Lecture Notes on Fluid Dynamics

Last edited: October 5, 2018

Contents

1 Warm-up: poor man’s approach to Fluid Dynamics 1

1.1 Leonardo’s Law: mass conservation . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Example 1: why is it always windy on Aarhus Ø? . . . . . . . . 1
1.1.2 Example 2: falling stream of liquid . . . . . . . . . . . . . . . . 2
1.1.3 Example 3: wave energy converter . . . . . . . . . . . . . . . . . 3
1.1.4 Example 4: wake flow behind a wind turbine and wind farm
optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.5 Power optimization of a two-turbine wind farm . . . . . . . . . 6

2 Derivation of Navier-Stokes equation 8

2.1 Pressure force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Friction force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Navier-Stokes equation in components . . . . . . . . . . . . . . . . . . 13
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Simplification of the Navier-Stokes equation 17

3.1 Simplification I: incompressible flows . . . . . . . . . . . . . . . . . . . 17
3.2 Simplification II: incompressible, ideal, stationary, irrotational flows . 17
3.3 Derivation of Bernoulli’s equation . . . . . . . . . . . . . . . . . . . . . 19
3.4 Example: Why does an airplane fly? . . . . . . . . . . . . . . . . . . . . 20

4 Ideal flow: planar 2-dimensional potential flow around cylinder 21

4.1 Stream line around a cylinder . . . . . . . . . . . . . . . . . . . . . . . . 23

5 More on ideal potential flows 28

6 Forces on a 2-dimensional body 33

6.1 Turbine blade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

i
 6.2 Sailing against the wind . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

7 Reynolds number 36

8 Viscous pipe flow 40

9 Boundary layers 43
9.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.2 Separation of boundary layers . . . . . . . . . . . . . . . . . . . . . . . 48
9.3 Solution of Prandtl equations for free boundary layers . . . . . . . . . 50

10 Small-amplitude surface waves 54

10.1 First boundary condition: (z = −d = −∞) . . . . . . . . . . . . . . . . 56
10.2 Second boundary condition: (surface) . . . . . . . . . . . . . . . . . . . 56
10.3 Kinematic boundary condition . . . . . . . . . . . . . . . . . . . . . . . 57

11 Sound waves 62
11.1 Outlook: shock waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

A Instabilities (not corrected) 66

ii
1 Warm-up: poor man’s approach to Fluid Dynamics

This simple approach is capable of quite a few important applications!

1.1 Leonardo’s Law: mass conservation

What streams into a volume has to stream out again.

Figure 1: An illustration of mass conservation.

Mass conservation means that the inflow on the left side must equal the outflow on
the right side. That is

∆m1 = ∆m2 (1.1)

⇓
ρA1 u1 ∆t = ρA2 u2 ∆t (1.2)
⇓
A1 u1 = A2 u2 (1.3)

Here (1.3) is known as Leonardo’s Law. It has the following properties:

A1 < A2 ⇒ u1 > u2 (1.4)

A1 > A2 ⇒ u1 < u2 . (1.5)

1.1.1 Example 1: why is it always windy on Aarhus Ø?

In front of the houses:

∆m1 = ρair A1 u1 ∆t (1.6)
Between the two houses:
∆m2 = ρair A2 u2 ∆t (1.7)

1
 Figure 2: The gap between adjacent apartment buildings seen from the side.

Equating (1.6) and (1.7) gives

A1
u2 = u1 (1.8)
A2
Plugging in "realistic" numbers:

65 m · h
u2 = · 10 m/s = 43.3 m/s (1.9)
15 m · h

Question: Why balconies?

Answer: Architects are not engineers/physicists!

1.1.2 Example 2: falling stream of liquid

Figure 3: A stream of falling liquid.

We use Leonardo’s law:

πR2 u1 = πr2 u2 (1.10)

2
Energy conservation tells us that the sum of kinetic and potential energy is con-
served:
m 2 m
u2 = u21 + mgh (1.11)
2 2
⇓
u22 = u21 + 2gh, (1.12)

where g = 9.82 m/S2 is the acceleration of gravity.

( )1 ( ) 12 ( ) 12
r πr2 2 A2 u1
= = = (1.13)
R πR 2 A1 u2
( 2 ) 14 ( ) 1
u1 u21 4
= = (1.14)
u22 u21 + 2gh

Remark: The narrowing of a falling stream of liquid holds only for the upper
part of the stream. At some height h the stream becomes too thin and drop
formation sets in.

1.1.3 Example 3: wave energy converter

Figure 4: Schematic of a wave energy converter.

The ocean waves induce an oscillating water surface height, which induces an os-
cillating air stream. A turbine is placed at the nozzle (with cross-section A1 ≪ A0 )
and extracts power from the moving air stream.
Oscillating height:

2π
h(t) = H sin ωt, ω = 2π f = , (1.15)
T
where f is the frequency, T is the oscillating period and ω is the angular frequency.

3
 dh(t)
u0 ( t ) = (1.16)
dt
= Hω cos ωt (1.17)

A0 u0 ( t ) = A1 u1 ( t ) (1.18)
⇓
A0
u1 ( t ) = u0 cos ωt (1.19)
A1

A1 ≪ A0 ⇒ u1 ( t ) ≫ u0 ( t ) (1.20)

1.1.4 Example 4: wake flow behind a wind turbine and wind farm optimization

Figure 5: The expanding wake behind a turbine.

Far-field modeling of a wake flow behind a wind turbine. We use the linear wake
expansion:
r = R + kx. (1.21)
We use the equation of continuity (Leonardo’s law):
 
∆m  ∆m 
 .
 = ρπR2 v(0+ ) + ρπ (r2 − R2 )u = ρπr2 v( x ) =  (1.22)
∆t x=0+ ∆t x

In words this equation states that the in-flow through the left side of the cylinder
is equal to the out-flow through the right side of the cylinder. Rearranging this
equation we can get an expression for the wind speed of the wake behind the

4
turbine:
R2 r 2 − R2 R2
v( x ) = v ( 0 + ) + u = u − (u − v(0+ )) (1.23)
r2 r2  r2

 v (0 ) 
1 − u+ 
= u 1− ( )2 . (1.24)

 

1 + kx
R

The ratio
v (0+ )
q= (1.25)
u
is called the axial induction factor.
Consistency checks:

lim v( x ) = v(0+ ) (1.26)

x → 0+

lim v( x ) = u (1.27)
x →∞

Betz theory: power produced by a wind turbine

Figure 6: The velocity deficit caused by the rotor disc.

A wind turbine extracts kinetic energy out of the wind flow:

m( 2 )
Eextracted = u − v 2 (0+ ) (1.28)
2
1 u + v (0+ )
= ρπR2 ∆t(u2 − v2 (0+ )) (1.29)
2 2

dEextracted
Pturbine = (1.30)
dt
{ }
ρπR2 u3 (1 + q) ( )
= 1−q 2
(1.31)
2 2

5
The term in front is the kinetic energy contained in the upstream wind (volume).
The term within the curly brackets is the efficiency of the wind turbine also called
the power coefficient:
(1 + q ) ( )
C p = C p (q) = 1 − q2 . (1.32)
2
The maximum efficiency of a turbine can be calculated by requiring
( )
dC p (q) d 1 !
= (1 + q)(1 − q ) = 0
2
(1.33)
dq dq 2
This gives the optimal q value
1
q= (1.34)
3
⇓
1
v ( 0+ ) = u. (1.35)
3
We can then calculate the power coefficient
1 16
max C p (q) = C p (q = )= ≈ 0.59. (1.36)
q 3 27
This is known as the Betz limit. Real turbines have about C p ≈ 0.40 − 0.50.

1.1.5 Power optimization of a two-turbine wind farm

Figure 7: A very simple wind farm consisting of two turbines. The wind is ap-
proaching from the left.

We look at a wind farm with two turbines and a wind direction aligned along the
connecting line. The total output of the two turbines is
ρπR2 ρπR2
P1+2 = C p1 ( q 1 ) u 3 + C p2 ( q 2 ) v 3 ( x ) (1.37)
2 2
There are no turbines behind turbine 2, so we configure it to extract the maximum
amount of energy from the wind
1 16
q2 = ⇒ C p2 ( q2 ) = (1.38)
3 27

6
Using previous expressions for C p (q) (1.33) and v( x ) (1.24) the total output of the
two turbines is
  3 

 
ρπR2 3 1 16  1 − q1  
P1+2 = u (1 + q1 )(1 − q1 ) +
2
1− ( ) (1.39)
2 
2 27 1 + kx 

R

Similar to before we find the optimal value of q1 by the requirement

dP1+2 !
= 0. (1.40)
dq1

We fix the values

k = 0.04
x
= 8.
R
The optimal q-value for turbine 1 is then

1
q1 = 0.58 > , (1.41)
3
so turbine 1 let’s through more wind.
1
Comparing this result with a q-value of 3 gives
( )
1
P1+2 (q1 = 0.58) = 1.07 · P1+2 q1 = , (1.42)
3

which is a 7% gain.

7
2 Derivation of Navier-Stokes equation

The goal of this section is to find an equation, which describes the spatio-temporal
evolution of the velocity field ⃗u(⃗r, t). This is the Navier-Stokes equation, which is
the most fundamental equation in Fluid Dynamics.

Figure 8: Velocity field around an object.

We want to describe the motion of a fluid particle. We start with Newton’s second
law of Classical Mechanics
⃗F = d (m⃗u) . (2.1)
dt
The equation states that the forces acting on the particle equals its change (the time
derivative) of momentum (the parenthesis).

Remark: in Fluid Dynamics we look not only at one fluid particle, but at all
fluid particles.

d dm d⃗u d⃗u
(m⃗u) = ⃗u + m = ρ∆V . (2.2)
dt dt dt dt
The mass of a fluid particle is constant and does not change over time, hence the
term dm/dt is zero. We also used the relation

m = ρ∆V. (2.3)
d⃗u
Given the field description ⃗u = ⃗u(⃗r, t), we have to be a little careful with dt . The
following is wrong:

d⃗u d⃗u(⃗r, t) ⃗u(⃗r, t + ∆t) − ⃗u(⃗r, t)

= = lim (2.4)
dt dt ∆t→0 ∆t
See the example in Figure 9.
The correct approach is to follow one fluid particle on its pathline (trajectory) ⃗r =
⃗r (⃗r0 , t0 ; t). See Figure 10.

8
Figure 9: Two fluid particles can have the same coordinate vector ⃗r but for different
times.

Figure 10: The path and changing coordinates of a single fluid particle.

d⃗u d⃗u (⃗r (⃗r0 , t0 ; t) , t)

= (2.5)
dt dt
d⃗u( x (⃗r0 , t0 ; t), y(⃗r0 , t0 ; t), z(⃗r0 , t0 ; t))
= (2.6)
dt
∂⃗u dx ∂⃗u dy ∂⃗u dz ∂⃗u dt
= + + + (2.7)
∂x dt ∂y dt ∂z dt ∂t dt
( )
∂ ∂ ∂ ∂⃗u
= u x + uy + uz ⃗u + (2.8)
∂x ∂y ∂z ∂t

Short notation for partial derivative:

∂
= ∂x . (2.9)
∂x
Here x can be replaced by y, z or t.
Short notation for velocity:
dx
= ux (2.10)
dt

( )
The terms in parentheses is the dot product between ⃗u and ∇
⃗ = ∂ x , ∂y , ∂z , calcu-

9
lated using the chain rule of differentiation:

du( f (t)) ∂u d f
= (2.11)
dt ∂ f dt
du( f (t), g(t)) ∂u d f ∂u dg
= + (2.12)
dt ∂ f dt ∂g dt
du( f (t), g(t), h(t)) ∂u d f ∂u dg ∂u dh
= + + . (2.13)
dt ∂ f dt ∂g dt ∂h dt

This leads to the final expression

d⃗u ∂⃗u ( ⃗ )
= + ⃗u · ∇ ⃗u. (2.14)
dt ∂t

The material derivative is defined as

D
∂t + ⃗u · ∇
⃗ = = Dt . (2.15)
Dt
Whenever we have the time derivative of a field, like ⃗u(⃗r, t), then we have to "go
with the fluid particle" and use the material derivative.

We now go back to Newton’s second equation:

d ( )
(m⃗u) = ρ∆V ∂t + ⃗u · ∇
⃗ ⃗u = ⃗F = ⃗Fexternal + ⃗Fsurrounding (2.16)
dt
The surrounding force can be decomposed into

⃗Fsurrounding = ⃗Fpressure + ⃗Ffriction (2.17)

The surrounding fluid particles push the "sandwiched" fluid particle around; they
exert pressure. Mutual friction between neighboring fluid particles due to relative
and rotational motion, deformation and compression.

Example: The force of gravity is an example of an external force:

⃗Fexternal = ⃗Fgrav = ρ∆V ⃗g = ρ⃗g∆V, (2.18)

|{z}
m

with the gravitational constant defined as

 
0
⃗g = −  0  = − g⃗ez . (2.19)
9.81 ms

10
 Figure 11: Illustration of the pressure force on each side of a fluid particle.

2.1 Pressure force

( )
top
⃗Fpressure = ⃗Fpressure bottom
+ ⃗Fpressure (2.20)
z
( ) ( )
∆z ∆z
= − p x, y, z + ∆x∆y + p x, y, z − ∆x∆y (2.21)
2 2
( )
∂p( x, y, z) ∆z
= − p( x, y, z) + ∆x∆y
∂z 2
( ( )) (2.22)
∂p( x, y, z) ∆z
+ p( x, y, z) + − ∆x∆y
∂z 2
∂p( x, y, z)
=− ∆x∆y∆z (2.23)
∂z
Using ∆x∆y∆z = ∆V the pressure force density is

⃗
⃗f pressure = Fpressure = −∇
⃗ p(⃗r, t) (2.24)
∆V

2.2 Friction force

Figure 12: Illustration of the friction force on each side of a fluid particle.

We consider the neighboring fluid particles above and below as sketched in Fig-

11
ure 12.
( )
⃗Ftop+bottom = µ ∆x∆y (u x ( x, y, z + ∆z) − u x ( x, y, z))
friction x ∆z (2.25)
µ
+ ∆x∆y (u x ( x, y, z − ∆z) − u x ( x, y, z))
∆z
If u x ( x, y, z + ∆z) > u x ( x, y, z), then the fluid particle above pulls the sandwiched
fluid particle with it.
Taylor series expansion up to second-order terms:
( ) { }
top + bot u x ( x, y, z + ∆z ) − u x ( x, y, z ) u x ( x, y, z − ∆z ) − u x ( x, y, z )
⃗F = µ∆x∆y +
friction x ∆z ∆z
{
µ∆x∆y ∂u x ( x, y, z) ∂ u x ( x, y, z) ∆z2
2
= u x ( x, y, z) + ∆z + − u x ( x, y, z)
∆z ∂z ∂z2 2
}
∂u x ( x, y, z) ∂2 u x ( x, y, z) (−∆z)2
+ u x ( x, y, z) + (−∆z) + − u x ( x, y, z)
∂z ∂z2 2
(2.26)
∂2 u x ( x, y, z)
= µ∆x∆y∆z (2.27)
∂z2
Front + back:
( )
⃗Ffront+back = µ∆V ∂ u x ( x, y, z)
2
(2.28)
friction x ∂y2
Most general expression of the friction force:
( )
⃗Ffriction ∂2 u i
= ⃗f friction = ⃗f friction (∇
⃗ ,∇
⃗ , ⃗u) (2.29)
∆V ∂xk ∂xl

The task is to build a vector ⃗f from a combination of three vectors ⃗a = ∇

⃗ ,⃗b =
∇, ⃗c = ⃗u, such that
⃗
( ) ( )
⃗f = α ⃗a · ⃗b ⃗c + β (⃗a ·⃗c)⃗b + γ ⃗b ·⃗c ⃗a (2.30)

The solution: ( ) ( ) ( )
⃗f friction = µ ∇ ⃗ ⃗u + µv + µ ∇
⃗ ·∇ ⃗ ∇⃗ · ⃗u , (2.31)
3
where µ is the shear (dynamic) viscosity and µv is the compression (bulk) viscosity.
Navier-Stokes equation:
( ( )) ( ) ( ) ( )
ρ ∂t + ⃗u · ∇
⃗ ⃗u = ⃗f ext − ∇
⃗ p+µ ∇ ⃗ ⃗u + µv + µ ∇
⃗ ·∇ ⃗ ∇⃗ · ⃗u (2.32)
3
where

⃗u = ⃗u (⃗r, t) (2.33)
p = p (⃗r, t) (2.34)
ρ = ρ (⃗r, t) (2.35)
⃗f ext = ⃗f ext (⃗r, t) (2.36)

12
2.3 Navier-Stokes equation in components

( ) 
 ∂u  ∂ ∂ ∂  ext   ∂p 
x u x ∂x + uy ∂y + uz ∂z ux
∂t ( )  fx ∂x
 ∂u y  ∂ ∂ ∂   ext   ∂p 
ρ  ∂t  + ρ  u x ∂x + uy ∂y + uz ∂z uy  = f y −  ∂y 
∂uz ( )  ∂p
∂t
∂
u x ∂x ∂
+ uy ∂y ∂
+ uz ∂z uz f zext
∂z
( 2 ) 
∂ ∂2 ∂2
+ + u
∂z ) x 
 ( ∂x22 ∂y 2 2
 ∂ ∂2 ∂2 
+ µ  ∂x2 + ∂y2 + ∂z2 uy 
( 2 ) 
∂ ∂2 ∂2
∂x2
+ ∂y2
+ ∂z2
u z
 ( )
∂ ∂u x ∂uy ∂uz
∂x ( ∂x + ∂y + ∂z ) 
( µ) 
 ∂ ∂ux ∂uy ∂uz 
+ µv +  + ∂y + ∂z 
3  ∂y ( ∂x )
∂ ∂u x ∂uy ∂uz
∂z ∂x + ∂y + ∂z
(2.37)

Remark: There are only a few exact analytical solutions; many approximate
analytical solutions (guided by intuition). Computational fluid dynamics can
give us "exact" numerical solutions for approximations to the Navier-Stokes
equation.

We now have three coupled differential equations for five fields: u x (⃗r, t), uy (⃗r, t),
uz (⃗r, t), p (⃗r, t), and ρ (⃗r, t). This means we are missing two equations.

The first missing equation

From thermodynamics we have an equation of state

g( p, ρ) = 0. (2.38)

For an incompressible flow, the equaiton of state is simply

ρ = constant. (2.39)

For a compressible flow, the equation of state can be found with the law of ideal
gases:
pV = NkT, (2.40)
from which we get
N 1
ρ= = p (2.41)
V kT
p
= kT = constant. (2.42)
ρ

This only holds if the temperature is constant.

13
The second missing equation

Equation of continuity, local mass conservation.

Figure 13: A simple pipe flow to illustrate mass conservation.

Example: Simple pipe flow. See Figure 13.

Min = ρS1 v1 ∆t (2.43)

Mout = ρS2 v2 ∆t (2.44)

Mass conservation:

Min = Mout (2.45)

⇓
v 1 S1 = v 2 S2 (2.46)

This is Leonardo’s law.

Local mass conservation in a small volume element:

Figure 14: Local mass conservation in a small volume element.

14
Mass flux through surface of volume ∆V = ∆x∆y∆z in x-direction:
( ) ( )
dMxS ∆x ∆x
= ρ x+ , y, z, t u x x + , y, z, t ∆y∆z
dt 2 2
( ) ( )
∆x ∆x
−ρ x− , y, z u x x − , y, z ∆y∆z
2 2
( )( )
∂ρ( x, y, z) ∆x ∂u x ( x, y, z) ∆x
= ρ( x, y, z) + u x ( x, y, z) + ∆y∆z
∂x 2 ∂x 2
[ ( )] [ ( )]
∂ρ( x, y, z) ∆x ∂u x ( x, y, z) ∆x
− ρ( x, y, z) + − u x ( x, y, z) + − ∆y∆z
∂x 2 ∂x 2
∂ρ( x, y, z) ∂u x ( x, y, z)
= u x ( x, y, z)∆x∆y∆z + ρ( x, y, z) ∆x∆y∆z
∂x ∂x
∂ (ρ( x, y, z, t)u x ( x, y, z, t))
= ∆V
∂x
(2.47)

Mass flux in y and z-direction:

dMyS ∂(ρ( x, y, z)uy ( x, y, z))

= ∆V (2.48)
dt ∂y
dMzS ∂(ρ( x, y, z)uz ( x, y, z))
= ∆V (2.49)
dt ∂z
Sum of mass fluxes through volume in all directions:
S
dMS dMxS dMy dMzS
= + + (2.50)
dt dt dt dt
∂(ρu x ) ∂(ρuy ) ∂(ρuz )
= ∆V + ∆V + ∆V (2.51)
∂x ∂y ∂z
   
∂
∂x ρu x
∂ 
=  ∂y  · ρuy  ∆V (2.52)
∂ ρuz
∂z
=∇
⃗ · (ρ( x, y, z)⃗u( x, y, z)) ∆V (2.53)

Increase of mass within fixed volume ∆V:

∂MV ∂(ρ(⃗r, t)∆V ) ∂ρ(⃗r, t)
= = ∆V (2.54)
∂t ∂t ∂t
Local mass conservation
dMV dMS
=− (2.55)
dt dt
If mass within the volume increases, then less has to flow out of the surface than to
flow in
∂ρ(⃗r, t) ⃗
+ ∇ · (ρ(⃗r, t)⃗u(⃗r, t)) = 0 (2.56)
∂t
This is the equation of continuity.

15
2.4 Summary

Navier-Stokes equation:
( ( )) ( ) (
∂ µ ) ⃗ (⃗ )
ρ + ⃗u · ∇ ⃗u = f ext − ∇ p + µ ∇ · ∇ ⃗u + µv +
⃗ ⃗ ⃗ ⃗ ⃗ ∇ ∇ · ⃗u (2.57)
∂t 3

⃗u = ⃗u(⃗r, t) = ⃗u( x, y, z, t) (2.58)

ρ = ρ(⃗r, t) (2.59)
p = p(⃗r, t) (2.60)

Equation of continuity:
∂ρ ⃗
+ ∇ · (ρ⃗u) = 0 (2.61)
∂t
Equation of state:
g( p, ρ; T ) = 0 (2.62)
Heat equation (if the temperature also becomes a field T (⃗r, t)):
( ( )) ( )
∂
+ ⃗u · ∇
⃗ T (⃗r, r ) = κ ∇⃗ ·∇
⃗ T (⃗r, t), (2.63)
∂t

where κ is the thermal diffusion.

16
3 Simplification of the Navier-Stokes equation

3.1 Simplification I: incompressible flows

Incompressiblity is a very good approximation for most liquids, including water.

In 1000 m depth the density of seawater is only 0.4% larger than at the surface.
For gas flows incompressiblity is also a good approximation as long as |⃗ugas | ≪
speed of sound. Compressibility becomes important when discussing e.g. sound
waves.
Incompressibility means that the density of a fluid particle (moving along its path-
line) remains constant.

dρ(⃗r, t) ∂ρ ( ⃗ )
0= = + ⃗u · ∇ ρ (3.1)
dt ∂t ( )
= −∇
⃗ (ρ⃗u) + ⃗u · ∇
⃗ ρ (3.2)
( ) ( ) ( )
= − ⃗u · ∇
⃗ ρ−ρ ∇ ⃗ · ⃗u + ⃗u · ∇
⃗ ρ (3.3)
( )
= −ρ ∇ ⃗ · ⃗u (3.4)
= −ρ div ⃗u (3.5)

In the first line we used the material derivative. In the step to the second line we
used continuity equation. For incompressibility the divergence must be zero:

∇
⃗ · ⃗u = 0 (3.6)

Navier-Stokes equation for incompressible flows (3 equations):

( )
∂⃗u ( ⃗ ) ( )
ρ + ⃗u · ∇ ⃗u = f ext − ∇
⃗ p+µ ∇ ⃗ ·∇
⃗ ⃗u (3.7)
∂t

Remark: It looks simple, but these nonlinear differential equations remain a

formidable challenge to engineers, physicists and mathematicians.

Fourth equation:
∇
⃗ · ⃗u = 0. (3.8)
Fifth equation: equation of state in the simplest form with constant density

p = p(ρ) ⇒ ρ = ρ0 = constant. (3.9)

3.2 Simplification II: incompressible, ideal, stationary, irrotational flows

We use the incompressibility result:

∇
⃗ · ⃗u = 0 (3.10)

17
Ideal means no friction. To eliminate friction forces we set µ = 0.
Euler equation: ( )
∂⃗u ( ⃗ )
ρ0 + ⃗u · ∇ ⃗u = ⃗f ext − ∇
⃗p (3.11)
∂t
stationary:

⃗u(⃗r, t) = ⃗u(⃗r ) (3.12)

⇓
∂⃗u
=0 (3.13)
∂t
( )
⃗ ⃗u = ⃗f ext − ∇
ρ0 ⃗u · ∇ ⃗p (3.14)

no external forces: ⃗f ext = 0

( )
ρ0 ⃗u · ∇
⃗ ⃗u = −∇
⃗p (3.15)

We now look at the convective term on the lefthand side (see the proof below):
( ) ( )
⃗ ⃗u = 1 ∇
⃗u · ∇ ⃗ (⃗u · ⃗u) −⃗u × ∇⃗ × ⃗u (3.16)
2 | {z }
⃗u2
⇓
(ρ ) ( )
0
∇
⃗ ⃗u2 + p = ρ0⃗u × ∇⃗ × ⃗u (3.17)
2

Assuming irrotational flow: ∇

⃗ × ⃗u = 0.
( )
⃗ ρ0 ⃗u2 + p = 0
∇ (3.18)
| 2 {z }
constant

Bernoulli’s equation
ρ0 2
⃗u + p = constant (3.19)
2
∇
⃗ · ⃗u = 0 (3.20)
∇
⃗ × ⃗u = 0 (3.21)

Given all the assumptions, this set of equations is equivalent to the Navier-Stokes
equation.

Proof of ( ) ( )
( )
⃗u · ∇⃗ ⃗u = 1 ∇
⃗ ⃗u2 − ⃗u × ∇ ⃗ × ⃗u . (3.22)
2
First we look at the x-component of the left-hand side:
[( ) ] ( )
⃗u · ∇
⃗ ⃗u = u x ∂ x + uy ∂y + uz ∂z u x (3.23)
x

18
 Now we look at the rightmost term on the right-hand side:
 
⃗ex ⃗ey ⃗ez 
  ( ) ( )
⃗ × ⃗u =  ∂ x ∂y ∂z  = ∂y uz − ∂z uy ⃗ex + (∂z u x − ∂ x uz )⃗ey + ∂ x uy − ∂y u x ⃗ez
∇  
u u u 
x y z
(3.24)
Now we can show that the x-component of the right-hand side is equal to the
x-component of the left-hand side:
[ ( )]
1 ⃗ ( 2) 1 ( )
∇ ⃗u − ⃗u × ∇ × ⃗u
⃗ = ∂ x u2x + u2y + u2z
2 x 2
 
 ⃗ex ⃗ey ⃗ez 
 
−  ux uy uz 

∂ u − ∂ u ∂ u − ∂ u ∂ u − ∂ u 
y z z y z x x z x y y x x
(3.25)
( )
= u x (∂ x u x ) + uy ∂ x uy + uz (∂ x uz )
( ) (3.26)
− uy ∂ x uy − ∂y u x + uz (∂z u x − ∂ x uz )
( )
= u x (∂ x u x ) + uy ∂y uy + uz (∂z uz ) (3.27)
[( ) ]
= ⃗u · ∇ ⃗ ⃗u (3.28)
x

3.3 Derivation of Bernoulli’s equation

The equation ( ) ( )
⃗ ρ0 ⃗u2 + p = ρ0⃗u × ∇
∇ ⃗ × ⃗u (3.29)
2
is (scalar) multiplied with d⃗s ∥ ⃗u, where d⃗s describes an increment of a specific
streamline (here pathline since ⃗u(⃗r, t) = ⃗u(⃗r )).
[ ( )]
d⃗s · ⃗u × ∇ ⃗ × ⃗u = 0 (3.30)
( )
Since d⃗s ∥ ⃗u it must be that d⃗s ⊥ ⃗u × ∇⃗ × ⃗u .

19
Subsequent path-integration along a streamline yields
∫ (ρ )
! 0 2
0= ∇
⃗ ⃗u + p ·d⃗s (3.31)
streamline | 2 {z }
W
 ∂W   
∫ ∂x dx
 ∂W   
=  ∂y  · dy (3.32)
streamline ∂W
∂y dz
∫ ( )
∂W ∂W ∂W
= dx + dy + dz (3.33)
streamline ∂x ∂y ∂z
∫ ∫ (ρ )
0 2
= dW = d ⃗u + p (3.34)
streamline streamline 2
⇓
ρ0 2
⃗u + p = constant (along a streamline) (3.35)
2
For another streamline the constant might in principle be different. Often the veloc-
ity in the far-field regime (away from the obstacle) is everywhere the same. The
same holds true for the pressure. Then the "Bernoulli constant" has to be ev-
erywhere (far and near-field) the same. From here we then also conclude that
∇
⃗ × ⃗u = 0 everywhere.

3.4 Example: Why does an airplane fly?

Figure 15: The profile of the wing of an airplane.

According to the Bernoulli’s equation, the wind speed difference between the top
and bottom of the wing creates a pressure difference:

utop > ubottom (3.36)

⇓
ptop < pbottom . (3.37)

This results in a lifting force.

20
4 Ideal flow: planar 2-dimensional potential flow around
cylinder

For further details see sections 4.3, 4.9 and 7.1-6 in the KCD book

For a planar two-dimensional stationary flow the velocity field

 
u x ( x, y)
⃗u =  uy ( x, y)  (4.1)
0

does not depend on z and t and the vector does not have a z-component. The flow
is defined to be "ideal" once the viscosity µ = 0 is put to zero. The mass density
ρ = ρ0 is assumed to be constant. We will determine the two velocity components
u x ( x, y) and uy ( x, y) from the two equations ∇
⃗ · ⃗u = 0 and ∇
⃗ × ⃗u = 0 defining
incompressible and irrotational flows. The pressure field p( x, y) is then determined
via Bernoulli’s equation.

Figure 16: 2-dimensional ideal flow around a cylinder.

Questions to Figure 16:

1. How does the velocity field ⃗u = ⃗u( x, y) look like?

2. How do the pathlines (streamlines) look like?

∇
⃗ × ⃗u = 0 ⇒ ⃗u(⃗r ) = ∇
⃗ ϕ(⃗r ) (4.2)
where ϕ(⃗r ) is the velocity potential.

∇
⃗ · ⃗u = 0 (4.3)
⇓
( )
∂2 ∂2
∇
⃗ ·∇
⃗ ϕ(⃗r ) = + ϕ( x, y) = 0 (4.4)
∂x2 ∂y2

This second order differential equation is the Laplace equation.

Question: How does the cylinder (obstacle) enter in solving the Laplace equation?

21
There are two boundary conditions.
First boudary condition:

⃗u(|⃗r | → ∞) = ⃗u∞ = u∞⃗ex (4.5)

⇓
ϕ(|⃗r | → ∞) = u∞ x + constant. (4.6)

Second boundary condition:



0 = ⃗usurface · ⃗n = ∇ϕ 
⃗ ·⃗n (4.7)
surface

The fluid particle does not flow into/out of the surface; only tangential component.

Figure 17: Tangential components of the velocity field.

For the flow around the cylinder the solution of the Laplace equation with the two
boundary conditions "falls from the sky" (for the moment):
( )
R2
ϕ( x, y) = u∞ x 1 + 2 . (4.8)
x + y2

It fulfills Laplace’s equation and the two boundary conditions.

Velocity field: ( ∂ϕ )
( )
ux ∂x
⃗u = =∇
⃗ ϕ( x, y) = ∂ϕ (4.9)
uy ∂y

( )
R2 ( y2 − x 2 )
u x = u∞ 1+
( x 2 + y2 )2
(4.10)
2xyR2
uy = −u∞ 2
( x + y2 )2

22
 Figure 18: Streamlines around the cylinder.

Examples:
 
 
⃗u  = u∞⃗ex = ⃗u 
 (4.11)
x →±∞ y→±∞


⃗u  ( ) = 2u∞⃗ex
0
(4.12)
⃗r =
R
 
 
⃗u  ( )
R
= 0 = ⃗u  (
−R
) (4.13)
⃗r = ⃗r =
0 0

The two points in (4.13) with ⃗u = 0 are called stagnation points.

4.1 Stream line around a cylinder

Figure 19: Path of fluid particles around the cylinder.

23
Top part of Figure 19 shows the Lagrangian picture: a particle is followed through
all times. The bottom part shows the Eulerian picture, which is a snapshot of all
fluid particles at one particular time.
Connection between Lagrangian and Eulerian picture:

d⃗r
= ⃗u(⃗r, t) (4.14)
dt
Given the snapshots ⃗u(⃗r, t), we can calculate the pathlines. Given the pathlines, we
can construct the snapshots. For stationary flows

⃗u(⃗r, t) = ⃗u(⃗r ) ⇒ pathline = streamline. (4.15)

Question: How to calculate the streamlines?

First approach: Definition of streamline:

d⃗s ∥ ⃗u, (4.16)

⃗ is a line element of a streamline.

where ds
 
 0 0 ⃗ez 
 
0 = d⃗s × ⃗u = dx dy 0  = (uy dx − u x dy)⃗ez (4.17)
u u 0
x y

⇓
dy uy 2xyR2
= =− 2 . (4.18)
dx ux ( x + y ) + R2 ( y2 − x 2 )
2 2

We will not try to solve this ugly non-linear differential equation.

Second approach: introduce the streamfunction ψ( x, y).
Incompressibility gives us:

⃗ · ⃗u = ∂u x + ∂uy = 0
∇ (4.19)
∂x ∂y

which leads to the ansatz

∂ψ ∂ψ
ux = , uy = − (4.20)
∂y ∂x

d⃗s × ⃗u = (uy dx − u x dy)⃗ex (4.21)

( )
∂ψ ∂ψ
= − dx − dy ⃗ez (4.22)
∂x ∂y
!
= −dψ⃗ez = 0 (4.23)

The streamfunction is constant along a streamline. This represents an isopotential

line of the streamfunction and describes a streamline.

24
From the defining functions of the streamfunction in (4.20) and the u x , uy solu-
tion for the ideal flow around a cylinder in (4.10), we can determine ψ by partial
integration
( )
R2
ψ( x, y) = v∞ y 1 − 2 (4.24)
x + y2
( )
R2
= v∞ sin ϕ r − (4.25)
r
In the last step we have introduced the cylindrical coordinates x = r cos ϕ and
y = r sin ϕ. The intermediate steps of the partial integration have been left out.

Remark: relationship between velocity potential and streamfunction

ϕ = constant, ψ = constant represent an orthogonal set of curves, because
( ∂ϕ ) ( ∂ψ ) ( ) ( )
( ) ( ) ux −uy
∂x ∂x
∇ϕ · ∇ψ = ∂ϕ · ∂ψ =
⃗ ⃗ · =0 (4.26)
∂y ∂y
uy ux

Back to the derivation of (4.8): two dimensional potential flow around an infinitely
long cylinder. Because of cylinder symmetry we can transform from Cartesian to
cylindrical coordinates

x = r cos ϕ (4.27)
y = r sin ϕ (4.28)

Φ( x, y) → Φ(r, ϕ) (4.29)
()
∂2 ∂2
∆Φ( x, y) = + 2 Φ( x, y) (4.30)
∂x2 ∂y
[ ( ) ]
1 ∂ ∂ 1 ∂
= r + 2 Φ(r, ϕ) (4.31)
r ∂r ∂r r ∂ϕ
=0 (4.32)
⇓
( )
∂ ∂Φ ∂2 Φ
r r =− (4.33)
∂r ∂r ∂ϕ2
Ansatz: factorization
Φ(r, ϕ) = f (r ) g(ϕ) (4.34)
( )
1 ∂ ∂( f (r ) g(ϕ)) 1 ∂2 ( f (r ) g(ϕ))
r r =− (4.35)
f · g ∂r ∂r f ·g ∂ϕ2
⇓
( )
1 ∂ ∂ f (r ) 1 ∂2 g ( ϕ ) ! 2
r r =− =m (4.36)
f (r ) ∂r ∂r g(ϕ) ∂ϕ2

25
Left part depends only on r, and the middle part depends only on ϕ. As a conse-
quence, both have to be equal to a constant, which does neither depend on r nor
ϕ.
∂2 g ( ϕ )
= − m2 g ( ϕ ) (4.37)
∂ϕ2
⇓
g(ϕ) = e imϕ
= cos mϕ + i sin mϕ (4.38)

Requirement:

g(ϕ) = g(ϕ + 2π ) (4.39)

⇓
eimϕ = eim(ϕ+2m) (4.40)
⇓
e2πim = 1. (4.41)

This fixes m to integer values:

m = . . . , −2, −1, 0, 1, 2, . . . (4.42)

( )
∂ ∂
r r f (r ) = m2 f (r ) (4.43)
∂r ∂r
Polynomial ansatz:
f (r ) = r α (4.44)

∂ ( α −1 ) !
r rαr = α2 r α = m2 r α (4.45)
∂r
⇓
α = ±m (4.46)

Φ(r, ϕ) = f (r ) g(ϕ) = r ±m eimϕ (4.47)

Since the Laplace equation is linear in Φ, the most general solution for Φ is a linear
superposition of all possible solutions:
∞ ( )
Φ(r, ϕ) = ∑ am r m + bm r −m eimϕ (4.48)
m=−∞
∞ [( ) ( ) ]
= ∑ am r m + bm r −m eimϕ + cm r m + dm r −m e−imϕ (4.49)
m =1

Remark:

Φm=0 = a0 + b0 = constant (4.50)

⇓
⃗um=0 = ∇
⃗ Φ m =0 = 0 (4.51)

26
Remark: Another m = 0 solution is ϕm=0 = c ln r. It fulfills (4.43). However, it is
not able to fulfill the boundary condition at r = R, and has to be discarded.
Determination of amplitudes am , bm , cm , and dm via boundary conditions:

Φ(r → ∞, ϕ) = u∞ x = u∞ r cos ϕ (4.52)

∞ ( )
Φ(r → ∞, ϕ) = ∑ am r m eimϕ + cm r m e−imϕ (4.53)
m =1
!
= u∞ r cos ϕ (4.54)
eiϕ + e−iϕ
= u∞ r (4.55)
2

a2 = a3 = · · · = c2 = c3 = · · · = 0 (4.56)

u∞
a1 =
= c1 (4.57)
2
∞ ( )
eiϕ + e−iϕ bm imϕ dm −iϕ
Φ(r, ϕ) = u∞ r + ∑ e + e (4.58)
2 m =1
rm rm

∂Φ 
⃗u ·⃗er |r= R = ∇Φ ·⃗er |r= R =
⃗ =0 (4.59)
∂r r= R

  ( )
∂Φ(r, ϕ)  eiϕ + e−iϕ ∞
(−m) 

∑ bm eimϕ + dm e−imϕ = 0
!
 = u ∞ +  (4.60)
∂r r=R 2 m =1
r m +1 r=R

b2 = b3 = · · · = d2 = d3 = · · · = 0 (4.61)

u∞ b u∞ d
− 12 = 0 = − 12 (4.62)
2 R 2 R
⇓
u ∞ R2
b1 = d1 = (4.63)
2

eiϕ + e−iϕ u∞ R2 eiϕ + e−iϕ

Φ(r, ϕ) = u∞ r + (4.64)
2 2 2
where
eiϕ + e−iϕ
= cos ϕ (4.65)
2
( ) ( )
R2 R2
Φ(r, ϕ) = u∞ r cos ϕ 1 + 2 = u∞ x 1 + 2 = Φ( x, y) (4.66)
r x + y2

27
5 More on ideal potential flows

Opening remark: 2-dimensional potential flow solutions will often look like

Φ( x, y) = u∞ x + f ( x, y). (5.1)

Any function f ( x, y), which fulfills

∂2 f ∂2 f
+ =0 and f (| x |, |y| → ∞) = 0, (5.2)
∂x2 ∂y2

describes a flow around some obstacle. The question is: which obstacle? Let’s
play around with f ( x, y).

Example 1: √
m
Φ( x, y) = ln x2 + y2 (5.3)
2π
represents the radial flow resulting from a source with strength m. See Figure 20.

Figure 20: Two-dimensional radial flow resulting from a source line along the z-
axis.

∂Φ m x m cos ϕ
ux = = = (5.4)
∂x 2
2π x + y 2 2π r
∂Φ m y m sin ϕ
uy = = = (5.5)
∂y 2
2π x + y 2 2π r

Example 2 (method of images): Source flow in front of a wall. See Figure 21.
Boundary condition: no flow through the wall; only tangential component.
√ √
m m
ϕ( x, y) = ln ( x + a)2 + y2 + ln ( x − a)2 + y2 (5.6)
2π 2π

28
 Figure 21: Source flow in front of a wall.

Example 3: flow past a 2-dimensional half-body. See Figure 22.

√
m
Φ = u∞ x + ln x2 + y2 (5.7)
2π
⇓
∂2 Φ ∂2 Φ
+ 2 =0 (5.8)
∂x2 ∂y

m x
u x ( x, y) = u∞ + (5.9)
2π x + y2
2
m y
uy ( x, y) = u∞ + (5.10)
2π x + y2
2

Engineering flow interpretations:

1. An example of the beginning of the half-body is the leading edge of an airfoil

2. pedestrian on a bridge looking down: front part of a bridge pier

3. flow over a cliff

Example 4 ("beauty of mathematics"): Conformal mappings.

Complex potential
w(z) = ϕ( x, y) + iψ( x, y) (5.11)
where z = x + iy and i2 = −1.
Velocity:

dw(z) dw(z)  ∂ϕ ∂ψ
= = +i = u x − iuy (5.12)
dz dz dz=dx ∂x ∂x

dw(z)  ∂ϕ ∂ψ ∂ψ ∂ϕ
= = +i = −i = u x − iuy (5.13)
dz  i∂y
dz=idy i∂y ∂y ∂y

29
 Figure 22: Flow past 2-dimensional half-body.

Example 4.1:
w(z) = u∞ z = u∞ ( x + iy) = u∞ x + iu∞ y (5.14)
describes the constant flow ⃗u = u∞⃗ex .
Example 4.2:
m m
w(z) = ln z = ln( x + iy) (5.15)
2π ( 2π )
m
= ln reiθ (5.16)
2π
m m
= ln r + ln eiθ (5.17)
2π √ 2π
m m
= ln x2 + y2 + i θ (5.18)
2π 2π
The two terms in the last line are the velocity potential and the stream function of
a radial source flow (see "Example 1").
Example 4.3:

A 2 A
w(z) = z = ( x + iy)2 (5.19)
2 2
A 2
= ( x − y2 ) + iAxy (5.20)
2

30
 Figure 23: Stream line around a corner

Example 4.4: flow around cylinder with radius R

( ) ( )
R2 R2
w(z) = ϕ( x, y) + iψ( x, y) = u∞ x 1 + 2 + iu∞ y 1 − 2 (5.21)
x + y2 x + y2
x − iy
= u∞ ( x + iy) + u∞ R2 2 (5.22)
x + y2
u ∞ R2
= u∞ ( x + iy) + (5.23)
x + iy
( )
R2
= u∞ z + (5.24)
z

Example 4.5:
Change of variable (z → z̃):
z = z(z̃) (5.25)

1
z̃ = (z + z0 ) + (5.26)
z + z0
⇓
wnew obstacle (z̃) = wcylinder (z(z̃)) (5.27)

31
Figure 24: Several examples for different z0 and R. e) is known as Joukowski’s
airfoil.

32
6 Forces on a 2-dimensional body

With the Bernoulli equation,

( )
⃗ ρ ⃗u2 + p = 0.
∇ (6.1)
2
we can calculate the force on the "obstacle" (Figure 25) from the surrounding flow:
∫ ∫ ( )
⃗ =− ρ
⃗F = (− p(⃗r ))d A p0 − ⃗u2 (⃗r ) d A
⃗ (6.2)
S
∫ S 2
ρ ⃗
= ⃗u2 (⃗r )d A (6.3)
2 S
= L⃗ey + D⃗ex . (6.4)
|{z}
=0

⃗L is the lift force and D

⃗ is the drag force. The last term equals zero because there is
no friction in ideal flows.

Figure 25: Lift and drag forces.

Figure 26: Angle of attack.

Lift coefficient
L
CL = ρ 2 (6.5)
2 v∞ bc
Drag coefficient (for the case that friction is non-zero):
D
CD = ρ 2 (6.6)
2 v∞ bc

6.1 Turbine blade

The lift force pulls the rotor blade of a wind turbine forward. See Figure 28.

33
 Figure 27: Generic lift and drag coefficients vs. angle of attack.

Figure 28: Lift force on the rotor blade of a wind turbine.

6.2 Sailing against the wind (KCD 14.9)

People have sailed without the aid of an engine for thousands of years and have
known how to reach an upwind destination. Actually, it is not possible to sail
exactly against the wind, but it is possible to sail at ≈ 40-45◦ to the wind. Figure 29
shows how this is made possible by the aerodynamic lift on the sail, which is a
piece of stretched and stiffened cloth. The wind speed is U, and the sailing speed is
V, so that the apparent wind speed relative to the boat is Ur . If the sail is properly
oriented, this gives rise to a lift force perpendicular to Ur and a drag force parallel
to Ur . The resultant force F can be resolved into a driving component (thrust) along
the motion of the boat and a lateral component. The driving component performs
work in moving the boat; most of this work goes into overcoming the frictional drag
and in generating the gravity waves that radiate outward from the hull. The lateral
component does not cause much sideways drift because of the shape of the hull.
It is clear that the thrust decreases as the angle θ decreases and normally vanishes
when θ is ≈ 40-45◦ . The energy for sailing comes from the wind field, which loses
kinetic energy after passing the sail.

34
Figure 29: Principle of sailing against the wind. A small component of the sails lift
pushes the boat forward at an angle θ < 90◦ to the wind. Thus by traversing a zig-
zag course at angles ±θ, a sailboat can reach an upwind destination. A sailboats
keel may make a contribution to its upwind progress too.

35
7 Reynolds number

Fluid around a cylinder can create several real flow patterns. See Figure 30.
Questions:

1. Why so many different flows?

2. What causes and characterizes them?

Certainly friction, i.e. viscosity, will have something to do with it.

[ ]
∂⃗u ( ⃗ ) ( )
0 = ρ0 + ⃗u · ∇ ⃗u + ∇ ⃗ p−µ ∇ ⃗ ·∇⃗ ⃗u (7.1)
∂t
[ ]
U ∂⃗u′ U 2 ( ′ ⃗ ′ ) ′ ρ0 U 2 ⃗ ′ ′ U ( ⃗ ⃗ )′ ′
= ρ0 + u
⃗ · ∇ u
⃗ + ∇ p − µ ∇ · ∇ ⃗u (7.2)
T ∂t′ L L L2
{ }
U 2 ∂⃗u′ ( ′ ⃗ ′ ) ′ ⃗ ′ ′ µ ( ⃗ ⃗ )′ ′
= ρ0 + ⃗u · ∇ ⃗u + ∇ p − ∇ · ∇ ⃗u (7.3)
L ∂t′ ρ0 LV

where L is the characteristic length, U is the characteristic velocity, T = L/U is the

characteristic time, and P = ρ0 U 2 the characteristic pressure.
Reynolds number:
ρ0 LU
Re = (7.4)
µ

Remark: law of similarity

If two flows have the same geometry (object) and the same Reynolds number,
but a different absolute scale it means that the two flows are similar (identical,
except for a scale transformation). Applications of this is wind tunnel experi-
ments of air wings, wind turbine blades, cars, etc.

The Reynolds number

ρ0 LU ρ0 U 2 /L
Re = = , (7.5)
µ µU/L
is the inertia force density divided by the friction force density.

36
Figure 30: Flow around a cylinder. From laminar to turbulent with increasing
velocity.

37
Figure 31: Extreme examples of Reynolds number.

38
Figure 32: Schematic real flow patterns around a cylinder with different Reynolds
numbers.

39
8 Viscous pipe flow

Figure 33: Ideal and viscous pipe flow.

Task: calculate velocity profile uz = uz (r ) for the viscous pipe flow.

Navier-Stokes equation
∂⃗u ( ) ( )
ρ0 +ρ0 ⃗u · ∇
⃗ ⃗u = ⃗f ext −∇
⃗ p+µ ∇⃗ ·∇
⃗ ⃗u (8.1)
∂t
|{z} | {z } |{z}
=0 =0
=0
The second term on the left-hand side vanishes because of the incompressibility
condition
0=∇ ⃗ · ⃗u = ∂ x u x + ∂y uy + ∂z uz = ∂z uz (8.2)
with u x = uy = 0 and
 
( ) 0
⃗ ⃗u = uz ∂z  0  = 0.
⃗u · ∇ (8.3)
uz
( )
∂2 ∂2
∇
⃗ p=µ + uz⃗ez (8.4)
∂x2 ∂y2
⇓
( ) ( )
∇
⃗p = ∇
⃗ p =0 (8.5)
x y

⇓
p = p(z) (8.6)

( ) ( )
µ ∂ ∂uz (r )
µ ∂2x + ∂2y uz (r ) = r (8.7)
r ∂r ∂r
∂p(z) !
= = constant (8.8)
∂z

40
 ∂p(z)
=c (8.9)
∂z
⇓
p(z) = cz + d (8.10)
p ( z = L ) − p ( z = 0)
= z + p ( z = 0) (8.11)
L
∆p
= − z + p ( z = 0) (8.12)
L

( )
µ ∂ ∂uz (r ) ∆p
r =c=− (8.13)
r ∂r ∂r L
⇓
∂uz (r ) ∆p r2
r =− + D1 (8.14)
∂r µL 2
⇓
∆p 2
u z (r ) = − r + D1 ln r + D2 (8.15)
4µL
(8.16)

D1 and D2 are determined from the boudnary conditions uz (r = R) = 0 (no slip

condition) and |uz (r = 0)| < ∞ (finiteness).

∆p ( 2 )
u z (r ) = R − r2 (8.17)
4µL

Fluid mass per time passing through pipe cross-section:

∫ R
dM πρ0 R4 ∆p
= ρ0 uz (r )2πrdr = (8.18)
dt 0 8µL

This is the Hagen-Poiseuille law.

Remark: this law allows to determine the viscosity:

 

 

 dM 
ρ0 , R, L , ∆p, ⇒µ (8.19)

 | {z } | {zdt}

 known 
measured

41
Remark: "Ohm’s Law"

dM
∆p = ∆U , =I (8.20)
dt
⇓
πρ0 R4 ∆U
I= ∆U = (8.21)
8Lµ R0
8Lµ
where R0 = = pipe resistance (8.22)
πρ0 R4
Reynolds number in Equation 7.4 (8.23)
ρ0 2Rūz ρ0 2R dM/dt 2
Re = = = I (8.24)
µ µ ρ0 πR2 πµR
(8.25)

R = R0 · f ( Re) (8.26)
with
f ( Re → 0) = 1. (8.27)
Turbulence occurs when the velocity becomes large; it increases the pipe resis-
tance. This is important for the operation of oil and gas pipelines.

42
9 Boundary layers

Figure 34: Example of boundary layer for flow around a cylinder.

Idea of boundary layer theory:

1. within the boundary layer the velocity increases from zero to the ideal flow
velocity

2. inside the boundary layer we use the Navier-Stokes equation (with friction)

3. outside the boundary layer we use the Euler equation without friction i.e.
ideal potential flow

4. at the boundary surface we match the inside solution with the outside solution

Derivation of the (laminar) boundary layer equations. Approximation to the Navier-

Stokes equation inside the boundary layer. See Figure 35.

Figure 35: Laminar boundary layer.

In the following we determine δ = δ( x ) (Figure 35) without solving the Navier-

Stokes equation, and derive Prandtl’s laminar boundary layer equations.

43
Incompressible flow:

⃗ · ⃗u = ∂u x + ∂uy = 0
∇ (9.1)
∂x ∂y
⇓
(u ) (u )
∞ y
O +O =0 (9.2)
L δ
⇓
u∞ δ
O(uy ) = δ = u∞ (9.3)
L L

Navier-Stokes equation (x-component):

∂u x ∂u x 1 ∂p µ ∂2 u x µ ∂2 u x
ux + uy =− + + (9.4)
∂x ∂y ρ0 ∂x ρ0 ∂x2 ρ0 ∂y2

The two terms on the left-hand

( )side and the first term on the right-hand side are of
u2∞
the order of magnitude O L . Compared to the last term of the right-hand side

( by ()δ/L) and can be neglected. The last term is of

the second term is suppressed 2
µ
the order of magnitude of O ρ0 uδ∞2 .
( ) ( )
u2∞ ! µ u∞
O =O (9.5)
L ρ0 δ2

( )2
δ µ 1
∼ = (9.6)
L ρ0 Lu∞ Re
The larger the Reynolds number, the thinner the boundary layer. This holds for
Re ≤ 1 × 105 − 1 × 106 , above that the boundary layer becomes turbulent, and is no
longer laminar.
√
µx
δ( x ) ∼ (9.7)
ρ0 u ∞

Navier-Stokes equation (y-component):

∂uy ∂uy 1 ∂p µ ∂2 u y µ ∂2 u y
ux + uy =− + + (9.8)
∂x ∂y ρ0 ∂y ρ0 ∂x2 ρ0 ∂y2

The first term on the right-hand side is the only large term. All other terms are
suppressed and can be neglected. This leads to

∂p
= 0 ⇒ p = p( x ) (9.9)
∂y

44
 Prandtl equations: laminar boundary layer equations

∂u x ∂u x 1 ∂p( x ) µ ∂2 u x
ux + uy =− + (9.10)
∂x ∂y ρ0 ∂x ρ0 ∂y2
∂u x ∂uy
+ =0 (9.11)
∂x ∂y

9.1 Example

Solution of Prandtl equations for laminar boundary flow above semi-infinite plate

p = p( x ) ⇒ p( x )|inside = p( x )|outside (9.12)

Outside boundary layer:

u x |outside = u x = u∞ = constant (9.13)
uy |outside = 0 (9.14)

Bernoulli equation:
ρ0 2
p( x ) + u = constant (9.15)
2 x
⇓
p( x ) = constant (9.16)
⇓
∂p( x )
=0 (9.17)
∂x
Prandtl equations inside boundary layer:
µ 2
u x ∂ x u x + uy ∂y uy = ∂ ux (9.18)
ρ0 y
∂ x u x + ∂y uy = 0 (9.19)

Solution: similarity ansatz.

( )
y
u x ( x, y) = u∞ g (9.20)
δ( x )
Except for a rescaling with δ( x ) the velocity u x ( x, y) looks like the same for all x.
Question: does the similarity ansatz work?
∂u x ∂uy
+ =0 (9.21)
∂x ∂y
⇓
∂ψ ∂ψ
ux = , uy = − (9.22)
∂y ∂x

45
Figure 36: Similarity ansatz for the laminar boundary layer.

( )
y
ψ( x, y) = u∞ δ( x ) f (9.23)
δ( x )

∂ψ d f (z) dz
ux = = u∞ δ( x ) (9.24)
∂y dz dy
d f (z) 1
= u∞ δ( x ) (9.25)
dz δ( x )
d f (z)
= u∞ (9.26)
dz
!
= u∞ g(z) (9.27)

d f (z)
f′ = = g(z) (9.28)
dz

∂ψ
uy = − (9.29)
∂x
dδ( x ) d f (z) dz
= −u∞ f (z) − u∞ δ( x ) (9.30)
[ dx ] dz dx
y f ′ dδ( x )
= u∞ − f + (9.31)
δ dx

( )
∂u x ( ′′
) −y dδ( x ) u∞ y f ′′ dδ( x )
= u∞ f = − (9.32)
∂x δ2 dx δ2 dx
∂u x 1 u∞ f ′′
= (u∞ f ′′ ) = (9.33)
∂y δ δ
∂2 u x u∞ f ′′′
= (9.34)
∂y2 δ2

46
 ( )
∂u x ∂u x µ ∂2 u x ′ u∞ y f ′′ dδ( x )
ux + uy − = −( u ∞ f )
∂x ∂y ρ0 ∂y2 δ2 dx
( [ ′ ] )( )
y f dδ( x ) u∞ f ′′
+ u∞ − f + (9.35)
δ dx δ
( ′′ )
µ u∞ f
−
ρ0 δ2
u2∞ dδ ′′ µ u∞ ′′′
=− ff − f (9.36)
δ dx ρ0 δ2
=0 (9.37)

ρ0 u ∞ dδ( x ) f ′′′ (z) !

δ( x ) =− ′′
= c21 (9.38)
µ dx f (z) f (z)

dδ 1 dδ2 µ
δ = = c21 (9.39)
dx 2 dx ρ0 u ∞

µ
δ2 = 2c21 x + c2 (9.40)
ρ0 u ∞
c2 = 0 since δ( x = 0) = 0.
√
2µ
δ ( x ) = c1 x (9.41)
ρ0 u ∞
√
µ
δ( x ) = x (9.42)
ρ0 u ∞
Freedom of choice c1 = √1 because of arbitrary definition of δ; for example, u x (y =
2
δ) = 0.99u∞ or u x ( x = δ) = 0.95u∞ .
This is the same result as the order of magnitude calculation when we calculated
the x-component of the Navier-stokes equation earlier in this section.

1
f ′′′ (z) + f (z) f ′′ (z) = 0 (9.43)
2

d2 f ( z ) d3 f ( z )
f (z) + 2 =0 (9.44)
dz2 dz3
This is Blasius’ equation. It is a special case of the more general Falkner-Skan
equation. The Blasius equation can only be solved numerically. The boundary
conditions for the solutions are:
u x (y = 0) = 0 ⇒ f ′ (0) = 0, (9.45)
uy (y = 0) = 0 ⇒ f (0) = 0, (9.46)
′
u x (y = ∞) = u∞ ⇒ f (∞) = 1. (9.47)
The solution is sketched in Figure 37.

47
 Figure 37: Solution to the Blasius equation.

9.2 Separation of boundary layers

Boundary condition at the wall:

u x ( x, y = 0) = uy ( x, y = 0) = 0 (9.48)

Prandtl equation (with pressure)

∂u x ∂u x 1 ∂p µ ∂2 u x
ux + uy =− + (9.49)
∂x ∂y ρ0 ∂x ρ0 ∂y2

If we are very close to the wall, the two terms on the left-hand side are equal zero.
We then have
∂p( x, y = 0) ∂2 u x ( x, y = 0)
=µ (9.50)
∂x ∂y2

Figure 38: Shape of the boundary layer for different pressure profiles.

48
 Figure 39: Detachment point of the boundary layer.

Example: flow around cylinder

Figure 40: Separation of boundary layer and detachment point for flow around a
cylinder.

The detachment point is at the separation of the boundary layer. When v ≈ 0 there
is no kinetic energy to run against the pressure gradient.

Remark: separation of boundary layers is a big issue in mechanical engineering;

for example: design of airfoils, wind-turbine blades etc.

In case of separation the lift decreases, which can lead to airplane crash. It
would also lead to a substantial rise in the overall drag (more friction). This
would require more engine power and therefore more fuel for an airplane. For
a wind turbine it would means less power generation.
Engineer’s dream: construct airfoils without turbulent wake, with e.g. shark
skin or fish scales.

49
9.3 Solution of Prandtl equations for free boundary layers

Figure 41 shows a 2-dimensional laminar jet flow generated from a flow through a
long slit streaming into a resting fluid.

Figure 41: 2-dimensional laminar jet flow.

Assuming δ( x ) ≪ x, we can use the Prandtl equations with the similarity ansatz:
( )
y
u x ( x, y) = umax ( x ) g z = (9.51)
δ( x )

∂u x ∂uy ∂ψ ∂ψ
+ = 0 ⇒ ux = , uy = − (9.52)
∂x ∂y ∂y ∂x
( )
y
ψ = umax ( x )δ( x ) f (9.53)
δ( x )

∂ψ 1
= umax δ f ′ = umax f ′ = umax g = u x (9.54)
∂y δ

∂ψ ′ (−y)δ′
uy = − = −umax δ f − umax δ′ f − umax δ f ′ (9.55)
∂x δ2

′ (−y)δ′
∂ x u x = umax f ′ + umax f ′′ (9.56)
δ2
1
∂y u x = umax f ′′ (9.57)
δ
umax ′′′
∂2y u x = f (9.58)
δ2

50
Insertion into Prandtl’s equation with p( x ) = constant:
{ }
µ 2 ′ ′ ′ yδ′ ′′
u x ∂ x u x + uy ∂y u x − ∂y u x = umax f umax f − umax 2 f
ρ0 δ
{ }
′ ′ yδ′ ′ 1
− umax δ f + umax δ f − umax f umax f ′′ (9.59)
δ δ
µ umax ′′′
− 2 2 f
ρ0 δ
′ δ′
= umax umax f ′2 − umax umax
′
f f ′′ − u2max f f ′′
δ (9.60)
µ umax ′′′
− f
ρ0 δ2
!
=0 (9.61)

All four terms in (9.60) have the form αi ( x ) β i (z) for (i = 1, ..., 4). The sum of these
four terms has to be zero. This means that α1 ( x ) ∼ α2 ( x ) ∼ α3 ( x ) ∼ α4 ( x ).
Ansatz:

umax ( x ) = c1 x m (9.62)
δ ( x ) = c2 x n
(9.63)

Remark: we expect m < 0 (decreasing velocity with penetration depth) and

n > 0 (increasing thickness of jet with penetration depth).

Sum of the 4 terms:

n µ c1 x m ′′′
c21 mx2m−1 ( f ′2 − f f ′′ ) − c21 x2m f f ′′ − f =0 (9.64)
x ρ0 c22 x2n

2m − 1 = m − 2n ⇒ m + 2n = 1 (9.65)

µ 1 ′′′
m( f ′2 − f f ′′ ) − n f f ′′ − f =0 (9.66)
ρ0 c1 c22

This differential equation determines the velocity profile

( ) ( )
y ′ y
g = f (9.67)
δ( x ) δ( x )
of the jet. We are not going to solve this, but we want to know m and n, because
they determine umax ( x ) and δ( x ). We need a second equation for m and n.
Second equation: conservation of momentum flux. Momentum flux through the
red plane in Figure 42 is identical to the momentum flux through the blue plane.
This means that the integrated momentum flux does not depend on x.

51
 Figure 42: Momentum flux for 2-dimensional laminar jet flow.

momentum = ρ0 ∆V · u x (9.68)
= ρ0 ∆Au x ∆tu x (9.69)

Momentum flux:
momentum
= ρ0 u2x (9.70)
∆A∆t

Proof of conservation of momentum flux

If ∫ ∞
ρ0 u2x ( x )dy = constant (9.71)
−∞
then ∫ ∞
d
ρ0 u2x ( x )dy = 0 (9.72)
dx −∞

∫ ∞ ∫ ∞ ( )
d ∂u x
ρ0 u2x ( x )dy = 2ρ0 ux dy (9.73)
dx −∞ −∞ ∂x
∞ ∫ ∞
∂u x  ∂u x
= 2µ  − 2ρ 0 uy dy (9.74)
∂y −∞ −∞ ∂y
∞ ∫ ∞
 ∂uy
= −2ρ0 u x uy  +2ρ0 ux dy (9.75)
−∞ −∞ ∂y
∫ ∞
∂u x
= −2ρ0 ux dy (9.76)
−∞ ∂x
=0 (9.77)

Prandtl’s equation (9.61) has been inserted into (9.73) to obtain (9.74). The first
term in (9.74) is zero because u x (y) = constant for y → ±∞. Partial integration
has been used to arrive at (9.75). The incompressibility condition ∂ x u x + ∂y uy
has been used to get (9.76). Since (9.76) is equal to (9.73) the integral has to be
zero.

52
 ∫ ∞ ∫ ∞ ( )
y
ρ0 u2x dy = ρ0 u2max ( x ) g2 dy (9.78)
−∞ −∞ δ( x )
∫ ∞
= ρ0 u2max ( x )δ( x ) g2 (z)dz (9.79)
−∞
!
= constant (9.80)

u2max ( x )δ( x ) = constant (9.81)

c21 x2m c2 x n = constant (9.82)
2m + n = 0 (9.83)

m + 2n = 1 , 2m + n = 0 (9.84)
⇓
1 2
m=− , n= (9.85)
3 3

1
umax ( x ) ∼ (9.86)
x1/3
δ( x ) ∼ x2/3 (9.87)

Remark: negative jet flow

Wake behind a wind turbine can be modeled as a negative jet.

( )
r
u x ( x, r ) = u∞ − ucenter ( x ) g (9.88)
δ( x )

53
10 Small-amplitude surface waves

Figure 43:

Figure 43 shows a surface wave. Here k is the wave number, τ the oscillation period,
and ω the circular frequency.

2π
k= (10.1)
λ
2π
ω= = 2π f (10.2)
τ
λ ω
c = = λf = (10.3)
τ k
Questions:

1. how does c depend on λ, d, a, ...

2. how does the fluid particles (below the surface) move? Pathlines? (expecta-
tion: more fluid motion close to the surface than in great depth.)

Assumptions: Incompressibility:

ρ = ρ0 , ∇
⃗ · ⃗u = 0 (10.4)

No friciton. Euler equation:

( )
∂⃗u ( ⃗ )
ρ0 + ⃗u · ∇ ⃗u = ⃗f ext − ∇
⃗p (10.5)
∂t

Gravitational force density:

⃗f ext = −ρ0 g⃗ez = −∇(
⃗ ρ0 gz) (10.6)

small amplitude waves:

a ≪ λ, d (10.7)

54
deep water waves:
λ≪d (10.8)

Hierarchy:
a≪λ≪d (10.9)

 
 ∂⃗u  ∆u u−0 u a
 
 ∂t ≈ ∆t ≈ τ/4 ≈ τ ≈ τ 2 (10.10)

⃗ u| ≈ u ∆u
|(⃗u · ∇)⃗ (10.11)
∆x
u u2 1 a2
≈u = = (10.12)
λ λ λ τ2

1 a2
|(⃗u · ∇)⃗
⃗ u| λ τ2 a
≈ = ≪1 (10.13)
|∂⃗u/∂t| a
τ2
λ

"Surviving" part of the Euler equation:

( )
∂⃗u p
= −∇ gz +
⃗ (10.14)
∂t ρ0

We do the curl ( )
∂⃗u p
∇×
⃗ = −∇ × ∇ gz +
⃗ ⃗ =0 (10.15)
∂t ρ0

⃗ × ⃗u = constant =! 0
∇ (10.16)
This constant has to be the same everywhere:

constant(z = 0) = constant(z = −∞) = 0 (10.17)

⃗u = ∇
⃗Φ (10.18)

with incompressibility
( )
∂2 ∂2
0 = ∇⃗
⃗u=∇
⃗ ·∇
⃗Φ= + Φ( x, z, t) (10.19)
∂x2 ∂z2

Question: how shall we solve this linear differential equation?

We solve it via factorization:

Φ( x, z, t) = X ( x ) Z (z) T (t) (10.20)

55
 [ 2 ]
1 ∂ ∂2
( X ( x ) Z (z) T (t)) + 2 ( X ( x ) Z (z) T (t))
X ( x ) Z (z) T (t) ∂x2 ∂z
(10.21)
1 ∂ X (x)
2 1 ∂2 Z ( z )
= +
X ( x ) ∂x2 Z (z) ∂z2
=0 (10.22)

1 ∂2 X ( x ) 1 ∂2 Z ( z )
= − k 2
= (10.23)
X ( x ) ∂x2 Z (z) ∂z2

∂2 X ( x )
+ k2 X ( x ) = 0 (10.24)
∂x2
∂2 Z ( z )
− k2 Z ( z ) = 0 (10.25)
∂z2

X ( x ) = e±ikx (10.26)
Z (z) = e±kz (10.27)

Φ( x, z, t) = Ae±ikx e±kz T (t) (10.28)

10.1 First boundary condition: (z = −d = −∞)

⃗u|z=−∞ = 0 (10.29)
⇓
Φ(z = −∞) = constant = 0. (10.30)

The second part of our general solution

Φ( x, z, t) = f + ( x, t)ekz + f − ( x, t)e−kz (10.31)

does not fulfill this boundary condition, and has to be set to zero, i.e. f − ( x, t) = 0.

Φ( x, c, t) = Ae±ikx ekz T (t) (10.32)

10.2 Second boundary condition: (surface)

p( x, z, t)|z=η ( x,t) = p0 (10.33)

56
 ( )
∂⃗u ∂ (⃗ ) p
= ∇Φ = −∇
⃗ gz + (10.34)
∂t ∂t ρ0
⇓
( )
∂Φ p
∇
⃗ + gz + =0 (10.35)
∂t ρ0
⇓
( ) ( ) 
∂Φ p  ∂Φ  p0
+ gz + = + gz  + = α(t) (10.36)
∂t ρ0 z=η (x,t) ∂t 
z=η ( x,t) ρ0

Gauge transformation:
∫ t
p0
Φ → Φ+ α(t′ )dt′ − t (10.37)
0 ρ0
We are allowed to do this because the velocity field ⃗u = ∇
⃗ Φ does not change with
this transformation.

( )
∂Φ 
+ gz  =0 (10.38)
∂t 
z=η ( x,t)

∂Φ 
= − gη ( x, t) (10.39)
∂t z=η (x,t)

If we want to determine Φ, i.e., T (t), we need another equation relating Φ and η,

so that we get rid of η. This is going to be the kinematic boundary condition.

10.3 Kinematic boundary condition

∂η
∆η = uz ∆t − u x ∆t (10.40)
∂x
⇓
∂η ∆η
≈ (10.41)
∂t ∆t
∂η
= uz − ux (10.42)
∂x
≈ uz . (10.43)

In the last step we have used

∂η a
∼ ≪ 1. (10.44)
∂x λ

57
 Figure 44:

 
∂Φ   ∂η ( x, t)
 = uz  = (10.45)
∂z z=η z=η ( x,t) ∂t
(  )
∂ (−1) ∂Φ 
= (10.46)
∂t g ∂t z=η

1 ∂2 Φ 
=− . (10.47)
g ∂t2 z=η

Again a ≪ λ:  
∂Φ  1 ∂2 Φ 

 +  =0 (10.48)
∂z z≈0 g ∂t2 z ≈0

Now we can determine T (t) by inserting the expression for Φ in (10.32).

 
∂Φ  
 = Ae ke T (t) 
±ikx kz
(10.49)
∂z z=0 z =0
= Ake±ikx T (t) (10.50)

! 1 ∂2 Φ  1 ±ikx ∂2 T (t)
=− = − Ae (10.51)
g ∂t2 z=0 g ∂t2

∂2 T ( t )
+ gkT (t) = 0 (10.52)
∂t2
√
T ( t ) = e ±i gkt
= e±iωt (10.53)

Φ( x, z, t) = Aekz e±ikx e±iωt (10.54)

58
Using Euler relations one of the four combinations can be rewritten as

Φ( x, z, t) = Aekz sin(kx − ωt). (10.55)

From this expression for the velocity potential we can calculate the surface function
η ( x, t) via the relation (10.39):

(−1) ∂Φ 
η ( x, t) = (10.56)
g ∂t z=η ( x,t)

(−1) 
kz 
= A(−ω ) cos(kx − ωt)e  (10.57)
g z=η
ω
= A cos(kx − ωt)ekη (x,t) (10.58)
g
ω
= A cos(kx − ωt) (10.59)
g

In the last step we have used ekη = e2πη/λ ≈ e0 = 1 and η/λ ≪ 1. This is the
expression as in Figure 43. The wave velocity is given by
√ √ √
ω gk g gλ
c= = = = (10.60)
k k k 2π
Surface waves with a large wave length propagate faster than those with a short
wave length.
Question: how does the motion of the fluid particles look like?

∂Φ
ux = = Akekz cos(kx − ωt) (10.61)
∂x
∂Φ
uz = = Akekz sin(kx − ωt) (10.62)
∂z
ω
Order of magnitude estimate for the velocity amplitude ( a = g A ):

g gk2 ω
O(u x ) = O(uz ) = Ak = ak = a 2 (10.63)
ω ω k
gk2
= a √ 2 c = akc (10.64)
gk
2π a
= a c = 2π c (10.65)
λ λ

O(u x ) = O(u x ) ≪ c (10.66)

Fluid particle is not moving with the wave velocity. Its velocity is much smaller.
The fluid particle is more or less at a fixed position. It is oscillating around this
fixed position with a small amplitude:

u x ≈ Akekz0 cos(kx0 − ωt) (10.67)

uz ≈ Akekz0 sin(kx0 − ωt) (10.68)

59
Pathline of a fluid particle

dx
ux = (10.69)
dt
dz
uz = (10.70)
dt

Ak kz0
x − x0 = − e sin(kx0 − ωt) (10.71)
ω
Ak kz0
z − z0 = e cos(kx0 − ωt) (10.72)
ω
Fluid particle moves on a circle (see Figure 45):
( )2
Ak kz0
( x − x0 ) + ( z −
2
z20 ) = e (10.73)
ω

with radius
Ak kz0 g k kz0
R= e = ae (10.74)
ω ωω
= aekz0 . (10.75)

Figure 45:

See Figure 45, so far we have discussed a pure wave, which

√ is characterized by
√
g gλ
the wave number k and which travels with c = k = 2π . The most general
solution describing the spatio-temporal dynamics of the water surface is obtained
by a superposition of different pure waves:
∫ ∞ [
Φ( x, z, t) = dk A(k)ei(kx+ωt) + B(k)e−i(kx+ωt)
0
] (10.76)
+ C (k)ei(kx−ωt) + D (k)e−i(kx−ωt)

60
 Remark: perturbation of a flat surface
A wave generated by this disturbance contains several different wave numbers k.
Initial wave form is not stable. It decays since different k components propagate
with different phase velocities.

Remark: small-amplitude surface waves at a finite depth (d < ∞)

√
g
c= tanh(kd) (10.77)
k
√ ( )
gλ d
= tanh 2π (10.78)
2π λ

√
gλ
d ≫ λ : tanh( x ≫ 1) = 0 ⇒ c = (10.79)
2π
√
d ≪ λ : tanh( x ≪ 1) = 0 ⇒ c = gd (10.80)

More surface waves:

• non-linear waves (λ ≫ d): solitons

• monster waves

• wind-wave interaction

• wave energy

61
11 Sound waves

Sound waves are pressure (density) waves in a compressible medium.

p = p0 + p̃, p̃ ≪ p0 (11.1)
ρ = ρ0 + ρ̃, ρ̃ ≪ ρ0 (11.2)

Equation of state:
ρ = ρ0 (1 + κ ( p − p0 )) (11.3)
The parameter κ is called compressibility.

ρ̃ = ρ0 κ p̃ (11.4)

Assumption: small velocities.

⃗u = 0 + ⃗u˜ . (11.5)

Euler equation without friction:

( )
∂⃗u ( ⃗ )
ρ + ⃗u · ∇ ⃗u = −∇
⃗p (11.6)
∂t

∂⃗u˜ ( )
(ρ0 + ρ̃) + (ρ0 + ρ̃) ⃗u˜ · ∇
⃗ ⃗u˜ = −∇
⃗ p̃ (11.7)
∂t

∂⃗u˜ 1⃗
=− ∇ p̃ (11.8)
∂t ρ0

Equation of continuity:

∂ρ ⃗
+ ∇ (ρ⃗u) = 0 (11.9)
∂t
⇓
∂ρ̃ ⃗ ( )
+ ∇ (ρ0 + ρ̃)⃗u˜ = 0 (11.10)
∂t
⇓
∂ρ̃ ⃗ ⃗u˜ = 0
+ ρ0 ∇ (11.11)
∂t
Divergence of (11.8):
˜
⃗ ∂⃗u = − 1 ∇
∇ ⃗ ·∇
⃗ p̃ (11.12)
∂t ρ0
Time derivative of (11.11):
∂2 ρ̃ ∂⃗ ˜
+ ρ0 ∇ ⃗u = 0 (11.13)
∂t 2 ∂t

62
 ∇
⃗ ·∇
⃗ p̃ = −ρ0 ∇ ⃗ ∂⃗u (11.14)
∂t
∂2 ρ̃
= 2 (11.15)
∂t

Wave equation:
⃗ p̃ = ρ0 κ ∂ p̃ = 1 ∂ p̃
2 2
∇
⃗ ·∇ (11.16)
∂t2 c2 ∂t2
In 1+1 dimensions:
∂2 p̃ 1 ∂2 p̃
= (11.17)
∂x2 c2 ∂t2
Plane-wave solution:
p̃ = A p cos( x − ct) (11.18)

Here c = √1 is the speed of sound. Examples:

ρo κ

cair = 340 m/s (11.19)

cwater = 1500 m/s (11.20)

"Density wave":
∂2 ρ̃ 1 ∂2 ρ̃
= ⇒ ρ̃ = Aρ cos( x − ct) (11.21)
∂x2 c2 ∂t2
Longitudinal "velocity wave":

∂ũ x 1 ∂ p̃ Ap
=− = sin( x − ct) (11.22)
∂t ρ0 ∂x ρ0
⇓
Ap
ũ x = cos( x − ct) = Au cos( x − ct) (11.23)
ρ0 c

Validity of approximation, which has neglected small quadratic terms in the Euler
equation:
( )
| ⃗u · ∇
⃗ ⃗u|
|u x ∂u
∂x | A2u |⃗u|
x
Au
∂⃗u
= ∂u x
≈ = ≈ (11.24)
| ∂t | | ∂t | A u c c c

Typical velocity oscillations are much smaller than the speed of sound:

|⃗u|
≪ 1. (11.25)
c

63
Example: loudspeaker

f = 100 − 2000 Hz ⇒ f typical = 1000 Hz (11.26)

Amplitude of membrane displacement ∆x ≈ 1 mm

∆x 2 × 10−3 m
∆u ≈ ≈ (11.27)
∆t/2 10−3 s
= 2 m/s ≪ cair = 340 m/s (11.28)

64
11.1 Outlook: shock waves

Figure 46: Sketches of shock waves.

65
A Instabilities (not corrected)

Motivation: Navier-Stokes equation in non-dimensional units with no external

forces:
∂⃗v ( ⃗ ) ⃗ p+ 1 ∇
+ ⃗v · ∇ ⃗v = −∇ ⃗ 2⃗v (A.1)
∂t Re
Reynolds number:

ρLV ρV 2 /L inertia force density

Re = = = (A.2)
µ µV/L 2 friction force density

As the Reynolds number increases, a stable flow becomes unstable, and a new stable
flow emerges.
Big question: when does a stable flow become unstable?
Answer: linear stability analysis

Figure 47:

Sketch of linear stability analysis

Simplifying assumptions: stable flows are steady (stationary)

⃗ (⃗r ) ,
⃗v(⃗r, t) = U p(⃗r, t) = P(⃗r ) (A.3)

Introduce perturbations

⃗ (⃗r ) + ⃗u(⃗r, t)
⃗v(⃗r, t) = U (A.4)
p(⃗r, t) = P(⃗r ) + p̃(⃗r, t) (A.5)

⃗ + ⃗u) [
∂ (U ] ( )
⃗ + ⃗u) · ∇
+ (U ⃗ (U ⃗ P + p̃) + 1 ∇
⃗ + ⃗u) = −∇( ⃗ ·∇
⃗ (U⃗ + ⃗u) (A.6)
∂t Re

66
Figure 48:

67
 ∂U
⃗
⃗ + ∇)
⃗ U⃗ + ∂⃗u + (U
⃗ + ∇)⃗
⃗ u
+ (U
∂t ∂t
1 (⃗ ⃗ ) ⃗
+(⃗u + ∇)
⃗ U⃗ + (⃗u + ∇)⃗
⃗ u = −∇
⃗P ∇·∇ U (A.7)
Re
( )
⃗ p̃ + 1 ∇
−∇ ⃗ ·∇
⃗ ⃗u
Re

∂⃗u ( ⃗ ⃗ ) ( )
⃗ p̃ + 1 ∇
( )
+ U · ∇ ⃗u + ⃗u · ∇
⃗ U ⃗ = −∇ ⃗ ·∇
⃗ ⃗u (A.8)
∂t Re

∇⃗· ⃗u = 0 (A.9)

⃗u(⃗r, t) = eλt⃗u(r,⃗ 0) (A.10)

1 (⃗ ⃗ ) ( ) ( )
λ⃗u(⃗r, t) = ∇ · ∇ ⃗u(⃗r, t) − ∇
⃗ p̃ − U⃗ ·∇
⃗ ⃗u(⃗r, t) − ⃗u · ∇
⃗ U ⃗ (⃗r, t) (A.11)
Re

∇
⃗ · ⃗u(⃗r, t) = 0 (A.12)
This is an eigenvalue equation.

λ = λ( Re, U
⃗) (A.13)
depends on Re and U ⃗ (⃗r ) and ⃗u(⃗r, 0).
If λ < 0: U
⃗ (⃗r ) is stable and the perturbations damps out. If λ > 0: U
⃗ (⃗r ) is unstable
and the perturbation grows.
Linear stability analysis of the poor man’s Navier-Stokes equation

∂⃗v [( ) ]
= − ⃗v · ∇ ⃗p + 1 ∇
⃗ ⃗v + ∇ ⃗ 2⃗v + ⃗f (A.14)
∂t Re
"analogy"
vt+1 − vt = −2v2t − vt + 1 (A.15)

Oversimplification: discrete time steps ∆t = 1, no spatial structure, no vector.

vt+1 = 1 − 2v2t , −1 ≤ v t ≤ 1 (A.16)

substitution

vt = 2xt − 1 (A.17)
⇓
xt+1 = 4xt (1 − xt ) , 0 ≤ xt ≤ 1 (A.18)

68
generalization
xt+1 = rxt (1 − xt ) , 0≤r≤4 (A.19)
This is the famous logistic (quadratic) map of deterministic chaos.
The order parameter r is analogous to Re: different dynamic (temporal) patterns
for different r.
As r increases an "old pattern" becomes unstable, and a "new pattern" emerges.

Figure 49:

Linear stability analysis

xt+1 = x ∗ + δxt+1 = f ( xt ) = f ( x ∗ + δxt ) (A.20)

≈ f ( x ∗ ) + f ′ ( x ∗ )δxt 0 + · · · (A.21)

Perturbations are only kept up to first order (linearization). Quadratic and

higher order terms are neglected.
 
 δxt+1 
  ′ ∗
 δxt = | f ( x )| (A.22)

| f ′ ( x ∗ )| < 1 : |δxt | = | f ′ ( x ∗ )|t |δx0 | = eλt |δx0 (A.23)

| f ′ ( x ∗ )| > 1 : |δxt | = eλt |δ( x0 )| (A.24)

69
 f ( x ) = rx (1 − z) (A.25)
⇓
′
f ( x ) = r [(1 − x ) − x ] = r (1 − 2x ) (A.26)
⇓
| f ′ ( x ∗ = 0)| = r (A.27)
⇓
∗
x =0 (A.28)

Stable fixed point for r < 1. Unstable fixed point for r > 1.
Question: what happens for r > 1?

Figure 50:

Fixed points:

x ∗ = f ( x ∗ ) = rx ∗ (1 − x ∗ ) (A.29)
⇓
r−1
x0∗ = 0 , x1∗ = (A.30)
r

Stability:
f ′ ( x0∗ ) = r > 1 (A.31)
which means that x0∗ is an unstable fixed point.
 ( ) 
 r−1 
|f ′
( x1∗ )| 
= r 1 − 2  = |2 − r | (A.32)
r 

| f ′ ( x1∗ )| < 1 ⇒ 1 < r < 4 (A.33)

which means that x1∗ is a stable fixed point.
Question: what happens for r > 3?

70
Figure 51:

71
 two cycle (r1 = 3 < r < r2 =?): jumps between two values x1∗ and x2∗ .

x2∗ = f ( x1∗ ) , x1∗ = f ( x2∗ ) (A.34)

⇓
x2∗ = f ( f ( x2∗ )) , x1∗ = f ( f ( x1∗ )) (A.35)

f (2) ( x ) = f ( f ( x )) = r f ( x )(1 − f ( x )) (A.36)

= r2 x ( x − 1)[1 − rx (1 − x )] (A.37)

 
 d f (2) ( x ) 
  < 1 ⇒ r1 = 3 < r < r2 = ? (A.38)
 dx 

stable two-cycle

r2 < r < r3 is a stable four-cycle

f (4) ( x ) = f ( f ( f ( f ( x )))) (A.39)

 
 d f (4) ( x ) 
 <1 (A.40)
 dx 

3 < r ≤ r∞ = 3.57... (A.41)

2-cycle, 4-cycle, 8-cycle, ...

3.57... ≤ r ≤ 4 (A.42)
Mostly chaotic.

72
 Figure 52:

Question: how can we characterize deterministic chaos?

Liapunov exponent:

Figure 53:

δx0 enλ(x0 ) = | f (n) ( x0 + δx0 ) − f (n) ( x0 )| (A.43)


1  f (n) ( x0 + δx0 − f (n) ( x0 )
λ( x0 ) = lim lim ln  (A.44)
n→∞ δx0 →0 n δx0
1
= lim ln |( f (n) ( x0 ))′ | (A.45)
n→∞ n
1
= lim ln | f ′ ( xn−1 ) · f ′ ( xn−2 ) · ... · f ′ ( x1 ) · f ′ ( x0 )| (A.46)
n→∞ n

73
 1 n −1
λ( x0 ) = lim
n→∞ n
∑ ln | f ′ (xi )| (A.47)
i =0

Attractive stable limit cycle (with period T = m)

1 m −1
ln | f ′ ( xi∗ )| < 0
m i∑
λ ( x0 ) = (A.48)
=0

Neighboring phase space trajectories with nearly identical initial conditions con-
verge towards each other.
Deterministic chaos:
λ ( x0 ) > 0 (A.49)
neighboring trajectories diverge

Figure 54:

Special case (of the logistic map): r = 4

Observation: λ( x0 ) > 0 leads to divergence of neighboring trajectories. All values
0 ≤ xt ≤ 1 occur (as t runs its course).

74
Question: what is the probability density for a specific x-value?

xt+1 = 4xt (1 − xt ) (A.50)

Substitution:
1
xt = [1 − cos(2πyt )] (A.51)
2

1
t t +1 =[1 − cos(2πyt+1 )] = 4xt (1 − xt ) (A.52)
2 [ ]
4 1 1
= [1 − cos(2πyt )] 1 − + cos(2πyt ) (A.53)
2 2 2
= [1 − cos(2πyt )][1 + cos(2πyt )] (A.54)
= 1 − cos (2πyt
2
(A.55)
1 1
= + [sin2 (2πyt ) + cos2 (2πyt )] − cos2 (2πyt ) (A.56)
2 2
1 1
= + [sin2 (2πyt ) − cos2 (2πyt )] (A.57)
2 2
1
= [1 − cos(4πyt )] (A.58)
2

Solution:
yt+1 = 2yt , 0 ≤ yt ≤ 1 (A.59)

y t = 2t y0 (A.60)

Figure 55:

75
 Once again: sensitive dependence on initial conditions

∞
y 0 = α 1 2−1 + α 2 2−2 + α 3 2−3 + · · · = ∑ α i 2− i (A.61)
i =1

y1 = 2y0 (A.62)
−2 −3 −4
= α2 2 + α3 2 + α4 2 +... (A.63)
∞
= ∑ α i +1 2− i (A.64)
i =1+1

∞
y2 = · · · = ∑ α i +2 2− i (A.65)
i =1+1
..
. (A.66)
∞
yt = ∑ α i + t 2− i (A.67)
i =1+1

y0 = 0.α1 α2 . . . αn αn+1 . . . (A.68)

y0′ = 0.α1 α2 . . . αn α′n+1 . . . (A.69)

1
|y0 − y0′ | ≤ 0.00
| {z . . . 01} = (A.70)
2n
(n−1) times

During the first n time steps the two trajectories yi = f (i) (y0 ) and yi′ = f (i) (y0′ )
stay close together; thereafter they separate completely. Sensitive dependence
on initial condition.

∞
y0 = 0.α1 α2 α3 · · · = ∑ α i 2−i (A.71)
i =1

with random numbers αi = {0, 1}.

{yt = f (t) (y0 )} uniformly distributed on [0, 1]
{
T −0 1, 0≤y≤1
1
ρ(y) = lim
T →∞ T
∑ δ(y − yt ) = 0, else
(A.72)
t =0

Back to our original question:

76
 Figure 56:

∫ 1 ∫ 1 ∫ 1
2
1= ρ(y)dy = dy = 2 dy (A.73)
0 0 0
∫ 1 ∫ 1
dx 2 dx
=2 = √ (A.74)
0 π sin(2πy) π 0 1 − cos2 (2πy)
∫ 1 ∫ ∫ 1
2 dx 2 1 dx 1
= = = dx (A.75)
π 0 [1 − (1 − 2x )2 ]1/2 π 0 [4x − 4x ] 2 1/2 0 π [ x (1 − x )]1/2

1
ρ( x ) = (A.76)
π [ x (1 − x )]1/2
invariant measure for the map xt+1 = 4xt (1 − xt ).
Transparency:

ρ( x ) , xt+1 = 4xt (1 − xt ) (A.77)

ρ(v) , v t +1 = 1 − 2v2t (A.78)

77
Figure 57:

Figure 58:

78