMULATION (LES)

Figure 6.18: Time series of velocity components.

185

186

CHAPTER 6. TURBULENCE MODELLING

Figure 6.19: Sequence of velocity proles over half period of the motion.

Figure 6.20: Sequence of velocity proles over half period of the motion. Semi-log
plot.

6.5. LARGE EDDY SIMULATION (LES)

Figure 6.21: Sequence of turbulence proles.

187

188

CHAPTER 6. TURBULENCE MODELLING

Figure 6.22: Shear stress proles.

rst over the x z plane at a given y location, and then over 3 periods was
su cient to achieve reliable ensemble averages.
Fig. 6.19 gives a sequence of velocity proles at di erent phase values t
over one half period of the motion. The distance from the wall is normalized
T
by the amplitude of the free-stream motion, a(= U0m
). The gure also gives
2
two sets of other data: one from the numerical study of Justesen (1988) where
a turbulence-energy equation model, namely the k
model, was used; and
the second from the experimental work of Jensen et al. (1989); the latter
has been described in greater detail in Chapter 5. Fig. 6.19 shows that the
LES model does a better job than the k
model. In Fig. 6.20, the velocity
information is plotted on a semi-logarithmic graph. The gure illustrates how
the LES solution behaves for small values of the distance from the bed in
relation to the experimental data. From the gure, although superior to the
k
model, the LES model underpredicts the velocities for small y values.
Figs. 6.21 and 6.22 present the turbulence data obtained by the LES
model along with the experimental data. Although there are discrepancies
between the model prediction and the experiment, the model predicts the

6.6. REFERENCES

189

turbulence fairly well. Yet, one must note that the model near the bed fails
to some extent to predict the turbulence accurately. This may be a major
shortcoming when processes (to be studied by the model) have signicant
elements related to the turbulence near the bed, particularly related to the
y and z components of turbulence.

6.6

References

1. Absi, R. (2005): Comment on Turbulent di usion of momentum and

suspended particles: A nite-mixing-length theory (Physics of Fluids,
16, 2342 (2004)). Physics of Fluids, vol. 17, 079101-079102.
2. Cebeci, T. and Chang, K.C. (1978): Calculation of incompressible
rough-wall boundary-layer ows. AIAA Journal, vol. 16, No. 7,
p.730.
3. Dixen, M. (2007): Ph.D. Study. Technical University of Denmark,
MEK, Coastal, Maritime and Structural Engineering Section. Ph.D.
study is in progress, and undertaken under the supervision of B. M.
Sumer and J. Fredse.
4. Jensen, B. L., Sumer, B. M. and Fredse, J.: Turbulent oscillatory
boundary layers at high Reynolds numbers. Journal of Fluid Mechanics, Vol. 206, p. 265, 1989.
5. Justesen, P. (1988): Turbulent wave boundary layers. Series Papers
43, Institute of Hydrodynamics and Hydraulic Engineering, Lyngby,
Denmark. A Ph.D. study supervised by J. Fredse and B.M. Sumer.
6. Lohmann, I.P., Fredse, J., Sumer, B.M. and Christensen, E.D. (2006):
Large eddy simulation of the ventilated wave boundary layer, J. Geophysical Research, vol. 111, C06036, doi:10.1029/2005JC002946, 1-21.
7. Meneveau, C. and Katz, J. (2000): Scale-Invariance and turbulence
models for Large-Eddy Simulation. In: Annual Rev. Fluid Mech., vol.
32, 1-32.
8. Menter, F. R. 1993 Zonal two equation k- turbulence models for aerodynamic ows. Paper AIAA-93-2906, AIAA 24th Fluid Dynamic Conference, July 6-9, 1993, Orlando, Florida.

190

CHAPTER 6. TURBULENCE MODELLING

9. Michelsen, J. A. 1992 Basis3D - a platform for development of multiblock PDE solvers. Technical report, Dept. of Fluid Mechanics, Technical University of Denmark, AFM 92-05, ISSN 0590-8809.
10. Nezu, I. 1977 Turbulent Structure in Open Channel Flows. Ph.D.
thesis, Kyoto University, Japan.
11. Nielsen, P. and Teakle, A.L. (2004). Turbulent di usion of momentum
and suspended particles: A nite-mixing-length theory. Physics of
Fluids, vol. 16, Number 7, p. 2342.
12. Rodi, W. (1992): On the simulation of turbulent ows past blu bodies. J. Wind Eng., No. 52, August, pp.1-16.
13. Rogallo, R.S. and Moin, P. (1984): Numerical simulation of turbulent
ows. Annual Review of Fluid Mechanics, vol. 16, pp. 99-137.
14. Roulund, A. 2000 Three-dimensional modelling of ow around a bottommounted pile and its application to scour. Ph.D. thesis, Department
of Hydrodynamics and Water Resources, Technical University of Denmark, Supervisors B.M. Sumer, and J. Fredse.
15. Roulund, A., Sumer, B.M., Fredse, J. and Michelsen, J.: Numerical
and experimental investigation of ow and scour around a circular pile.
J. Fluid Mechanics, vol. 534, 351-401, 2005.
16. Srensen, N. N. 1995 General purpose ow solver applied to ow over
hills. Ph.D. thesis, Ris National Laboratory, Roskilde, Denmark, RisR-827(EN).
17. Sumer, B. M., Chua, L. H. C., Cheng N.-S. & Fredse, J. 2003 The
inuence of turbulence on bedload sediment transport, J. Hydraulic
Engineering ASCE, 129, No. 8, August Issue.
18. Sumer, B. M., Cokgor, S., & Fredse, J. 2001 Suction removal of sediment from between armour blocks. Journal of Hydraulic Engineering,
ASCE, 127, No. 4, pp. 293-306, 2001.
19. van Driest, E.R. (1956): On turbulent ow near a wall. J. Aeronautical Sciences, vol. 23, p.1007.

6.6. REFERENCES

191

20. Wilcox, D.C. (1994): Turbulence Modelling for CFD. DCW Industries,
Inc., La Canada, California, 460 p.
.