Lecture notes on Gasdynamics, AE4-140

by

Prof. Peter Bakker (Delft University of Technology)

and

Prof. Bram van Leer (University of Michigan)

September 3, 2019
2
Contents

1 Introduction 7
1.1 Introductory comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Notations, definitions, relations for gases . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Equation of state (EOS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Integral form of the conservation laws . . . . . . . . . . . . . . . . . . . 14
1.4.2 Differential form of the flow equations . . . . . . . . . . . . . . . . . . . 16
1.5 Euler equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.6 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.7 Discontinuities in compressible flow . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7.1 Shock wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7.2 Shear wave discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.7.3 Contact discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.7.4 Examples of discontinuities in steady flows . . . . . . . . . . . . . . . . . 21
1.8 One-dimensional unsteady jump equations . . . . . . . . . . . . . . . . . . . . . 22
1.9 Multi-dimensional form of the jump relations . . . . . . . . . . . . . . . . . . . 24
1.10 Conservation law of entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.11 Too many jump equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.12 Moving shocks; (t, x)-plane, entropy condition II . . . . . . . . . . . . . . . . . 28
1.12.1 (t, x)-plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.12.2 Shock speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.12.3 Entropy condition II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.13 Numerical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2 Linearized flow equations 37

2.1 1D acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.1 Relation with the wave equation . . . . . . . . . . . . . . . . . . . . . . 38
2.1.2 Riemann invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 Method of Characteristics, M.O.C. . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2.1 Forward M.O.C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2.2 Backward M.O.C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.3 Initial value problem for the wave equation . . . . . . . . . . . . . . . . 41
2.2.4 Discontinuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 Piston problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3.1 Specific Example: A sinusoidal excursion of the piston for only one single
period. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3
4 CONTENTS

2.3.2 Example: a tube with two pistons . . . . . . . . . . . . . . . . . . . . . 50

2.4 Riemann’s initial-value problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.4.1 Example 1: a density discontinuity; (M, S̃)-diagram . . . . . . . . . . . 52
2.4.2 Example 2: two colliding flows . . . . . . . . . . . . . . . . . . . . . . . 55
2.4.3 Example 3: gas streams running away from each other . . . . . . . . . . 55
2.4.4 Example 4: Moving piston . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3 One-dimensional unsteady non-linear flows 61

3.1 Characteristic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Simple waves, non-linear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3 Simple waves created by a moving piston . . . . . . . . . . . . . . . . . . . . . . 66
3.4 Centered expansion wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4.1 Flow variables in the centered expansion wave . . . . . . . . . . . . . . . 70
3.4.2 Particle path in a centered simple wave . . . . . . . . . . . . . . . . . . 72
3.5 Riemann problem, non-linear . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.5.1 Shock moving into a gas; Hugoniot curve . . . . . . . . . . . . . . . . . . 74
3.5.2 Expansion moving into a gas; Poisson curve . . . . . . . . . . . . . . . . 76
3.5.3 Example of a Riemann problem . . . . . . . . . . . . . . . . . . . . . . . 77
3.5.4 Properties of Hugoniot and Poisson curves . . . . . . . . . . . . . . . . . 77
3.5.5 Solving Riemann problems with a (p, u)-graph . . . . . . . . . . . . . . . 80
3.6 The Method of Characteristics for one-dimensional unsteady flow . . . . . . . . 82
3.7 Simple compression waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.8 Wave interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.8.1 Two right running shocks . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.8.2 Two right running expansion waves . . . . . . . . . . . . . . . . . . . . . 91
3.8.3 Shock interacting with expansion . . . . . . . . . . . . . . . . . . . . . . 92
3.8.4 Shock interacting with expansion . . . . . . . . . . . . . . . . . . . . . . 93
3.8.5 Inteaction of shocks of different family . . . . . . . . . . . . . . . . . . . 93
3.8.6 Interaction of two expansions of different family . . . . . . . . . . . . . . 95

4 Burgers’ equation for simple waves 97

4.1 Shock description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.2 Shock formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3 Shock formation from a triangular wave . . . . . . . . . . . . . . . . . . . . . . 102
4.4 Shock formation from a parabolic curve . . . . . . . . . . . . . . . . . . . . . . 104
4.5 Entropy conditions revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.6 Wave interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5 Traffic waves 115

5.1 Definitions, traffic equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.2 Characteristics and Discontinuities . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.3 Traffic light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.4 A chain collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
CONTENTS 5

6 Two-Dimensional Inviscid Flow 123

6.1 Governing flow equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.2 Characteristic form for 2D unsteady flow . . . . . . . . . . . . . . . . . . . . . . 126
6.3 Wave propagation in two-dimensional flow . . . . . . . . . . . . . . . . . . . . . 130

7 Steady Two-Dimensional Flow 141

7.1 Characteristic directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
7.2 "Time-like" and "space-like" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.3 Equations in flow aligned coordinates . . . . . . . . . . . . . . . . . . . . . . . . 147
7.4 Characteristic equations, compatibility relations . . . . . . . . . . . . . . . . . . 148
7.5 Method of Characteristics; M.O.C. . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.6 Method of Waves; M.O.W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.6.1 Description of the method, simple waves. . . . . . . . . . . . . . . . . . 159
7.6.2 Interaction of waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.6.3 M.O.W. for isentropic flows, a calculation scheme . . . . . . . . . . . . . 164
7.6.4 Design of a supersonic windtunnel nozzle by M.O.W. . . . . . . . . . . . 167

8 Two-dimensional steady potential flows 171

8.1 Crocco’s Theorem, Velocity Potential, Potential Equation. . . . . . . . . . . . . 171
8.2 Linearized flow; the Prandl-Glauert equation . . . . . . . . . . . . . . . . . . . 174
8.2.1 Linearized pressure coefficient . . . . . . . . . . . . . . . . . . . . . . . . 175
8.2.2 Linearized boundary condition . . . . . . . . . . . . . . . . . . . . . . . 176
8.2.3 Example 1: a subsonic flow along a wavy wall . . . . . . . . . . . . . . . 177
8.2.4 Example 2: Supersonic flow along a wavy wall. . . . . . . . . . . . . . . 179
8.3 Characteristics and compatibility relations for potential flow . . . . . . . . . . . 180
8.3.1 Prandtl-Glauert equation . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.3.2 The Full Potential Equation . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.4 Two-dimensional flow in the throat of a Laval nozzle . . . . . . . . . . . . . . . 184
8.4.1 Mass flow through a two-dimensional nozzle . . . . . . . . . . . . . . . . 190
8.4.2 Literature on the subject . . . . . . . . . . . . . . . . . . . . . . . . . . 193

9 The ’first non linear’ description of steady two dimensional supersonic flow195
9.1 Inviscid Burgers equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
9.2 Discontinuities of the inviscid Burgers equation . . . . . . . . . . . . . . . . . . 197
9.3 Shock pattern created by a biconvex airfoil . . . . . . . . . . . . . . . . . . . . . 202
9.4 Wave interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
9.4.1 Shock-expansion interaction . . . . . . . . . . . . . . . . . . . . . . . . . 208
9.4.2 Backward facing slope . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

10 Steady one-dimensional viscous flow 211

10.1 Channel flow: assumptions and basic equations . . . . . . . . . . . . . . . . . . 211
10.2 Constant-area channel flow; Fanno line . . . . . . . . . . . . . . . . . . . . . . . 215
10.3 A channel with a variable cross-sectional area . . . . . . . . . . . . . . . . . . . 219
10.4 Internal structure of a shock wave . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10.4.1 Introductory comments . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10.4.2 Navier-Stokes equations for one-dimensional flow . . . . . . . . . . . . . 224
10.4.3 Equations of shock structure . . . . . . . . . . . . . . . . . . . . . . . . 225
6 CONTENTS

10.4.4 An estimate for the shock-thickness . . . . . . . . . . . . . . . . . . . . . 226

10.4.5 Entropy production in the shock zone . . . . . . . . . . . . . . . . . . . 228
10.4.6 More about the shock structure . . . . . . . . . . . . . . . . . . . . . . . 228
10.4.7 Special solutions for the shock structure problem . . . . . . . . . . . . . 231
10.4.8 Entropy behaviour in the shock zone . . . . . . . . . . . . . . . . . . . . 234
Chapter 1

Introduction

1.1 Introductory comments

Compressibility of the fluid =⇒ density is variable

Density = mass/unit volume

What makes compressibility of real importance?

Incompressible flow : fluids (water)

gases (low speed M ≤ 0.3)
Compressible flow : propagation of sound
wave drag
shock waves
high temperature
combustion
re-entry aerodynamics
high speed flows (M >> 1)
Energy needed to accelerate fluid particles is extracted from two sources:

1. work done by pressure forces

2. decrease of internal energy resulting in a decrease in temperature

7
8 CHAPTER 1. INTRODUCTION

1.2 Notations, definitions, relations for gases

Quantity Symbol Comment

Density ρ mass/unit volume

1
Specific volume v= ρ volume/unit mass

Pressure p

Velocity V̄ , q̄ momentum/unit mass

|V̄ | ≡ V
|q̄| ≡ q
Specific internal energy e energy/unit mass

p
Specific enthalpy h h=e+ ρ

Specific total energy E E = e + 12 q 2

p
Specific total enthalpy H H = h + 21 q 2 = E + ρ

Specific total entropy s probability of state

Temperature T

∂e


Specific heat at const v,p cv ,cp cv = ∂T
v
∂h
cp = ∂T p

cp
Ratio of specific heats γ γ= cv (= 1.4 for air)

Specific gas constant R R = 287 kgJK (air)

1.3 Equation of state (EOS)

• "State principle" (empirically found):
"If the chemical composition of the fluid is fixed then the local thermodynamic state is
determined completely by two independent thermodynamic variables".
So, for example:
p = f (ρ, e) or p = f (ρ, h). (1.1)

• Thermally perfect gas: cv and cp only depend on T , so

cv = cv (T ) and cp = cp (T ).
Equation of state:
p = ρRT (1.2)
1.3. EQUATION OF STATE (EOS) 9

R is constant for

a particular gas.
∂e
Since cv = ∂T v :
Z
e = cv (T )dT (1.3)

∂h


and cp = ∂T p :
Z
h= cp (T )dT (1.4)

Thus h = e + ρp :
Z Z
cp (T )dT = cv (T )dT + RT

R = cp (T ) − cv (T ) (1.5)
R γ(T )R
cv (T ) = , cp (T ) = (1.6)
γ(T ) − 1 γ(T ) − 1
• Calorically perfect gas: cv and cp are constants
EOS: ρp = RT, R = cp − cv
RT p
e = cv T = γ−1 = ρ(γ−1) =⇒

p = (γ − 1)ρe (1.7)
γR γ p
h = cp T = γ−1 T = γ−1 ρ = γe =⇒
γ−1
p= ρh (1.8)
γ
Air is perfect for temperatures up to 2500 K
For T < 800 K air is calorically perfect.
For 800 K < T < 2500 K air is thermally perfect.
• Molecular degrees of freedom
From kinetic gas theory we know that the energy is divided according to the “Principle
of Equipartition”. This means that: “Internal energy of a gas is equally distributed over
the available degrees of freedom”. Let n be the number of degrees of freedom, then
1
e = nRT ; (1.9)
2
If n is constant, e.g. n is no fucntion of the temperature T , then:
 
∂e 1
cv = = nR (1.10a)
∂T v 2
1
cp = R + cv = ( n + 1)R (1.10b)
2
When combining (1.10a) and (1.10b), we find that the specific heat ratio is directly
related to the degrees of freedom:
cp n+2
γ= = (1.11)
cv n
10 CHAPTER 1. INTRODUCTION

Examples:

1. monatomic gas (He, dissociated gas):

3 degrees of freedom, n = 3 so γ = 5/3.
2. diatomic gas (O2 , N2 , air at standard conditions):
a diatomic gas has 5 degrees of freedom: 3 translational and 2 rotational. Then
n = 5 and γ = 7/5 = 1.4.

Figure 1.1: Diatomic gas with 5 degrees of freedom, halter model

At higher temperatures due to vibrational excitation two extra degrees of freedom

(in the form of a potential and kinetic contribution) become available such that
n = 7 and γ = 9/7, see NACA report 1135.
3. monatomic gas in one-dimensional world (only of mathematical interest!):
here there is only one degree of freedom possible, so n = 1 and γ = 3.

• Influence of chemistry on R
The value of R depends on the type of gas because of the different number of particles
(molecules, atoms and ions) per unit mass. Multiplying the perfect gas law by the
molecular weight µ yields:

µpv = µRT,

or,

pV = <T ,

where V is the volume of a mole of gas and < is the universal gas constant (gas constant
of Regnault):

< = 8314 J/(mole K)

1.3. EQUATION OF STATE (EOS) 11

The specific gas constant R then becomes: R = </µ.

For air (78% N2 , 21% O2 , 1% Ar) with µ = 29 kg/mole it follows that R = 8314/29 =
287.04 J/(kg K).
When chemical reactions occur, the value of R changes. Some examples are:

 
O2 → 2O µ is halved
dissociation:
N2 → 2N R doubles

 
N + O → NO 
µ rises

reaction: N O + O → N O2
R drops
N + 2O → N O2
 

µ drops
ionization: O → O+ + e−
R rises

• Entropy

To derive expressions for the entropy we start with the First Law of Thermodynamics:
de = đQ + đW
For a closed system (no mass transfer!) the increase of internal energy de is found as the
sum of heat added to the system đQ and the work performed on the gas in the system
đW .
Here the notation đ indicates that the increments đQ and đW depend on the type of
process (e.g. isothermal, isobaric, isochoric, etc) thus Q and W are no state variables!
However, entropy is a state variable, the entropy difference ∆s between two states only
depends on the thermodynamical (state) properties of the gas in both states.

The first law

de = đQ + đW
may be written as
de = đQ − pdv.
For reversible processes đQ = T ds and the First Law can be rewritten as:
T ds = de + pdv.
For caloric perfect gasses de = cv dT such that:
T ds = cv dT + pdv
pdv
ds = cv dT
T + T 
= cv dp dv
p + v + R
v dv
= cv dp dv
p + cp v
12 CHAPTER 1. INTRODUCTION
 
dp dv
ds = cv +γ (1.12)
p v
This equation can be integrated, resulting in:
pv γ
 
s − s0 = cv ln
p 0 v0 γ

If the process is also adiabatic : đQ = 0 =⇒ ds = 0 =⇒ s = constant.

A reversible adiabatic process is called an isentropic process having :

s = constant =⇒ pv γ = constant

or

p = Cργ (1.13)

which is called Poisson’s relation.

For an irreversible process: Second Law of Thermodynamics =⇒ T ds > đQ. This

implies for adiabatic flow (đQ = 0) that ds > 0.

• Speed of sound

The propagation of sound waves through a gas is a reversible adiabatic process. The
speed of sound is denoted by a.
For sound waves it is assumed that:

– perturbations in density, pressure and temperature are infinitely small, and

– the propagation is reversible and adiabatic ⇒ isentropic.

Although it is mentioned here as an assumption, one can actually prove that when
considering small perturbations, a soundwave is actually isentropic.
The speed of sound is defined by:
 
2 ∂p
a = ; (1.14)
∂ρ s
where the subscribt ‘s’ expresses that sound propagation is assumed to be an isentropic
process.

Theorem

   
2 ∂p p ∂ρ
a = + 2 (1.15)
∂ρ e ρ ∂e ρ

If the equation of state is given as p = p(ρ, e) then it is logical to have an

equation for the sound velocity wherein ρ and e are the independent variables.
1.3. EQUATION OF STATE (EOS) 13

Proof
Sound propagation is isentropic, so ds = 0, therefore we can write that
 
1
de + pd = 0.
ρ

     
1 ∂e ∂e p
de + pd = dp + dρ − 2 dρ = 0,
ρ ∂p ρ ∂ρ p ρ

or
  (  )
∂e ∂e p
dp + − 2 dρ = 0,
∂p ρ ∂ρ p ρ

so
 −1 (   )
dp ∂e ∂e p
= − + (∗)
dρ ∂p ρ ∂ρ p ρ2

dp
Since we have assumed that ds = 0 we may conclude that dρ is actually
   
dp ∂p
dρ ds=0 or ∂ρ s .
Which results in the following equation:
   −1 (   )
∂p ∂e ∂e p
= − +
∂ρ s ∂p ρ ∂ρ p ρ2

Multivariable calculus teaches us that if p, ρ and e are contstrained by f (p, ρ, e) =

0 then
     
∂p ∂e ∂ρ
· · = −1,
∂e ρ ∂ρ p ∂p e

from which we can derive

 
∂e −1
=    ,
∂ρ p ∂p ∂ρ
∂e ρ ∂p e

 
∂e
Inserting ∂ρ p in (∗) results into

     
2 ∂p ∂p p ∂p
a = = + 2 ,
∂ρ s ∂ρ e ρ ∂e ρ

So equation (1.15) is proven. 2

14 CHAPTER 1. INTRODUCTION

Figure 1.2: Control volume

Using the equation of state in the form

p = (γ − 1)ρe

we can now easily determine the sound velocity from equation (1.14):
p
a2 = (γ − 1)e + 2 (γ − 1)ρ
ρ
 
p
= (γ − 1) e +
ρ
= (γ − 1)h
= (γ − 1)cp T
= γRT
γp
=
ρ

1.4 Conservation laws

1.4.1 Integral form of the conservation laws
Consider a control volume V fixed in space with surface S = ∂V , see figure 1.2. Let dV be
a small volume element and let dS be a small surface element. Let n̄ be the outward unit
normal on dS. The integral form of the conservation laws for mass, momentum and energy
are then written as follows.

• Mass conservation
ZZZ ZZ
d
ρdV + ρV̄ · n̄dS =0 (1.16)
dt
| {zV } | S
{z }
time rate of change of mass inside V net mass flow out V through surface S
1.4. CONSERVATION LAWS 15

Figure 1.3: Example of an external force (F̄external ) on a control volume

• Momentum conservation
ZZZ ZZ
d
ρV̄ dV + ρV̄ V̄ · n̄dS +
dt V S
| {z } | {z }
time rate of change
ZZ of momentum inside V net flow Z
ofZmomentum
Z across S

p · n̄dS = ρf¯dV + (1.17)

S | {z }
| {z }
surface force due to pressure body forces
F̄visc + F̄external
|{z} | {z }
viscous forces external forces (strut force, enclosure force)

Comments

1. Example of F̄external
Consider a windtunnel model as shown in figure 1.3. The force F̄ on the model
acts as an external force F̄external = −F̄ on the control volume V . The force
F̄external = −F̄ is also formed as the integral of the surface forces (both due to
pressure as well due to viscous stresses) along the profile contour.
2. F̄visc contains the contribution of the viscous forces
ZZ
F̄visc = τ̄¯ · n̄dS (1.18)
∂V

where τ̄¯ (x̄, t) is the viscous stress tensor.

For a Newtonian fluid we have
 
∂ui ∂uj ∂uk
τij = µ + + λδij (1.19)
∂xj ∂xi ∂xk

here µ and λ are viscosity coefficients which in general depend on temperature.

16 CHAPTER 1. INTRODUCTION

• Energy equation
ZZZ ZZ
d
ρEdV + ρE V̄ · n̄dS +
dt V S
| {z } | {z }
time rate of
Z change
Z of energy inside V net Z Z Zof energy across S
flow

pV̄ · n̄dS = ρf¯ · V̄ dV +

S | {z }
| {z }
work on fluid by pressure force work done by body forces
Q̇ + Ẇvisc + Ẇext
|{z} | {z }
heat added to the fluid; conduction, radiation, condensation work done by viscous and external forces
(1.20)

Comments

1. Since the control volume is fixed in space, the time derivative can be brought into
∂
the integral as ∂t .
2. If f¯ = −∇Φ (x̄) then we have a conservative force field. In that case the body force
term in the energy equation can be accounted for by including the potential energy
Φ (x̄) in E and H.
3. Q̇ has a volume (e.g. combustion) and a surface component (e.g. conduction).
ZZZ ZZ
Q̇ = ρcdV − q̄ · n̄dS (1.21)
V S
| {z } | {z }
volumetric heating heat flux due to conduction

Fouriers law:
q̄ = −k∇T (1.22)
with k as heat conduction coefficient.
The integral forms of the conservation laws have a wide range of applicability. Actually the
only assumption to be made is that the integrals are integrable in V and over ∂V . The
integrands may be discontinuous.

1.4.2 Differential form of the flow equations

To derive the differential form of the flow equations (PDE’s) the surface integrals in the in-
tegral form have to be transformed into volume integrals. This can be done by applying the
divergence and the gradient theorem.

Divergence theorem:
ZZZ ZZ
∇ · ĀdV = Ā · n̄dS. (1.23)
V S

Gradient theorem:
ZZZ ZZ
∇pdV = pn̄dS. (1.24)
V S
1.5. EULER EQUATIONS 17

Combining the volume integrals and recalling that V is fixed in space and arbitrarily chosen,
the integrand has to be zero for all points in space. When the contributions of viscosity and
heat transfer are omitted, the following set of partial differential equations results:

Continuity equation
∂ρ
+ ∇ · ρV̄ = 0 (1.25)
∂t
Momentum equation
∂
ρV̄ + ∇ · ρV̄ V̄ + ∇p = ρf¯ (1.26)
∂t
Energy equation
∂
ρE + ∇ · ρV̄ E + ∇ · pV̄ = ρV̄ · f¯ (1.27)
∂t

p
With H=E+ ρ we have ∇ · ρV̄ E + ∇ · pV̄ = ∇ · ρH V̄ .

1.5 Euler equations

The equations (1.25), (1.26) and (1.27) describe a mathematical model of compressible flows
where the effects of viscosity, heat conduction and external heating have been neglected. If
also external forces are neglected (f¯ = 0) the Euler equations result:

∂ρ
+ ∇ · ρV̄ = 0 (1.28)
∂t
∂
ρV̄ + ∇ · ρV̄ V̄ + ∇p = 0 (1.29)
∂t
∂
ρE + ∇ · ρH V̄ = 0 (1.30)
∂t

These equations can also be written in the compact form:

∂U
e ∂ Fe ∂ G
e ∂H e
+ + + =0 (1.31)
∂t ∂x ∂y ∂z
with
       
ρ ρu ρv ρw

 ρu 


 p + ρu2 


 ρvu 


 ρwu 

U
e =
 ρv ,
 Fe = 
 ρuv ,
 G
e=
 p + ρv 2 ,
 H
e =
 ρwv .

 ρw   ρuw   ρvw   p + ρw2 
ρE ρuH ρvH ρwH

U
e is the vector of state quantities and Fe, G
e and H
e are flux vectors.
Equation (1.31) gives the Euler equations in the so-called conservative form.
18 CHAPTER 1. INTRODUCTION

Non-conservative forms of the Euler equations follow by expanding the derivatives and intro-
ducing the substantial derivative:
D ∂
= + V̄ · ∇. (1.32)
Dt ∂t
Which results in:
Dρ
+ ρ∇ · V̄ = 0, (1.33)
Dt
DV̄
ρ + ∇p = 0, (1.34)
Dt
DE
ρ + ∇ · pV̄ = 0. (1.35)
Dt

1.6 Entropy
For a compressible flow governed by the Euler equations it can be shown that the entropy s
is constant when moving with a particle.

The first law of thermodynamics đQ = T ds gives

 
1
T ds = de + pd .
ρ
When following a fluid particle, Ds
Dt determines the value of s when time progresses and the
particle is convected with the flow,
 
Ds 1 De p Dρ
= − .
Dt T Dt ρ2 Dt
Using (1.33) and (1.35) follows
 
Ds 1 1 DV̄ p
= − ∇ · pV̄ − V̄ · + 2 ρ∇ · V̄ .
Dt T ρ Dt ρ

Introducing (1.34) for DDtV̄ :

 
Ds 1 p V̄ ∇p p
= − ∇ · V̄ − · ∇p + V̄ · + ∇ · V̄ .
Dt T ρ ρ ρ ρ
or
Ds
=0 (1.36)
Dt
So the entropy is constant when moving with a fluid particle. For steady flows this implies
that the entropy is constant along a streamline.

The fact that entropy is convected with a fluid particle is a typical result for Euler flows
where non-adiabatic processes (external heating), heat conduction and viscous effects are ab-
sent.
1.7. DISCONTINUITIES IN COMPRESSIBLE FLOW 19

1.7 Discontinuities in compressible flow

The Euler equations allow for the appearance of discontinuities which are the so called ’weak
solutions’. Across such a discontinuity, certain jump relations hold. These jump relations
will be derived in section 1.8, in this paragraph the physical aspects of three different types
of discontinuities will be first discussed on a more qualitative basis. They are represented as
plane waves appearing in a two-dimensional unsteady flow. The following dicontinuities can
be identified:

• Shock wave

• Shear wave

• Contact discontinuity

1.7.1 Shock wave

First we consider the shock wave discontinuity, see figure 1.4.
We see a shock wave that moves with velocity cs . The shock separates the flow domain into

Figure 1.4: Shock wave (viewed in laboratory frame)

two parts having different fluid velocities: V1 (components Vn1 and Vt1 ) and V2 (components
Vn2 and Vt2 ). The velocities are defined w.r.t a fixed reference frame called the ’laboratory
frame’. For steady shocks it can be shown that the tangential velocity component is constant
over the shock while the normal componnent decreases in flow direction, whereas the thermo-
dynamic variables p, ρ, e, T and s increase in flow direction. For steady normal shocks the
pre-state is always supersonic and the post-state is always subsonic.

For moving shocks this poses a problem. Take the general situation of a moving shock as
depicted in figure 1.4. Where do we find the pre and post-state?
To answer this question the shock is brought into a reference frame that moves with the shock;
this frame is called the ’shock frame’. Figures 1.5 and 1.6 show the situation for a steady flow
and a non-moving shock. Viewing figures 1.5 and 1.6, the normal velocities in domain À and
Á are respectively U1 = Vn1 − cs and U2 = Vn2 − cs . Two different possibilities occur.
20 CHAPTER 1. INTRODUCTION

Figure 1.5: U1 > 0, 1 is pre-state Figure 1.6: U1 < 0, 1 is post-state

Case 1: U1 > 0 (figure 1.5)

If U1 > 0 then the steady shock relations tell us that also U2 > 0. Thus domain À
appears to be the pre-state where the normal flow component is supersonic; so U1 > a1
and the thermodynamic variables satisfy: p2 > p1 , ρ2 > ρ1 , T2 > T1 , e2 > e1 and s2 > s1 .

Case 2: U1 < 0 (figure 1.6)

If U1 < 0 then domain À appears to be the post-state of the shock and domain Á is now
the pre-state of the shock. This means that domain Á is supersonic and domain À is
subsonic (|U1 | < a1 ). The thermodynamic variables satisfy: p2 < p1 , ρ2 < ρ1 , T2 < T1 ,
e2 < e1 and s2 < s1 .

The tangential velocities Vt1 and Vt2 are equal, so Vt1 = Vt2 both in the steady and in the
unsteady situation.

1.7.2 Shear wave discontinuity

A moving shear wave discontinuity is depicted in figure 1.7. The shear wave moves with
a velocity cs normal to its front. The fluid on either side of the shear wave has a normal
velocity equal to the shear wave velocity: Vn1 = Vn2 = cs . This implies that the shear wave
is convected with the fluid. However, there is a tangential velocity jump across the wave:
Vt1 6= Vt2 . By definition the thermodynamic variables p, ρ, e, T and s are continuous across
a shear wave.

Figure 1.8: Contact discontinuity (may be

Figure 1.7: Shear wave moving)
1.7. DISCONTINUITIES IN COMPRESSIBLE FLOW 21

1.7.3 Contact discontinuity

A contact discontinuity that moves w.r.t the laboratory frame is shown in figure 1.8. Typical
for a contact discontinuity is that ρ, e and s jump across the discontinuity while the velocity
and pressure are continuous. This implies that a contact discontinuity convects with the fluid,
thus Vn1 = Vn2 = cs . Often a contact discontinuity and a shear wave are superimposed.

A summary of the properties of various discontinuities can be found in table 1.1.

Table 1.1: Summary of the properties of various discontinuities

vn vt c p ρ e s

Shock wave J C cs J J J J

Shear wave C J vn C C C C

Contact discontinuity C C vn C J J J

Shear + contact C J vn C J J J

1.7.4 Examples of discontinuities in steady flows

Flow over a ramp in a channel
In this flow field an oblique shock reflects on a wall, causing a lambda-shock. At the
so-called triple point a superposition of a shear wave and a contact discontinuity occurs.
Across this discontinuity the pressure is continuous, the entropy (s), the density (ρ) and
the tangential velocity (vt ) jump, see figure 1.9.

Figure 1.9: Flow over a ramp in a channel

Flow in a channel with two unequal ramps

The ramps generate two shocks which run downstream and intersect (figure 1.10). At
22 CHAPTER 1. INTRODUCTION

the intersection the shocks interact and again two shocks are produced together with
a shear wave/contact discontinuity. This fully supersonic steady 2D flow has a strong
analogy in 1D unsteady flow (see next chapters).

Figure 1.10: Flow in channel with two unequal ramps

Flow over two successive ramps

Due to the presence of the ramps, two shocks of the same family are produced (figure
1.11). They converge and at the intersection a much stronger single shock is formed
combined with a shear wave/contact discontinuity. Also in this case a 1D unsteady
analogy exists.

Figure 1.11: Flow over two successive ramps

1.8 One-dimensional unsteady jump equations

Now that we have had a qualitative discussion of discontinuities in compressible flows, let’s
have a closer look at the governing relations for shock waves. To that end, consider a shock
discontinuity D that appears in a one-dimensional unsteady flow. The shock path is depicted
by the curve D in the (t, x)-plane, see figure 1.12. The flow on either side of the discontinuity
is governed by the Euler equations (see equation (1.31), omitting the tildes ‘˜’):
∂U ∂F
+ = 0, (1.37)
∂t ∂x
where
   
ρ ρu
U =  ρu  , F =  p + ρu2  . (1.38)
ρE ρuH
1.8. ONE-DIMENSIONAL UNSTEADY JUMP EQUATIONS 23

Figure 1.12: Shock discontinuity in the (t, x)-plane

The question is now, which jump corresponds to the continuous solutions in À and Á.
To find the answer we take a small element δl along the shock path and we assume that δl
is so small that the shock path can be approximated by a straight line element. This shock
element is surrounded by the closed contour C that encloses an area A. Next we apply the
divergence theorem to the area A in the space-time domain:
I x
Ā · n̄ dl = ∇ · Ā dA
c A

with
∂
     
U nt ∂t
Ā = , n̄ = , and ∇ = ∂ .
F nx ∂x

Now we can write

I  
U x  ∂U ∂F

· n̄dl = + dA = 0
c F ∂t
A
∂x

To obtain the jump relations across the discontinuity D choose the contour C tightly around
an element of D:
   
U U
· n̄2 δl + · n̄1 δl = 0.
F F
Since n̄1 = −n̄2 , we can write
 
U2 − U1
· n̄2 = 0 or,
F2 − F1
(U2 − U1 )n2t + (F2 − F1 )n2x = 0.
where the subscribt ‘t’ denotes a temporal component of the normal vector, and the subscript
‘x ’ denotes a spatial component. This can be written as:
n2
(U2 − U1 ) t + (F2 − F1 ) = 0
n2x
24 CHAPTER 1. INTRODUCTION

and since n̄2 is normal to D:

 
n2t dt
· = −1.
n2x dx D
dt


Since dx D is the shock velocity VD we finally get that:

−VD (U2 − U1 ) + (F2 − F1 ) = 0,

or

VD [U] = [F], (1.39)

where [·] = (·)2 − (·)1 expresses the difference between state 2 and state 1. Equation (1.39) is
the jump relation for one-dimensional steady flow.

Consider the special case that the discontinuity is steady, i.e. VD = 0. Equation (1.39)
then reduces to:

[F ] = 0, or F2 = F1

and if we substitute the elements of F from equation 1.38:

ρ2 u2 = ρ1 u1 , (1.40a)
p2 + ρ2 u22 = p1 + ρ1 u21 , (1.40b)
ρ2 u2 H2 = ρ1 u1 H1 . (1.40c)

we get the well-known Rankine-Hugoniot relations (1.40a)–(1.40c) for steady normal shock
waves! In the (t, x)-plane this steady situation looks like figure 1.13.

Figure 1.13: Steady shock in the (t, x)-plane

1.9 Multi-dimensional form of the jump relations

Now that we have established the jump relations in 1D we can also have a look at the multi-
dimensional form. Consider a discontinuity D moving with velocity V̄D in a three-dimensional
space. Let n̄D be the unit vector in space normal to D pointing from À to Á, see figure 1.14.
1.9. MULTI-DIMENSIONAL FORM OF THE JUMP RELATIONS 25

Figure 1.14: Discontinuity in a three-dimensional space

The governing equations for this unsteady three-dimensional problem are

∂ Ũ ∂ F̃ ∂ G̃ ∂ H̃
+ + + = 0.
∂t ∂x ∂y ∂z

To find the multi-dimensional jump relation, we apply equation (1.39) to an element of D.

This element moves with a velocity (V̄D · n̄D ) in the direction of its normal, so that
h i h i h i h i
(V̄D · n̄D ) Ũ = F̃ nDx + G̃ nDy + H̃ nDz . (1.41)

With the definitions of Ũ , F̃ , G̃ and H̃ this can be rearranged into:

 
ρ(V̄ − V̄D ) · n̄D =0 (1.42a)
 
ρV̄ (V̄ − V̄D ) · n̄D + pn̄D =0 (1.42b)
 
ρE(V̄ − V̄D ) · n̄D + pV̄ · n̄D = 0. (1.42c)

Using discrete algebra:

   
ā · b̄ = [ā] hb̄i + b̄ hāi,

where

[ā] = (ā)2 − (ā)1 is the difference and

1
hāi = (ā2 + ā1 ) is the mean value.
2
The multi-dimensional jump relations can be brought in the form
 
ρ(V̄ − V̄D ) · n̄D =0 (1.43a)
 
ρ(V̄ − V̄D )(V̄ − V̄D ) · n̄D + pn̄D =0 (1.43b)
h  2  i
ρ h + 12 V̄ − V̄D 


V̄ − V̄D · n̄D = 0. (1.43c)

The jump relations given in equations (1.43a)–(1.43c) are again the well-known Rankine-
Hugoniot relations.
26 CHAPTER 1. INTRODUCTION

Reversibility of jump conditions

The jump relations are invariant under exchange of states À and Á. Particularly with regard to
shocks, the jump relations cannot distinguish between the pre-shock state and the post-shock
state. Since in a shock irreversible processes (viscous dissipation, heat conduction) take place,
a criterion is needed to indicate the direction of time.

When considering a shock wave, we have seen that there is no heat exchange which means
that the shock process is adiabatic. For an irreversible adiabatic flow we know that ds > 0,
so entropy can be used to keep track of the direction of time in the process: a particle that
crosses a shock experiences a rise of entropy so the condition takes the form of an inequality,
which we will call “the entropy condition”.

Summary

For the unambiguous description of an inviscid flow we need:

1. PDE’s for the flow away from discontinuities, e.g. given as equation (1.31);

2. jump equations, valid across discontinuities e.g. given as equations (1.43a) to (1.43c);

3. an entropy condition; various formulations are in use, the first condition reads as:

Enropy condition I

A fluid particle passing through a shock increases its entropy.

1.10 Conservation law of entropy

For an adiabatic reversible flow we have that:
Ds
=0
Dt
thus, entropy is convected with a fluid particle (see also section 1.5). In case of a one-
dimensional unsteady flow this reduces to:
∂s ∂s
+u =0 (1.44a)
∂t ∂x
Let us combine this equation with conservation of mass:
∂ρ ∂ρu
+ = 0. (1.44b)
∂t ∂x
Multiply (1.44a) with ρ and (1.44b) with s and add:

∂(ρs) ∂
+ (ρus) = 0 (1.45)
∂t ∂x
This equation expresses conservation of the entropy density ρs. Be careful here! (1.45) is not
valid for irreversible processes, so it cannot be used to calculate the entropy jump across a
1.11. TOO MANY JUMP EQUATIONS 27

shock. The entropy jump [s] follows from the state variables determined by the other jump
relations, e.g. (1.39). Recall: the entropy is not a state variable.

In the case of an irreversible flow (1.45) should be replaced by:

(ρs)t + (ρus)x ≥ 0

leading to the jump inequality:

−VD [ρs] + [ρus] ≥ 0. (1.46)

Using discrete algebra and the density jump equation

VD [ρ] = [ρu]

we can reduce equation (1.46) to

 
hρui
− VD [s] ≥ 0. (1.47)
hρi

Check
Assume steady flow, i.e. VD = 0; then forward flow with hρui > 0 means that [s] > 0 or
s2 > s1 , a result as expected! Backward flow with hρui < 0 means s2 < s1 !

1.11 Too many jump equations

For steady flow the jump equation [F ] = 0 results into the so called Rankine-Hugoniot rela-
tions:

ρ1 u1 = ρ2 u2 (1.48a)
p1 + ρ1 u21 = p2 + ρ2 u22 (1.48b)
ρ1 u1 H1 = ρ2 u2 H2 (1.48c)

From (1.48a)–(1.48c) many useful relations can be derived, e.g.

p2 2γ
=1+ (M 2 − 1) (1.49a)
p1 γ+1 1
ρ2 (γ + 1) M12
= (1.49b)
ρ1 2 + (γ − 1) M12

2 + (γ − 1)M12
 
T2 2γ 2
= 1+ M1 − 1 (1.49c)
T1 γ+1 (γ + 1)M12
u2 2 (M12 − 1)
=1− (1.49d)
u1 γ + 1 M12

where M1 = u1 /a1 is the Mach number of the flow in the pre-shock state.
For unsteady flows with moving shock waves these relations are less useful since they contain
the Mach number M1 which is not a priori known in the moving shock case. Moreover
28 CHAPTER 1. INTRODUCTION

equations (1.49a) to (1.49d) are only valid for steady flows. In order to use them for a
moving shock one has to go to the shock frame, find the appropriate velocities, etc. This is a
cumbersome job and not always straightforward.
For moving shocks it is therefore much more convenient to have a jump relation that relates
only thermodynamic variables. Starting with

VD [U ] = [F ]

and eliminating VD one can derive the Hugoniot relation

1
hpi∆( ) + ∆e = 0. (1.50)
ρ

The proof is left as a homework problem. For a perfect gas with e = cv T and p = (γ − 1)ρe
we can rewrite equation (1.50):
γ+1 p2
ρ2 1 + γ−1 p
= γ+1 p21 . (1.51)
ρ1 γ−1 + p 1

p2
Observe that for strong shocks with p1 → ∞, the density ratio remains finite and takes the
value
 
ρ2 γ+1
= .
ρ1 max γ − 1

The maximum density ratios obtainable depend on γ, for example:

 
5 ρ2
γ= : =4
3 ρ1 max
 
7 ρ2
γ= : = 6.
5 ρ1 max

1.12 Moving shocks; (t, x)-plane, entropy condition II

1.12.1 (t, x)-plane
One-dimensional unsteady flow phenomena such as moving shocks can be visualised very well
in the (t, x)-plane. To become familiar with this plane, three examples of shocks separating
two uniform states are discussed, see table 1.2.

Case A shows a steady shock situation. We notice the following:

• the pictures in the laboratory frame and the shock frame are identical
• in the (t, x)-plane the shock path is a line for which x = constant.
• since u1 > u2 particles move faster in À. So dx


dt of the particle path is larger in
dt


À meaning that the slope dx of the path is lower.
• a slower particle has a steeper particle path in the (t, x)-plane
1.12. MOVING SHOCKS; (T, X)-PLANE, ENTROPY CONDITION II 29

Table 1.2: Three examples of shocks

Case A: Case B: Case C:
shock created by gas shock created in piston
steady shock
hitting a wall driven gas

picture in laboratory
picture in laboratory picture in laboratory
frame, cs < 0, u1 = 0,
frame, cs = 0 frame, cs < 0, u2 = 0
u2 < 0

picture in shock frame picture in shock frame picture in shock frame

picture in (t, x)-plane picture in (t, x)-plane picture in (t, x)-plane

• due to conservation of mass, at time t1 the mass contained between A1 and B1 is

ρ1 u1 ∆t. When this amount of mass is moved to the interval A2 –B2 at time t2 , it
is equal to ρ2 u2 ∆t. So
ρ1 u1 ∆t = ρ2 u2 ∆t,
| {z } | {z }
∆x1 ∆x2

therefore
∆x1 > ∆x2 ⇒ ρ2 > ρ1
When crossing a shock in forward time direction, particle paths get closer!
Case B represents a flow with velocity u1 > 0 that runs into a tube which is closed off at the
end. A shock runs into the moving gas bringing it to standstill, i.e. u2 = 0. We notice
the following:
• The pictures in the laboratory frame and the shock frame now differ! Calling the
pre- and post state velocities in the shock frame v1 and v2 respectively, we have:
v1 = u1 + |cs |, v2 = |cs |.
30 CHAPTER 1. INTRODUCTION

• since u1 > 0, particle paths in region À have a positive slope; since u2 = 0 particle
paths in Á have
 
dt
= ∞,
dx 2

therefore in region Á the particle paths are given by x = constant.

• from conservation of mass it follows that particle paths get closer when crossing
the shock in forward direction.
• this case has an analogue in steady two-dimensional flow: a shock reflection at a
wall, see figure 1.15.

Figure 1.15: Case B two-dimensional analogue; Shock reflection at a wall.

Case C shows a piston with velocity up that runs into a tube filled with a gas at rest. The
piston creates a shock wave moving with velocity cs < 0 into the gas at rest. We notice
the following:

• The pictures in the laboratory frame and the shock frame differ! Calling the pre-
and post state velocities v1 and v2 respectively, there follows:

v1 = |cs |, v2 = |cs | − u2 .

• u1 = 0, therefore
 
dt
=∞
dx 1

so particle paths in À are given by x = constant.

• in Á u2 is equal to the velocity of the piston up , so the paricle paths in Á are
parallel to the piston path
• conservation of mass yields again that particle paths get closer when crossing a
shock in forward time direction.
• also for this case there is a two-dimensional analogue: a shock induced by a ramp
in a supersonic flow, see figure 1.16.
1.12. MOVING SHOCKS; (T, X)-PLANE, ENTROPY CONDITION II 31

Figure 1.16: Case C two-dimensional analogue; Shock induced by a ramp in a supersonic flow.

Figure 1.18: Picture in shock frame

Figure 1.17: Shock moves to the left with
velocity cs

1.12.2 Shock speed

As shown in case C a piston with velocity up running into a gas at rest creates a shock that
moves with the speed cs , see figure 1.17. Over the shock a pressure rise ∆p = p2 − p1 exists. It
is convenient to write the shock speed in terms of the pressure rise p2 − p1 . Recall the already
have the relation (1.49a):

p2 2γ
=1+ (M 2 − 1); (1.49a)
p1 γ+1 1

which is only applicable in steady flow. Therefore we view the problem in shock frame, see
figure 1.18 where the Mach number in front of the shock is:

v1 |cs |
M1 = = .
a1 a1

So from equation (1.49a) follows that:

 2 
p2 2γ cs
=1+ −1
p1 γ + 1 a21

Solving for cs we find

r
γ + 1 p2 − p1
|cs | = a1 1 + (1.52)
2γ p1

Note that |cs | > a1 so the shock travels faster than the speed of sound!

First consider the weak shock limit:

∆p
 1, where ∆p = p2 − p1
p1
32 CHAPTER 1. INTRODUCTION

Assuming ∆p small w.r.t p1 in equation (1.52), the shock speed can be approximated by:
 
γ + 1 ∆p
|cs | = a1 1 + + O(∆p2 ) (1.53)
4γ p1
Observe from this formula that a sound wave is an infinitesimal small shock (∆p → 0).

Now consider the strong shock limit:

∆p
1
p1
then (1.52) can be approximated by:
r
γ + 1 p2
|cs | ≈ a1 ,
2γ p1
or
p2 2γ c2s
≈ . (1.54)
p1 γ + 1 a21
From equation (1.51) we observe that in the strong shock limit the density remains finite and
takes the value (also see the previous section):
ρ2 γ+1
= ,
ρ1 γ−1
therefore in the strong shock limit the temperature and internal energy behave as:
T2 e2 2γ(γ − 1) c2s
= ≈ .
T1 e1 (γ + 1)2 a21
So for strong shocks the pressure ratio p2 /p1 and the temperature ratio T2 /T1 increase with
γ+1
the square of the shock speed cs while the density ratio ρ2 /ρ1 approaches γ−1 (in the steady
case), so
v2 ρ1 γ−1
= →
v1 ρ2 γ+1

This implies the following velocity ratios in the cases A, B and C for the strong shock limit:
1.13. NUMERICAL APPROACH 33

1.12.3 Entropy condition II

For case A we know from elementary gasdynamics (compressible aerodynamics course) that
the flow enters the shock supersonic and exits subsonic, see table 1.2. For a steady shock we
therefore have u1 > a1 , u2 < a2 or

(u1 − a1 ) > 0 > (u2 − a2 ). (1.55)

This relation is a special case of a more general inequality:

(u1 − a1 ) > cs > (u2 − a2 )

Consider a moving shock in lab frame as shown in figure 1.19 (left) and its transformation in
shock frame as shown in figure 1.19 (right). Application of (1.55) for the situation in shock

Figure 1.19: Moving shock in lab frame (left) and transformed situation in shock frame (right)

frame gives:

(u1 − cs − a1 ) > 0 > (u2 − cs − a2 )

or:

(u1 − a1 ) > cs > (u2 − a2 ) (1.56)

A moving shock violating this inequality is not physically valid, so we may regard relation
(1.56) as another form of the entropy condition, without mentioning the entropy explicitly.

1.13 Numerical approach

Consider the Euler equations for one-dimensional unsteady compressible flows:
∂U ∂F
+ =0
∂t ∂x
For numerically simulating inviscid compressible flows, the most popular methods nowadays
are dissipative finite volume methods. These methods are consistent with the integral conserva-
tion laws, so the jump equations are automatically satisfied. To get insight into the derivation
and origin of these methods, consider a small domain in (one-dimensional) space–time, see
figure 1.20.
Now we integrate the conservation law Ut + Fx = 0 over this domain:

Zx2 Zt2 Zt2 Zx2

Ut dt dx + Fx dx dt = 0,
x1 t1 t1 x1
34 CHAPTER 1. INTRODUCTION

Figure 1.20: A finite domain in space–time

and get
t2 x2
Zx2  Zt2 
 
U dx + F dt = 0.
 (1.57)
 
x1 t1 t1 x1

Furthermore integrate over spatial coordinate to define a spatial mean:

Z x2
1
U= U dx,
∆x x1

and integrate over time to define a temporal mean:

Z t2
1
hF i = F dt.
∆t t1

Inserting the spatial mean and the temporal mean in equation (1.57) results in:

U (t2 ) − U (t1 ) hF (x2 )i − hF (x1 )i

+ = 0. (1.58)
∆t ∆x
This looks like a finite-difference approximation of Ut + Fx = 0, but this formula is still ex-
act: no numerical errors have been introduced! Such errors come into play when the average
time-integral of the fluxes is approximately predicted from the solution at the discrete times
t1 and/or t2 .

Equation (1.58) is an update scheme: once the solution at time level t1 is known, equa-
tion (1.58) yields the solution at time level t2 , etc., etc. To proceed, one needs to know the
fluxes or the flux differences. So, one has to find the flux formulae that are consistent with the
equations up to terms of the order O(∆t, ∆x) or higher, see the works of Prof. B. van Leer
and Prof. Ph. Roe.
The update from t to t + ∆t has to be stable not just if updates are small (in the continuous
part of the solution) but also in the presence of strong shock waves, see figure 1.21.
1.13. NUMERICAL APPROACH 35

Figure 1.21: A shock intersecting with the finite time-space domain

To achieve this, the flux formulae are designed in such a way that the dominant numerical
error produced by the scheme (the ‘truncation’ error) has the appearance of dissipation term.
With such an ‘artificial dissipation’ the update is close to equations of the type:
Ut + Fx = (D2 Ux )x (1.59)
or
Ut + Fx = (D4 Uxxx )x
where
D2 = O(∆t, ∆x)
and
D4 = O(∆t2 , ∆t2 ∆x, ∆t∆x2 , ∆x3 ).
Adding dissipation terms ensures time irreversibility, and production of entropy no matter
how small ∆x and ∆t are taken.
For ∆x → 0, ∆t → 0 the numerical solution approaches the solution of the inviscid
equation Ut + Fx = 0 including all physically admissable shocks.
Dropping viscosity leads to simpler equations (Euler) but one loses physics (viscous dis-
sipation). To retain flow phenomena related to viscous dissipation (e.g. shocks) dissipation
is added artificially to the governing equations; see examples in equation (1.59) by designing
appropriate flux formulae in numerical schemes.
This approach is based on the most fundamental form of the Entropy condition, which we
call ‘Entropy condition III’.

Entropy condition III

Only those inviscid solutions are physically acceptable that can also be obtained
from the full viscous and heat conducting equations in the limit of vanishing vis-
cosity and heat conduction.
The limiting process mentioned in entropy condition III is depicted in figure 1.22.
36 CHAPTER 1. INTRODUCTION

Figure 1.22: Artifical viscosity ensures that only physical solutions of the Euler equations will
be obtained.
Chapter 2

Linearized flow equations

2.1 1D acoustics
In this paragraph we will study the physics of acoustic wave propagation in a fluid at rest. In
real-life situations fluids occupy a three-dimensional space but the three-dimensional effects
are not relevant to understand the principles of acoustics; therefore, it is sufficient to consider
the one-dimensional case. All phenomena take place in a constant area tube.
Let the background state in the tube be given by

ρ0 , p0 , a0 and u0 = 0. (2.1)

Note that the background state is a fluid state at rest. Let perturbations be denoted by ∆ρ,
∆p and ∆u, so that

ρ = ρ0 + ∆ρ, p = p0 + ∆p, and u = u0 + ∆u = ∆u. (2.2)

Let the entropy s = s0 be kept constant uniformly in space (x) and time (t). This
assumption is justified since acoustic wave propagation is understood to be a reversible process.
Thus entropy s is equal to s0 for all particles at any time and place. Then we have

∆p = a20 ∆ρ, (2.3)

which is valid not along particle paths only, but for any pair of nearby states! Now introduce
the relative density change or “condensation” 1 ,
∆ρ
S̃ = , (2.4)
ρ0
and the Mach number
∆u u
M= = . (2.5)
a0 a0

Substitute (2.4) and (2.5) into the governing equations for 1D unsteady flow:

Ut + Fx = 0. (2.6)
1
Here we will follow the notation from Liepmann & Roshko, don’t confuse S̃ with S = entropy.

37
38 CHAPTER 2. LINEARIZED FLOW EQUATIONS

From the continuity equation follows

∂ρ ∂(ρu) ∂ ∂
+ = 0, (ρ0 + ∆ρ) + ((ρ0 + ∆ρ)u) = 0, (2.7)
∂t ∂x ∂t ∂x
or
∂∆ρ ∂u ∂
+ ρ0 + (u∆ρ) = 0. (2.8)
∂t ∂x |∂x {z }
second order perturbation
Neglecting second order perturbations, we find

∂ S̃ ∂M
+ a0 = 0. (2.9)
∂t ∂x
From the momentum equation there is obtained:
∂ ∂
p + ρu2 = 0.


(ρu) +
∂t ∂x
Inserting perturbations:
∂ ∂
p0 + ∆p + ρ0 u2 + u2 ∆ρ = 0,


(ρ0 u + u∆ρ) +
∂t ∂x
∂u ∂∆p ∂ ∂ ∂
u2 ρ0 + u2 ∆ρ = 0.



ρ0 + + (u∆ρ) +
∂t ∂x |∂t ∂x {z ∂x }
2nd and 3rd order perturbations
Neglecting 2nd and 3rd order terms yields
∂u ∂∆p
ρ0 + = 0,
∂t ∂x
or with equation (2.3):

∂M ∂ S̃
⇒ + a0 = 0. (2.10)
∂t ∂x
The equation Ut + Fx = 0 contains also the energy equation, however this gives no extra
information since it is replaced by the assumption of reversibility ds = 0.
Equations (2.9) and (2.10) are fundamental for the description of small perturbations
appearing in a gas at rest. Actually they govern wave propagation and therefore it may be
expected that there is a relation between the well-known wave equation and the two equations
(2.9) and (2.10) derived here.

2.1.1 Relation with the wave equation

Differentiate equation (2.9) with respect to time (or space) and differentiate equation (2.10)
with respect to space (or time), eliminate cross derivatives and we find:
∂ 2 S̃ 2
2 ∂ S̃
− a 0 = 0, (2.11a)
∂t2 ∂x2
2
∂ M 2
2∂ M
− a0 = 0, (2.11b)
∂t2 ∂x2
2.1. 1D ACOUSTICS 39

i.e. both S̃ and M satisfy a wave equation with a0 as the wave speed.
The general solution for the wave equation is given by d’Alembert:

S̃(x, t) = F (x − a0 t) + G(x + a0 t), (2.12a)

M (x, t) = f (x − a0 t) + g(x + a0 t), (2.12b)

where F (X), f (X), G(Y ) and g(Y ) are arbitrary functions (twice differentiable) of the single
arguments X = x − a0 t and Y = x + a0 t. Equations (2.12a) and (2.12b) show that there
are components in the solution which are constant along X = x − a0 t and others which are
constant along Y = x + a0 t; F (X) and f (X) represent the so-called right-running waves and
similarly G(Y ) and g(Y ) represent the so-called left-running waves. Both waves preserve their
exact shape while they propagate with speed a0 through the fluid along straight lines X =
const, or Y = const in the (t, x)-plane, see figure 2.1.

Figure 2.1: Propagation of S̃ in the (t, x)-plane

This fenomenon, so typical for variables satisfying the wave equation can alse be observed
from the set of equations (2.9) and (2.10).
Adding and substracting both equations yields:

(M + S̃)t + a0 (M + S̃)x = 0, (2.13)

(M − S̃)t − a0 (M − S̃)x = 0. (2.14)

Hence, M + S̃ is constant along a path in (t, x)-space with dx dt = +a0 , this path is called the
‘characteristic’ Γ+ , and M − S̃ is constant along a path in (t, x)-space with dx dt = −a0 , this
−
path is called the ‘characteristic’ Γ .
These characteristic curves appear to be straight lines in the above case of linear acoustics;
for non-linear gasdynamics they are generally curved.

2.1.2 Riemann invariants

Now we introduce the Riemann invariants J + and J − which are defined as

J + = M + S̃, J − = M − S̃. (2.15)

Equations (2.13) and (2.14) now result into

∂J ± ∂J ±
± a0 = 0. (2.16)
∂t ∂x
So J + is constant along Γ+ : dx
dt = +a0 and J − is constant along Γ− : dx
dt = −a0 .
40 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.2: Characteristics in the (t, x)-plane

Equations (2.16) will be called the “characteristic equations”. In other texts the equations
describing the characteristic curves, i.e.

dx
= ±a0 along Γ± (2.17)
dt
are called “characteristic equations” and the conditions on the flow variables, along the char-
acteristics

J ± = M ± S̃ = constant along Γ± (2.18)

are called “compatibility relations”.

2.2 Method of Characteristics, M.O.C.

The equations (2.13) and (2.14) or (2.15) and (2.16) are very well suited for obtaining solutions
whether analytically or, in non-linear cases, numerically. There are two forms, although very
similar, of the “method of characteristics”: the forward M.O.C. and the backward M.O.C.

2.2.1 Forward M.O.C

Assume that on the x-axis (t = 0) initial values are given, in particular in the points A
and B. Draw the Γ+ characteristic through A: Γ+ A forward (in time) and draw also the Γ
−
−
characteristic through B: ΓB forward (in time). They intersect in P (see figure 2.3). The
solution in P follows from the observation that:

along Γ+
A : (M + S̃)A = (M + S̃)P (2.19)
along Γ−
B : (M − S̃)B = (M − S̃)P (2.20)
2.2. METHOD OF CHARACTERISTICS, M.O.C. 41

Solving for the unknows MP and S̃P we find

MA + S̃A MB − S̃B
M (xp , tp ) = + (2.21)
2 2
MA + S̃A MB − S̃B
S̃(xp , tp ) = − (2.22)
2 2
Observe that the solution in P only depends on the initial data in A and B itself and not on
the initial data between A and B.

Figure 2.3: Forward method of character- Figure 2.4: Backward method of charac-
istics teristics

2.2.2 Backward M.O.C

Backward M.O.C. and forward M.O.C. are very similar. In backward M.O.C one has a method
that is applicable when one seeks the solution in a particular point P (xp , tp ), whereas the initial
values of M and S̃ are known on the line t = 0 (the x-axis in the (x, t)-plane).
Draw Γ+ and Γ− backwards through P, intersecting the x-axis in A and B respectively,
see figure 2.4. Here the initial data for M and S̃ are given.
The solution in P is now determined by the formulae already given as equations (2.21) and
(2.22).

2.2.3 Initial value problem for the wave equation

Consider the wave equation for S̃:
∂ 2 S̃ 2
2 ∂ S̃
+ a 0 = 0, (2.23)
∂t2 ∂x2
and assume that initial conditions are given on the x-axis (t = 0) as follows:
S̃(x, 0) = f (x), (2.24)
∂ S̃
(x, 0) = g(x). (2.25)
∂t
How does the solution S̃(x, t) depend on the initial data f (x) and g(x)?
The solution of d’Alembert, equation (2.12a) and (2.12b) gives
S̃(x, t) = F (x − a0 t) + G(x + a0 t), (2.26)
0 0
S̃t (x, t) = −a0 F (x − a0 t) + a0 G (x + a0 t) (2.27)
42 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.5: Domains of dependence and influence

Evaluating both expressions on the inital data line t = 0 yields

S̃(x, 0) = F (x) + G(x) = f (x), (2.28)

S̃t (x, 0) = −a0 (F 0 (x) − G0 (x)) = g(x). (2.29)

The last equation can be integrated with respect to x:

Z x
1
F (x) − G(x) = − g(ξ) dξ. (2.30)
a0 0

Now there are two equations for the unknown functions F (x) and G(x); they can be solved
to give
Z x
1 1
F (x) = f (x) − g(ξ) dξ, (2.31)
2 2a0 0
Z x
1 1
G(x) = f (x) + g(ξ) dξ. (2.32)
2 2a0 0

The general solution for S̃(x, t) is now found as

x+a
Z 0t
1 1
S̃(x, t) = {f (x − a0 t) + f (x + a0 t)} + g(ξ) dξ. (2.33)
2 2a0
x−a0 t

Equation (2.33) presents the solutions of S̃(x, t) in an arbitrary point (x, t). It depends on the
initial data lying on the interval bounded by the intersecting points with the characteristics
drawn backward through (x, t).
In figure 2.5 the initial data between A and B determine the solution in the point P (x, t).
The area enclosed by the characteristics Γ+ , Γ− and the initial line t = 0 is called the Domain
of Dependence (D.O.D.) of the point P . The area above P between Γ− and Γ+ is called the
Domain of Influence (D.O.I) of the point P .
In the domain of dependance we find all points on which the solution in P depends. In the
domain of influence we find all points whose solution is influenced by the solution in P .
2.2. METHOD OF CHARACTERISTICS, M.O.C. 43

Figure 2.6:

2.2.4 Discontinuities
Linear acoustics is governed by the set of equations

∂ S̃ ∂M
+ a0 = 0, (2.9)
∂t ∂x
∂M ∂ S̃
+ a0 = 0, (2.10)
∂t ∂x
whose solutions are characterised by the fact that M + S̃ is constant along Γ+ and M − S̃ is
constant along Γ− . The solution

(M + S̃)(x, t) = (M + S̃)(x − a0 t, 0), (2.34)

(M − S̃)(x, t) = (M − S̃)(x + a0 t, 0), (2.35)

obtained from the differential equations, is still valid even when the differentiability of the
individual variables M and S̃ is not fulfilled.
To show this, consider a charasteristic Γ+
1 in the (t, x)-plane, and let the Riemann invariant
on this characteristic have the value J1+ . So J1+ = M + S̃ and is constant along Γ+ 1.
Although M + S̃ is constant, the individual variables M and S̃ may vary along Γ+ 1 , and
the variations may still contain discontinuities.
Assume that M is discontinuous showing a jump ∆M between A and B, see figure 2.6, so
that MB = MA + ∆M .
To satisfy the condition J1+ is constant, the variable S̃ also jumps between A and B:

S̃B = S̃A + ∆S̃, with ∆S̃ = −∆M. (2.36)

‘ ’ This behaviour is sketched in figure 2.7, which shows J1+ , M and S̃ along Γ+
1.
Discontinuous solutions of the system (2.9) - (2.10) have to satisfy the jump relation

∆S̃ + ∆M = 0 on Γ+ , (2.37)

and similarly

∆S̃ − ∆M = 0 on Γ− . (2.38)
44 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.7:

Figure 2.8:
2.2. METHOD OF CHARACTERISTICS, M.O.C. 45

The discontinuity that is observed in point A, B on Γ+ 1 is not isolated in the (t, x)-plane
−
but is carried along the Γ -characteristic going through A and B, see figure 2.8. This is
immediately clear if one applies the condition that M − S̃ is constant along Γ− characteristics
Γ− − − −
A and ΓB . Take the neighbouring points P and Q on ΓA and ΓB respectively. This gives
the following equations:
along Γ−
A MP − S̃P = MA − S̃A , (2.39)
along Γ−
B MQ − S̃Q = MB − S̃B = MA + ∆M − S̃A − ∆S̃. (2.40)
Substraction gives:
MQ − MP − S̃Q + S̃P = ∆M − ∆S̃. (2.41)
Since ∆S̃ + ∆M = 0, this relation becomes
MQ − MP − S̃Q + S̃P = 2∆M. (2.42)
Since P and Q are on the same Γ+ -characteristic, we have
MP + S̃P = MQ + S̃Q ,
or
MQ − MP + S̃Q − S̃P = 0. (2.43)
Combining (2.42) and (2.43), there follows
MQ − MP = ∆M, (2.44)
and similarly
S̃Q − S̃P = ∆S̃. (2.45)
Conclusion:
Characteristics serve as discontinuity lines.
To get a formal affirmation of this conclusion consider again the set of equations (2.9) and
(2.10), and apply the jump equation for one-dimensional unsteady flow: equation (1.39). The
jump equation applied to S̃t + a0 Mx = 0 gives:
VD ∆S̃ = a0 ∆M ; (2.46)
the jump equation applied to Mt + a0 S̃x = 0 gives:
VD ∆M = a0 ∆S̃. (2.47)
Combining both equations and solving for the discontinuity velocity VD yields:
VD2 = a20 , (2.48)
or
VD = ±a0 . (2.49)
Conclusion:
In linear acoustics, characteristics are candidates to be discontinuity line D.
46 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.9: Piston problem

2.3 Piston problem

Consider a constant area tube, filled with gas at rest. The rest conditions, labelled with
subscribt ‘0’ are pressure p0 , density ρ0 , sound velocity a0 , temperature T0 , etc. The tube is
closed off by a movable piston at the left end. The right end of the tube extends to infinity.
The piston, at rest at t < 0, starts moving at t = 0. The piston velocity Up (t) and the
piston position Xp (t) may depend on time and are assumed to be small. Further, the position
of the piston Xp at time t is given by
Z t
Xp (t) = Up (τ )dτ. (2.50)
0

Problem:
Find for a prescribed piston movement, for example if the piston is oscillating, the
impact it has on the gas that was initially at rest. How does the gas come into
motion and what is the resulting flow in the tube?
To discuss this problem, draw a picture in the (t, x)-diagram, see figure 2.9. The first message
telling the gas at x > 0 that the piston has started moving is the sound wave with path
x − a0 t = 0.
The gas contained at x − a0 t > 0 is still undisturbed, but the gas contained at x − a0 t < 0
experiences the moving (oscillating) piston and is disturbed. So in particular on t = 0 we have
M = 0 and S̃ = 0.
Let us find the flow variables (M and S̃) in a point À in the disturbed zône, using the
backward M.O.C. method.
Drawing Γ− through À backward in time to t = 0 goes as usual, yielding the intersection
point Á, but drawing Γ+ backward in time we hit the piston in point Â. Unlike the x-axis
where M and S̃ are known (i.e. M = 0 and S̃ = 0), at the piston path only one flow variable
2.3. PISTON PROBLEM 47

U
is known: the piston velocity Up or Mp = a0p . Fortunately the missing piece of information is
obtained by drawing a Γ− characteristic through (3) backward in time to t = 0 onto point Ã
on the initial value line.
This provides enough information to find M and S̃ in point À. Writing down the equations
valid along characteristics:

along Γ−
(1) : M1 − S̃1 = M2 − S̃2 = 0 (2.51a)
along Γ+
(1) : M1 + S̃1 = M3 + S̃3 (2.51b)
along Γ−
(3) : M3 − S̃3 = M4 − S̃4 = 0 (2.51c)
on piston path: M3 = Mp (t3 ) (2.51d)

The actual position of the points Á and à are immaterial as long as they lie in the undisturbed
zône.
Rt
The point (x3 , t3 ) is the intersection of the piston path: Xp (t) = 0 Up (τ ) dτ with the
Γ+ -characteristic going through À:

x − a0 t = x1 − a0 t1 (2.52)

In general, finding (x3 , t3 ) results in solving a trancedental equation. Fortunately, we are

considering small disturbance theory (linear acoustics) so that it is allowed to assume that the
piston excursion x3 is small. Therefore we may take x3 = 0 and we find from equation (2.52):
x1
t3 = t1 − . (2.53)
a0

Now, system (2.51a)-(2.51d) can be solved:

step 1: t3 is inserted in (2.51d) ⇒ Mp ⇒ M3
step 2: (2.51c) ⇒ S̃3 = M3
step 3: (2.51a) and (2.51b) ⇒ M1 − S̃1 = 0, M1 + S̃1 = 2M3
This leads to the solution M1 = M3 , S̃1 = M3 or,
   
x x
M1 (x, t) = Mp t − , S̃1 (x, t) = MP t −
a0 a0

which tells us that the values of the state variables at the piston in  are transported without
change along Γ+ into the gas. In other words, the velocity in a certain position x at time t is
just equal to the velocity of the piston at time t − ax0 ; ax0 is precisely the traveling time for a
sound wave propagating from the piston to the location x.
Notice that in this case, not just J + = M + S̃ is constant along Γ+ but the individuals M
and S̃ are constant too along the Γ+ -characteristic.
A flow in which there is really only one Riemann invariant varying is called a “simple wave”.
Indeed in the example discussed above we have a simple wave since the Riemann invariant
J − = M − S̃ is constant throughout the flow since all characteristics emanate from the initial
line t = 0 where the conditions are constant (i.e. M = 0, S̃ = 0 → J − = 0). Thus J − is
constant in the whole (t, x)-domain, only J + is varying.! (More on simple waves may be found
in chapter 3 of the lecture notes).
48 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.10:

2.3.1 Specific Example: A sinusoidal excursion of the piston for only one
single period.
The theory just presented enables us to describe the unsteady gas motion resulting from a
prescribed piston movement. To be specific, assume:
(
a0 sin ωt, 0 ≤ t ≤ 2π
ω ,
Up = (2.54)
0, for any other time,

with  small. Equation (2.54) can be written as

2π
Mp =  sin ωt, 0≤t≤ . (2.55)
ω

The piston path

Between t = 0 and t = 2π ω the piston path becomes:
Z t
a0
xp (t) = up (τ ) dτ = (1 − cos ωt). (2.56)
0 ω
The piston moves forward, then backward returning to its starting position at x = 0. When
the piston stops at t = 2π
ω , again an undisturbed region (i.e. Â) results, see figure 2.3.1.
The solution in the various domains becomes:
domain À : (x − a0 t) > 0 : S̃ = 0, M = 0; (2.57)
x
domain Á : −a0 2π
ω < (x − a0 t) < 0 : M (x, t) =  sin ω(t − ), S̃ = M ; (2.58)
a0
domain  : (x − a0 t) < −a0 2π
ω : M = 0, S̃ = 0. (2.59)

Particle paths
Consider a particle ‘i’ at the initial postion xi (0) = xi,0 . The position of the particle can be
described by:
2.3. PISTON PROBLEM 49

xi,0
for 0 < t < : xi (t) = xi (0) = xi,0 (2.60)
a0
Zt Zt Zt   
xi,0 xi,0 2π xi,0
for <t< + : xi (t) = ui udτ = a0 Mi dτ = a0 ε sin ω τ − dτ
a0 a0 ω a0
xi,0 xi,0 xi,0
a0 a0 a0

(2.61)
or
   
a0 ε xi,0
xi (t) = xi,0 + 1 − cos t − ω (2.62)
ω a0

All paths have similar shapes; there is only a time shift. Note that the piston path is found
as the path of the particle having xi,0 = 0.

Disturbed gas motion at t = 2π/ω

Figure 2.11: Sinusoidal wave introduced by the piston

The piston introduces a sinusoidal wave into the gas at rest; the wave travels to the right
with the speed of sound a0 . This process is depicted in figure 2.11. This figure enables us to
describe what an observer located at time x = a0 2π ω experiences as time progresses.
Until t = 2π
ω the gas is still at rest and the observer is not aware of the fact that the piston
has started moving already 2π ω time units ago. However at t = 2π
ω things are changing since
2π
the gas at x = a0 ω starts moving. The observer experiences a gas flow moving to the right
until t = 3π
ω . At that particular time instant the gas flow is stopped and now starts moving
to the left. Until t = 4πω the observer experiences this left moving flow. Beyond t = ω the
4π

gas is at rest again.

From figure 2.11 we observe the following:

1. discontinuities (in slope) are carried along the characteristics;

50 CHAPTER 2. LINEARIZED FLOW EQUATIONS

2. discontinuities are not smoothened during travelling; there is no physical mechanism

such as viscous dissipation that could do that job. In absence of viscous dissipation
discontinuities survive.

2π
3. An initial wave, e.g. t = ω cannot be steady; it has to move with the speed a0 .

2.3.2 Example: a tube with two pistons

Figure 2.12:

The Method of Characteristics is generally applicable. For example in the case of a finite
tube with length L having pistons at both ends. If both pistons follow a prescribed path the
motion of the gas between both pistons can be determined using the M.O.C.
In the (t, x)-plane several domains can be distinguished; domain À is undisturbed; domains
Á and  are simple waves since J − and J + are constant in Á and  respectively. Domain Ã
is a non-simple region; here both Riemann invariants J + and J − vary. In a non-simple region
the variables M and S̃ are not constant along Γ+ or Γ− .

2.4 Riemann’s initial-value problem

Riemann’s initial-value problem, or Riemann’s problem shortly, is one of the most fundamental
problems in Gasdynamics and in particular in the description of unsteady gas motions. Apart
from its great value to understand these physical phenomena it has also a dominant role in
the development of CFD methods to solve these problems.
Riemann’s problem is formulated as follows: on the initial line e.g. t = 0 two uniform
2.4. RIEMANN’S INITIAL-VALUE PROBLEM 51

states are separated by a discontinuity, for example at x = 0; thus

x < 0 :M = M1 , S̃ = S̃1 ;
(2.63)
x > 0 :M = M4 , S̃ = S̃4 .

Take an arbitrary point P (t, x) and use backward M.O.C. to get the solution:
1 1
MP = (M + S̃)A + (M − S̃)B ;
2 2 (2.64)
1 1
S̃P = (M + S̃)A − (M − S̃)B ,
2 2
where A and B are points on the initial value line, see figure 2.13.
Let us evaluate this solution in the (t, x)-plane. Depending on the location of P in the
(t, x)-plane, we observe three different solutions.

Figure 2.13:

If P (t, x) lies in the domain À then A and B are both on x < 0. This implies:

MA = MB = M1

and

S̃A = S̃B = S̃1 .

If P (t, x) lies in domain à then A and B are both on x > 0, implying

MA = MB = M4 ,

and

S̃A = S̃B = S̃4 .

Finally if P lies in the domain above the Γ+ and Γ− characteristics emanating from the origin
x = 0, then xA < 0 and xB > 0. Then we have MA = M1 , S̃A = S̃1 , MB = M4 and S̃B = S̃4 .
The preliminary notation for this domain is 2+3 .
52 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Using the general solution given in (2.64) the solution of Riemann’s problem becomes:

domain À : MP = M1 ,
S̃P = S̃1 ;
M1 + M4 S̃1 − S̃4
domain Á + Â : MP = + ,
2 2 (2.65)
S̃1 + S̃4 M1 − M4
S̃P = + ;
2 2
domain à : MP = M4 ,
S̃P = S̃P .

Let us now discuss some examples of Riemann problems.

2.4.1 Example 1: a density discontinuity; (M, S̃)-diagram

The first example concerns the time evolution of a density discontinuity in a gas at rest. The
initial values at t = 0 are shown in figure 2.14.

Figure 2.14: Initial values at t = 0

These initial data resemble the following physical situation present in a tube at t ≤ 0 as
shown in figure 2.15.
We observe two uniform states in a tube being separated by a diaphragm at x = 0, the
standard setup for the so-called shock tube problem. Left from the diaphragm the pressure is
p0 and the density is ρ0 ; right from the diaphragm the pressure is p0 + ∆p and the density is
ρ0 + ∆ρ. Because of the assumption of reversibility ∆p and ∆ρ are coupled by ∆p = a20 ∆ρ.
On both sides of the diaphragm the gas is at rest: u = 0 ⇒ M = 0.

Question: What happens if the diaphragm at x = 0 is withdrawn suddenly at t = 0?

2.4. RIEMANN’S INITIAL-VALUE PROBLEM 53

Figure 2.15: Initial flow situation for t ≤ 0

Answer: The answer to this question is found by solving the Riemann problem with initial
data M1 = M4 = 0, S̃1 = 0, S̃4 = ε. Using (2.65) one finds:

domain À M = 0, S̃ = 0;
1 1
domain Á and  M = − ε, S̃ = + ε;
2 2
domain à M = 0, S̃ = ε.

The solution tells us that in domains À and à the gas remains unperturbed with respect
to its initial conditions. In domain 2+3 the gas moves with velocity u = a0 M = −a0 ε/2.
The (t, x)-diagram together with the behaviour of M (x) and S̃(x) at a certain time level,
t∗ , is shown in figure 2.16.

Figure 2.16: Evolution of M and S̃ after diaphragm withdrawal

The figure also shows some particle paths: in À and à the gas is at rest, so the particle
paths are straight lines, described by x = constant. In 2+3 particles move with velocity
54 CHAPTER 2. LINEARIZED FLOW EQUATIONS

u = a0 M = −a0 ε/2 to the left. This makes sense since the pressure drops from right to
left. Again we see straight lines but now they are inclined to the left.
The path of the particle initially at the discontinuity x = 0 divides 2+3 in separate (sub-)
domains Á and Â. Domain Á contains the gas that was originally at x < 0. Domain Â
contains the gas that was originally at x > 0.
Note that the characteristic (or wave) separating À and Á is a compression wave; it
moves to the left into À and compresses the gas from state À into state Á. Similarly,
the characteristic (or wave) separating  and à is an expansion wave; it moves to the
right while expanding state à into state Â.
The conditions in the perturbed states Á and  are equal; gases in Á and  are in the
same state. This is a result typical for linear acoustics. In a nonlinear treatment (see
chapter 3) the states Á and  are generally different and the solution of the Riemann
problem then generates four uniform domains.

Now we will solve this Riemann problem differently by using (M, S̃)-diagrams. Let us

Figure 2.17: (t, x)-diagram Figure 2.18: (M, S̃)-diagram

consider the (t, x)-diagram (see figure 2.17), and observe that:

• À, Á,  and à are uniform states. Uniform means there are no changes of M and S̃ in
the same region due to the constant conditions at the initial line.

• Domains À, Á and  are connected by Γ+ characteristics along which M + S̃ = constant.

This implies that the images (locis) of the domains À, Á and  all lie on the same straight
line in a (M, S̃)-diagram. The slope of the line is −1.

• Similarly, domains Á,  and à are connected by Γ− -characteristics on which M − S̃ is

constant. Domains Á,  and à have images (loci) in the (M, S̃)-diagram that lie on the
same straight line with slope +1.

• Since we are dealing with homentropic linear theory, the states Á and  are equal, so
they have the same spot in the (M, S̃)-diagram.

The solution of the Riemann problem with initial data M1 = 0, S̃1 = 0, M4 = 0, S̃4 = ε
is found as the intersection of the locus of all possible states Á obtainable from À, with the
2.4. RIEMANN’S INITIAL-VALUE PROBLEM 55

locus of all possible states  obtainable from Ã, see figure 2.18. From the figure the solution
is easily obtained as:
1
M2 = M 3 = − ε
2
1
S̃2 = S̃3 = ε
2

2.4.2 Example 2: two colliding flows

Next consider a Riemann problem with initial values at t = 0 as shown in figure 2.19:
State 1: x < 0, M1 = +ε, S̃1 = 0,
(2.66)
State 4: x > 0, M4 = −ε, S̃4 = 0,
These initial data resemble the flow situation that occurs if two gases in the same state

Figure 2.19: Two colliding flows; initial data

(S̃1 = S̃4 = 0) move toward each other with the ’same’ velocity u1 = +εa0 , u4 = −εa0 , and
meet at x = 0. This situation can be maintained in a tube for t ≤ 0 by means of a diaphragm,
but it can be created dynamically for one instant (t = 0) as explained under example 4. Also
the solution of this Riemann problem will be found by using the (M, S̃)-diagram technique.
The initial data M1 = ε, S̃1 = 0, M4 = −ε, S̃4 = 0 result into the solution M2 = M3 = 0
and S̃2 = S̃3 = ε, see figure 2.20 It is seen that the two colliding gas streams create two
compression waves; in between these, the motions of the gases cancel each other while the
density rises.

2.4.3 Example 3: gas streams running away from each other

This case is similar to example 2, but with the flow velocities reversed; the gases run away
from each other with equal speeds but opposite directions.
Initial values (see figure (2.21):
State 1: x < 0, M1 = −ε, S̃1 = 0,
(2.67)
State 4: x > 0, M4 = +ε, S̃4 = 0,
56 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.20: Flow collision depicted in (M, S̃)-diagram

The solution in shown in figure 2.22. The motion of the gases creates two expansion waves;
in between these the motions of the gases again cancel each other while the density is now
reduced.
2.4. RIEMANN’S INITIAL-VALUE PROBLEM 57

Figure 2.21: Gas streams running away; initial data

2.4.4 Example 4: Moving piston

In this example a Riemann problem is created dynamically in a tube of length l by the action
of pistons at both ends of the tube (initially at x = ± 2l ). These pistons are set in motion
instantaneously at t = − 2al 0 . This creates sound waves which meet in the middle of the tube at
t = 0. At that instant a Riemann problem is created, see figure 2.23. Take for exampe the left
and right piston speeds equal to UP,L = 2εa0 and UP,R = εa0 , acting on an initially uniform
fluid at rest. Calling this initial state 0 we have M0 = 0, S̃0 = 0. The states created by the left
and the right pistons are labelled Àand Ãrespectively, in anticipation of the resulting Riemann
problem. The states can be found in the (M ,S̃)-diagram, see figure 2.24 by the intersection
of lines. To find state À, draw the locus of all possible states obtainable from state 0 by a
right moving wave, and intersect this by the vertical line M = MP,L = 2ε. Similarly, state
Ãis found by drawing the locus of all possible states obtainable from 0 by a left running wave
and intersecting this by the vertical line M = MP,R = ε.
The solution of the ensuing Riemann problem, i.e. finding states Áand Âcan be completed in
the same diagram. The results are:

State 1: M1 = 2ε, S̃1 = 2ε,

State 4: M4 = −ε, S̃4 = −ε (2.68)
State 2,3: M2,3 = 3ε, S̃2,3 = ε.

Note that states Áand Âhave a larger velocity than either piston. This is caused by the extra
pressure difference (p1 > p4 ) created by the pulsing/pulling actions of the pistons.
58 CHAPTER 2. LINEARIZED FLOW EQUATIONS

Figure 2.22: Gas streams running away

2.4. RIEMANN’S INITIAL-VALUE PROBLEM 59

Figure 2.23: Flow caused by the action of two pistons, (t, x)-diagram

Figure 2.24: Flow caused by the action of two pistons; (M ,S̃-diagram)

60 CHAPTER 2. LINEARIZED FLOW EQUATIONS
Chapter 3

One-dimensional unsteady non-linear

flows

3.1 Characteristic equations

Let us start from the one-dimensional time-dependent Euler equations in non-conservative
form, see (1.33), (1.34) and (1.35). Further, we assume reversibility (smooth flows) so that
according to (1.36) the entropy is constant when moving with a fluid particle. Along a particle
path the entropy is constant but it may differ from one particle to another. Due to the
assumption of reversibility, the energy equation (1.36) is replaced by (1.38) expressing s =
constant along a particle path.

ρt + uρx + ρux = 0, (3.1)

1
ut + uux + px = 0, (3.2)
ρ
st + usx = 0. (3.3)
Since ds ∝ (dp − a2 dρ), see (1.12), equation (3.3) may be written in terms of p and ρ:
pt − a2 ρt + u(px − a2 ρx ) = 0,
or
pt + upx − a2 (ρt + uρx ) = 0. (3.4)
Using (3.1) the ρ-derivatives can be eliminated, yielding
pt + upx + a2 ρux = 0. (3.5)
We would like to make combinations of this equation with (3.2), as we did in the linear case.
Therefore, divide (3.5) by ρa, so that it gets the same physical dimension as (3.2), yielding:
pt px
+ u + aux = 0,
ρa ρa
repeat:
px
ut + uux + a = 0; (3.2)
ρa

61
62 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

adding and substracting these equations yields:

 
pt px
ut + + (u + a) ux + = 0, (3.6)
ρa ρa
 
pt px
ut − + (u − a) ux − = 0. (3.7)
ρa ρa

Introducing the Riemann invariants J + and J − by:

dp
dJ + = du + , (3.8)
ρa
dp
dJ − = du − , (3.9)
ρa

then (3.6) and (3.7) become

∂J + ∂J +
+ (u + a) =0, (3.10)
∂t ∂x
∂J − ∂J −
+ (u − a) =0. (3.11)
∂t ∂x
Apparently there are still two quantities J + and J − transported unchanged along the char-
acteristics Γ+ and Γ− respectively; the characteristic speeds are now (u + a) and (u − a),
indicating that the signals travel with speeds ±a with respect to the gas which has a speed u.
J + and J − are the Riemann invariants, in general they cannot be given in integrated form,
so we have to regard J + and J − as functions of the independent variables x and t; J + and
J − are determined at most by differential relations as given in equations (3.8) and (3.9), Γ+
and Γ− are characteristics defined by:
dx
Γ± : = u ± a. (3.12)
dt
Since the characteristic speeds are no longer constant but depend on the solution, the char-
acteristics are generally curved in the non-linear case.
In the homentropic case all particles have the same entropy, the relation between pressure
and density is given by Poisson’s equation (1.13):

p = Cργ .
dp
In that case the term ρa can be written as

dp 2 da
= .
ρa γ−1

So for homentropic flow equations (3.8) and (3.9) can be integrated:

2
J± = u ± a. (3.13)
γ−1

Since the entropy equation (3.3) has the same form as the equations for the Rieman invariants,
we consider that equation also among the characteristic equations and call the particle path
3.1. CHARACTERISTIC EQUATIONS 63

a charcteristic Γ0 . The total set of characteristic equations may therefore be written as:

 
∂ ∂
+ (u − a) J − = 0, (3.14)
∂t ∂x
 
∂ ∂
+ (u) s = 0, (3.15)
∂t ∂x
 
∂ ∂
+ (u + a) J + = 0, (3.16)
∂t ∂x

or

− dx
dJ = 0 along Γ−
with = u − a, (3.17)
dt
 
0 dx
ds = 0 along Γ with = u, (3.18)
dt
 
dx
dJ − = 0 along Γ+ with = u + a. (3.19)
dt

The latter equations can also be used if the inital values and also the corresponding solution
are ∈ C 0 only in certain points; the entropy distribution may actually contain discontinuities.
The characteristics Γ± will abruptly change their slope when they cross a Γ0 that carries an
entropy jump. Three characteristics can be drawn through an arbitrary point in the (t, x)-
space, see figure 3.1.

Figure 3.1: The three characteristics

64 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.2: Domain of influence of P (x, t)

Domain of influence (DOI) and domain of dependence (DOD)

Figure 3.3: Domain of dependence of Q(x, t).

In linear acoustics (chapter 2) we have seen that a point in (t, x)-space influences only the
solution along the two characteristics Γ+ and Γ− through that point. In the non-linear case
however the influence spreads over a 2-D domain due to the dependence of the characteristic
slopes on the solution itself.
Figure 3.2 shows the domain of influence in the (t, x)-space of an arbitraty point P (x, t).
Any point in the domain of influence of P (x, t) is affected by an infinitesimal perturbation of
variables in P . In figure 3.3 the domain of dependence of an arbitrary point Q(x, t) is shown.
Any point in the domain of dependence of Q(x, t) affects the solution in Q.
Both domains are interrelated as follows: the domain of dependence of Q contains all
points which have Q in their domain of influence; the domain of influence of P contains all
points that have P in their domain of dependence.
3.2. SIMPLE WAVES, NON-LINEAR 65

3.2 Simple waves, non-linear

Simple waves are a special class of solutions of the characteristic equations. As in linear theory
a simple wave appears if one of the Riemann invariants J + or J − is constant in a certain
domain. Let us examine some properties of simple waves and assume that J − = constant;
furthermore we assume homentropic flow:
2a
J− = u − = J0− = const. (3.20)
γ−1
Observe that a linear relation exists between the variables u and a. The Riemann invariant
J + becomes
2a 4a
J+ = u + = 2u − J0− = + J0− . (3.21)
γ−1 γ−1

Thus, in a simple wave along the characteristic Γ+ both u and a must be constant. Hence,
along Γ+ , u + a is constant as well, so we may write:
∂ ∂
(u + a) + (u + a) (u + a) = 0, (3.22)
∂t ∂x
which exactly states that in a simple wave having J − = constant the state variable (u + a) is
constant along curves: Γ+ : dx
dt = u + a.
This implies that the characteristic Γ+ itself must be a straight line.

Conclusion: In a simple wave having J − constant the Γ+ characteristics are straight lines
and similarly: in a simple wave having J + constant the Γ− characteristics are straight lines.
The ’J − = constant’ simple waves are called forward running waves. The flow variables in
the simple wave are governed by equation (3.22). Using the state quantity v = u+a, being the
characteristic speed of Γ+ , this equation is the inviscid form of the so called Burgers equation:

vt + vvx = νvxx ; (3.23)

the right-hand-side expresses the contribution of viscosity.

So in a forward simple wave (J + is the only variable) the 1-D Euler equations reduce to a
single equation having the form

vt + vvx = 0. (3.24)

The same equation with w = u − a is obtained for a backward running simple wave (J − is
variable and J + is constant):

wt + wwx = 0 (3.25)

In this case the Γ− characteristics are straight lines in the (t, x)-plane.

Curiosity In the unrealistic situation that γ = 3 the characteristic equations (3.14) and
(3.16) take the form of two decoupled inviscid Burgers equations because J ± = u ± a in that
case. Hence for γ = 3 the Γ+ and Γ− characteristics are all straight lines. (Useful test case
for numerical gas dynamics!)
66 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

3.3 Simple waves created by a moving piston

Consider a semi-infinite constant area tube filled with a gas at rest, labelled “0”. At the tube’s
left end (x = 0) a moving piston is located; at the right the tube extends to infinity:

Figure 3.4: Simple wave in a one-dimensional flow

The piston is smoothly accelerated backwards (x < 0) from t = 0 to t = T , at which time

it has reached a negative speed UT . From that time on the piston moves with constant speed
UT further backward. Figure 3.4 shows the (t, x)-diagram.
The first signal telling the gas that the piston started to move travels along the Γ+ -
characteristic x = a0 t. The last signal telling the gas that the piston stopped accelerating
travels along the Γ+ characteristic going through the point A (see figure 3.4) located on the
path of the piston. This characteristic has a slope dx
dt = UT + aT . The flow between the first
−
and the last signal is a simple wave because J is constant in the whole domain.
All Γ+ -characteristics are straight lines. Beyond the last signal a domain appears having
constant conditions: u = UT , a = aT . The value of aT follows from

2a 2aT 2a0
J− = u − = UT − =− ,
γ−1 γ−1 γ−1

yielding

γ−1
at = a0 + UT . (3.26)
2
The slope of the Γ+ -characteristics in the uniform domain is UT + aT , which yields that
 
dx γ+1
= UT + aT = a0 + UT . (3.27)
dt Γ+ 2

The solution in the simple wave region is governed by the following set of equations, which
is obtained as follows; take an arbitrary point À in the simple wave and let point  be the
intersection of the Γ+ -characteristic through À and the piston path. Assume that the velocity
of the piston up (t) is prescribed; so if t3 should be known then u3 = up (t3 ) is also known. We
3.3. SIMPLE WAVES CREATED BY A MOVING PISTON 67

already knew that u and a are constant along Γ+ , hence

u1 = u3 = up (t3 ) (3.28)
γ−1
a1 = a3 = a0 + up (t3 ) (3.29)
2
The solution given by equations (3.28) and (3.29) depends on the actual value of t3 .
How to find t3 for a given point À ? From figure 3.4 we observe that point  is found
as the intersection of the piston path xp (t) with the Γ+ -characteristic through À: The piston
path:
Z t
xp (t) = up (τ ) dτ (3.30)
0

The Γ+ (1) characteristic is given by:

xp − x1 = (up + a3 )(t3 − t1 ) (3.31)

Knowing the prescribed function up (t) and using (3.29) (to get a3 ) the equations (3.30) and
(3.31) can be solved to find t3 .
Note: if up (t) is linear in t, for 0 ≤ t ≤ T , then xp (t) is quadratic in t and one has to
solve a quadratic equation to find t3 . But in general solving (3.30) and (3.31) will require a
numerical rootfinder.
In the case t3 ≥ T , we always have

u3 =up (t ≥ T ) = UT , (3.32)
γ−1
a3 =aT = a0 + UT . (3.33)
2
Equation (3.33) leads to the interesting observation that if UT takes the value
2
UT = a0 (3.34)
γ−1
we get a3 = 0 and also p3 = 0 and ρ3 = 0. What does that mean?
The piston has been pulled so fast that the gas has expanded itself to the limit of vacuum.
Pulling the piston away even faster will create a vacuum between the piston and the gas, since
the latter can no longer follow the piston, see figure 3.5.
2
The speed γ−1 a0 is the maximum speed that a gas particle from its rest state (u = 0, a =
a0 ) can reach. For air (γ = 1.4) the maximum speed is umax = 5a0 .

Comments
• Piston withdrawal in a tube filled with gas at rest creates an expanding flow being a
simple wave. Simple waves can also be created by moving the piston into the gas. Then
a compression simple wave appears, such a wave would not last for ever but a shock is
formed which violates the prerequisites (constant entropy and J − ) for a simple wave.

• A simple wave-region always borders at least a uniform region; the boundary is always
a straight characteristic. Notice that in Fig. (3.4) the simple wave is bordered by two
uniform regions, i.e. the undisturbed region u = 0, a = a0 and the region beyond the
last signal having u = uT , a = aT = a0 + γ−1
2 UT .
68 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.5: Expansion caused by a piston withdrawal

• Uniform regions can be distinguished with respect to Mach number: M = u/a and its
influence on the slope of the Γ± characteristics: dx dt = u ± a respectively. The various
possibilities that can appear in the (t, x)-plane are shown in the following diagram.
3.3. SIMPLE WAVES CREATED BY A MOVING PISTON 69

Table 3.1: Various uniform flow regions in the (t, x)-plane

Type of flow u u + a u − a Characteristics in (t, x)-plane

Supersonic
flow to the u>a >0 >0
right

Sonic
flow to the u=a >0 =0
right

Subsonic flow −a < u < a >0 <0

Sonic
u = −a =0 <0
flow to the left

Supersonic
u < −a <0 <0
flow to the left
70 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

3.4 Centered expansion wave

3.4.1 Flow variables in the centered expansion wave
A centered expansion wave results if the piston is withdrawn instantaneously to the speed UT .
The centred expansion may be regarded as the limit case of the previous case when T is small
with respect to time intervals we are interested in, i.e. extend figure 3.4 to large x and t and
look at the expansion wave from far. Then we observe a picture that is shown in figure 3.6.

Figure 3.6: Centered expansion wave due to instantaneous piston withdrawal

The center of the wave at O is now a singularity; because J + is constant on the Γ+

characteristics, J + is multi-valued in O; together with J + also the flow variables u, a, p, ρ are
multi-valued.
The solution in the centered wave region, i.e. domain À, is found by considering a Γ− -
characteristic going through a point 1 in domain À:

2 2a0
Γ−
(1) : u1 − a1 = constant = − . (3.35)
γ−1 γ−1

Furthermore we use the fact that 1 lies on a characteristic Γ+ having a slope

dx x1
= u1 + a1 = . (3.36)
dt t1

From equations (3.35) and (3.36) we find the solution:

 
2 x1
u1 = − a0 . (3.37)
γ+1 t1
3.4. CENTERED EXPANSION WAVE 71

Figure 3.7: u(x) and a(x) at constant t

This shows that the disturbance of u is linear in x for constant t. A similar conclusion results
for the variable a, see figure 3.7.
For an arbitrary point in domain À we now omit the subscript 1; the solution for u and
also for other flow-variables in terms of x/t becomes:
2 x 
u= − a0 (3.38)
γ+1 t
γ−1x 2
a= + a0 (3.39)
γ+1 t γ+1
x
u+a= (slope of Γ+ ) (3.40)
t
3−γx 4
u−a= − a0 (slope of Γ− ) (3.41)
γ+1 t γ+1
From equation (3.39) and constant entropy we find the distibutions for ρ and p:
  2
ρ(x, t) a γ−1
= (3.42)
ρ0 a0
  2γ
p(x, t) a γ−1
= (3.43)
p0 a0
The Mach number in the simple wave is now equal to:
 2 x
 
x
 γ+1 t − a0   2 t − 2a0 
  
M =  γ−1 x = x (3.44)
 + 2 a0   (γ − 1) t + 2a0 
γ+1 t γ+1

From this formula it is evident that at

x
= a0 ⇒ M =0
t
x
=0 ⇒ M =1
t
x
≤0 ⇒ M ≥1
t
72 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

3.4.2 Particle path in a centered simple wave

The gas particle residing at x0 is hit by the first wave of the expansion fan when t = t0 where
t0 = x0 /a0 , see figure 3.8.

Figure 3.8: Particle path

At that time instant the particle starts to move. As long as the particle is in the expansion
wave its path X : X(t) follows from the differential equation:
 
dX 2 X
=u= − a0 (3.45)
dt γ+1 t
with initial condidtion
x0
X = x0 at t = = t0 .
a0
Experience in solving this type of equation teaches us that the solution is a linear combi-
nation of linear terms in t (caused by the constant in the right hand side term) and a t2/(γ+1)
term (caused by the x/t term). Introduce the dimensionless quantities:
X t
ξ= = X/(a0 t0 ), τ= ; (3.46)
x0 t0
Equation (3.45) may then be written as:
dξ 2 ξ
= ( − 1). (3.47)
dτ γ+1 τ
To solve (3.47) try the following solution:
2
ξ = ατ γ+1 + βτ
3.5. RIEMANN PROBLEM, NON-LINEAR 73

γ+1 2
and verify that α = γ−1 and β = − γ−1 satisfy the initial conditions, ξ = 1 and τ = 1.
Hence the particle that starts moving at t = t0 from its position x0 follows the path:
  2
X(t) γ+1 t γ+1 2 t
= − . (3.48)
a0 t0 γ−1 t0 γ − 1 t0

This path is followed as long as the Γ+ -characteristic with slope UT + aT is not crossed, or in
mathematical terms:
 
γ+1
UT + a0 t ≤ X(t) ≤ x0 . (3.49)
2
dx
After crossing the particle path is a straight line with slope dt = UT .

3.5 Riemann problem, non-linear

We encountered the Riemann problem in the case of linear acoustics. In non-linear theory it
is formulated in a similar way and is stated as follows: find the solution of the 1-D, unsteady
non-linear flow equations with constant initial conditions except for a single jump discontinuity
(say at t = 0).
Initial data at t = 0:
 
u1 for x<0 a1 for x<0
u(x, 0) = , a(x, 0) =
u4 for x>0 a4 for x>0

The solution of the Riemann problem in the full non-linear case differs from the linear case in
three aspects:

1. There appears an extra entropy wave carrying an entropy jump. This wave travels with
characteristic speed u so that its path is a characteristic Γ0 . There are now really four
different domains: the pre-state domains À and à and the post-state domains Á and
Â. The entropy wave separates Á and Â. Frequently these domains are also labelled:

L ≡ À, L∗ ≡ Á, R∗ ≡ Â, R ≡ Ã.

The entropy wave appears as a contact discontinuity.

2. The two acoustic waves are now non-linear, meaning that they appear either as a shock
(discontinuity) or as an expansion.

3. The expansion is a fan occupying some domain in space-time, so it is not a discontinuity.

The solution techniques used for linear and for non-linear Riemann-problems are remark-
able similar. In linear theory the (M, S̃) diagram technique was useful; in the non-linear case
the ‘velocity-pressure’ or (u, p)-plane is very appropriate. The reason that we now use (u, p)
instead of (M, S̃) is that the post-states Á and  must have the same velocity and the same
pressure because Á and  are seperated by a contact discontinuity.
In preparing ourselves to solve the non-linear Riemann-problem, let us first consider shocks
and expansions running into a prescribed ‘pre-state’ (À or Ã) and let us look in particular for
equations involving the post-wave pressure.
74 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.9: Shock advancing in a gas

3.5.1 Shock moving into a gas; Hugoniot curve

Consider a shock wave, for example generated by a moving piston, that advances with speed
cs into a gas which is in the ’pre-state’ Ã having prescribed a pressure p4 and a sound speed
a4 .
Assume for now that the gas in the ‘pre-state’ Ã is at rest: u4 = 0.
Due to the moving shock, see figure 3.9, the pressure rises and the gas is brought into
motion, thus: p3 > p4 and u3 > 0.
How is the pressure rise ∆p = p3 − p4 be related to the velocity increase ∆u = u3 − u4 ?
Consider the process in a shock frame and apply the Rankine-Hugoniot relations for mass
and momentum (for now take u4 = 0):

mass: ρpre cs = ρpost (cs − u3 ) = m;

momentum: ppre + ρpre c2s = ppost + ρpost (cs − u3 )2 .

Inserting the mass flux m into the momentum equation gives:

ppre + mcs = ppost + m(cs − u3 ),

or

∆p = ±m∆u (3.50)

The + sign regards a shock moving to the right and the − sight is for a shock moving to the
left.
From equation (1.50) the shock velocity cs is calculated from the pressure rise as:
s
γ + 1 ∆p
|cs | = apre 1 + (3.51)
2γ ppre
3.5. RIEMANN PROBLEM, NON-LINEAR 75

so that the mass flux m is determined as

s
γ + 1 ∆p
m = ρpre |cs | = ρpre apre 1+ (3.52)
2γ ppre

Equations (3.50) and (3.52) together describe the relation between ∆p and ∆u for a shock
moving with speed cs . Thus for a shock moving to the right, with mass flux mR , we have
found:

p3 − p4 = mR (u3 − u4 ) (3.53)
r
γ + 1 p3 − p4
mR = ρ4 a4 1 + . (3.54)
2γ p4

The two relations above give all possible ‘post shock’ states  that are obtainable from the
‘pre-shock’ state à by passing a shock!
The curve p3 = f (u3 ; u4 , p4 ) in the (p, u)-plane is called the Hugoniot curve; properties of
the Hugoniot curves are discussed in section 3.5.4. Contrary to linear theory only the part of
the curve with p3 > p4 can be used.
Consider the Hugoniot curve in the weak shock limit. Assume a weak shock with (p3 −
p4 )  p4 ; then mR ≈ ρ4 a4 and therefore

∆p = ρ4 a4 ∆u,

or

∆p
∆u − = 0. (3.55)
ρ4 a4

Notice that equation (3.55) is a discretisation of the characteristic equation along a Γ− char-
acteristic (see e.g. 3.9) that would cross the shock!!!
Introducing

∆u ∆ρ ∆p
M= and S̃ = = ,
a4 ρ4 ρ4 a24

equation (3.55) can be reduced to

ρ4 a4 S̃
a4 M − = 0,
ρ4 a4

or

M − S̃ = 0.

In the weak shock limit the shock jump reduces to the condition that the Riemann invariant
J − in the linear theory must be constant crossing the shock.
76 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.10: Expansion advancing in a gas

3.5.2 Expansion moving into a gas; Poisson curve

Remember we are preparing ourselves to solve the non-linear Riemann problem; to that end we
now consider an expansion into a prescribed ‘pre-state’. Consider a right running expansion
wave that runs into the gas with prescribed pressure p4 and prescribed sound speed a4 , e.g.
it runs into ‘pre-state’ Ã. The (t, x)-diagram as it is in figure 3.10 shows the process as it is
generated by a piston moving to the left.
Note that the expansion is a continuous process that is homentropic because all particles
have the same initial entropy s4 and there are no shocks so that entropy is kept constant at
the initial level.
Due to the expansion the pressure decreases and the gas accelerates until it attains the
values p3 and u3 at the ‘post-state’ respectively. Due to expansion the pressure decreases:
p3 < p4 .
As in the shock case, we want to relate the pressure difference ∆p = p3 − p4 to the velocity
difference u3 − u4 .
Going along a Γ− -characteristic that runs from state à to state  we have:
2 2
u3 − a3 = u4 − a4 ,
γ−1 γ−1
or, with the aid of isentropic relations:
(  γ−1 )
2 p3 2γ
u3 − u4 = − 1 a4 . (3.56)
γ−1 p4

This relation can also be brought into the form

∆p = ±m∆u (3.57)

with
p
γ−1 1 − ppost
pre
m = ρpre apre  γ−1 . (3.58)
2γ 
ppost 2γ
1 − ppre
3.5. RIEMANN PROBLEM, NON-LINEAR 77

The + sign is chosen for a right running wave, the − sign for a left running wave. Equations
(3.57) and (3.58) together describe the relation between ∆p and ∆u over an expansion fan;
they give all possible ‘post-expansion’ states that are obtainable from the ‘pre-state’ using an
expansion process.
The curve ppost = f (upost ; ppre , upre ) determined by (3.57) and (3.58) in the (p, u)-plane is
called the Poisson curve.
The mass flux in (3.58) must be regarded as some average over the expansion wave.

3.5.3 Example of a Riemann problem

An example of a Riemann problem with initial conditions uL = uR = 0 ,pL > pR results
into a solution having a left moving expansion and a right moving shock; the corresponging
(t, x)-diagram is shown below in figure 3.11.

Figure 3.11: Solution of a Riemann problem; UL∗ = UR∗ , pL∗ = pR∗

3.5.4 Properties of Hugoniot and Poisson curves

When discussing the Riemann problem in section (3.5) we encountered the Hugoniot (H) curve
and the Poisson (P) curve. Both are curves in the (p, u)-plane. In short: Hugoniot describes
shocks and Poisson describes expansions.
Both the ‘H-curve’ and the ‘P-curve’ are described formally by the relation
∆p = m∆u,
where the mass flux m differs for shocks and expansions — compare (3.52) and (3.58).
Let us write down explicitly the formula for an ‘H-curve’ and a ‘P-curve’ describing all
possible ‘post-states’ Â obtainable from a prescribed ‘pre-state’ Ã. That means, let us consider
right running shocks (H-curve) and right running expansions (P-curve). The formulae are
r
γ + 1 p3 − p4
H : p3 − p4 = ρ4 a4 1 + (u3 − u4 ), (3.59)
2γ p4
  γ−1
p3 2γ γ − 1 u3 − u4
P : =1+ . (3.60)
p4 2 a4
Figure 3.12 now shows some properties of the Hugoniot and Poisson curves in the (p, u)-plane,
both going through the same reference (p4 , u4 ).
78 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.12: Comparison of the Hugoniot and Poisson curve

Some observations:
1. The Poisson curve lies above the Hugoniot curve. For the same ∆u, isentropic compres-
sion is more effective than compression by a shock.
2. In the limit p3 → ∞ there results for:
H :p3 ∝ (u3 − u4 )2 (3.61)
2γ
P :p3 ∝ (u3 − u4 ) γ−1 (3.62)
The Hugoniot curve goes less fast to ∞ than the Poisson curve does.
3. Vacuum conditions: p3 = 0 leads to
s
2
H :u3 = u4 − a4 (3.63)
γ(γ − 1)
2
P :u3 = u4 − a4 (3.64)
γ−1
Since γ > 1 the H-curve intersects the level p3 = 0 at higher values of u3 than the
P-curve does.
4. In the reference point à the H-curve and the P-curve have equal slope and equal curva-
ture. A straightforward calculation shows that

   
dp dp
= = ρ4 a4 (3.65a)
du H du P
and
d2 p d2 p
   
γ+1
= = ρ4 (3.65b)
du2 H du2 P 2
The third derivative of the H- and P-curve differ in the reference point which shows that
the entropy rise due to a shock is a third-order effect.
3.5. RIEMANN PROBLEM, NON-LINEAR 79

5. According to linear theory the pressure change ∆p is proportional to ∆u: no distinction

is made between expansions and real shocks.
In non-linear theory however, there is a substantial difference since both Poisson- and
Hugoniot curve share only the slope with the linear graph, see figure 3.13; so the devia-
tion of the pressure from linear theory is O(∆u2 ).
From equations (3.65a) and (3.65b) the difference in pressure between Poisson and Hugo-
niot is O(∆u3 ), which reflects the influence of the entropy rise in the Hugoniot case.

Figure 3.13: Accuracy of linear theory, Poisson and Hugoniot

For an O(∆u2 ) accuracy it is still acceptable to use the Poisson-curve for compressions.

6. The expansion branch of the H-curve is not physically valid because expansions are isen-
tropic. If only expansion appear, the P-curve gives a valid description. Since shocks are
not isentropic, it is clear that in the case of shocks P-curves can give only an approximate
description; for an exact description of shocks, the H-curve has to be used.

7. From the jump equation ∆p = ±m∆u it follows that m is represented in the (p, u)-graph
as the chord between pre- and post-wave states, shown in figure 3.14.

Figure 3.14: Geometrical representation of the mass flux

80 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

The properties of right running waves have just been discussed. Shocks are represented
by Hugoniot curves in the (p, u)-plane, while expansions are represented by Poisson curves.
In the case of left running waves, the graphs in the (p, u)-plane are very similar; they only
differ because the velocities change sign. Therefore, the (p, u)-graph for left running waves is
just the mirror image of the (p, u)-graph of right running waves. Figure 3.15 summarises the
results for both left running and right running waves.

(a) Right running waves (b) Left runnig waves

Figure 3.15: (p, u)-graphs for right- and left running waves

3.5.5 Solving Riemann problems with a (p, u)-graph

In the previous sections the necessary tools to solve non-linear Riemann problems were pre-
sented. Now, three examples will be given here.

Example 1: two shocks

The states À and Ã, or (L) and (R) respectively, are prescribed as shown in the (p, u)-plane
given in figure 3.16.

Figure 3.16: Two shocks; (p, u)-plane and (t, x)-plane

The solution of the Riemann problem is found as the intersections of two H-curves, one
going through à and one going through À. The intersection point Á =  represents the states
3.5. RIEMANN PROBLEM, NON-LINEAR 81

Figure 3.17: Expansion and shock; (p, u)-plane and (t, x)-plane

Figure 3.18: Two expansions; (p, u)-plane and (t, x)-plane

Á and  in the (t, x)-plane having equal pressure p2 = p3 and equal velocity u2 = u3 ; states
Á and  are separated by a contact discontinuity (c.d.) which is parallel to the particle paths
in Á and Â.
States Á and  have different values for the entropy, density and internal energy, so the
contact disconuity is realy a discontinuity. In this example the particle paths in the (t, x)-plane
have a positive slope.

Example 2: a shock and an expansion

The states À and à are given in the (p, u)-plane as shown in figure 3.17.
In this example the solution is found by intersecting the H-curve through à with the P-
curve through À. The ‘pre-states’ À and à are gases at rest but p1 > p4 . An expansion wave
travels into Á and a shock wave travels into Ã. The states Á and  have equal pressure and
equal velocity. The densities, entropy values and internal energy levels are different in Á and
Â. This example resembles the flow in a shock tube.

Example 3: a double expansion

The ‘pre-states’ À and à are now given in such a way that two expansion waves appear, see
figure 3.18.
The solution is found as the intersection of two P-curves. The flow in the expansion fans
itself is also part of the solution and is determined by the P-curves. Since expansions are
82 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

isentropic, the entropy jump over the contact discontinuity is equal to the entropy difference
between the ‘pre-states’ À and à .
It can happen that the ‘pre-states’ À and à are chosen in such a way that the P-curves
intersect at a negative value of the pressure. This is physically impossible; if P-curves pass
the zero-pressure level, the flow expands to vacuum conditions and cavitation occurs.

3.6 The Method of Characteristics for one-dimensional unsteady

flow
The same method that we used to obtain the simple-wave solution analytically can be used
more generally to obtain solutions numerically on a discrete grid. The grid is built up by
characteristics i.e. the grid points are intersections of characteristic curves. Let us start with
the homentropic case.
In homentropic flow the entropy is uniform and the Riemann invariants are written in
algebraic form:

2
J± = u ± a. (3.66)
γ−1

The Riemann-invariants J + and J − are constant along the characteristic curves Γ+ and Γ−
respectively.

Figure 3.19: M.O.C to find the solution in P , homentropic flow

The solution in a point P in the (t, x)-plane follows entirely from the known solution in
the points A and B if P lies at the intersection of Γ+ going through A and Γ− going through
B (see figure 3.19);

2 2
along Γ+ (A) :JA+ = uA + aA = uP + ap (3.67)
γ−1 γ−1
2 2
along Γ− (B) :JB− = uB − aB = uP − aP (3.68)
γ−1 γ−1
3.6. THE METHOD OF CHARACTERISTICS FOR ONE-DIMENSIONAL UNSTEADY FLOW83

Solving for uP and aP yields:

1 +
JA + JB−


uP = (3.69)
2
γ−1 +
JA − JB−


aP = (3.70)
4
Now we have the solution in P, but we we still have to determine the location of P . The
point P can be found, at least approximately, by taking the average direction of Γ+ in A and
P and also the average direction of Γ− between B and P . Then the approximate location
(tP , xP ) of P follows from the system:
 
dx
xP − xA = (tP − tA ), (3.71a)
dt AP
 
dx
xP − xB = (tP − tB ), (3.71b)
dt BP

where
 
dx uA + aA + uP + aP
= ,
dt AP 2

and
 
dx uB − aB + uP − aP
= .
dt BP 2

Note that the flow variables in the point P are exactly known; on the other hand, the position
of P can not be determined exactly. This altogether makes the numerical implementation of
the M.O.C. an approximate method in finding the flow variables.

Let us now turn to the non-homentropic case. In non-homentropic flow without shock dis-
continuities, the entropy s remains constant when following a particle; however, s may differ
for different fluid particles. So s is not necessarily uniform in the (t, x)-domain. Due to this
non-uniform entropy the Riemann invariants can not be integrated and must be approximated
from their differential form.

pP − pA
along Γ+
A : uP − uA + 1 =0 (3.72)
2 (ρP aP + ρA aA )
pP − pB
along Γ+
B : uP − uB − 1 =0 (3.73)
2 (ρP aP + ρB aB )

Here we have two equations for the three unknowns uP , pP and ρP . How to solve this
problem? There are several answers; one is explained in figure 3.20.
Observe the particle path through P and assume that this path also goes through C being
the intersection of Γ− +
A and ΓB . In other words: it is assumed that the gas particle in P comes
from C. Then, the entropy in P has the same value as the entropy in C, so that

sP = sC ,
84 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.20: M.O.C to find the solution in P , non-homentropic flow

or

pP ργC = pC ργP (3.74)

Equation (3.74) is the third equation that closes the system (3.72), (3.73) and (3.74), because
the point C and the flow variables in that point are already calculated in a previous step in
the computational process.

Comment To close (3.72) and (3.73) with a third equation a possible method is given above.
There are other more accurate methods available. But does it make sense to use such a more
accurate third equation?
Well, that depends on how each equation contributes to the accuracy of the whole system.
Here we touch on the important issue of consistency. To improve the order of accuracy all
equations have to be improved in a consistent way. For example in our case: if the order of
accuracy of the equations (3.72), (3.73) and (3.74) is the same then it is not useful to look for
an improvement of a single equation, e.g. (3.74).

3.7 Simple compression waves

In section 3.5.1 we studied, as part of the non-linear Riemann problem, the instantaneous
movement of a piston into a gas at rest. There it was seen, see figure 3.9, that a shock is
created moving into a gas with a shock speed cs that is related to the pressure rise. Since the
piston had an infinite acceleration to its final speed up = u3 , the shock immediately (at t = 0)
starts moving from the piston’s position (at x = 0).
Let us now consider the case that the piston is accelerated at a finite rate until the piston
has reached its maximum velocity umax at t = T , see figure 3.21.
Considering such a piston that is accelerated and advancing into a gas at rest, a simple
comprssion wave will be created.
The characteristics emanating from the piston path are now converging forward in time
and sooner or later a fan of neighbouring characteristics will intersect. The solution gradient
3.7. SIMPLE COMPRESSION WAVES 85

Figure 3.21: Shock development by an accelerating piston

has become infinite at that point; a shock must be introduced now. The shock is building up
in strength by characteristics intersecting the shock from behind.
As long as the shock is still absent, the flow is isentropic and since all particles emanate
from the rest state (labelled ‘0’) the entropy is uniform. The question now is how long it lasts
before the compression waves start intersecting.
To answer this, let us assume that the piston has a constant acceleration, u̇p :
dup
u̇p = ,
dt
so that
1
up = u̇p t, xp = (u̇p ) t2 . (3.75)
2
Take a short excursion of the piston to the point A with xA = ∆x and tA = ∆t, ∆x and ∆t
are small, see figure 3.22.
Let point I be the intersection of the two Γ+ characteristics, one coming from origin and
the other coming from A:
first characteristic: xi = a0 ti (3.76)
characteristic through A: (xi − ∆x) = (uA + aA )(ti − ∆t) (3.77)
The characteristic speed uA + aA follows from the fact that the Γ− -characteristic through A
starts in the domain where the gas is at rest. Along Γ−
A:

2 2a0
uA − aA = − ,
γ−1 γ−1
or
γ−1
aA = a0 + uA ,
2
so that
γ+1 γ+1
uA + aA = uA + a0 = u̇p ∆t + a0 . (3.78)
2 2
86 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.22: Intersecting Γ+ characteristics

Then equation (3.77) becomes

 
γ+1
xi − ∆x = u̇p ∆t + a0 (ti − ∆t) (3.79)
2
Inserting (3.75) and (3.76) into (3.79) yields
 
1 2 γ+1
a0 ti − u̇p (∆t) = u̇p ∆t + a0 (ti − ∆t).
2 2
Solving for ti yields
a0 γ
ti = γ+1 + ∆t.
2 u̇p
γ+1
For ∆t → 0, we find that the first signal and its neigbouring characteristic intersect at:
a0
ti = γ+1 . (3.80)
2 u̇p

The faster the piston accelerates the sooner the characteristics intersect. Equation (3.80) can
be interpreted as follows. Therefore we generalize the nominator and denominator of equation
(3.80).
Since a0 is constant, the denominator may be written as:
   
γ+1 d γ+1 d
u̇p = u̇p + a0 = (up + ap ) ,
2 dt 2 dt 0

and since u0 = 0, the nominator may be written as:

a0 = u0 + a0 ,
Then equation (3.80) can be interpreted as
u0 + a0 initial characteristic speed
ti = d

= d
.
dt (up + ap ) 0 initial dt characteristic speed
3.7. SIMPLE COMPRESSION WAVES 87

A typilcal example of a simple compression wave appears if the piston path is such that the
characteristics Γ+ all intersect in one focal point. See figure 3.23. Such a compression wave is
well known as a centered compression wave.

Figure 3.23: Γ+ characteristics intersecting in one focal point

Let us consider in more detail the various domains that can be distinguished in the (t, x)-
plane if an accelerating piston causes the development of a shock wave. The fact that various
domains of different properties appear is clear because of shock development and its influence
on the entropy in the flow.
Consider now figure 3.24 featuring the shock development process. Domain A represents
the fluid at rest. The domains B, C, D, E and F lie left from the particle path Γ0 going through
the starting point of the shock; the domains G, H and I are to the right of this particle path.

Figure 3.24: Various flow domains under shock development

88 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Domain A: uniform conditions; the gas is at rest. It is bounded by the first signal and the
shock curve.

Domain B: is a simple wave because all Γ− characteristics originate from t = 0 and have the
same JB− = JA− . All gas particles have the same entropy, so the entropy is uniform and
2a0
Riemann invariants are integrable. The Riemann invariant J + varies form J + = γ−1 on
2a
the first signal to J + = up + γ−1
p
on the Γ+ characteristic going through P. Domain B
is bounded by the Γ+ characteristic OS, the Γ− characteristic SR, the Γ+ characteristic
RP and the piston path OP.

Domain C: is bounded by the Γ+ characteristic PR, the Γ− characteristic RQ and the piston
path QP with up = constant. All particles in C start in A and they have crossed no
shock, so the entropy is uniform and the Riemann invariants are integrable. All Γ−
characteristics issue from t = 0 and do not cross a shock, so JC− = JA− = constant.
2a 2
On each Γ+ characteristic through C the relation u+ γ−1 = up + γ−1 ap is satisfied. Since
−
JC = constant there is a linear relation between up and ap and since up is constant in
C, ap is constant as well. However, up and ap constant implies that JC± is constant and
so u and a are constant in domain C, therefore domain C is uniform.

Domain D: is bounded by the particle path Γ0 : ST, the Γ+ characteristic RT and the Γ−
characteristic RS. All particles in D arrive from A without having crossed a shock,
therefore the entropy is uniform, so the Riemann invariants J + and J − are integrable.
+ −
However, JD = JB+ and they vary. Also JD varies because the Γ− characteristics have
crossed a shock with varying strength, so domain D is a non-simple domain.

Domain E: is bounded by the Γ+ characteristics RT and QU and by the particle path TU

and the Γ− characteristic QR. All particles in E arrive from A with out having crossed
a shock, therefore J + and J − are integrable. JE+ = JC+ is constant. JE− varies because
the Γ− characteristics have crossed a shock wave with varying strength. Conclusion: E
is a simple wave region.

Domain F: is bounded by the Γ+ characteristic QU, the particle path Γ0 and the piston path
described by up = constant. All gas particles in F start in A and have not crossed a
shock, therefore J + and J − are integrable. Since Γ− characteristics have crossed a shock
of varying strength, JF− varies. Then ap (on the piston path) also varies and therefore
JF+ varies. Conclusion: F is a non-simple region.

Domain G: is bounded by the shock segment SV, the particle path Γ0 and the Γ+ charac-
teristic TV. Particles arriving in G must have crossed the shock, therefore the entropy is
+
not uniform and the Riemann invariants are not integrable. Furthermore, JG is equal to
+ − −
JB which both vary; JG also varies since the Γ characteristics have crossed the shock.
Conclusion: G is a non-simple domain.

Domain H: is bounded by Γ+ characteristics TV and the Γ+ characteristic through U, and

by the particle path TU and a shock segment. Particles in H have crossed a shock,
therefore the entropy is not uniform and J + and J − are not-integrable. Furthermore,
+ −
JH = JC+ is constant; JH varies, so domain H is non-simple.
3.8. WAVE INTERACTIONS 89

Domain I: is bounded by the particle path Γ0 , a shock segment and the Γ+ characteristic
through U. Particles in I have crossed a shock, therefore the entropy is not uniform.
Furthermore, JI+ = JF+ which varies, and also JI− varies. So domain I is a non-uniform,
non-simple region.

The properties of each domain just discussed are summarised in table 3.2.

Domain J+ J− S comment
A u u u uniform, i
B v u u simple wave, i
C u u u uniform, i
D v v u non-simple, i
E u v u simple(!), i
F v v u non-simple
G v v v non-simple
H u v v non-simple
I v v v non-simple

Table 3.2: Properties of flow domains appearing in the shock development process;
key: u: uniform, v: varies, i: integrable

3.8 Wave interactions

Until now, the one-dimensional unsteady flow phenomena we have encountered were expansion
waves, contact discontinuities, compression waves and shocks. In addition, two types of waves
were distinguished: left running and right running waves. They always appeared in a single
manner, but what if they interact, what can happen? In this section some examples of
interactions are discussed.

3.8.1 Two right running shocks

Consider a semi-infinite constant area tube closed off by a piston at its left end and filled with
a gas at rest. At t = 0 the piston accelerates instantaneously and acquires the speed up1 ;
then, at t = t1 it instantaneously switches to the speed up2 with up2 > up1 , see figure 3.25.
Two shock waves are created; they intersect at time t2 .
At t2 a Riemann problem between the states Á (left) and 0 (right) appears. The first
question is: do shocks really intersect? To answer this, we go to the shock frame and view the
first moving shock.
The first moving shock viewed in the shock frame shows the following picture:
90 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.25: Two intersecting shocks of the same family

The entropy condition II states (a + u)0 < cs1 < (a + u)1 . The second moving shock in the
shock frame:

The entropy condition II states (a + u)1 < cs2 < (a + u)2 .

Combining the two inequalities gives cs1 < cs2 . Hence the second shock moves faster and
reaches the first shock at t = t2 .
The entropy condition II also shows that Γ+ characteristics cannot cross the right running
shock. The shock is hit by characteristics from both sides. A shock is a barrier for character-
istics, so they terminate there. On the shock the characteristic speed is double valued. One
can say that the right running shock belongs to the “+” family. In a similar way one can say
that a left running shock belongs to the “-” family implying that Γ− characteristics cannot
cross left-running shocks.
The question we still have is how to solve the Riemann problem appearing at t2 . The
answer can be found by sketching the interaction process in the (p, u)-plane, see figure 3.26.
State À lies on the Hugoniot curve H0 going through the point 0 . The slope in 0 of the
H0 -curve is ρ0 a0 . State Á lies on the H1 curve going through À; the slope of the H1 curvce
À is ρ1 a1 . State Á is found where u2 has the prescribed velocity up2 . State à also has to
lie on the H0 curve because the merged shock transfers the gas from the ‘pre-state’ 0 to the
‘post-state’ Ã. Since H1 lies above H0 a Poisson-curve is needed to find a valid solution for
state Â. Hence an expansion fan appears between states Á and Â. Furthermore, a contact
discontinuity with characteristic speed ucd > up2 separates Á and Â. That H1 lies above H0
can be proved for 1 < γ < 2 by straightforward algebra.
Compare this unsteady interaction with a corresponding 2D steady supersonic flow, sketched
3.8. WAVE INTERACTIONS 91

Figure 3.26: (p, u)-plane

in figure 3.27. The weak wave between  and à can be either an expansion or a shock de-
pending on the free stream Mach number and the two ramp angles δ1 and δ2 .

Figure 3.27: Supersonic flow along two successive ramps

3.8.2 Two right running expansion waves

Consider again a semi-infinite constant area tube closed off by a piston at its left end and filled
with a gas at rest. At t = 0 the piston is instantaneously withdrawn and acquires a speed up1
(< 0), at t = T it instantaneously switches to the speed up2 < up1 < 0. Two expansion waves
are created, see figure 3.28.
As we see these expansions do not interact. The last signal of the first expansion fan and
the first signal of the second expansion fan have the same characteristic speed.
The whole process in the (t, x) domain is mapped onto one single graph in the (p, u) plane:
the Poisson curve for the right running waves that starts in the state 0 .
Compare this with two subsequent expansions in 2-D steady supersonic flow (see figure
3.29). They also do not interact.
92 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.28: Two expansions of the same family

Figure 3.29: Supersonic 2-D steady flow along a plane wall with two successive kinks

3.8.3 Shock interacting with expansion

A piston moves with a constant speed up > 0 into a gas at rest. At time ti the piston is
instantaneously decelerated to standstill. This creates the interaction of a shock wave and an
expansion wave both of the same family. This situation is sketched in figure 3.30.
Interaction starts at (xi , ti ). Due to interaction the shock slows down; its strength decreases
and its entropy jump decreases too. The particle path going through the point i separates
two domains in the expansion fan. On the left the entropy is uniform; on the right of the
particle path however, the entropy changes. There is no analytical solution available for this
fully non-linear case. Fortunately, there is a first non-linear approximation using the inviscid
Burgers’ equation (see chapter 4).
Compare this interaction with 2-D steady supersonic flow figure (3.31).
3.8. WAVE INTERACTIONS 93

Figure 3.30: Shock-expansion interaction

Figure 3.31: Shock-expansion interaction in supersonic flow

3.8.4 Shock interacting with expansion

A second example of shock-expansion interaction where both waves are of the same family
(e.g. both are right running waves) appears if the piston is instantaneously withdrawn with
speed up < 0. Later, at time ti the piston is instantaneously brought to standstill. The initial
withdrawal creates an expansion fan, the acceleration to standstill creates a shock wave. Both
waves interact at (xi , ti ), see figure 3.32.
Due to interaction the shock accelerates. “Behind” the accelerated shock the entropy is
non-uniform.
Compare this interaction to the 2D steady supersonic interaction as sketched in figure
3.33.

3.8.5 Inteaction of shocks of different family

The interaction of two shocks, one left running and the other right running, is given in figure
3.34. Interactions happen at (xi , ti ). At t = ti a Riemann problem has to be solved between
the states À and Ã.
From the (p, u) plane one concludes that after interaction two shocks appear. This inter-
action problem can be solved easily by a numerical Riemann solver.
94 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS

Figure 3.32: Unsteady flow, shock–expansion interaction

Figure 3.33: steady flow: expansion–shock interaction

Figure 3.34: Two interacting shocks

3.8. WAVE INTERACTIONS 95

3.8.6 Interaction of two expansions of different family

This situation is sketched in figure 3.35. The interaction can be calculated with analytical
methods. The interaction pattern displays a non-simple region; both J + and J − vary.

Figure 3.35: Two interacting expansion fans

96 CHAPTER 3. ONE-DIMENSIONAL UNSTEADY NON-LINEAR FLOWS
Chapter 4

Burgers’ equation for simple waves

4.1 Shock description

In chapter 3 we encountered the inviscid form of the Burgers’ equation

vt + vvx = 0 (4.1)

for the description of a simple wave having J − = constant; v denotes the characteristic speed
v = u + a.
In a similar way the Burgers’ equation

wt + wwx = 0 (4.2)

is valid for simple waves having J + = constant; w denotes the characteristic speed w = u − a.
Equations (4.1) and (4.2) are applicable in the continuous part of the flow; once a shock
is formed the solutions break down because the assumption that both entropy (s) and one of
the Riemann invariants (J − or J + ) are uniform is violated. For weak shocks however, the
error in the solution for v is small: O(∆v) and acceptable.
Consider the conservative form of (4.1):
1
vt + ( v 2 )x = 0, (4.3)
2
and apply the jump relations. The corresponding jump equation reads:
 
1 2
Vshock [v] = v . (4.4)
2

With [v] = vpost − vpre the jump equation becomes

1
Vshock = (vpre + vpost ), (4.5)
2
or
1
Vshock = ((u + a)pre + (u + a)post ), (4.6)
2
Note that the approximate shock speed is equal to the mean of the characteristic speeds of
the pre- and post state, see figure 4.1

97
98 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.1: Shock running into a gas at rest

Let us determine the order of accuracy in which equation (4.5) or (4.6) approximates the
exact value of the shock speed.
For a shock running into a gas at rest (state u0 = 0, p0 , ρ0 ) we know from equation (1.50)
that the exact speed of the shock in terms of the pressure rise: p1 − p0 is given as:
r
γ + 1 p1 − p0
|cs | = a0 1 + . (4.7)
2γ p0

In terms of the post state velocity u1 , the shock speed cs can be written as

γ+1
c2s = a20 + cs u1 . (4.8)
2

To find this result, consider the flow situation of figure 4.2 in the shock frame and apply
conservation of mass and momentum across the shock:

ρ0 cs = ρ1 (cs − u1 )
p0 + ρ0 c2s = p1 + ρ1 (cs − u1 )2 .

Figure 4.2: Shock in shock frame

Eliminating ρ1 yields:

p1 − p0 = ρ0 cs u1 (4.9)
4.2. SHOCK FORMATION 99

Inserting (4.9) into (4.7) gives

 
2 2 γ + 1 ρ0 cs u1
cs = a0 1 +
2γ p0
2
γ + 1 a0 ρ0
= a20 + cs u1
2γ p0
γ+1
= a20 + cs u1
2
For weak shocks (u1  cs ) the exact shock speed (4.8) may be expanded in terms of u1 as:

γ+1 (γ + 1)2 2
cs = a0 + u1 + u1 + O(u41 ). (4.10)
4 32a0

The approximate shock speed according to (4.6) with (u + a)pre = a0 and (u + a)post = u1 + a1
becomes
1
cs ≈ Vs = (a0 + u1 + a1 ). (4.11)
2
From J − is constant (simple wave!) follows:

2a1 2a0
u1 − =− ,
γ−1 γ−1
or
γ−1
a1 = a0 + u1 .
2
Inserting this expression for a1 in (4.11) we find
 
1 γ−1
cs ≈ a0 + a0 + u1 + u1
2 2
or
γ+1
cs ≈ a0 + u1 . (4.12)
4
Comparing (4.10) and (4.12) we conclude that using the simple wave approximation the shock
speed is accurate to first order.

4.2 Shock formation

The Burgers’ equation vt + vvx = 0 has characteristics Γ+ : dx
dt = v along which v is constant.
Consequently, the characteristics of the Burgers’ equation are straight lines in the (t, x)-plane.
Burgers’ equations can be used to study the development of a shock wave.
To that end, let us consider an initial value problem where the v-distribution with x is
prescribed at t = 0. Assume that the initial values of v vary linearly between v1 at x = −1
and v0 at x = +1; take v1 > v0 > 0. As a consequence of the linear distribution of the initial
data the characteristics in the compression domain focus in the single point F, see figure 4.3.
100 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.3: Shock development

Let the initial distribution be given by:

x < −1 : v(x) = v1
 
v0 + v1 v1 − v0
−1 ≤ x ≤ 1 : v(x) = − x
2 2
x ≥ 1 : v(x) = v0

Characteristics emanating from x0 on the initial line t = 0, −1 < x0 < 1 are given by:

1
t= (x − x0 ).
v(x0 )

The focal point F is located at

2
tF = , xF = 1 + v0 tF .
v1 − v0

At time level t = tF the shock is developed at its full strength instantaneously. Once it appears
it moves to the right with velocity Vshock = 21 (v0 + v1 ) which equals the characteristic speed
at x = 0.
Now we consider a non-linear distribution of the initial value v(x) between v1 at x = −1
and v0 at x = +1. Due to this non-linearity a shock develops gradually. The location where
shock formation starts depends strongly on the initial data v(x, 0) = v(x0 ).
Consider two neighbouring points x0 and x0 + ∆x0 on the initial line. The characteristic
through x0 is given by:

1
Γx0 : t = (x − x0 ).
v(x0 )
4.3. SHOCK FORMATION FROM A TRIANGULAR WAVE 101

Similarly the characteristic through x0 + ∆x0 is the straight line

1
Γx0 +∆x0 : t = dv
(x − x0 − ∆x0 )
v(x0 ) + dx 0
∆x 0

Both characteristics intersect at

−1 −1
ti = = .
dv v 0 (x


dx x0 0)

So, shock formation does not occur if v 0 (x0 ) > 0. Intersecting characteristics announce shock
development, which happens first (in time) at

−1
tF = (4.13)
v 0 (x
0 )min

where tF is the time level at which the shock formation starts. The x position is given by

xF = x̂0 + tF v(x̂0 ),

where x̂0 denotes the location where v(x0 ) attains its largest negative slope.

Example

Case A: Take as initial value at t = 0:


v1 = 3,
 x < −1,
1
v(x, 0) = 2 − x + 2 (x2 − 1), −1 < x < 1,

v0 = 1, x > 1,


see figure 4.4. The slope v 0 (x) = −1+x attains its minimal value at xˆ0 = −1; here v 0 min = −2.
Shock formation starts at time level tF = 12 at position xF = 12 . The initial shock velocity
Vs,i is just equal to the characteristic speed at x = −1, so Vs,i = 3. The final shock velocity
is Vs,f = 2; it indicates that the shock slows down when moving to the right.

Case B: Take as initial value at t = 0:


v1 = 3,
 x < −1,
v(x, 0) = 2 − x − 21 (x2 − 1), −1 < x < 1,

v0 = 1, x > 1,


see figure 4.5. The slope v 0 (x) = −1 − x attains its minimum at x̂0 = +1. Here v 0 min = −2
indicating that shock formation starts at tF = + 12 and xF = 32 . The initial shock speed is
Vs,i = 1, the final shock speed is Vs,f = 2, the shock speeds up when moving to the right, see
figure 4.5.
102 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.4: Shock development

Figure 4.5: Shock development

4.3 Shock formation from a triangular wave

We now look at how the Burgers’ equation gives rise to shock development when we start with
a triangular wave, see figure 4.6.
4.3. SHOCK FORMATION FROM A TRIANGULAR WAVE 103

Figure 4.6: Shock formation from a triangular wave

The initial conditions are:



 0, x < −L2 ,

4 L −L −L


L x + 2 , 2 < x < 4 ,



v(x, 0) = −4 L x,
−L
4 < x < 4,
L
 4 L

L L
L x− 2 , 4 < x < 2,





0, x > L2 ,


The initial values between − L4 and + L4 form a centered compression wave that focuses at
x = 0, t = L4 . After shock formation at t = L4 the shock remains at x = 0 and decays due
to the expansion fans that arrive at the location x = 0. To get the decay we need to know v
for increasing time at each side of the shock (say at x = 0− and x = 0+ ). Use the method of
characteristics to obtain v(x, t) as:

v(x, t) = v(x − vt, 0).

For the domain covered by the characteristics emanating from − L2 < x − L

4 we then find
 
4 L
v(x, t) = x + − vt
L 2
or
4x + 2L
v(x, t) = .
L + 4t
104 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

L
In a similar way for the domain that is covered by the characteristics emanating from 4 < x L2
can be found that:
4x − 2L
v(x, t) = .
L + 4t
L
So, at a certain time level t ≥ 4, the value of v at each side of the shock at x = 0 is given by
−2L 2L
v(0+ , t) = , and v(0− , t) = .
4t + L 4t + L
The shock strength ∆v = v(0+ , t) − v(0− , t) becomes:
−4L
∆v = .
4t + L
For large t, we have
lim ∆v = 0;
t→∞

the shock strength decays asymptotically as t−1 .

4.4 Shock formation from a parabolic curve

Figure 4.7: Parabolic initial data

Now let us consider initial data that has a parabolic shape. Take initial values v(x, 0)
consisting of two parabolic curves defined on the interval
−L/2 ≤ x ≤ L/2 as follows:

16 L
vL (x, 0) = − L2 x(x + 2 )
 on − L2 ≤ x ≤ 0,
v(x, 0) = vR (x, 0) = + L162 x(x − L2 ) on 0 ≤ x ≤ L2 , (4.14)

0 outside the interval.


The graph of the initial values is shown in figure 4.7. At x = 0, vL (x) and vR (x) together
dv
with their first derivatives match such that v(x, 0) and dx (x, 0) are continuous. The first
derivatives:
 
dvL 16 L
= − 2 2x + ,
dx L 2
 
dvR 16 L
= + 2 2x −
dx L 2
4.5. ENTROPY CONDITIONS REVISITED 105

attain their largest minimum value: −8/L at x = 0. So shock formation starts at time level
t = L8 at x = 0. Since v(x, 0) is an odd function, the shock velocity

1
vs = (vL + vR )
2
L
is zero. The shock remains at x = 0 for all t > 8, see figure 4.8.

Figure 4.8: Formation of a shock from parabolic initial data

2
The shock strength defined as ∆v = vL − vR can be found as ∆v = Lt − 8tL
2 . The shock
strength increases until t = L/4, thereafter it decays. The maximum shock strength equals
(∆v)max = 2. Shock decay for large t appears asymprotically: ∆v ∝ t−1 .

4.5 Entropy conditions revisited

In chapter 1 we discussed the need of an entropy condition in modelling non-viscous discon-
tinuities properly. Three different but equivalent formulations were presented:

I a fluid particle passing through a shock increases its entropy;

II for a physically valid moving shock the inequality (u1 − a1 ) > cs > (u2 − a2 ) holds;

III only those inviscid solutions are physically acceptable that can be obtained also from
the fully viscous and heat-conducting equations in the limit of vanishing viscosity and
heat conduction.

Condition II was derived for the situation in which the shock velocity cs of a right running
shock is smaller than the gas velocity u2 in the domain in front of the moving shock. So
domain Á is the post-shock state and domain À is the pre-shock state. Let us rephrase this
condition for the case in which a right running shock moves into the pre-shock domain (R);
see figure 4.9.
Domain R is the pre-shock state, so it follows that cs − uR > aR ; similarly, domain L is
the post shock state, so cs − uL < aL . Combining these conditions there results

uL + aL > cs > uR + aR (4.15)

106 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.9: Moving shock in laboratory frame (left) and shock frame(right)

or

vL > cs > vR .

Although the entropy is uniform in the Burgers flow description, we can recognise physical
and unphysical shocks by the characteristic pattern in which they appear, see figure 4.10.

Figure 4.10: Physical and un-physical discontinuities

For physically valid shocks the characteristics of the same family merge or terminate at the
shock discontinuity. If characteristics emerge from a discontinuity, an un-physical phenomenon
occurs which should be replaced by an expansion fan.
Concerning entropy condition III let us study travelling-wave solutions of the viscous
Burgers’ equation:

vt + vvx = µvxx (4.16)

in the limit µ → 0 (vanishing viscosity).

We assume two different initial value distributions v(x, 0) shown in figure 4.11. There is
an interval ∆ where v(x, 0) varies between VL and VR ; outside the interval ∆, v(x, 0) takes
the constant value VL or VR .
In the non-viscous approximation one finds:
 
dx vL + vR
case A: Vshock = = , ⇒ vL > Vshock > vR
dt 2
 
dx vL + vR
case B: Vshock = = , ⇒ vL < Vshock < vR
dt 2

Using (4.15) case A thus evolves into a physically valid shock whereas case B evolves into an
unphysical situation.
4.5. ENTROPY CONDITIONS REVISITED 107

Figure 4.11: Initial values for travelling wave solutions of the viscous Burgers’ equations. Case
A: vL > vR , case B: vL < vR .

Let us now study what happens if viscosity is incorporated into the model. Introduce the
new variables:

z = x − Vshock · t (4.17)
w = v − Vshock (4.18)

Using
     
∂ ∂ ∂
= − Vshock ,
∂t x ∂t z ∂z t
   
∂ ∂
=
∂x t ∂z t

it follows that (4.16) becomes

wt + wwz = µwzz . (4.19)

The initial condition is transformed into

vL + vR
w(x, 0) = v(x, 0) − , (4.20)
2
thus the limit values wL and wR are
vL − vR vR − vL
wL = , wR = .
2 2
In the non-viscous approximation, the shock velocity becomes
 
dz 1
= (wL + wR ) = 0,
dt 2

indicating that the shock ultimately evolves into a steady position at z = 0. The steady
situation happens first at t = ∆/(vL − vR ); thereafter the steady shock persists for all time
⇒ wt = 0.
To study viscous effects we may also assume wt = 0, then (4.19) can be written as
 
1 2
w = (µwz )z (4.21)
2 z
108 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.12: Graph of w(1) and w(2)

having the first integral

1 2 dw
w (z) = µ − µc (4.22)
2 dz
where c is an integration constant. This differential equation can be written in the form
dw dz
= (4.23)
w2 + 2µc 2µ
which has two different solutions for c 6= 0 : w(1) and w(2) :
s
(1)
p |c|
c < 0 : w = − 2µ|c| tanh z; (4.24)
2µ
s
p |c|
c > 0 : w(2) = − 2µ|c| tg z; (4.25)
2µ

the integration constant c can be interpreted as the value of dw

dz at w = 0. A graph of the
solution w(1) and w(2) is given in figure 4.12.
From this figure we observe that w(1) fulfills the boundary conditions:
p
z → −∞ :w(1) = + 2µ|c| = wL ;
p
z → +∞ :w(1) = − 2µ|c| = wR ; .

2µ|c| = wL −w
p p
If 2µ|c| is written in terms of wL and wR : 2
R
, then the solution for w(1)
becomes
  
(1) wL − wR wL − wR
w =− tanh z . (4.26)
2 4µ

In the limit µ → 0 w(1) approaches the non-viscous solution with a shock discontinuity
where w jumps from wL to wR .
Figure 4.12 also makes clear that the w(2) solution cannot satisfy the boundary conditions;
thus the assumption that a travelling wave with wR − wL > 0 exists has been proven to
be invalid. ‘Expansion shocks’ do not appear as solutions of the Navier-Stokes equations for
decreasing viscosity; so ‘expanding shocks’ must be regarded as unphysical.
4.6. WAVE INTERACTION 109

4.6 Wave interaction

The inviscid Burgers equation enables us to study various types of wave interactions. Some
examples will be elaborated upon in more detail in the sequel.

Example 1: expansion weakens shock

This situation can be viewed in the t, x-plane by uniformly moving a piston into a gas and
then stopping it. Assume initial values along the entire x-axis at time t = 0 (being the time
instant at which the piston stops):

0,
 for x < 0,
v(x, 0) = 1, for 0 < x < 1,

0, for x > 1,


see figure 4.13.

The discotinuity at x = 1 represents the shock, the abrupt change in v at x = 0 generates
a centered expansion created by the stopping piston.

Figure 4.13: Initial values at t = 0

The shock produced by the uniformly moving piston has a velocity

 
dx 1 1
= Vs = (vpre + vpost ) = .
dt s 2 2

The first characteristic of the centered expansion wave has a velocity v = 1 and hits the shock
at t = 2. From that instant on the interaction starts. The time evolution of the shock is
shown in figure 4.14.
For t ≤ 2 the shock is a straight line with a slope dx 1
dt = 2 . For t > 2 the shock is curved.
s

Its shape can be determined from the O.D.E.:

dxs 1 1 x x
= (vpre + vpost ) = (0 + ) = . (4.27)
dt 2 2 t 2t
With the boundary conditions: x = 2, t = 2 there follows (for t ≥ 2):
√
xs = 2t (4.28)

The shock has a parabolic shape in the (t, x)-plane. The shock strength for t ≥ 2 can be
calculated as
x
∆v = vpost − vpre = ; (4.29)
t
110 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.14: Expansion eats shock

Figure 4.15: Shock development

using (4.28) this becomes

√ r
2t 2
∆v = = . (4.30)
t t
√
The shock strength decays with increasing time to zero as 1/ t. The time evolution of v(x)
is shown in figure 4.15. Observe that the area A of the v-disturbance is constant in time:
A = xs ∆V
2 = 1.

Example 2: a shock accelerated by an expansion

Assume initial conditions v(x, 0) such that:
x < −1 : v(x, 0) = 1,
−1 ≤ x < 0 : v(x, 0) = 0,
x > 0 : v(x, 0) = 1.
It represents a shock at x = −1 and a centered expansion at x = 0, see figure 4.16.
The shock at x = −1 has a speed Vs = 12 . It hits the last characteristic of the expansion
fan at t = 2. For t > 2 the shock weakens and accelerates; its path may be found as a solution
of the O.D.E.:
dxs 1 x
= 1+ (4.31)
dt 2 t
4.6. WAVE INTERACTION 111

Figure 4.16: Shock accelerated by expansion

with boundary condition

x = 0, t = 2.

The solution is:

√
xs = − 2t + t, for t ≥ 2. (4.32)

The shock accelerates but it never moves faster than the characteristic speed v = 1. The
shock strength for t > 2 can be calculated as

xs
∆v = vpost − vpre = 1 − (4.33)
t

With (4.32) this becomes:

r
2
∆v = , (4.34)
t

indicating that the shock strength decays asymptotically to zero.

The spatial distance between the shock and the first characteristic of the expansion:
√ √
xchar − xshock = t + 2t − t = 2t (4.35)

grows with increasing time. At large t the spatial v-distribution at constant t looks as shown
in figure 4.17. This figure shows that there is always a weak shock
√ left.
p The disturbance in the
1
v-distribution has a triangular shape of which the area A = 2 2t 2/t = 1 does not change
in time.
112 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.17: v-distrubution at large t.

Example 3: expansion eats shock

Assume initial conditions v(x, 0) such that:

x < 0 : v(x, 0) = 0,
0 < x < 1 : v(x, 0) = 2,
x > 1 : v(x, 0) = 1.

It represents a shock at x = 1 and a centered expansion at x = 0, see figure 4.18.

Figure 4.18: Expansion eats shock

The shock at x = 1 has a velocity vs = 23 . It meets the first characteristic of the expansion
fan at t = 2, x = 4. For t = 2 expansion-shock interaction starts. The path of the shock is
found as the solution of the O.D.E.:
dxs 1 x
= 1+ (4.36)
dt 2 t
with boundary condition x = 4, t = 2. The solution is (for t > 2):
√
xs = t + 2t. (4.37)
4.6. WAVE INTERACTION 113

The shock slows down, the shock speed is found as

dx 1
= Vs = 1 + √ . (4.38)
dt 2t
Observe that the shock speed never drops below the value 1. This implies that part of the
expansion wave, where 0 < v < 1, is not affected by the shock.
The shock strength for t ≥ 2 can be calculated as
xs
∆v = vpost − vpre = − 1. (4.39)
t
Using (4.37) this becomes:
r
2
∆v = . (4.40)
t
The shock strength decays asymptotically to zero. For large t the spatial v-distribution is as
shown below in figure 4.19. There is always a weak shock travelling to the right.

Figure 4.19: Spatial v-distrubution for large t

Example 4: shock eats expansion

Assume initial conditions such that:

x < −1 : v(x, 0) = 2,
−1 < x < 0 : v(x, 0) = 0,
x > 0 : v(x, 0) = 1.

It represents a shock at x = −1 and a centered expansion wave at x = 0, see figure 4.20.

The shock at x = −1 has a velocity Vs = 1. It meets the last characteristic of the expansion
wave at t = 1, x = 0. Due to the interaction the shock accelerates. The shock path xs (t) is
determined by the O.D.E.:
dxs 1 x
= 2+ (4.41)
dt 2 t
with boundary condition x = 0, t = 1. The solution is:
√
xs = −2 t + 2t (4.42)

which is valid for 2 ≤ t ≤ 4. At t = 4 the shock meets the first characteristic v = 1 of the
expansion fan. For t > 4 the shock speed is constant having the value Vs = 23 . The shock
moves uniformly; all information about what happened for t < 4 is lost.
114 CHAPTER 4. BURGERS’ EQUATION FOR SIMPLE WAVES

Figure 4.20: Shock eats expansion

The shock strength for 2 < t < 4 can be calculated as

xs
∆v = vpost − vpre = 2 − .
t
Using (4.42) there follows:
2
∆v = √ , (4.43)
t
indicating shock decay from ∆v = 2 at t = 1 to ∆v = 1 at t = 4.
Chapter 5

Traffic waves

5.1 Definitions, traffic equation

As an application of non-linear unsteady waves we will study traffic flow on a one way, one-
lane road. A traffic equation is derived assuming a continuum model of traffic flow.
Consider a one-lane highway with cars having an average length l. Let the traffic speed be
u (x, t), depending on time (t) and location (x), see figure 5.1.

The general conservation law for a quantity f (x̄, t) reads

∂ y x
f (x̄, t) dV + f (x̄, t) ū · n̄dS = s˙f (5.1)
∂t
V ∂V

s˙f is a source term denoting the production of f per unit time, V is the control volume fixed
in space domain.
In case of a one-dimensional (spatial) problem with x ∈ {x1 , x2 } as control space this results in:
Z x2
∂
f (x, t) dx + f (x, t) u (x, t) kxx21 = s˙f . (5.2)
∂t x1
Regarding traffic flow f (x, t) is equal to the number of cars per unit length, or the car density:
ρ(x, t)
number of cars in control space
ρ(x, t) =
length of control space
For ρ we have the limit values: ρmax = 1/l meaning bumper to bumper traffic and ρmin = 0
meaning no traffic.

Figure 5.1: Traffic flow on a one-lane road.

115
116 CHAPTER 5. TRAFFIC WAVES

If the conservation law of equation (5.2) with s˙ρ = 0 (no cars are destroyed nor created) is
applied to an infinitesimal small control space: x1 = x; x2 = x + dx, the result is given by:
∂ρ ∂
= (ρu) = 0 (5.3)
∂t ∂x
The term ρu is identified as the car flux: the number of cars passing a fixed location per unit
time.
To solve the problem a relation between ρ and u is needed. A simple model for the density-
speed relation is the linear relation:
u
= 1 − ρl (5.4)
V
Here V is the speed limit of the cars. If ρ = 1/l there is no free space between the successive
cars and the car flow stagnates (u = 0); if ρ → 0 no one else is on the road and cars approach
the speed limit.
Of course the density-speed relation can be chose more realistic e.g. by accounting for the fact
that the speed limit already appears at ρ = ρ∗ 6= 0. Another improvement of the density-speed
relation follows if one takes the fact into account that the minimum intermediate space between
successive cars depends on the velocity. The higher the velocity the larger the necessary
intermediate space to decelerate a car to stand still.
Inserting equation (5.4) into (5.3) the traffic equation in dimensional form becomes:
∂ρ ∂
+ {ρV (1 − ρl)} = 0. (5.5)
∂t ∂x
Let us introduce the dimensionless variables

ũ = u/V, ρ̃ = ρ/(1/l),

and the dimensionless coordinates

x̃ = x/l, t̃ = t/(l/V ).

Then the traffic eqaution simplifies to

∂ ρ̃ ∂
+ {ρ̃(1 − ρ̃)} = 0 (5.6)
∂ t̃ ∂ x̃
and the speed-density relation becomes

ũ = 1 − ρ̃. (5.7)

This relation is shown in figure 5.2.

Another interesting variable is the car flux: ρu. In dimensionless form we have:

ρ̃ũ = ρ̃(1 − ρ̃). (5.8)

The car flux attains a maximum value at ρ̃ = 1/2 and ũ = 1/2, (see figure 5.2). So the most
efficient speed that allows the highway to be used at its maximum capacity (maximum car
flux) is not the speed limit but (in the case of the linear speed-density relation) just the half
of this speed limit. (So to use the road at its maximum capacity don’t drive at maximum
speed!)
5.2. CHARACTERISTICS AND DISCONTINUITIES 117

Figure 5.2: Speed-density relation and car flux versus car desity

5.2 Characteristics and Discontinuities

Take the traffic equation in the form of equation (5.6)
∂ ρ̃ ∂
+ {ρ̃(1 − ρ̃)} = 0,
∂ t̃ ∂ x̃
and write it in the form of a non-linear convection equation i.e. (omitting tildes):
∂ρ ∂ρ
+ c(ρ) = 0, (5.9)
∂t ∂x
where c(ρ) is the characteristics speed given by

c(ρ) = 1 − 2ρ. (5.10)

Equation (5.9) is interpreted such that along curves dx dt = c(ρ), ρ is constant.

dx
The curves dt = c(ρ) are characteristics of equation (5.9); since the slope of the charactristics
depends solely on ρ they must be straight lines in the (t,x)-plane. This slope, 1 − 2ρ, depends
on the solutions ρ(x, t); so their graph in the (t,x)-plane is not known a priori.
On the interval 0 ≤ ρ ≤ 1, c(ρ) varies between the values c(0) = 1 and c(1) = −1.
Observe that
d(ρu)
c(ρ) = , (5.11)
dρ
hence the characteristic speed is found as the slope of the graph ρu = f (ρ) as shown in figure
5.2. Assume that the initial values of ρ(x, t) are given at: t = 0 as ρ(x, 0) = φ(x).
The solution ρ(x, t) is determined as follows; draw the straight characteristic through (x, t)
backwards. This line is determined by the equation: x − c(ρ)t = constant, so:

ρ(x, t) = ρ(x − ct, 0) = φ(x − ct).

Using c = 1 − 2ρ there follows the implicit expression for ρ(x, t):

ρ(x, t) = φ(x − (1 − 2ρ)t). (5.12)

Equation (5.12) describes a closed form solution for continuous traffic flow satisfying the initial
conditions ρ(x, 0) = φ(x).
118 CHAPTER 5. TRAFFIC WAVES

Figure 5.3: Characteristic

Apart from these continuous solutions also discontinuities may appear. Apply the jump re-
lations for 1D steady flow: apply equation (1.39) to equation (5.6) which yields (omitting
tildes):

cs [ρ] − [ρ(1 − ρ)] = 0, (5.13)

with cs being the jump velocity and [−] denoting the jump: (−)R − (−)L , so one finds:

cs (ρR − ρL ) = ρR (1 − ρR ) − ρL (1 − ρL )
= (ρR − ρL )(1 − ρR − ρL ),

or

cs = 1 − (ρR + ρL ). (5.14)

Now compare the shock velocity cs with the characteristic speed on either side of the shock
discontinuity. At the left side : c = cL = 1 − 2ρL ; on the right side: c = cR = 1 − 2ρR .
Expressing cs in terms of cR and cL one obtaines

1
cs = (cL + cR ), (5.15)
2
reflecting the property that the shock speed cs is the mean value of the characteristic speeds
cL and cR .
Returning to equation (5.13) and remembering that ρ(1−ρ) denotes the car flux ρu, the shock
speed cs can also be viewed as

[ρu]
cs = , (5.16)
[ρ]

being the slope of the chord LR in the graph (ρu) vs (ρ), see figure 5.4; L and R are two points
on this graph representing the left- and right side of the discontinuity respectively.
5.3. TRAFFIC LIGHT 119

Figure 5.4: Shock speed Cs in graph (ρu) vs ρ

5.3 Traffic light

Consider the traffic situation where a train of cars is lined up at a red traffic light. This
situation is modelled by the Riemann problem with initial values (t ≤ 0):

1 , x <0
ρ(x, 0) =
0 , x >0

Let the light turn green at t = 0.

View the (t,x)-plane. The characteristics emanating from x < 0 (t = 0) have the slope
dx dx
dt = −1 and those emanating from x > 0 (t = 0) have the slope dt = +1.
From x = 0 a fan of straight characteristics appear. Along characteristics the density is
constant, so the solution for t 6= 0 can be constructed very easily.
Take a certain time level t = t1 and let us find the solution for ρ(x, t1 ) and u(x, t1 ). A single
characteristic in the fan is given as:

x = ξt.

The particular value of ξ identifies a particular characteristic. The slope of the characteristic
is:
dx
= ξ = 1 − 2ρ.
dt
So the traffic density depends on the slope, yielding

1−ξ 1 − x/t t−x

ρ= = = . (5.17)
2 2 2t
Since u = 1 − ρ one obtaines
t+x
u= . (5.18)
2t
120 CHAPTER 5. TRAFFIC WAVES

Figure 5.5: Distribution of ρ(x) and u(x) at t = t1

Observe that u and ρ are both linear functions of x at constant t1 see figure 5.5. When time
progresses the original discontinuity at x = 0 relaxes into a continuous distribution along the
x-axis at an arbitrary time instant.

What is the trajectory of a car? Take the particular car lined up at x = −1. This car starts
moving at t = 1 the time instant that the characteristic x = −t is passed. The trajectory of
the car is found as the solution of the ODE
dxc 1 x
=u=1−ρ= 1+ , (5.19)
dt 2 t
with the initial conditions x = 1, t = 1.
The solution of equation (5.19) that satisfies these conditions is
√
xc (t) = t − 2 t, t ≤ 1. (5.20)

For t → ∞ the car path becomes parallel to the characteristics dx

dt = +1. The car attains its
c

maximum speed (u = 1) as t → ∞. A picture of the car path is shown in figure 5.6.

5.4 A chain collision

Consider the Riemann problem with initial values

1/2 , x < 0
ρ(x, 0) =
1 , x>0
which models a traffic jam.
5.4. A CHAIN COLLISION 121

Figure 5.6: Car path

Figure 5.7: Cars backing up

The characteristics emanating from x < 0 have a slope dx dt = 0 and those emanating from
x > 0 have a slope dx
dt = −1. Both sets of characteristics intersect; a shock has to be formed.
1
The shock speed is cs = 1 − (ρL + ρR ) = − 2 .
Observe that the shock travels ’upstream’. It acts as a carrier of information telling the on-
coming cars that a traffic jam is ahead, see figure 5.7.
1
Car paths in domain Àare straight lines with slope dx dt = u = 1 − ρ = 2 . When the shock
has passed, the cars are decelerated abruptly to stand still. The shock represents the front of
cars coming to a stand still.
122 CHAPTER 5. TRAFFIC WAVES
Chapter 6

Two-Dimensional Inviscid Flow

6.1 Governing flow equations

In paragraph (1.5) we have derived the Euler equations which are the appropriate equa-
tions to model compressible flows where the effects of viscosity, heat conduction and exter-
nal forces have been neglected. These equations for two dimensional flows written in the
conservative form (omitting tildes) are
∂U ∂F ∂G
+ + =0 (6.1)
∂t ∂x ∂y
with
     
ρ ρu ρv
 ρu   p + ρu2
 , G =  ρvu 
  
U =
 ρv  , F =  ρuv
 
  p + ρv 2 
ρE ρuH ρvH
U is the vector of state variables, F and G are flux vectors.
The non-conservative form of these equations read
∂ρ ∂ρ ∂ρ
ρ ∂u ∂v 
∂t + u ∂x + v ∂y + ∂x + ρ ∂y = 0 





∂p
ρ ∂u + ρu ∂u ρv ∂u

∂t ∂x + ∂y + ∂x = 0 


(6.2)
∂p
ρ ∂v ∂v ∂v 
∂t + ρu ∂x + ρv ∂y + ∂y = 0 







ρ ∂E + ρu ∂H + ρv ∂H = 0

∂t ∂x ∂y

For the analysis of the vector equation (6.1) it is useful to define the Jacobian matrices
(U )
Ac = dFdU ∂Fi

with (Ac )i,j = ∂U 
j 
(6.3)
dG(U ) ∂Gi 
Bc = dU with (Bc )i,j = ∂Uj


Since F = F (U ) we can write

∂F dF ∂U
= , (6.4)
∂x dU ∂x

123
124 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

and in a similar way

∂G dG ∂U
= . (6.5)
∂y dU ∂y

Inserting (6.4) and (6.5) into (6.1), the last equation can rewritten as

∂U ∂U ∂U
+ Ac + Ac =0 (6.6)
∂t ∂x ∂y

In order to compute Ac and Bc the components of F (U ) and G(U ) have to be expressed in

terms of the components of U = (U1 , U2 , U3 , U4 )T .
A straightforward analysis leads to

F1 = U2 




2 2 2
n o 
U2 U3 U2 
F2 = (γ − 1) U4 − 2U1 − 2U1 + U1 



(6.7)
F3 = UU2 U1 3








2 2
n o 
U2 U U 
F =
4 U1 γU − (γ − 1)
4
2
− (γ − 1)
2U1
3 
2U1

and

G1 = U3 




U2 U3

G2 =


U1 


n
U22 U32
o
U32
(6.8)
G3 = (γ − 1) U4 − 2U1 − 2U1 + U1







o 
U2 U2
n 
U3 
G4 = U1 γU4 − (γ − 1) 2U21 − (γ − 1) 2U31 

Equation (6.6) is a non-conservative form of the flow equations expressed in terms of the
state vector Ū = (ρ, ρu, ρv, ρE)T . However in a lot of applications it is usefull to work with
equations expressed in terms of the state vector of primary variables W = (ρ, u, v, p)T . Assume
that the equations expressed in terms of W have the form

Wt + A(W )Wx + B(W )Wy = 0 (6.9)

Can we find A(W ) and B(W )?

Considering (6.2) it appears that it is almost expressed in these primary variables apart from
the last equation which still contains the non-primary variables E (total energy) and H (total
enthalpy).
But this equation can be replaced by the energy equation for adiabatic flow

∂p ∂p ∂u ∂p ∂v
+u + ρa2 +v + ρa2 =0 (6.10)
∂t ∂x ∂x ∂y ∂y
6.1. GOVERNING FLOW EQUATIONS 125

Which is a direct consequence of the adiabatic flow assumption:

DS
=0
Dt
expressing that entropy is constant along a particle path.
From DS
Dt = 0 it follows in 2D flows

∂S ∂S ∂S
+u +v =0 (6.11)
∂t ∂x ∂y
Since the entropy change dS is proportional to dp − a2 dρ, equation (6.11) may be written as:
   
∂p 2 ∂ρ ∂p 2 ∂ρ ∂p 2 ∂ρ
−a +u −a +v −a =0
∂t ∂t ∂x ∂x ∂y ∂y
All ρ-derivatives can be eliminated using the unsteady form of the continuity equation
∂ρ ∂ρ ∂ρ ∂u ∂v
+u +v = −ρ −ρ
∂t ∂x ∂y ∂x ∂y
Then equation (6.10) is obtained.

From (6.2) and (6.10) the matrices A(W ) and B(W ) are easily found as
   
u ρ 0 0 v 0 ρ 0
 0 u 0 ρ1  , B(W ) =  0 v 0 01 
 
A(W ) =  0 0  0 0 v (6.12)
u 0  ρ

u ρa2 0 u 0 0 ρa v 2

Of course it is also possible to retrieve the primitive form: (6.9) from the conservative form:
(6.1) or (6.6) by noting that from the relation W = W (U ) it follows
dW dU
dW = dU = T−1 dU, T= (6.13)
dU dW
Formally (6.6) is written as
∂W ∂W ∂W
T + Ac T + Bc T =0
∂t ∂x ∂y
or, if T is non-singular
∂W ∂W ∂W
+ T−1 Ac T + T−1 Bc T =0 (6.14)
∂t ∂x ∂y
Apparently we can obtain A and B from Ac and Bc by the relations

A = T−1 Ac T, B = T−1 Bc T (6.15)

with
1 0 0 0
 
 u ρ 0 0 
T=
 v 0 ρ 0

 (6.16)
u2 +v 2 1
2 ρu ρv γ−1
126 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

6.2 Characteristic form for 2D unsteady flow

Now we have the objective to find, if possible, the characteristics and their related invariants
for 2D unsteady flow expressed in primary variables.
Remembering for 1D unsteady flows we found
∂J + ∂J +
∂t + (u + a) ∂x = 0,
∂S ∂S
∂t + u ∂x = 0,
∂J − ∂J −
∂t + (u − a) ∂x = 0,

with
dp
dJ ± = du ± .
ρa
Can we find something similar for 2D unsteady flows?

If so, then it should have the form

Vt + AVx + BVy = 0 (6.17)

Where A and B are diagonal and V is a new state vector related to W .

Apparently W itself is not a proper state vector because of A(W ) and B(W ) contain some
off-diagonal terms. They are definitely not-diagonal.
How to get the proper state vector V and the corresponding matrixes A and B?
Previously in the 1D case we used the "add & substract" method and we relatively easy
found the diagonalized form by trial and error. Using this technique here does not result into
a complete succes. For instance, try to combine the x-momentum equation (6.2) with the
energy equation in the form (6.10); first multiply the latter with (ρa)−1 to get the proper
dimensions
∂u ∂u 1 ∂p
x-momentum : ∂t + u ∂x + ρ ∂x + v ∂u
∂y = 0
1 ∂p u ∂p ∂u v ∂p ∂v
energy equation : ρa ∂t + ρa ∂x + a ∂x + ρa ∂y + a ∂y = 0

If we add them together we obtain:

   
1 1 1
ut + pt + (u + a) ux + px + v uy + py + avy = 0
ρa ρa ρa
∂ ∂ ∂
We observe the appearance of the partial derivatives ∂t , ∂x and ∂y of a single term of which
1
du + ρa dp is the total differential, however the extra term avy has appeared, which destroys
the diagonal form. Of course, this was just a first try motivated by our previous succes in 1D;
we may have overlooked other possibilities.
It is now time to develop a straightforward method of diagonalization.
Let us develop this technique first for the 1D unsteady case; here we already know the outcome!

Diagonalization for unsteady 1D flow The unsteady 1D flow field is governed by (6.9)
and (6.12)

Wt + A(W )Wx = 0 (6.18)

6.2. CHARACTERISTIC FORM FOR 2D UNSTEADY FLOW 127

in which W is the state vector of primary variables W = (ρ, u, p)T and A is given by
 
u ρ 0
A =  0 u ρ1  (6.19)
0 ρa u2

Introduce the change of state variables from W to V by

dW = SdV (6.20)

and find S so that it diagonalizes the coefficient matrix in the resulting equation

SVt + A(W )SVx = 0

or if S is invertible

Vt + S−1 ASVx = 0 (6.21)

where S−1 AS is now a diagonal matrix

 
λ1 0 0
S−1 AS = Λ =  0 λ2 0  (6.22)
0 0 λ3

whose non zero entries lie on the diagonal and follows from the eigenvalues λ i , i = 1, 2, 3 of
the matrix A. The eigenvalues of A are
λ1 = u − a (6.23)
λ2 = u (6.24)
λ3 = u + a (6.25)
From linear algebra we know that the matrix S is build up by the matrix of right (column)
eigenvectors of A.
Right eigenvectors Ri of A satisfy the relation

ARi = λi Ri (6.26)

With the eigenvalues found in (6.23) the eigenvectors (apart from an arbitrary factor) are
easily obtained

T
λ1 = u − a ⇒ R1 = ρ, −a, ρa2 ,
λ2 = u ⇒ R2 = (1, 0, 0)T , (6.27)
T
ρ, a, ρa2 ,


λ3 = u + a ⇒ R3 =

So the matrix S is found as

 
ρ 1 ρ
S =  −a 0 a  . (6.28)
ρa2 0 ρa2

With the help of (6.20) the new state vector V is determined from

dV = S−1 dW, (6.29)

128 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

which shows that the inverse of S: S−1 has to be determined.

If S is given by (6.28) then S−1 is
 
0 −ρa 1
1 
S−1 = 2ρa2 0 −2ρ  , (6.30)
2ρa2
0 ρa 1
so that the new state vector V becomes
    
0 −ρa 1 dρ −ρadu + dp
dV ∝  2ρa2 0 −2ρ   du  =  2ρa2 dρ − 2ρdp 
0 ρa 1 dp ρadu + dp
or
−ρadJ −
−ρadJ −
   

dV ∝  −2ρ dp − a2 dρ  =  −2ρp
cv dS
 (6.31)
ρadJ + +
ρadJ

Here J − , S and J + are the characteristic variables or invariants appearing in 1D unsteady

flow.
If the eigenvectors R1 , R2 and R3 are multiplied by an appropriate but inconsequential factor
the previously defined state variables dJ + , dS and dJ − are obtained immediately.
Anyhow, the diagonalization technique, just introduced appears to be a straightforward method
to find the characteristic speeds and the components of the new state vector (or invariants)
dV .
However to attain dV we had to find the inverse S−1 . This gives some labour; a faster method
is to go immediately for S−1 which can be accomplished by working with the left (row) eigen-
vectors of A. This will be accomplished now
We want to diagonalize the equation

Wt + AWx = 0,

assume

dW = L−1 dV, (6.32)

where V is the new state vector and L−1 is the matrix that diagonalizes the origonal equation.
Inserting (6.32) into (6.18) one obtaines

L−1 Vt + AL−1 Vx = 0

or

Vt + LAL−1 Vx = 0 (6.33)

The objective is to find L such that LAL−1 is diagonal,

LAL−1 = Λ. (6.34)

From Linear Algebra we know that the rows of L are the left (row) eigenvectors L i of A
satisfying the relation

Li A = λi Li (6.35)
6.2. CHARACTERISTIC FORM FOR 2D UNSTEADY FLOW 129

Proof Arrange the row vectors as the rows of matrix L, so we have

LA = ΛL1
Multiplying with L−1 there results
LAL−1 = ΛL1 L−1 = Λ
Q.E.D
When L is obtained, dV follows explicitly from (6.32).
Let us now demonstrate the left eigenvector method to determine dV directly, e.g. for the
unsteady 1D case. Matrix A is already given in (6.14), its eigenvalues were found (6.23) as
λ1 = u − a, λ2 = u, λ3 = u + a
The left eigenvectors Li (corresponding to λi )of A are obtained as follows.
For λ1 = u − a assume L1 = {`11 , `12 , `13 }.
With (6.35) we have:
 
u ρ 0
(`11 , `12 , `13 )  0 u ρ1  = (λ1 `11 , λ2 `12 , λ3 `13 ) .
0 ρa2 u
or
`11 a = 0
`12 ρ + `12 a + `13 ρa2 = 0
`12 ρ1 + `13 a = 0
Solving for `11 , `12 and `13 one obtains (apart from an arbitrary factor):
`11 = 0, `12 = −ρa, `13 = 1,
making the left eigenvector L1 = (0, −ρa, 1).
In a similar way the left eigenvectors L2 and L3 can be calculated.
Summarizing the results gives:
λ1 = u − a → L1 = (0, −ρa, 1)
λ2 = u → L2 = (−a2 , 0, 1)
λ3 = u + a → L3 = (0, ρa, 1)
The matrix L has the form
 
0 −ρa 1
L =  −a2 0 1  (6.36)
0 ρa 1
and dV = LdW gives the result:
−ρadJ −
    
0 −ρa 1 dρ
dV =  −a2 0 1   du  =  cpv dS  (6.37)
0 ρa 1 dp ρadJ +
1
Observe that for slightly different normalisation of the left eigenvectors, i.e. L̃1 = − ρa L1 ,
L̃2 = cpv L2 and L̃3 = ρa
1
L3 will yield the previously defined state variables J + and J − directly.
The left eigenvector method is a powerful tool that enables us to find directly the new state
variables in the diagonalized formulation. So let us apply this technique now in the case of a
two dimensional unsteady flow.
130 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

Diagonalization of the 2D flow equations Unsteady 2D inviscid and non-heat conduct-

ing adiabatic flow is governed by the vector equation

Wt + A(W )Wx + B(W )Wy = 0 (6.38)

W is the state vector of primary variables W = (ρ, u, v, p) and A and B are given by
   
u ρ 0 0 v 0 ρ 0
 0 u 0 ρ1   0 v 0 0 
A=  0 0
, B=  0 0 v 1
 (6.39)
u 0  ρ

0 ρa2 0 u 0 0 ρa2 v

The objective is to find a transformation of variables dV = LdW such that the coefficient
matrices in the transformed equation are both diagonalized. Substitute in (6.38) now

dW = L−1 dV,

and multiply with L; this gives

Vt + LAL−1 Vx + LBL−1 Vy = 0 (6.40)

From linear algebra we remember that A and B can be simultaneously diagonalized by the
same transformation L if A and B have the same eigenvectors. In that case A and B commute,
ie AB = BA.
However from (6.39) where A and B are defined it is easily checked that A and B do not
commute!
This disappointing but important result indicates that no set of equations of the form
∂ ∂Vi ∂Vi
Vi + λi + µi =0 (6.41)
∂t ∂x ∂y
exists which is equivalent with the unsteady Euler equations.
Physical experience teaches us the well known fact that sound propagates in all directions. If
(6.41) where true then sound would travel only in discrete directions indicated by the velocity
vectors (λ i , µi ). In reality there are no such directions; sound propagates omni-directionally.

6.3 Wave propagation in two-dimensional flow

Since we know from the previous paragraph that sound travels in all directions, let us study
the behaviour of plane-wave solutions of the Euler-equations.
Assume in two dimensions that plane waves travel with velocity λ in a direction ` (having an
angle θ with respect to the x-axis) through a fluid which itself moves with a velocity q in the
direction s (having an angle ϕ with respect to the x-axis).
Decomposing q into its components along the x- and y-axis yields u = q cos ϕ, v = q sin φ,
figure 6.1 shows this wave propagation set up.
The flow situation sketched in figure 6.1 is an unsteady wave phenomenon that has to be a
solution of
∂W ∂W ∂W
+A +B =0 (6.42)
∂t ∂x ∂y
6.3. WAVE PROPAGATION IN TWO-DIMENSIONAL FLOW 131

Figure 6.1: Travelling planar acoustic waves

The assumption of plane acoustic waves suggests solutions of the type:

W = W0 f (x cos θ + y sin θ − λ(θ)t) (6.43)
where f is an arbitrary function of the argument X = x cos θ + y sin θ − λ(θ)t. Equation (6.43)
represents wave phenomena having W is constant along lines
x cos θ + y sin θ − λ(θ)t = const (6.44)
in the x,y-plane. These straight lines travel with the speed λ(θ) in the direction ` when time
elapses.
In order to ensure that such waves can exist in the real world they have to satisfy equation
(6.42). From equation (6.43) we obtain the partial derivation
df df df
Wt = −W0 λ, Wx = −W0 cos θ, Wy = −W0 sin θ.
dX dX dX
Substitution of these expressions in (6.42) yields
 
df 
+W0 −λI + A| cos θ {z
+ B sin θ} = 0
dX
Aθ

or
df
+W0 (Aθ − λI) = 0 (6.45)
dX
To have non-trivial solutions of this homogeneous set of equations we need
det (Aθ − λI) = 0 (6.46)
or
qθ − λ ρ cos θ ρ sin θ 0
 
1
 0 qθ − λ 0 ρ cos θ 

det 
0 0 qθ − λ 1 =0 (6.47)
ρ sin θ

0 ρa2 cos θ ρa2 sin θ qθ − λ
132 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

Where qθ = u cos θ + v sin θ denotes just the projection of the flow speed q in the ` direction.
Solving (6.47) for λ we obtain
n o
(qθ − λ)2 (qθ − λ)2 − a2 = 0 (6.48)

or

λ1 = qθ − a 
λ2 = λ3 = qθ (6.49)
λ4 = qθ + a


This result tells us that physical plane waves propagate either with the acoustic speed with
respect to the flow (λ1 and λ4 ) or they propagate with the flow (λ2 = λ3 = qθ ).
The first two are just the forward and backward acoustic waves. The last two are convected
with the flow; see also the graph in figure 6.2.

Figure 6.2: Propagation of wave front

Upstream and downstream moving waves (6.49) demonstrates clearly that the travel-
ling speed λ of a plane wave is not just a constant but it depends on the direction e.g. θ in
which the wave front travels. Let us study this dependancy, first for acoustic waves and then
also for convecting waves.
The travelling speed of acoustic signals is given by

λ1,4 = qθ ∓ a = u cos θ + v sin θ ∓ a

Introducing the flow angle ϕ by u = q cos ϕ, v = q sin ϕ we found λ1,4

λ1,4 = q cos(ϕ − θ) ∓ a (6.50)

Depending on the Mach number M = aq , the polar graph λ(θ) shows three distinct graphs,
one for M < 1, one for M = 1 and one for M > 1, see figure 6.3.
6.3. WAVE PROPAGATION IN TWO-DIMENSIONAL FLOW 133

Figure 6.3: Graphs for λ1,4(θ)

The graph of (6.50) is known as the "cardiode", a figure that has the shape of a heart.
From figure 6.3 we draw the following conclusions. Acoustic waves always travel with the
speed of sound with respect to the fluid.
Since λ1 and λ4 are travelling speeds in the laboratory frame we see fast and slow moving
waves and also forward and backward moving waves.
Consider figure 6.3 part a being the subsonic case. If l and q have the same direction e.g.
θ = ϕ then λ1 = q − a and λ4 = q + a. Since q < a this implies λ1 < 0 and λ4 > 0; λ1
represents the backward moving waves and λ4 represents the forward running waves.
If l and q are perpendicular to each other e.g. θ = ϕ + π2 then (6.50) says λ4 = ∓a, λ1 = ±a;
both waves move transversally w.r.t. the fluid flow with the sound velocity a. Observe that
this result is valid irrespective of the fact whether the flow is subsonic, sonic or supersonic.
Take for simplicity ϕ = 0, then from (6.50) one obtains λ1,4 = q cos θ ∓ a. Since λ4 is always
positive, it represents forward moving waves regardless the Mach number M = aq . On the
other hand λ1 waves can be either forward moving waves (q cos θ < a) or backward running
waves (q cos θ > a).
In subsonic flow q < a, λ1 waves move backward but in supersonic flows q > a, the λ1 wave
can move backwards also if the propagation direction e.g. the angle θ exceeds a certain value
θ∗ . In case ϕ
= 0 we have λ1 = q cos θ − a. The critical value of
θ appears at λ1 = 0, so
θ∗ = cos−1 M 1
. In that case
the wave front has an angle sin−1 M 1
with respect to the flow
−1 1
direction. However sin M is just the Mach angle being the angle of a Mach line with
respect to the flow direction, see figure 6.4.
Since λ1 is zero in this perpendicular case, it tells us that wave fronts inclined at the Mach
angle µ do not propagate nor upstream nor downstream.
On the other hand if the propagation angle |θ| > θ∗ then λ1 < 0, and λ1 waves will move
upstream (backwards) regardless of the fact that the flow is supersonic.
Based on this observation we may draw the very important conclusion that wave fronts inclined
at angles smaller than µ with respect to the flow direction, will travel upstream in a supersonic
134 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

Figure 6.4: Wave front inclined at Mach angle /mu

flow, see figure 6.5.

A direct result of this phenomenon is the well-known effect that upstream travelling waves

Figure 6.5: Wave front travelling upstream

generated in the wake of an airfoil at supercritical conditions do not necessarily build up a

shock wave when entering the local supersonic zone because these waves can travel through
this zone provided they are inclined at angles smaller than the Mach angle.
Finally let us consider the convecting (entropy) waves represented by λ 2 = λ3 = qθ , using
u = q cos ϕ, v = q cos ϕ we find for λ2,3

λ2,3 = q cos(θ − ϕ) (6.51)

The polar graph of this equation is a circle with radius 2q having its center at u v



2, 2 ; the
graph is depicted in figure 6.6.
Figure 6.6 shows that the circle goes throught the origin u = 0,v = 0.
6.3. WAVE PROPAGATION IN TWO-DIMENSIONAL FLOW 135

Figure 6.6: Graph λ(θ) for convecting waves and entropy waves

Hyperbolicity Because all plane waves admitted by (6.42) are real for any value of θ e.g.
the propagation direction, this equation is truly hyperbolic and the initial value problem is
well-posed.
Well-posedness means that the solution has to satisfy the requirements:

1. that a solution exists,

2. that the solution is unique and

3. that the solution depends continuously on the boundary conditions.

Enveloping waves It is a known fact that travelling waves can coalesce to form envelopes.
A well-known example is a plane wave which can be regarded as the envelope of circular waves
emitted by a row of point sources, figure 6.7. From the plane-wave speeds we can also derive
the shape of wave fronts caused by these point disturbances.
According to Huygens’ principle, the envelope of all possible plane waves initially going

Figure 6.7: Plane wave as envelope

through one point, determine the wave front at later times.

Consider a point source where at t = 0 waves in all directions are present.
Let us first pay attention to entropy waves λ2 = λ3 = qθ . For reasons of simplicity we take
q q

ϕ = 0; in that case the polar graph λ2,3 (θ) is a circle with radius 2 and centre 2 . The
entropy wave pattern at t = 0 is depicted in figure 6.8.
In particular the waves 1, 2, 3, 4 and 5 are shown all intersecting each other in the common
point P . Where do we find these waves one unit of time later? Since we are dealing with
entropy waves they are convected with the speed λ(θ) = q θ = q cos θ.
So at t = 1 the wave front has travelled a distance d(θ) = λ(θ) × 1; d(θ) = q cos θ.
Consequently at t = 1 all waves now intersect each other at the point Q(q, θ), see figure 6.9.
136 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

Figure 6.8: Entropy waves at t = 0, all intersecting at common point P in the 2D space

The envelope of all plane waves is the point Q; it refects the known property that entropy is
constant along a particle path. Since ϕ = 0 the line P Q is a particle path so the entropy in
P at t = 0 is equal to the entropy in Q at t = 1.
Consider again the point source P at t = 0; now the waves which are present in all directions
are identified as acoustic waves. These waves travel with the speed λ1,4 = q cos θ ∓ a; observe
that we have again taken ϕ = 0. Where do we find these waves one unit of time later? As
in the case of the entropy waves we can determine the travel distance of each particular wave
front easily by the formula:

d∓ = d1,4 (θ) = q cos θ ∓ a (6.52)

Assume that q is constant in the whole domain then one can proof that the wave fronts build
up the envelope:

(x − q)2 + y 2 = a2 (6.53)

being a circle with radius a and centered at the point x = q, y = 0.

The proof is left as a homework problem!
Figure 6.10 shows the enveloping wave fronts for the supersonic case q > a.
The slow moving waves are indicated with the superscript "-", the fast moving waves carry
  of the circular envelope going through the source point P
the supersript "+". The tangent
−1 a
makes the angle µ = sin q with the flow direction. It is the well known Mach angle.
Therefore the envelope that is formed by all planar waves emanating from the point source P
(at t = 0) is nothing else then the location onto sound is propagated at time t = 1. It is the
response of the fluid to the point disturbance at P .

Determination of envelopes To determine the geometrical shape of an envelope of 2D

plane waves, consider two neighbouring plane wave fronts at time level t. Assume a wave front
L1 that travels with speed λ(θ) in the direction θ; at time t its position is represented by the
6.3. WAVE PROPAGATION IN TWO-DIMENSIONAL FLOW 137

Figure 6.9: Individual entropy wave fronts at t = 1

Figure 6.10: Individual acoustic waves at t = 1, supersonic main flow q > a

graph:

L1 : x cos θ + y sin θ = λt (6.54)

138 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW

Let wave front L2 have a slightly different speed λ + dλ and a slightly different propagation
direction θ + dθ. At time t its position is represented by:

L1 : x cos(θ + dθ) + y sin(θ + dθ) = (λ + dλ)t (6.55)

The intersection point (xI , yI ) of L1 and L2 can be obtained by solving the following two
equations:

xI cos θ + yI sin θ = λ
0 (6.56)
−xI sin θ + yI cos θ = dλ dθ = λ

The latter is obtained if (6.54) is substracted from (6.55) and writing cos(θ + dθ) = cos θ −
dθ sin θ, sin(θ + dθ) = sin θ + dθ cos θ.
The solution of (6.56) may be written in the vector/matrix form
    
xI cos θ − sin θ λ(θ)
= (6.57)
yI sin θ cos θ λ0 (θ)

This is the equation for the wave front envelope.

Another way of finding the point response is to look at the group speed of the plane waves;
for this purpose we write the plane wave information as a dispersion relation:

ω̄ = λ(θ)k̄,

here k is the unit vector in the propagation direction (see figure 6.11). λ(θ) is the "phase
speed":

|ω| ω
λ(θ) = =
|k| k

To get the "group speed" in polar coordinates take the derivatives

Figure 6.11: k in propagation direction

∂ω
vgk = ∂k = λ(θ)
(6.58)
vgθ = 1 ∂ω
k ∂θ = ∂λ
∂θ = λ0 (θ)
6.3. WAVE PROPAGATION IN TWO-DIMENSIONAL FLOW 139

In Carthesian coordinates:
    
vgx cos θ − sin θ λ(θ)
= (6.59)
vgy sin θ cos θ λ0 (θ)

This is exactly the same formula derived earlier for the wave front. Both procedures are
equivalent.
140 CHAPTER 6. TWO-DIMENSIONAL INVISCID FLOW
Chapter 7

Steady Two-Dimensional Flow

7.1 Characteristic directions

In chapter 6 we attempted to diagonalize the vector equation:

Wt + AWx + BWy = 0, (7.1)

in order to find the characteristic equations equivalent to the unsteady Euler equations; in
(7.1) W represents the state vector W = (ρ, u, v, p)T . Unfortunately it turned out that
diagonalization was impossible, because of the matrices A and B did not commute, e.g.
AB 6= BA.
Now we aim less ambitiously to diagonalize the steady Euler equations, (6.2);

∂ρ ∂ρ ∂u ∂v
u +v +ρ +ρ = 0, (7.2)
∂x ∂y ∂x ∂y
∂u ∂u ∂p
ρu + ρv + = 0, (7.3)
∂x ∂y ∂x
∂v ∂v ∂p
ρu + ρv + = 0, (7.4)
∂x ∂y ∂y
∂H ∂H
u +v = 0. (7.5)
∂x ∂y
The last equation is a direct consequence of adiabatic flow and tells that the total enthalpy
H is constant on a streamline. Another consequence of the adiabatic flow assumption is that
also the entropy is constant along a streamline.

Proof
By definition, the total enthalpy H may be written as
p 1 2
H =e+ + q . (7.6)
ρ 2
Inserting this into the energy equation (7.5) yields:
   
γ px p γ py p
u − 2 ρx + v − 2 ρy + uqqx + vqqy = 0, (7.7)
γ−1 ρ ρ γ−1 ρ ρ

141
142 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

using the momentum equations (7.3) and (7.4):

u v
uqqx + vqqy = u(uux + vvx ) + v(uuy + vvy ) = − px − py
ρ ρ

Inserting in equation (7.1) and rearranging the terms:

   
px ρx py ρy
u −γ +v −γ = 0. (7.8)
p ρ p ρ
 
using ds = cv dpp − γ dρ
ρ we obtain from (7.8) the result,

∂s ∂s
u +v = 0. (7.9)
∂x ∂y

telling us that the entropy s is constant along a streamline. 2

Since this result is obtained from the energy equation and the momentum equations it implies
that one of them, e.g. the y-momentum equation can be replaced by the entropy equation to
keep a closed system.
In order to work with a system that is as simple as possible we continue with the following
system:

uρx + ρux + vρy + ρvy = 0,

uux + vuy + ρ1 px = 0,
(7.10)
usx + vsy = 0,
uHx + vHy = 0.

Let us now choose as primary variables the two velocity components u and v and the ther-
modynamic variables s (entropy) and H (enthalpy) and introduce the new state vector :
W = (u, v, s, H)T .
So we have to get rid of the ρ- and p derivatives and try to express them in derivatives of W .
From the definition of H (7.6) and s there follows:
 
γ dp p
dH = − 2 dρ + qdq, (7.11)
γ−1 ρ ρ

ds dp dρ
= −γ (7.12)
cv p ρ
From equations (7.11) and (7.12) dρ and dp can be solved in terms of dH and dq:

dρ −qdq ds̄ dH
= 2
− + 2 (7.13)
ρ a γ−1 a

dp p
= −qdq − ds̄ + dH (7.14)
ρ ρ(γ − 1)
where ds̄ = ds
cv .
Observe that equation (7.14) implies the well known result that along a streamline dp+ρqdq =
7.1. CHARACTERISTIC DIRECTIONS 143

0. Let the pressure- and density derivatives in (7.10) be expressed in terms of velocity-,
entropy- and total enthalpy derivatives then the following system is obtained:

AWx + BWy = 0, (7.15)

with
 2 
1 − ua2 − uv
a2
u
− γ−1 u
a2
a2
0 −v − γ(γ−1) 1 
 
A= , (7.16)

 0 0 u 0 
0 0 0 u
2
− uv 1 − av 2 v v
 
a2
− γ−1 a2
 v 0 0 0 
B=
 0
. (7.17)
0 v 0 
0 0 0 v
and W is the state vector W = (u, v, ds̄, H)T . Let us now diagonalize (7.15). Assume A is
not-singular, e.g. A−1 exists then

Wx + A−1 BWy = 0. (7.18)

This equation looks very similar to the unsteady 1-D equations, with x playing the role of
time and y playing the role of the space coordinate.
If we want to take advantage of this form we must be sure that x is "time like" and y is "space
like", ("space like" and "time like" will be explained in paragraph 7.2).
For the moment we assume that x and y play their intended roles so that the well-known
unsteady 1-D analysis is applicable. The objective is to find a transformation of variables
dV = LdW such that equation (7.18) is transformed in the form

Vx + ΛVy = 0. (7.19)

matrix. Get the eigenvalues of A−1 B from det A−1 B − λI =

where Λ is the diagonal
det A−1 (B − λA) = 0. Since A−1 is not-singular this reduces to

det (B − λA) (7.20)

or,
a2
 
−uv − λ a2 − v 2 a2 − v 2 + λuv


γ−1 (−v + λu) v − λu
λ a2
−γ
 
det  v λv γ γ−1  = 0,
 
 0 0 v − λu 0 
0 0 0 v − λu
or,
n  u  v o
(v − λu)2 −λv λ + 2 (v − λu) − v 2 (λu − v) + 1 =0 (7.21)
a a
The eigenvalues of A−1 B are:
√
uv ∓ a2 M 2 − 1
λ1,4 = (7.22)
u2 − a2
144 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

v
λ2,3 = = tan ϕ (7.23)
u
√
2 2
where M is the Mach number: M = u a+v .
All the eigenvalues are dimensionless, they represent certain directions in the x,y plane. The
eigenvalues λ2,3 indicate the local flow direction, the eigenvalues λ1,4 indicate two directions
of steady acoustic waves.

λ1,4 = tan (ϕ ∓ µ) (7.24)

where µ is the Mach angle:

1 1
sin µ = , tan µ = √
M M2 − 1
The angle µ represents the directions of the Mach lines, e.g. the wave fronts created by an
infinitely small pointed obstacle in supersonic flow or those created by a ramp having an
infinitely small ramp angle, see figure 7.1.
The angle µ is real only for M ≥ 1. For M > 1, λ1 and λ4 represent two different directions;

Figure 7.1: Mach lines

for M = 1, λ1 = λ4 and the directions are parallel. The directions λ1 = tan (ϕ − µ), λ4 =
tan (ϕ + µ) are characteristic directions ; λ1 representing the direction of the Γ− characteristics
and λ4 representing the direction of the Γ+ characteristics. In a supersonic point P (x, y) in
the flow one can always draw the fundamental picture shown in figure 7.2.
The flow direction bisects the angle between the two characteristics Γ− and Γ+ .

7.2 "Time-like" and "space-like"

For supersonic flow the characteristic directions are real. Unfortunately this does not guarantee
that equation (7.6) is the proper form to use; if x is not "time-like" there is no advantage in
writing the equation this way. If Γ+ and/or Γ− are not pointing downstream the form of (7.6)
has no special preference.
To understand what is meant consider the domain of influence (D.O.I.) of a point P in the
x,y-plane, as shown in figure 7.3.
7.2. "TIME-LIKE" AND "SPACE-LIKE" 145

Figure 7.2: Characteristic directions in 2D supersonic flow

Figure 7.3: Domain of influence

Assume that λ1 and λ4 are such that Γ+ and Γ− point downstream. Then also the D.O.I.
of P points downstream. So we can specify initial values: W , e.g. at xp = constant and the
solution follows by marching (in space) downstream.
For example the solution at x = xp + ∆x can readily be obtained from the formulation
Wx + A−1 BWy = 0. Observe that x acts like a "time-like" parameter and y acts as the
"space-like" parameter.
However if one of the characteristic directions points upstream we have not the possibility nor
the freedom to specify initial values at xp is constant and to march downstream (in positive
x-direction) to find the solution that satisfies these initial values. The problem is that the
initial value line lies partly in the D.O.I. of point P . Figure 7.4 shows this situation in general
146 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

and illustrates it with the example of a ramp flow where the ramp angle δ and Mach angle µ2
are such that δ + µ2 > π2 .
When there are Mach lines travelling backwards, e.g. Γ+ in figure 7.4, then the x coordinate

Figure 7.4: Violation of the ’time-like’ character of the x coordinate

cannot be a "time-like" parameter because points on x = xp above P are in the domain of

influence of P . So it is not possible to specify initial conditions in P and Q independently.
For a coordinate line to be "space-like", it must run nowhere through the domain of influence
of any of its own points.
The possibility of Mach lines travelling backwards is nothing else than a projection effect.
It can be avoided by changing from the (x, y)-coordinate system to flow alligned coordinates
(s, n); s being a coordinate along the streamlines and n perpendicular to them, see figure 7.5.
From this definition s is always "time-like" and n is always "space-like".

Figure 7.5: Flow aligned coordinates

7.3. EQUATIONS IN FLOW ALIGNED COORDINATES 147

7.3 Equations in flow aligned coordinates

Take the coordinates (s, n) from:
s = x cos ϕ + y sin ϕ, (7.25)
n = −x sin ϕ + y cos ϕ, (7.26)
where ϕ is the local flow direction, see figure 7.6. The transformation from (x, y) to (s, n) is
a rotation with angle ϕ.
∂ ∂
The partial derivatives ∂s and ∂n follow from (7.25) as

Figure 7.6: Local rotation from (x, y) to (s, n)

∂ ∂ ∂
= cos ϕ + sin ϕ , (7.27)
∂s ∂x ∂y
∂ ∂ ∂
= − sin ϕ + cos ϕ . (7.28)
∂n ∂x ∂y
From (7.27) we can also obtain
∂ ∂ ∂
= cos ϕ − sin ϕ , (7.29)
∂x ∂s ∂n
∂ ∂ ∂
= sin ϕ + cos ϕ . (7.30)
∂y ∂s ∂n
Let us use the above expressions to transform the flow equations (7.15) to the (s, n) reference
frame; this yields

AWx + BWy = (A cos ϕ + B sin ϕ) Ws + (−A sin ϕ + B cos ϕ) Ws = 0,

or,

A∗ Ws + B∗ Wn = 0. (7.31)

p system still uses u −1

This and
v as the velocity variables but it is more natural to use q =
v
2 2
(u + v ) and ϕ = tan u .
To get rid of (u, v) in favour of (q, ϕ) we introduce the transformation

dW = QdZ, (7.32)
148 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

with

W = (u, v, s̄, H)T and Z = (q, ϕ, s̄, H)T . (7.33)

Z is the new state vector containing (q, ϕ) as the primary velocity variables. Using (7.32) into
(7.31) the state vector Z satisfies:

A∗ QZs + B∗ QZs = 0. (7.34)

The transformation matrix Q has the form:

 
cos ϕ −q sin ϕ 0 0
 sin ϕ q cos ϕ 0 0 
Q=
 0
. (7.35)
0 1 0 
0 0 0 1

e = A∗ Q and B
If we define A e = B∗ Q equation (7.34) becomes

AZ
e s + BZ
e n=0 (7.36)

where
−q q
1 − M2
 
0 γ−1 a2
2 cos ϕ
0 −q 2 sin ϕ − aγ(γ−1) cos ϕ 
 
A
e =
  (7.37)
 0 0 q 0 
0 0 0 q

and
 
0 q 0 0
a2 sin ϕ
q sin ϕ 0 − sin ϕ
 
B
e =
 γ(γ−1) .

(7.38)
 0 0 0 0 
0 0 0 0

Equation (7.36) is the desired result. It represents the conservation laws in very compact form
expressed in terms of flow alligned coordinates and using primary variables Z = (q, ϕ, s̄, H)T .

7.4 Characteristic equations, compatibility relations

Consider (7.36) and observe that the third and fourth row of matrix B e contain only zeroes.
The third and fourth equation of system (7.36) are therefore reduced to

q s¯s = 0, qHs = 0,

respectively. These equations tell us that entropy and total enthalpy are constant in s-direction
(streamline). Using this result system (7.36) can be simplified into:

AZ
e s + BZ
e n = 0, (7.39)
7.4. CHARACTERISTIC EQUATIONS, COMPATIBILITY RELATIONS 149

with
 
1 − M2 0 q 0 0
 
0 0 0
a2
0 −q 2 0 0  e  q 0 −1
   
A
e = ,B =  γ(γ−1) .

 0 0 1 0   0 0 0 0 
0 0 0 1 0 0 0 0

Let us now apply the left-eigenvector methos to diagonalize (7.39) in order to find the char-
acteristic equations. First we have to get the eigenvalues from:
 
det B e − λAe = 0,

or

λ4 M 2 − 1 − λ2 = 0,


(7.40)

yielding
 
dn 1
λ1,4 = = ∓√ = ∓ tan µ (7.41)
ds Γ∓ M2 − 1
 
dn
λ2,3 = =0 (7.42)
ds Γ0
The eigenvalues λ1 and λ4 represent the directions of the Mach lines in the (s, n)-plane. These
directions will be referred to as the characteristic Γ− : dn
ds
= − tan µ and Γ+ : dnds
= + tan µ.
The eigenvalues λ2 and λ3 are both zero, they represent the direction of the flow.

e −1 B
Let us now get the left (row) eigenvectors Li of A e by solving

e −1 B
Li A e = λ i Li (7.43)
e −1 B
Matrix A e has the form:
q
0 1−M 0 0
 
2
−1 −1 1
e −1 B

e = q 0 γ(γ−1)M 2 q2

A  0
. (7.44)
0 0 0 
0 0 0 0
The left eigenvectors satisfying (7.43) are,
√ √ √ 
q M 2 −1 − M 2 −1
λ1 = − tan µ ⇒ L1 = M 2 − 1, q, γ(γ−1)M 2 , q ,
λ2 = 0 ⇒ L2 = (0, 0, 1, 0) ,
λ3 = 0 ⇒ L3 = (0,
 0, 0, 1) , √
√ M 2 −1
√ 
λ4 = + tan µ ⇒ L4 = − M 2 − 1, q, −q
γ(γ−1)M 2
, M 2 −1
q .

Use these row vectors to form the matrix L

 √ √ √
q M 2 −1 − M 2 −1

M 2 − 1 q γ(γ−1)M 2 q
0 0 1 0
 
L= . (7.45)
 
0 0 √
0 1
√ √
 
2 −q M 2 −1 M 2 −1
− M − 1 q γ(γ−1)M 2 q
150 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

The new state quantities now follow from

dV = LdZ (7.46)

resulting into
√ q
√
M 2 −1
√
M 2 −1
dV1 = M 2 − 1dq + qdϕ + γ(γ−1) M 2 ds̄ − q dH,
dV2 = ds̄,
(7.47)
dV3 = dH,
√ q
√
M 2 −1
√
M 2 −1
dV4 = − M 2 − 1dq + qdϕ − γ(γ−1) M 2 ds̄ + q dH.

The corresponding characteristic equations are

∂Vi ∂Vi
+ λi = 0. (7.48)
∂s ∂n
So we may conclude from (7.47) and (7.48) and using more convenient notations: V − instead
of V1 and V + instead of V4 :

Along Γ∓ with dn ∓ = 0,


ds = ∓ tan µ : dV

dn
Along Γ0 with


ds =0 : dV2 = V3 = 0.

Equation (7.48) gives the compatibility relations along characteristics for the general case of
a non-isentropic flow. Although the flow is assumed to be adiabatic, enthalpy gradients can
appear, e.g. due to the fact that the enthalpy specified on the boundary is not uniform. In
this general case dV − and dV + cannot be brought in integral form, a direct consequence of
the presence of entropy- and/or enthalpy variations.

Let us consider the variables dV − and dV + in terms of the familiar variables q, ϕ, ρ and
p. So we substitute from (7.11) and (7.12):

a2
   
γ dp p dp dρ
dH = − 2 dρ + qdq = − + qdq, (7.49)
γ−1 ρ ρ γ−1 p ρ

dp dρ
ds̄ = −γ . (7.50)
p ρ
Then we find:
√ !
− M 2 − 1 dp
dV = q dϕ − , (7.51)
γM 2 p
√ !
M 2 − 1 dp
dV + = q dϕ + . (7.52)
γM 2 p
Compatibility relations in the above form are still valid fot the general case of adiabatic non-
isentropic (ds̄ 6= 0) and non-isenthalpic (dH 6= 0).

Special case: Homentropic flow (ds̄ = 0, dH = 0).

7.4. CHARACTERISTIC EQUATIONS, COMPATIBILITY RELATIONS 151

In homentropic flow, entropy and total enthalpy are constant throughout the whole flow do-
main. In the special flowfields the compatibility relations: dV − on Γ− and dV + on Γ+ can
be simplified considerably. If ds̄ = 0, the expression for dH (7.49) yields dH = dp
ρ + qdq; thus
in homentropic flow where we have in any direction dH = 0 it implies

dp = −ρqdq (7.53)

Therefore, after a suitable renormalisation of eigenvectors:

√
∓ M2 − 1
dV = dq ± dϕ. (7.54)
q
The first term of the right hand side may be recognized as the differential of the Prandtl-Meyer
function: ν(M )
√
M2 − 1
dν = dq, (7.55)
q
ν is the Prandtl-Meyer angle defined as:
Z M√ 2 r r
M −1 γ+1 −1 γ+1 p
ν(M ) = dq = tan (M 2 − 1) − tan−1 M 2 − 1. (7.56)
1 q γ−1 γ−1

So the compatibility equations for homentropic flow can now be written as

dV ∓ = dν ± dϕ, (7.57)

which can be integrated very easily resulting into

dy
V − = ν + ϕ is constant along Γ− with slope dx = tan(ϕ − µ)
dy
V + = ν − ϕ is constant along Γ+ with slope dx = tan(ϕ + µ)

with V − and V + being the invariants in homentropic flow.

In figure 7.2 we have presented a fundamental picture of the flow situation in and near an
arbitrary supersonic point P .
Since we now have the conditions along characteristics Γ∓ , this picture can be completed for
homentropic flows as depicted in figure 7.7.

Example 1: Centered Prandtl-Meyer expansion A uniform supersonic flow having a

Mach number M1 > 1 is deflected abruptly over a turning angle δ such that the gas expands.
At the turning point a Prandtl-Meyer expansion fan is centered and emanates from this point
into the flow field. Upstream of the fan the flow is uniform with conditions M = M1 and
ϕ1 = 0. Downstream of the fan the flow becomes uniform again; the conditions are M = M2
and ϕ = −δ see figure 7.8.
The expansion fan is build up by Γ+ characteristics; the first characteristic has an angle µ1
with the upstream flow direction. The last characteristic has an angle µ2 with respect to the
downstream flow direction.
We aim to determine the flow conditions in domain Áand use characteristics theory. Consider
152 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

Figure 7.7: Flow properties in 2D supersonic flow; homentropic case

Figure 7.8: Prandtl-Meyer expansion flow

an arbitrary Γ− -characteristic running from domain Àthrough the expansion fan into domain
Á. Observe that on this characteristic we have V − = ν1 + ϕ1 = ν2 + ϕ2 = constant. Since
ϕ1 = 0 and ϕ2 = −δ there results:

ν2 = ν1 + δ (7.58)

The Prandtl-Meyer angle in domain Áis now found and the remaining flow conditions in this
domain e.g. Mach number, temperature, pressure, density can be determined using isentropic
relations.
7.4. CHARACTERISTIC EQUATIONS, COMPATIBILITY RELATIONS 153

The well-known relation,

ν2 = ν1 + δ

finds his origin in the theory of characteristics!

Since all Γ− -characteristics issue from the uniform domain Àit implies that the invariant
ν + ϕ is a constant in the whole flow field. Thus the expansion wave is a simple wave and the
Γ+ -characteristics are straight lines. The proof is found in the next example.

Example 2: Simple waves Simple waves were already encountered in one-dimensional

unsteady flows. They have the property that one of the Riemann invariants J + or J − is
uniform. In 2D steady flows also simple waves appear, if one of the invariants V − or V + is
uniform.
Let us discuss the simple wave flow that appears if a uniform flow is expanded along a partly
curved wall as sketched in figure 7.9. Assume a uniform supersonic flow with Mach number
M1 which flows along the plane part (upstream of s).
Downstream of point S the wall bends away and the gas expands. Since all Γ− -characteristics

Figure 7.9: Simple wave flow

originate from the same uniform domain À, V − is constant throughout the flowfield. This
implies

νA + ϕA = νB + ϕB ,

for two arbitrary points A and B in the expansion region. If A and B are also taken on the
same Γ+ characteristic then the additional relation

νA − ϕA = νB − ϕB ,

holds. Both relations are satisfield if

νA = νB and ϕA = ϕB .

Thus along a Γ+ -characteristic all flow variables are constant. This implies that also ϕ + µ
is constant so that Γ+ is a straight line. However Γ− characteristics are curved but each Γ−
154 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

characteristic intersects a particular Γ+ characteristic at a constant angle. Of course this angle

differs from Γ+ characteristic to Γ+ characteristic.
Non-simple regions occur if both invariants V − and V + vary in the flow region. Analogous
to the 1D unsteady case: uniform regions are bordered by simple waves and simple waves are
bordered by either non-simple regions or uniform regions.

7.5 Method of Characteristics; M.O.C.

In 2D steady supersonic flow the characteristics are real curves in the flow domain. This
enables the set up of a characteristic method. Consider a flow situation as sketched in figure
7.10 where the flow variables are given on the initial value line.
In each point, for example A and B, we know the slopes of the characteristics Γ+ and Γ− ,

Figure 7.10: Constructing solution in P

the Mach number and the flow direction. Consider the Γ+ - characteristic emanating from
A and the Γ− -characteristic emanating from B; they intersect at point P . The solution in
P is found by applying the compatibility relations : V − = ν + ϕ = constant along Γ−B and
+ +
V = ν − ϕ = constant along ΓA . This yields two equations:

νA − ϕA = νP − ϕP ,
νB + ϕB = νP + ϕP ,
for the unknowns νP and ϕP . The solution is
νP = 21 (νB + νA ) + 12 (ϕB − ϕA )

(7.59)
ϕP = 12 (ϕB + ϕA ) + 12 (νB − νA )
The solution in P is exactly known and, without knowing the position of P in advance! The
position of P can be determined most accurately afterwards.
7.5. METHOD OF CHARACTERISTICS; M.O.C. 155

Point a b c d e f g i j l
ν 10 10 10 10 12 12 12 14 14 16
ϕ 0 4 8 12 2 6 10 4 8 6

Table 7.1: ν and ϕ in 2D diffusor

To find P draw straight lines at average angles 12 (ϕA + µA + ϕP + µP )

and 12 (ϕB − µB + ϕB − µB ) through A and B respectively. Intersecting these lines gives P ’s
location approximately. The larger the distance between A and B the less accurate the posi-
tion of the point P .
Using this method a set of points can be constructed on which the solution in known. This
set of points forms a new initial value line from which the solution process can proceed.
The M.O.C. is used in particular for analyzing supersonic windtunnel nozzles or to analyse
supersonic jets for example produced by rocket engines.

Example 1: Radial flow in a 2D diffusor Consider a supersonic flow entering a two-

dimensional conical diffusor having a certain diverging angle, say ϕ = 12 ◦ .
The flow enters radially with a constant speed q0 that corresponds to a Prandtl-Meyer angle
ν0 = 10. The flow expands in the diffusor to higher Mach numbers.
The method of characteristics is appropriate to analyse the expanding flow regime. Assume
that the initial value line is part of a circular arc on which ν = ν0 = 10 and on which the flow
angle ϕ ranges as 0◦ ≤ ϕ ≤ 12◦ , see figure 7.11.
Assume points a, b, c and d on the initial value line with ϕa = 0, ϕb = 4, ϕc = 8 and ϕd = 12◦ .

Figure 7.11: Radial flow in diffusor

Using M.O.C. the solution is constructed on all grid points in the characteristic triangle ”adl”.
The results are assembled in table 7.5. But, how to proceed the solution beyond the chracter-
istic triangle ”adl”?
To find the solution in the points ”h” or ”k” we cannot use M.O.C. directly because of the
value of one invariant is not known. However in ”h” and ”k” and also on all points lying on
the diffusor walls the flow angle in known. In ”h”, ϕh = 0◦ and in ”k”, ϕk = 12◦ . This enables
us to find the solution in ”h” to be νh = 14, ϕh = 0◦ and in ”k” to be νk = 14, ϕh = 12◦ .
Since ”h” and ”k” are now also known we have constructed a new initial value curve going
156 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

through the points ”h”, ”i”, ”j” and ”k”, on this line the Prandtl-Meyer angle is constant:
ν = 14 and the flow angle varies from 0◦ to 12◦ .
The solution procedure can now be repeated to find the solution in the characteristic triangle
”hkq”. In this way the whole flow domain downstream of the diffusor can be determined.
Remember: although the solution values are exactly known, the positions of the grid points
can only be calculated approximately.

Example 2: Under-expanded jet An under-expanded jet exits from a rocket nozzle. At

the exit the flow is parallel (ϕ = 0) and has a constant Mach number Mjet . The jet exits into
quiescent air having an ambient pressure pa .
The pressure in the nozzle exit is assumed to be pe > pa causing a further expansion of the
jet when flowing downstream.
A sketch of this flow situation is depicted in figure 7.12. The method of characteristics may
now be used to find the development of the jet outside the nozzle.
At the outer lip of the nozzle exit (point A) a centered expansion appears which reflects on

Figure 7.12: Under-expanded jet, pexit > pambient

the opposite jet boundary as a compression wave. In the expansion regime characteristics di-
verge, in the compression part thay converge to each other. Where converging characteristics
start to intersect a shock wave will appear and the M.O.C. method breaks down. Consider
those parts of the jet where no shocks have been formed yet. The flow is symmetric with
respect to the centerline and therefore the discussion can be restricted to the upper part only.
We observe several domains having different conditions. Domain ABC is a simple wave with
straight Γ− -characteristics. In domain BCE two simple waves interact and a non-simple re-
gion is present here. Domain ACD is uniform with a pressure that is equal to pa . Domain
CDEF is a simple wave with straight Γ+ -characteristics.
Domain DF G is non-simple because both invariants V + and V − vary. Domain EF H is uni-
form with a pressure lower than pa . Domain F GHI is a simple wave with V + is uniform;
the Γ− -characteristics are straight. Domain GIJ is uniform with a pressure equal to pa . In
domain HIK two simple waves interact and a non-simple region is formed. Table 7.5 sum-
marizes the typical conditions of each domain.
The jet boundary of the exhausting jet is a line of constant pressure: pa . There appears a
repeated pattern of divergence and convergence of the total jet area.

Well-posed problems (WPP’s) When working with the method of characteristics one
frequently encounters several so called well-posed problems (WPP’s). A WPP is a problem
7.5. METHOD OF CHARACTERISTICS; M.O.C. 157

Domain V+ V− comment
ABC u v simple wave
BCE v v non-simple
ACD u u uniform
CDEF v u simple wave
DFG v v non-simple
EFH u u uniform
FGHI u v simple wave
GIJ u u uniform
HIK v v non-simple

Table 7.2: Characteristics of exhausting jet, key: u: uniform, v: varies

regarding the solution of a differential equation that fulfils the requirements:

1. uniqueness of the solution,

2. existence of the solution and

3. solution depends continuously on the initial conditions.

The following three WPP’s are very common.

WPP 1: Cauchy initial value problem Initial data are given on a non-characteristic
curve AB.
The solution can be determined inside the characteristic quadrangle formed by the charac-

Figure 7.13: Cauchy problem

teristics: Γ+ and Γ− going through A and B.

WPP 2: Goursat problem Initial data are given on both characteristics: Γ+ and Γ−
going through a single point A.
158 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

Figure 7.14: Goursat problem

On Γ+ the data are specified from A to B. On Γ− the data are specified from A to C.
Now the solution can be determined in the characteristic quadrangle ABCD, CD is the Γ− -
characteristic emanating from C and BD is the Γ+ -characteristic that issues from point B.
Both characteristics intersect in D.

WPP 3: Combined initial value/boundary value problem In this problem initial

data are specified along a characteristic curve AB: ν and ϕ are known on curve AB, see figure
7.15.
In addition boundary conditions e.g. ν or ϕ are specified on a non-characteristic curve AC.
The solution can be determined in the triangle ABC formed by the characteristic Γ+ through
A and B, the characteristic Γ− going through BC and the boundary AC. Take a point 1 on
the Γ+ -characteristic through AB close to A. In 1 the variables ν and ϕ are assumed to be
known and consequently also the invariant V1− is known. Draw the Γ− -characteristic through
1 which intersects the boundary curve in 2. Now we have V1− = V2− = ν2 + ϕ2 . Since ν2 or ϕ2
is known because of 2 lies on the boundary curve all conditions in 2 can now be determined. So
if 2 is known then also V2+ and the characteristic direction ϕ2 +µ2 is known. This enables us to
calculate the Γ+ -characteristic going through 2 which then acts as a new initial value line like
AB. Now the whole calculation process can be repeated until the domain ABC is determined.

Observe that the determination of the radial flow in the diffusor (example 1 fig 7.11) is a
combination of a Cauchy problem (domain adl), a combined initial value/boundary
value problem (domain alo and alp) and a Goursat problem (domain olp).
7.6. METHOD OF WAVES; M.O.W. 159

Figure 7.15: Combined IV/BV problem

7.6 Method of Waves; M.O.W.

7.6.1 Description of the method, simple waves.
In contrast to the M.O.C. which provides exact data of the flow variables e.g. ν and ϕ on grid
points, the method of waves is essentially an approximate method that subdivides the flow
domain in a great number of small regions having constant properties.
Where the M.O.C. is very suitable for analysis of properties of supersonic flowfields, the
M.O.W. however is more favourite for design studies.
The key idea in the M.O.W. is the discretization of a continuous varying flow domain into
domains having constant properties. To elucidate the method consider a simple wave that
develops along a continuous curved contour.
In figure 7.16 a non-centered expanding simple wave flow is shown in subfigure a the wall
is continuously curved, the Γ+ characteristics are straight lines representing that the flow
expands in down-stream direction. The flow properties are exactly described by the simple
wave condition ν + ϕ = constant in the whole domain implying that on a particular Γ+
characteristic ν is constant and ϕ is constant. Thus the Γ+ characteristics issuing from the
curved wall form a bundle of diverging straight lines representing an expanding flow.
Hence the flow properties in an arbitrary point follow by considering the Γ+ characteristic
going through this point. This Γ+ characteristic intersects the boundary at P where the wall
has a known inclination ϕp . The flow properties on that particular Γ+ are then

ϕ Γ+ ν Γ+



P = ϕP , P = const − ϕP . (7.60)

Equation (7.60) is the exact solution of the expanding flow as depicted in figure 7.16 part a.
In the M.O.W. the continuous curved wall is replaced by the geometry shown in figure 7.16
160 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

Figure 7.16: Non-centered expansion

part b. The wall is approximated by successive straight segments 1, 2, 3, . . . having constant

direction angles ϕ1 , ϕ2 , ϕ3 , . . . respectively. Here ϕ1 > ϕ2 > ϕ3 > . . . . The difference in
direction angle between two consecutive wall segments ϕi+1 −ϕi = ∆ϕi may differ for different
pairs of wall segments. However in many practical applications ∆ϕi is taken constant ∆ϕ < 0
having an absolute value |∆ϕ|. Now consider the ’exact’ flow along the approximate wall
contour as depicted in figure 7.16 part b.
This flow is characterised by a collection of uniform domains separated by centered Prandtl-
Meyer expansion fans. The flow conditions (speed, direction, pressure, etc.) in these uniform
regions follow from Prandtl-Meyer theory. If ∆ϕ (∆ϕ < 0) is taken constant and the flow in
region Àis assumed to be known then the solution along the segmented contour becomes

ϕ2 = ϕ1 − |∆ϕ|, ϕ3 = ϕ1 − 2|∆ϕ|
(7.61)
ν2 = ν1 + |∆ϕ|, ν3 = ν1 + 2|∆ϕ|

If ∆ϕ is taken small (e.g. 0.01 rad.) then the Prandtl-Meyer fans are also small meaning that
the angle between the first and the last characteristic of the fan is also small (O(0.01 rad.)).
Therefore in the M.O.W. Prandtl-Meyer fans will be replaced by single straight lines being
called a ’wave’.
The slope of a wave is determined as the average of the slopes of the first and the last
characteristic.
For example the wave that separates the domains Àand Áhas a slope tan α 12 , with
1
α12 = (ϕ1 + µ1 + ϕ2 + µ2 ) (7.62)
2
Notice that the waves are not characteristics; they just represent a (weak) Prandtl-Meyer fan.
The difference between a charateristic and a wave is apparent in the case of uniform parallel
flow with Mach number M∞ > 1. This flow is supersonic so characteristics exist having a
slope ± tan µ with respect to the flow direction.
However there are no waves here because the flow is uniform. So waves and characteristics
are not identical but they are related. Therefore waves related to Γ+ characteristics will be
calles C + waves and those related to Γ− characteristics will be called C − waves.
We conclude that in the method of waves the continuous expansion is discretized into a number
of uniform regions bounded by waves across which the flow properties jump; see figure 7.17.
In order to develop the M.O.W. further, consider an isentropic compressing flow as this is
7.6. METHOD OF WAVES; M.O.W. 161

Figure 7.17: M.O.W. applied to a continuous expanding flow

generated by the curved boundary shown in figure 7.18. It presents the exact flow together
with the representation in the M.O.W.
In the M.O.W. the treatment of the isentropic compression and the continuous expansion is

Figure 7.18: M.O.W. applied to a compressing flow

very similar: the curved boundary is approximated by an appropriate number of successive

straight segments 1, 2, 3, . . . having constant slopes ϕ1 , ϕ2 , ϕ3 , . . . respectively. Again take the
difference in slope of the consecutive wall segments ϕi+1 − ϕi = ∆ϕi > 0 constant. Since ϕ is
counted positive counter clockwise ϕ1 < ϕ2 < ϕ3 and ∆ϕ > 0. Using Prandtl-Meyer theory
the flow conditions in the various uniform domains are:
ϕ2 = ϕ1 + |∆ϕ|, ν2 = ν1 − |∆ϕ|
(7.63)
ϕ3 = ϕ1 + 2|∆ϕ|, ν3 = ν1 − 2|∆ϕ|

Across the C + wave the flow properties show a jump; |∆ϕ| is taken as a measure of the jump
and it is called the strength of the wave.
The slope of the C + wave follows as the average slope of the Γ+ characteristics in the domains
that have C + as the common boundary. For example the C + wave between domains Áand
Âhas the slope tan α2,3 with
1
α2,3 = (µ2 + ϕ2 + µ3 + ϕ3 ) . (7.64)
2
Again observe the difference between waves and characteristics.
162 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

C + waves C − waves

expansions

compressions

Table 7.3: The effect of C + and C − waves on flow properties.

When overlooking the results for C + waves having a strength |∆ϕ| as given in equations
(7.61) and (7.62) we may conclude:
Expanding C + waves (|∆ϕ| < 0) do increase the Prandtl-Meyer angle and do decrease the flow
direction by the same amount |∆ϕ|, so ∆ν = |∆ϕ|, ∆ϕ = −|∆ϕ|.
Compressing C + waves (|∆ϕ| > 0) do decrease the Prandtl-Meyer angle and do increase the
flow direction both by the amount |∆ϕ|, so ∆ν = −|∆ϕ|, ∆ϕ = |∆ϕ|.
In case of C − waves we expect to find almost similar results, however there are some minor
differences regarding slope and strength. A C − wave may be related to a Γ− characteristic
and therefore the slope of a C − wave is found as the average slope of two appropriate Γ−
characteristics. For example a C − wave lying between domain Àand Áhas a slope tan β1,2
determined by
1
β1,2 = (ϕ1 − µ1 + ϕ2 − µ2 ) (7.65)
2
The difference with equation (7.64) is obvious. Concerning the strength observe that an
expanding C − wave gives an increase in flow direction, thus ∆ϕ > 0 and a compressing C −
wave yields a decreasing flow direction , thus then ∆ϕ < 0.
Very similar to C + waves we may now conclude for C − waves having a strength |∆ϕ|.
Expanding C − waves (|∆ϕ| > 0) do increase the Prandtl-Meyer angle and do increase the flow
direction by the same amount |∆ϕ|, so ∆ν = |∆ϕ|, ∆ϕ = |∆ϕ|.
Compressing C − waves (|∆ϕ| < 0) do decrease the Prandtl-Meyer angle and do decrease the
flow direction both by the amount |∆ϕ|, so ∆ν = −|∆ϕ|, ∆ϕ = −|∆ϕ|.
The results for C + waves and C − waves are summarized in table 7.6.1.

7.6.2 Interaction of waves

Consider a two-dimensional supersonic flow in a diverging channel having the special geometry
shown in figure 7.19. Two centered expansions are generated one at the kink B in the top wall
and one at the kink A in the bottom wall. Upstream of the kinks the top and bottom wall
are parallel and the flow is uniform with conditions ν1 and ϕ1 = 0. Figure 7.19 part a shows
7.6. METHOD OF WAVES; M.O.W. 163

Figure 7.19: Interacting expansion waves

the interacting wave pattern displaying the uniform domains À, Á, Âand Ã, the simple wave
regions S, T , U and V , and the non-simple wave domain W . Assume that the flow deflection
at kink B: ϕ1 − ϕ2 > 0 is known. Let us now apply the method of waves to this flow problem.
In the M.O.W. approximation the expansion fan from B is replaced by a C − wave, see figure
7.19 part b.
Observe that the non-simple domain W is reduced to the intersection W ∗ of the C + wave
and the C − wave. If the deflection angles ϕ2 and ϕ3 are different (irrespective their signs)
the C + and the C − waves will have different strength. The strength of the C + wave from A
to W ∗ is |∆ϕ|A = |∆ϕ3 |, (take ϕ1 = 0) and the strength of the C − wave from B to W ∗ is
|∆ϕ|B = |∆ϕ2 |. Taking into account that ϕ2 > 0 and ϕ3 < 0 and recalling table 7.6.1 both
the C + and the C − waves have to be expansion waves and therefore one may conclude:

ν2 = ν1 + |∆ϕ|B = ν1 + ϕ2
(7.66)
ν3 = ν1 + |∆ϕ|A = ν1 − ϕ3

This result from M.O.W. is still exact for those points in Áand Âwhich ly outside the expansion
fans. For points lying inside the fans (7.66) gives an approximation for the flow variables. The
accuracy of the solution depends on the wave strength; weaker waves will increase the accuracy.
Let us now try to determine the flow variables in domain Ã. Domain Ãis uniform and appears
when both waves (C + and C − ) have interacted. It is obvious that domain Ãcan be determined
if the strength of the C + wave between Áand Ãis known or if the strength of the C − wave
between Âand Ãis known. However these wave strengths are yet unknown since C + and C −
have been interacted at W ∗ . So the question remains what happens with the wave strength if
iteraction occurs?
The wave strength of the C − wave before interaction is |∆ϕ|pre = |ϕ2 | = ϕ2 , the wave strength
after interaction is |∆ϕ|post = |ϕ4 − ϕ3 |. How are |∆ϕ|post and |∆ϕ|pre related to each other?
Theorem: The strength of a wave (in terms of |∆ϕ|) is conserved during interaction with
waves of the opposite family.
Proof: Consider the ’exact’ flow pattern as sketched in figure 7.19 part a and use M.O.C. to
find the flow properties in domain Ã

alongΓ+
1→2 : ν1 − ϕ1 = ν2 − ϕ2 , (7.67)

alongΓ−
2→4 : ν4 + ϕ4 = ν2 + ϕ2 , (7.68)
alongΓ−
1→3 : ν1 + ϕ1 = ν3 + ϕ3 , (7.69)
164 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

alongΓ+
3→4 : ν4 − ϕ4 = ν3 − ϕ3 . (7.70)
Substracting (7.67) and (7.68) yields:

ν4 − ν1 + ϕ4 + ϕ1 = 2ϕ2 . (7.71)

Substracting (7.69) and (7.70) yields:

ν4 − ν1 − (ϕ4 + ϕ1 ) = −2ϕ3 . (7.72)

Substracting (7.71) and (7.72) yields:

2(ϕ4 + ϕ1 ) = 2ϕ2 + 2ϕ3 ,

or

ϕ4 − ϕ2 = ϕ3 − ϕ1 (7.73)

or

ϕ4 − ϕ3 = ϕ2 − ϕ1 (7.74)

Equation (7.73) tells that the strength of the C + wave is constant if a C − is crossed. Similarly,
equation (7.74) implies that the strength of the C − wave is constant if a C + wave is crossed.
Based on this result the flow properties in domain Ãmay be found as

ν4 = ν2 + |∆ϕ|A = ν1 + ϕ2 − ϕ3
(7.75)
ϕ4 = ϕ2 − |∆ϕ|A = ϕ2 + ϕ3

Comment: In the case that both waves: C + and C − have the same strength: |∆ϕ|A = |∆ϕ|B
then ϕ3 = −ϕ2 , ν4 = ν1 + 2ϕ2 and ϕ4 = 0.
Two interacting expansion waves of equal strength and opposite family (C + and C − ) will
cancel out deflections but strengthens the expansion rate.

7.6.3 M.O.W. for isentropic flows, a calculation scheme

The M.O.W. just introduced and explained for simple flow fields (e.g. a single compression or
a single expansion) can be generalized to calculated flows having a more complicated struc-
ture. Such flow may be encountered if compression waves together with expansion waves ,
irrespective of their type (C + or C − ) are present in the flow domain. Let us study some
examples.
Consider a uniform flow (conditions: ν1 > 0, ϕ1 = 0) entering a two dimensional diverging
channel, not necessarily symmetric. An example is shown in figure 7.20.
Let the walls be descretized so that all waves emanating from corner points have equal
strength: |∆ϕ|. Since the channel is diverging all waves are expansion waves. The pattern of
C + waves and C − waves is shown in figure 7.20, they intersect and subdivide the flow domain
in a number of uniform subdomains or cells. The flow properties (ν, ϕ) in a particular cell
may now be calculated simply by counting waves. Travelling from domain Àwhere we have
specified the entrance conditions to a particular cell : P one passes several C + waves and/or
C − waves. Once a wave is passed (in flow direction!) the Prandtl-Meyer angle (ν) increases
7.6. METHOD OF WAVES; M.O.W. 165

Figure 7.20: Wave pattern in a two-dimensional diverging channel

with the amount |∆ϕ|; the flow angle (ϕ) decreases if a C + wave is crossed and increases if a
C − wave is crossed both with the amount of |∆ϕ|.
Assume that for the cell P one has to cross a number of m+ + waves and m− C − waves.
p C p
The flow properties in cell P are then determined by
νp = ν1 + m− +


p + mp
|∆ϕ|,
(7.76)
ϕp = ϕ1 + m− +
p − mp |∆ϕ|.

The above result holds for expansion waves.

In case only compression waves appear a very similar expression can be derived. Assume that
all compression waves have equal strength: |∆ϕ|. For a particular cell P one has to cross a
number of n+ + − −
p C waves and a number of np C waves. Crossing a compression wave (in flow
direction) decreases the Prandtl-Meyer angle with the amount |∆ϕ|; it also decreases the flow
angle if the crossed wave is of C + type. Thus for compressing flows we find:
νp = ν1 − n− +


p + np
|∆ϕ|,
(7.77)
ϕp = ϕ1 − n− +
p − np |∆ϕ|.

The particular results for expanding regimes and compressing regimes may be combined to
obtain a general expression for the flow conditions in a particular cell P :
νp = ν1 + m− + − +


p + mp − np − np
|∆ϕ|,
(7.78)
ϕp = ϕ1 + m− + − +
p − mp − np + np |∆ϕ|.

This expression holds for isentropic supersonic flow domains, shocks may not occur.

Wave reflection on a plane wall In figure 7.20 we see C + waves running onto the top
wall and C − waves running onto the bottom wall. If these waves arrive at the opposite wall
they will be reflected into waves of the opposite type: C + waves will reflect into C − waves
and vice versa.
Let us study some details of the reflection process, for example if a C − expansion wave enters
a plane wall, see figure 7.21.
A C − expansion wave with strength |∆ϕ| is generated by deflecting the top wall at A with
an angle ∆ϕ. The C − wave runs onto the lower wall where it is reflected into a C + wave.
Question:
166 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

Figure 7.21: Reflection of an expansion wave

What are the flow conditions in domain Âand what is the character of the reflected
wave expanding or compressing?

In domain Âthe flow must be parallel to the wall so ϕ3 = ϕ1 = 0. Travelling from domain
Àto domain Âwe pass one C − wave and one C + wave, this makes m− −
3 = 1, n3 = 0 since the
character of the C + wave is as yet unknown the integers n+ +
3 and m3 cannot be determined.
Using (7.78) the flow angle ϕ3 follows as

ϕ3 = ϕ1 + 1 − m+ +


3 + n3 |∆ϕ|.

Taking into account the boundary condition ϕ3 = ϕ1 = 0 there is obtained

1 − m+ +


3 + n3 = 0.

Since m+ + + +
3 and n3 are non-negative integers there follows m3 = 1, n3 = 0 implying that the
reflected C + wave is an expansion wave with strength |∆ϕ|. The above example leads to
the more general observation that: Expansion waves will reflect on a plane wall as expansion
waves; during reflection the wave strength is unchanged.
A similar treatment for compression waves will lead to the observation that: Compression
waves will reflect on a plane wall as compression waves; during reflection the wave strength is
unchanged.

Wave reflection on a jet boundary Another interesting reflection phenomenon happens if

a C + wave (expansion or compression) reflects on a jet boundary. A jet boundary is a contact
discontinuity separating two flows having different conditions. Across the jet boundary the
pressure and the flow direction is continuous but flow velocity (Mach number), density and
entropy are discontinuous.
Since the flow direction is continuous the jet boundary is also a streamline. Consider a C −
expansion wave that hits a jet boundary, see figure 7.22. Assume that the jet boundary
separates the supersonic flow in domain À(conditions ν1 , ϕ1 = 0, p1 ) from a gas at rest
(M = 0) in domain Ã. The C − wave is reflected and a C + wave appears emanating from
point B. Across the jet boundary the pressure is continuous thus p1 = p4 = p3 and because
the flow is isentropic ν3 = ν1 . Travelling from domain Àto domain Âone C − wave and one
7.6. METHOD OF WAVES; M.O.W. 167

Figure 7.22: Wave reflecting on a jet boundary

C + wave is passed, thus m− −

3 = 1, n3 = 0. Since the character of C
+ is yet unknown the
+ +
integers m3 and n3 are left undetermined. From (7.78) we can determine the Prandtl-Meyer
angle in domain Âas
−
ν3 = ν1 + 1 + m+


3 − n3 .

Taking into account the condition ν3 = ν1 and that m+ +

3 and n3 have to be zero or positive
+ +
there is obtained m3 = 0, n3 = 1. The reflected wave is a compression wave and has strength
|∆ϕ|. The flow direction in region Âfollows (use again (7.78) with m− + +
3 = 1, m3 = 0, n3 = 0,
n+
3 = 1) as

ϕ3 = 2|∆ϕ|.

Due to the interaction the flow deflection doubles! Conclusion:

Expansion waves will reflect on a jet boundary with a gas at rest as compression waves and
compression waves will reflect as expansion waves.

7.6.4 Design of a supersonic windtunnel nozzle by M.O.W.

The main objective of a supersonic windtunnel is the realisation of a steady uniform flow having
specified conditions such as Mach number, pressure and temperature. From one dimensional
compressible flow theory it is well known that a nozzle throat is needed, for the transition from
subsonic to supersonic flow. In the nozzle throat sonic flow conditions appear. Downstream
of the throat in the diverging part a supersonic flow may be obtained.
The required steady uniform flow is found in the test section of the windtunnel. The area
ratio between test section and throat determines the test section Mach number. To guarantee
that the flow in the test section is exactly uniform a special contour between the throat and
the test section has to be designed. To obtain such a special contour, the method of waves is
a useful technique.
Assume a symmetric nozzle, a sketch of the upper part of the nozzle contour between throat
and test section is shown in figure 7.23. The flow enters the throat just at sonic conditions, so
the Prandtl-Meyer angle in the throat is ν = 0. To be specific we aim to design a nozzle contour
for a supersonic windtunnel with test section Mach number: MT = 1.916, the corresponding
168 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

Prandtl-Meyer angle in the test section is νT = 24. The nozzle contour is replaced by rectilinear
segments, such that at each corner point the flow direction changes by a constant amount of
2o .
Along the diverging part AB the slope angle increases stepwise untill the maximum angle of

Figure 7.23: Design of a nozzle contour for a supersonic windtunnel. Indicated values in the
flow field refer to the values of the Prandtl-Meyer angle ν, values along the contour refer to
the contour slope.

12o is reached along BC. From C to D the slope angle decreases until 0o beyond D. From
the diverging part AB, C − expansion waves are generated. Crossing these waves will increase
the Prandtl-Meyer angle and the flow direction by an amount of 2o . The C − waves reflect at
the centre line along EF into C + waves. The C + waves are also expansion waves, crossing
them will increase the value of ν but decreases the flow angle, both by the same amount of
2o . Each wave has a strength |∆ϕ| = |2o |.
In the actual design the part AB of the nozzle contour may be chosen more or less arbitrary;
for example a parabolic curve with zero slope at A. Using M.O.W. the flow conditions in each
cell in ADEF can now be determined, just by counting waves! For example the triangular
shaped domains along the centre line all have ϕ = 0 due to the fact that an equal number of
C + waves and C − waves have to be passed when travelling from the throat to a particular
domain. But since C + waves and C − waves both do increase the Prandtl-Meyer angle we see
∆ν = 2|∆ϕ| = 4o between consecutive triangles. We observe that ABEH is a simple wave and
that EHF is a non-simple region. Downstream of the C + wave F D the test section conditions
ν = 24, ϕ = 0 are reached. The slopes of the waves in ADEF can now be calculated. For
example the slope angle of AE may be determined as
1
βAE = (ϕpre − µpre + ϕpost − µpost ) .
2
Now we come to the determination of the shape of contour CD. Consider the flow near C in
more detail, see figure 7.24.
The C + wave emanating from E intersects the wall segment having ϕ = 12o at point C.
When passing this C + wave the ν increases from 12o to 14o and the flow angle decreases from
12o to 10o . At C the C + wave reflects and a C − wave runs downstream. This C − wave is an
expansion wave which increases the ν values by an amount of 2o . This process also happens on
the centre line downstream of F where the flow conditions already had met the requirements
in the test section.
Due to the C − wave emanating from C the flow conditions in the test section will be disturbed.
7.6. METHOD OF WAVES; M.O.W. 169

Figure 7.24: Wall adaption starting a C

To avoid this disturbing effect the C − wave reflecting from C has to be surpressed. This can
be done by properly adapting the nozzle contour and changing the slope of the nozzle contour
at C from 12o to 10o which gives the desired result and cancels the reflected wave. In a similar
way all the other C + waves arriving at the contour CD will be prevented from reflecting by
wall adaption. Along CD the nozzle contour shows a decreasing slope angle untill beyond D
the angle is 0o . Along part AB of the nozzle contour expansion waves are generated, along
part CD wave cancellation is accomplished by wall adaption. The arbritrariness in the choise
of the shape of nozzle part AB is corrected by a proper design of the nozzle part CD.
Comments:

1. The test conditions νT = 24, ϕT = 0 are realized by passing six C − waves and six
C + waves, each of them having a strength |∆ϕ| = 2o . The nozzle contour obtained in
figure 7.23 is not a unique contour but just one of the possible designs that realizes the
required test conditions. On the other hand it has the special property that it uses just
one reflection of C − waves on the centre line to acquire the prescribed test conditions
νT = 24, ϕT = 0.
Designs having more reflections are very well known, however ’the one reflection nozzle’
gives a short nozzle for a given throat height.

2. For a ’the one reflection nozzle’ the maximum deflection appears at the nozzle contour
where all C − waves and no C + waves are passed, thus if just the half of the total number
of waves are passed. So the maximum deflection angle ϕmax becomes ϕmax = 1/2νT , it
increases with increasing Mach number.
Due to viscous effects flow separation is likely to occur certainly if the deflection angle
becomes too large. In practise, a maximum deflection angle of ≈ 6o is acceptable. Thus
supersonic windtunnels designed for Mach numbers M > 1.5 (νT ≤ 12o ) are equipped
with nozzles showing more than one reflection.

3. The design of a ’one-reflection nozzle’ is a very good starting point to design nozzle
contours for the same Mach number but having more reflections because ’multi-reflection
nozzles’ are already incorporated in the ’one-reflection nozzle’ design. Such a ’multi-
reflection nozzle’ is obtained if one constructs in the ’one-reflection’ solution a streamline
going from throat to test section and that crosses the non-simple domain EF H (see figure
7.23). Such a streamline can be viewed as the contour for a new nozzle. Inside the new
nozzle the flow is accelerated from throat (M = 1, ν = 0) to test conditions (νT = 24).
Now we observe more reflections of C − waves along the centre line; the closer the nozzle
170 CHAPTER 7. STEADY TWO-DIMENSIONAL FLOW

contour is chosen to the centre line the more reflections appear and the smaller the
maximum deflection angle becomes.
So, once a ’one-reflection nozzle’ for a given test Mach number is obtained then a class
of ’multi-reflection nozzles’ having the same Mach number but different deflection angles
are available as part of the ’one-reflection’ solution.

4. Figure 7.23 shows the exact nozzle flow of a supersonic nozzle. The flow is definitely
two-dimensional in the domain between nozzle throat and test section. One of the
consequences of this two dimensional behaviour is that the Mach number at the centre
line already attains the test section value upstream of the test section. The test section
conditions are found not only in the test section itself but also in the diverging part of
the nozzle downstream of the wave F D. So in actual supersonic experiments one may
observe that windtunnel models are sometimes placed upstream of the test section.
Chapter 8

Two-dimensional steady potential

flows

8.1 Crocco’s Theorem, Velocity Potential, Potential Equation.

In chapter 7 we have considered the two-dimensional steady flow of an inviscid, non-heat
conducting fluid. Furthermore, we have neglected the influence of external force fields and
external heat sources.
So we have assumed adiabatic flow governed by the steady Euler equations (see (7.24)).
∂ ∂
continuity: (ρu) + (ρv) = 0 (8.1)
∂x ∂y
∂u ∂v ∂p
x-momentum: ρu + ρv + =0 (8.2)
∂x ∂x ∂x
∂v ∂v ∂p
y-momentum: ρu + ρv + =0 (8.3)
∂x ∂y ∂y
∂H ∂H
energy: u +v =0 (8.4)
∂x ∂y
The last equation expresses that the total enthalpy H is kept constant when travelling along
a streamline. A similar result for the entropy could be obtained (see equation (7.9)):
∂s ∂s
u +v = 0. (8.5)
∂x ∂y
In this chapter we will consider the special class of irrotational flow fields because they have
a wide range of applicability.
In irrotational flows the vorticity ω, defined as
ω̄ = ∇ × v̄, (8.6)
is everywhere zero in the flow field.
It is well known that vorticity appears in shear layers and/or boundary layers, thus in flows
where viscous effects are present. However, are there, apart from viscosity, other mechanisms
that can create vorticity in a flow field? To answer this question combine the momentum
equation for unsteady inviscid flow:
∂v̄ 1
+ (v̄ · ∇)v̄ + ∇p = 0, (8.7)
∂t ρ

171
172 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

and the first law in the form:

1 1
T ∇s = ∇e + p∇ = ∇h − ∇p. (8.8)
ρ ρ
Using the vector identity
1
(v̄ · ∇)v̄ = ∇( v̄ · v̄) − v̄ × ω̄, (8.9)
2
there follows from equation (8.7):
∂v̄ 1 1
+ ∇( v̄ · v̄) − v̄ × ω̄ + ∇p = 0. (8.10)
∂t 2 ρ
Eliminating ρ1 ∇p from equation (8.8) there results

∂v̄ 1
+ ∇( v̄ · v̄) − v̄ × ω̄ + ∇h − T ∇S = 0. (8.11)
∂t 2
Using the definition of the total enthalpy H:
1
H = h + v̄ · v̄, (8.12)
2
Crocco’s well-known result is now obtained:
∂v̄
+ ∇H − T ∇s = v̄ × ω̄. (8.13)
∂t
Crocco’s equation relates vorticity (ω̄) , entropy (s), total enthalpy (H) and unsteadiness ( ∂v̄
∂t )
together. It tells that in inviscid flow vorticity is created if entropy gradients, total enthalpy
gradients or unsteady effects appear. Three well-known examples of vorticity production are
• curved shocks (∇s 6= 0),
• jet boundaries (∇H 6= 0),
∂
• oscillating airfoil ( ∂t 6= 0)
From Crocco’s Theorem it is obvious that steady flows having uniform entropy and uniform
total enthalpy can be treated as irrotational.
If the entropy is uniform then equation (8.5) is identically fulfilled and the flow is called ho-
mentropic; similarly if the total enthalpy is uniform then equation (8.4) is identically fulfilled,
the flowfield is called isenthalpic.
So, steady homentropic and isenthalpic flows can be treated as irrotational.
For this class of flows the governing equations (8.1)-(8.3) can be simplified considerably
by introducing the velocity potential Φ(x, y, z) according to

v̄ = ∇Φ(x, y, z). (8.14)

In order to derive a simple equation for the potential function Φ we try to eliminate the
pressure p and density ρ from the system (8.1)-(8.3). First write the continuity equation in
the form
 
∂u ∂v ∂ρ ∂ρ
ρ + +u +v = 0. (8.15)
∂x ∂y ∂x ∂y
8.1. CROCCO’S THEOREM, VELOCITY POTENTIAL, POTENTIAL EQUATION. 173

Density gradients can be expressed in terms of pressure gradients by differentiating the isen-
thalpic condition:

γ p u2 + v 2
H= + = constant, (8.16)
γ−1ρ 2

with respect to x and with respect to y, yielding

a2 1 ∂ρ ∂ u2 + v 2
 
γ 1 ∂p
= + , (8.17)
γ − 1 ρ ∂x γ − 1 ρ ∂x ∂x 2
a2 1 ∂ρ ∂ u2 + v 2
 
γ 1 ∂p
= + . (8.18)
γ − 1 ρ ∂y γ − 1 ρ ∂y ∂y 2

With equations (8.2) and (8.3) the pressure derivatives can be substituted, yielding:

a2 1 ∂ρ ∂ u2 + v 2
   
−γ ∂u ∂u
= u +v + , (8.19)
γ − 1 ρ ∂x γ−1 ∂x ∂u ∂x 2
a2 1 ∂ρ ∂ u2 + v 2
   
−γ ∂v ∂v
= u +v + . (8.20)
γ − 1 ρ ∂y γ−1 ∂x ∂y ∂y 2

The density derivatives according to equations (8.19) and (8.20) can now be inserted in the
continuity equation in the form of equation (8.15); after some rearranging of terms there
results
 
2 2 ∂u ∂u ∂v ∂v
(a − u ) − uv + + (a2 − v 2 ) = 0. (8.21)
∂x ∂y ∂x ∂y

The equation just obtained contains only the velocity components u and v and the sound
velocity a, the latter can also be expressed in terms of the velocity components u, v by writing
a2 = γp/ρ and using equation (8.16) leading to the relation

a2 u2 + v 2
+ = constant. (8.22)
γ−1 2

From the definition of the velocity potential, equation (8.14), the velocity components are
expressed as:

∂Φ ∂Φ
u= , v= . (8.23)
∂x ∂y

Inserting them in equation (8.21) we find the well-known potential equation:

∂2Φ ∂2Φ 2
2 ∂ Φ
(a2 − u2 ) − 2uv + (a2
− v ) = 0. (8.24)
∂x2 ∂x∂y ∂y 2

The velocity potential Φ satisfies a quasi-linear P.D.E. of second order. The equation governs
compressible flows ranging from low speeds (subsonic) to high speeds (supersonic) as long as
the fluid may be regarded as a perfect gas.
Also the transonic speed regime where the transition from subsonic so supersonic and vice
versa appears us very well modelled by equation (8.24). Supersonic flows with strong shock
174 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

waves cannot be treated since the assumption of constant entropy is violated when crossing a
shock wave.
In the exceptional case of incompressible flow: a2  (u2 + v 2 ), equation (8.24) reduces to
the Laplace equation

∂2Φ ∂2Φ
+ = 0. (8.25)
∂x2 ∂y 2

Obviously incompressible potential flow is just a special case of a compressible flow.

8.2 Linearized flow; the Prandl-Glauert equation

Consider a steady two-dimensional flow and let this flow be described as a perturbation of
the uniform flow with velocity U∞ in x-direction. The velocities of the perturbed flow then
become:

u(x, y) = U∞ + u0 (x, y), (8.26)

0
v(x, y) = v (x, y). (8.27)

The perturbation velocities u0 and v 0 are assumed to be small with respect to the velocity
U∞ .:

u0 v0
 1,  1. (8.28)
U∞ U∞

A perturbation velocity potential φ(x, y) is now defined as:

Φ(x, y) = U∞ (x + εφ(x, y)), (8.29)

yielding the velocity components:

u = Φx = U∞ + εU∞ ∂φ
∂x ,
∂φ (8.30)
v = Φy = εU∞ ∂y .

Obviously the velocity perturbations are small so:

∂φ ∂φ
ε  1, and ε  1, (8.31)
∂x ∂y

where ε is a small quantity governing the magnitude of the perturbations.

Taking the magnitude O(ε) then

∂φ ∂φ
= O(1), = O(1). (8.32)
∂x ∂y

The velocity potential equation may now be linearized by taking into account the order esti-
8.2. LINEARIZED FLOW; THE PRANDL-GLAUERT EQUATION 175

mations:
u2
2
= 1 + 2εφx + O(ε2 ),
U∞
v2
2
= O(ε2 ),
U∞
uv
2
= εφx + O(ε2 ),
U∞
a2 1
2
= 2
− (γ − 1)εφx + O(ε2 ),
U∞ M∞
∂2Φ ∂2φ
= εU∞ 2 ,
∂x2 ∂x
∂2Φ ∂2φ
= εU∞ ,
∂x∂y ∂x∂y
∂2Φ ∂2φ
= εU∞ 2 .
∂y 2 ∂y
Inserting these order estimates into the velocity potential equation: (8.24) and neglecting
terms of O(ε2 ) and higher the Prandtl-Glauert equation:

2 ∂2φ ∂2φ
(1 − M∞ ) + 2 = 0, (8.33)
∂x2 ∂y
for the perturbation potential φ(x, y) is obtained. Here M∞ = U∞ /a∞ and a∞ is the sound
speed of the unperturbed flow.
The Prandtl-Glauert equation is a linear second order PDE with constant coefficients. For
subsonic free streams (M∞ < 1) the Prandtl-Glauert equation is elliptic, for supersonic free
stream (M∞ > 1) this equation is hyperbolic.
For transonic flow (M∞ ≈ 1) equation (8.33) cannot be used because the term (1 − M∞ 2 )

can no longer be viewed as an O(1) term but has to be considered as an O(ε) term if M → 1.

8.2.1 Linearized pressure coefficient

Once a solution of the Prandtl-Glauert equation is obtained the velocity components u and v
are known and the pressure can be calculated; in terms of pressure coefficient cp it leads to:
p − p∞
cp = 1 2
= −2εφx + O(ε2 ), (8.34)
2 ρ∞ U∞
which reveals that the first order pressure perturbation only depends on the perturbation
velocity in main flow direction.
Expression (8.34) can be verified by using the fact that in isentropic flow the sound velocity
is proportional to the pressure as:
γ−1
a2 ∝ p γ . (8.35)
If the coefficient of proportionality is evaluated using the uniform flow conditions (a∞ , p∞ , ρ∞ )
then there is obtained
 γ−1
a 2
  
p γ
= , (8.36)
a∞ p∞
176 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

with the help of the order estimate of the variable (a/U∞ )2 one finds
  γ−1  
p γ
2 a 2
= M∞ = 1 − (γ − 1)M∞ εφx + O(ε2 ), (8.37)
p∞ U∞
or
p 2
= 1 − γM∞ εφx + O(ε2 ). (8.38)
p∞
Using the definition of the pressure coefficient the result given in (8.34) is obtained. Notice
also the minus sign, velocity increase causes a pressure decrease and vice versa.

8.2.2 Linearized boundary condition

Assume that a solid surface y = εh(x) is present in the flow domain. The shape of this surface
depends on the small parameter ε. For ε = 0 this surface falls along the x-axis (y = 0)
being also a streamline of the undisturbed uniform flow. For ε 6= 0 the solid surface induces
a perturbation on this uniform flow.
Along the solid surfave the boundary condition ~v · ~n = 0 has to be fulfilled;
This boundary condition can also be linearized. Applying ~v ·~n = 0 on the surface y = εh(x)
there follows
dy v(x, y)
= εh0 (x) = (8.39)
dx u(x, y)
which has to be valid at y = εh(x), yielding

εh0 (x) · u(x, εh(x)) = v(x, εh(x)), (8.40)

or in terms of the perturbation potential:

εh0 (x) {U∞ + U∞ εφx (x, εh(x))} = εU∞ φy (x, εh(x)) (8.41)

For small ε the perturbation velocities φx (x, εh(x)) and φy (εh(x)) may be expanded as:
 
∂φx
φx (x, εh(x)) = φx (x, 0) + ε + O(ε2 )
∂ε ε=0
 
∂φy
φy (x, εh(x)) = φy (x, 0) + ε + O(ε2 )
∂ε ε=0

Inserting this expansion into Eq. (8.35) and neglecting terms of O(ε2 ) the linearized boundary
condition

h0 (x) = φy (x, 0) (8.42)

results. Equation (8.42) expresses that the line y = 0 (x-axis) experiences an outflow (φy (x =
0)) such that at the line y = 0 the velocity vector is tangential to the direction of the solid
surface (h0 (x)).
A small perturbation problem for potential flow may be summarized as follows: find a
solution of the Prandtl-Glauert equation:

2 ∂2φ ∂2φ
(1 − M∞ ) + 2 =0
∂x2 ∂y
8.2. LINEARIZED FLOW; THE PRANDL-GLAUERT EQUATION 177

which satisfies the boundary conditions

∂φ dh
= at y = 0.
∂y dx
When φ(x, y) is known, the pressure coefficient follows from
∂φ
cp = −2ε .
∂x

8.2.3 Example 1: a subsonic flow along a wavy wall

Consider a subsonic flow along an infinite solid surface which is wavy; for example having a
sinusoidal shape:

y = εh(x) = ε sin kx, (8.43)

k is the so called wave number which counts the number of waves on an interval of 2π length.
Assume that far away of this wall the flow is uniform with velocity u = U∞ parallel to the
x-axis, see figure 8.1. The linearized boundary condition becomes

Figure 8.1: Flow along a wavy wall

φy (x, 0) = k cos kx (8.44)

This is a Neumann type boundary conditon because it specifies the normal derivative (e.g.
∂φ
∂y ).
For subsonic flow M∞ < 1 the Prandtl-Glauert-equation is of elliptic type, implying that
boundary conditions have to be specified all along the boundary.
In this example we require that the perturbation potential φ(x, y) vanishes at large y (say
y → ∞) and further that the solution continues harmonically for |x| → ∞.
Let us perform the method of separation of variables and try a solution of the form:

φ(x, y) = X(x) · Y (y) (8.45)

where X(x) is a function of x and Y (y) is a function of y. Both have to be determined so that
φ(x, y) satisfies the boundary conditions. When (8.45) is substituted in the Prandtl-Glauert
equation there results

2 X 00 Y 00
(1 − M∞ ) =− , (8.46)
X Y
178 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

(where a prime indicates differentiation). The left-hand side of equation (8.46) is just a
function of x and the right-hand side of it is just a function of y; consequently both must be
equal to the same constant because otherwise a relation between the independent variables x
and y appears which is not allowed if x and y have to be independent. so the only possibility
left is that both sides are equal to the same constant. Since φ(x, y) is expected to be periodic
in x we propose this constant equal to a negative number, say −m2 . Then equation (8.40)
leads to the set of second-oder ODE’s:
2 )X 00 = −m2 X
(1 − M∞
(8.47)
Y 00 = m2 Y
The solution that satisfies the boundary conditions reads
1
φ(x, y) = − cos kx exp(−kβ∞ y) (8.48)
β∞
where we have taken
p
β∞ = 1 − M∞ 2 (8.49)
The perturbation potential is periodical in x-direction and in y-direction it shows exponential
decay for y → ∞. With the help of (8.34) we can calculate the pressure coefficient at the wall
as
εk
cp = −2εφx (x, 0) = −2 sin kx. (8.50)
β∞
We observe a periodic behaviour in x-direction with a wave number equal to that of the wavy
wall. The pressure coefficient varies linearly with the perturbation parameter ε. Furthermore
it appears that the pressure coefficient is proportional with the curvature of the wavy wall.
The solution given in equation (8.50) shows exponential decay in y-direction. Let us
introduce a characteristic length L by equating
y
exp(−kβ∞ y) = exp(− )
l
1
The characteristic length L = β∞ k is characteristic for the damping of the perturbations in y-
direction. Damping increases with increasing Mach numer and increasing wave number. This
implies that perturbations induced by a wall with long waves in a low subsonic flow damp
slowly.
Equation (8.50) shows that in the limit of incompressible flow (M∞ = 0, β∞ = 1) the
pressure coefficient becomes
cp incompressible = −2ε sin kx.
So if we express the pressure coefficient for incompressible flow (M∞ = 0) in terms of the
pressure coefficient for incompressible flow, the well-known Prandtl-Glauert rule results:
1
cpM∞ 6=0 = p cpM∞ =0 (8.51)
1 − M∞ 2
p
The factor 1/ 1 − M∞ 2 acts as a compressibility correction. In subsonic flow (M
∞ < 1) the
pressure coefficient increases due to compressibility effects. For M∞ close to 1 this correction
formula fails, the correction factor becomes very large which violates the small perturbation
assumption.
8.2. LINEARIZED FLOW; THE PRANDL-GLAUERT EQUATION 179

8.2.4 Example 2: Supersonic flow along a wavy wall.

Assume a wavy wall expressed by

y = εh(x) = ε sin kx,

and a supersonic flow (M∞ > 1) along it. Since M∞ ≥ 1 let us call
2 2
β∞ = M∞ − 1, (8.52)

so that the Prandtl-Glauert equation takes the form

2
β∞ φxx − φyy = 0, (8.53)

which is typically a 2D wave equation. The general solution is known as d’Alembert’s solutions:

φ(x, y) = F (x − β∞ y) + G(x + β∞ y), (8.54)

where F and G are arbitrary functions of the arguments x − β∞ y and x + β∞ y respectively.

F (x − β∞ y) represents left running waves and G(x + β∞ y) represents right running waves.
The straight lines x − β∞ y = constant and x + β∞ y = contant are left- and right running
characteristics. They represent Mach lines having an angle µ = ± sin−1 (1/M∞ ) with respect
to the undisturbed flow direction (the x-axis); see figure 8.2.

Figure 8.2: Characteristics of a linearized flow along a wavy wall

Since disturbances in a supersonic flow only propagate downstream the right running waves
do not carry any information; to account for this effect take G = 0 and the solution of the
Prandtl-Glauert-equation for the wavy wall problem has to be sought in the form

φ(x, y) = F (x − β∞ y). (8.55)

From equation (8.42) the boundary condition that has to be fulfilled reads:

φy (x, 0) = h0 (x) = k cos(kx) = −β∞ F 0 (x). (8.56)

Thus there follows

k
F 0 (x) = − cos(kx), (8.57)
β∞
or
k
F (x) = − sin(kx), (8.58)
β∞
180 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

or
1
F (x − β∞ y) = − sin(x − β∞ y). (8.59)
β∞
Hence the perturbation potential φ(x, y) that satisfies the Prandtl-Glauert equation and that
fulfills the boundary condition is then found as
1
φ(x, y) = − sin k(x − β∞ y). (8.60)
β∞
In order to obtain the pressure (e.g. in the form of the pressure coefficient) we have to calculate:
k
φx (x, 0) = − cos(kx). (8.61)
β∞
Using equation (8.34) we then find
εk
cp = 2 cos(kx). (8.62)
β∞
The pressure coefficient appears to be proportional to the slope of the wall. Everywhere where
the wall has a positive slope the pressure is increased with respect to the reference pressure p∞ .
Similarly, at negative slopes the pressure just decreases compared to the reference pressure
p∞ .
From equation (8.62) it can be deduced very easily that also in the supersonic case the
Prandtl-Glauert rule can be expressed as
1
cpM∞ 6=0 = cp √ (8.63)
β ∞ M∞ = 2
Note that in the transonic region (M∞ ≈ 1) the factor 1/β∞ becomes unbounded, the Prandtl-
Glauert rule becomes invalid and the small perturbation assumption is clearly not justified.
A qualitative picture of cp versus Mach numer shows this behaviour very well; see figure 8.3.

Figure 8.3: Prandtl-Glauert rule

8.3 Characteristics and compatibility relations for potential flow

In the previous section we have discussed the fully non-linear potential model (e.g. equation
(8.24)):
(a2 − u2 )Φxx − 2uvΦxy + (a2 − v 2 )Φyy = 0, (8.64)
8.3. CHARACTERISTICS AND COMPATIBILITY RELATIONS FOR POTENTIAL FLOW181

and also the linearized flow model (eg equation (8.33)):

2
(M∞ − 1)φxx − φyy = 0. (8.65)

The latter was referred to as the Prandtl-Glauert equation.

Both models are in the form of a second-order P.D.E. Now we attempt to find the charac-
teristics and the corresponding compatibility relations of both equations in order to compare
them with those of the fully non-linear equations obtained in chapter 7 (eg equation (7.54)).

8.3.1 Prandtl-Glauert equation

In order to obtain the characteristics of the PG equation we will apply the ‘left eigenvector’
method to the system

AUx0 + BUy0 = 0 (8.66)

which is obtained if the perturbation velocities u0 and v 0 are inserted into the PG equation.
Here U 0 = (u0 , v 0 )T is the state vector of the perturbation velocities u0 and v 0 according to

u0 = εU∞ φx ,
v 0 = εU∞ φy ,

and
 2   
β∞ 0 0 −1
A= , B= (8.67)
0 −1 1 0

where
2 2
β∞ = M∞ − 1. (8.68)

From det(B − λA) = 0 we obtain the eigenvalues

p
λ1 = − β1∞ = − 1/ M∞ 2 −1 (8.69)
p
λ4 = + β1∞ = + 1/ M∞ 2 −1 (8.70)

The eigenvalues are real only in the case of supersonic main flow (U∞ > a∞ ). The corre-
sponding left eigenvectors Li (where i = 1 . . . 4) are found from

Li A−1 B = λLi . (8.71)

Matrix A−1 B has the form

2
 
−1 0 −1/β∞
A B= (8.72)
−1 0

The left eigenvectors satisfying equation (8.71) are found as

λ1 = − β1∞ ⇒ L1 = (β∞ , −1),

λ4 = + β1∞ ⇒ L4 = (β∞ , +1),
182 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

The compatibility relations now follow as:

dy 1
along Γ− with =− :β∞ du0 + dv 0 = 0, (8.73)
dx β∞
dy 1
along Γ+ with =+ :β∞ du0 − dv 0 = 0. (8.74)
dx β∞

Since β∞ is a constant, the compatibility relations can be integrated into:

dy
β∞ u0 + v 0 = constant along = −1/β∞ = − tan µ, (8.75a)
dx
dy
β∞ u0 − v 0 = constant along = +1/β∞ = + tan µ, (8.75b)
dx
In the linearized flow model the characteristics appear as two bundles of parallel lines. One
bundle is the set of left-running characteristics with slope + tan µ∞ , the other bundle rep-
resents the set of right running characteristics with slope − tan µ∞ . Note that these slopes
are exactly the same as those in the undisturbed flow region. In the linearized flow model it
appears that perturbations do not influence the characteristic pattern of the flow field.
Let us now compare the compatibility relations as found in the linearized model (e.g.
equations 8.75a, 8.75b) with those of the fully non-linear model as they are given in equation
(7.54):
√
M2 − 1 dy
dq ± dφ = 0 along = tan(φ ∓ µ).
q dx

By remembering that q 2 = u2 + v 2 and tan φ = v/u the exact compatibility relations take the
form

(βu ∓ v)du + (βv ± u)dv = 0, (8.76)

with
p
β= M 2 − 1.

If we now introduce the perturbations u0 and v 0 according to

u(x, y) = U∞ + u0 (x, y), v(x, y) = v 0 (x, y)

and if these perturbations are assumed to be so small that

u0 v0
<< 1 and << 1 (8.77)
U∞ U∞

then equation (8.76) reduces to

dy
β∞ du0 ± dv 0 = 0 along = tan(∓µ∞ ) (8.78)
dx
which are precisely the linearized compatibility relations given in equations (8.75a),(8.75b).
8.3. CHARACTERISTICS AND COMPATIBILITY RELATIONS FOR POTENTIAL FLOW183

8.3.2 The Full Potential Equation

The full potential equation for two-dimensional steady compressible homentropic flow reads

(a2 − u2 )Φxx − 2uvΦxy + (a2 − v 2 )Φyy = 0 (8.79)

In order to obtain the characteristics and corresponding compatibility relations of this equation
we will again use the ‘left eigenvector method’. Equation (8.79) can be transformed into the
system

AUx + BUy = 0, (8.80)

where the state vector U is defined as

 
u
U= , (8.81)
v

and where the matrices A and B have the form

 2
a − u2 −uv −uv a2 − v 2
  
A= , and B = . (8.82)
0 −1 1 0

From det(B − λA) = 0 we obtain the eigenvalues:

√
uv ∓ a2 M 2 − 1
λ1,4 = , (8.83)
u2 − a2
which are exactly the same as those of the steady two-dimensional Euler equations, see equa-
tion (7.22). This implies that the characteristics of the full potential equation have a slope
dy
dx = tan(φ ∓ u). Physically it means that they have a slope-angle ∓µ with respect to the flow
direction.
The eigenvalues λ1,4 are real everywhere the local Mach number is larger than 1 (which is
the case for supersonic flow). In order to obtain the compatibility relations that correspond
to the real eigenvalues we have to determine the left eigenvectors Li (with i = 1 · · · 4) from

Li A−1 B = λLi . (8.84)

In the full potential flow model the matrix A1 B takes the form

a2 − u2
 
1 −2uv
A−1 B = 2 (8.85)
a − u2 −(a2 − u2 ) 0

The left eigenvectors that satisfy equation (8.84) are found to be

uv − βa2
λ1 = ⇒ L1 = (uv − βa2 , −(a2 − v 2 )),
u2 − a2
uv + βa2
λ4 = 2 ⇒ L4 = (uv + βa2 , −(a2 − v 2 )).
u − a2
From

dV = L dU, (8.86)
184 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

with
 
L1
L= (8.87)
L4

we obtain the compatibility relations in the full-potential model:

dV1,4 = (uv ∓ βa2 ) du − (a2 − v 2 ) dv = 0 (8.88)

dy
along dx = λ1,4 .
Using the definition:

q 2 = M 2 a2 = u2 + v 2 (8.89)

we may write

β(u2 + v 2 )
βa2 = (8.90)
β2 + 1

and equation (8.88) gets the form

dV1,4 = (β 2 uv + uv ∓ βu2 ∓ βv 2 )du + (β 2 v 2 − u2 )dv = 0,

or

(βu ∓ v)(βv ∓ u)du + (βv − u)(βv + u)dv = 0,

so that the compatibility relations for the full potential equation becomes
dy
(βu ∓ v)du + (βv ± u)dv = 0 along = λ1,4 = tan(φ ∓ µ) (8.91)
dx
Comparing the results with equation (8.76) leads to the conclusion that the compatibility
relations of the full potential equation and those of the Euler model for homentropic flow are
exactly the same. This result could be expected since for steady homentropic flow Crocco’s
theorem implies irrotational flow (∇ × ~v = ~0) which allows the introducion of potential flow.

8.4 Two-dimensional flow in the throat of a Laval nozzle

The Laval nozzle is an essential device in a supersonic windtunnel to produce supersonic flow
conditions downstream of the throat. From a one-dimensional point of view the transition
from subsonic to supersonic flow takes place just at the smallest cross-sectional area, the
throat of the nozzle. Upstream of the throat in the converging part, the flow is subsonic,
whereas downstream of the throat in the diverging part, the flow is supersonic, see figure 8.4.
Due to cross sectional area variations of the nozzle the quasi one-dimensional theory is not
an exact treatment of the flow problem. The quasi one-dimensional theory accounts for the
area effects on the mass flow by using the continuity equation in the form ρuA = constant but
this theory ignores traversal effectts such as transverse pressure gradients (due to streamline
curvatures) or transverse velocity variations. In order to get a more realistic view on the flow
properties in the nozzle a two-dimensional (planar or axi-symmetric) treatment is necessary.
8.4. TWO-DIMENSIONAL FLOW IN THE THROAT OF A LAVAL NOZZLE 185

Figure 8.4: Transition of a subsonic flow into a supersonic flow in a Laval nozzle (quasi one-
dimensional approximation)

Let us consider here the planar two-dimensional nozzle transporting a gas with ideal fluid
properties. Furthermore, we assume steady flow, no frictional effects and no heat transfer.
Also the flow is assumed to behave continuously in the sense that flow discontinuities such as
shock waves are absent. At the subsonic entrance of the nozzle the flow enters with uniform
conditions.
Summing up we are dealing with an isentropic and isenthalpic flow which is governed by
the full-potential equation:

(a2 − u2 )Φxx − 2uvΦxy + (a2 − v 2 )Φyy = 0, (8.92)

with

(u, v) = ∇Φ. (8.93)

The geometry of the Laval nozzle is shown in figure 8.5; only the upper part of the nozzle

Figure 8.5: Upper part of the symmetric Laval nozzle

contour is depicted because we have assumed a symmetric nozzle configuration. Now we

introduce the reference frame (x, y) with the x-axis on the center line of the nozzle. The
origin of the (x, y)-system is taken at the sonic point on the center line. At this particular
point, let the velocity components u and v take the values u = u∗ and v = v ∗ . Note that for
reasons of symmetry v ∗ = 0, hence u∗ is the local sound speed at the origin where the flow
is sonic. This specific value of the sound speed is referred to as a∗ . So at the origin we may
write u = u∗ = a∗ , v = v ∗ = 0.
In order to attain the details of the flow in the transition region we will assume that the
flow is a combination of a uniform flow (u = a∗ , v = 0) in x-direction and a non-uniform
186 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

velocity field u0 (x, y), v 0 (x, y) superimposed on it. Then the velocity field may be written as

u(x, y) = a∗ + u0 (x, y) (8.94)

0
v(x, y) = v (x, y) (8.95)

When taken in the neighbourhood of the throat, the perturbation terms u0 (x, y) and v 0 (x, y)
are small compared to the sonic sound speed a∗ , so that

u0 v0
 1 and  1. (8.96)
a∗ a∗
The transonic throat problem is concieved as a small perturbation of a uniform sonic flow. A
perturbation potential φ(x, y) may be introduced by

Φ(x, y) = a∗ (x + φ(x, y)). (8.97)

The perturbation velocities u0 (x, y) and v 0 (x, y) expressed in terms of the perturbation poten-
tial become

u0 (x, y) = a∗ φx
v 0 (x, y) = a∗ φy

So φx and φy may be seen as dimensionless velocities.

Let us now rewrite the full potential equation (8.92) in terms of u0 , v 0 and the perturbation
potential φ(x, y). The sound speed a may be also expressed in u0 and v 0 ; take the energy
equation in the form of equation (8.22) and evaluate the constant at at sonic conditions where
u = a∗ and v = v ∗ = 0:

a2 u2 + v 2 (γ + 1)(a∗ )2
+ = (8.98)
γ−1 2 2(γ − 1)

Now the following expressions can be obtained in a straightforward manner:

 a 2 γ−1
=1− (2φx + φ2x + φ2y ), (8.99)
a∗ 2 
a2 − u2

γ+1 2 γ−1 2
= − 2φ x + φ x + φ , (8.100)
(a∗ )2 2 γ+1 y
a2 − v 2
 
γ + 1 2(γ − 1) 2(γ − 1) 2 2
=1− φx + φ + φy . (8.101)
(a∗ )2 2 γ+1 γ+1 x

When these expressions are used in the full potential equation, the dominant terms are
obtained as

−(γ + 1)φx φxx − 2φy φxy + φyy + H.O.T. = 0 (8.102)

Equation (8.102) is a second-order P.D.E. for the perturbation potential φ(x, y). A solution
φ(x, y) is sought in the form of a series expansion in even powers of y:

φ(x, y) = f (x) + g(x)y 2 + h(x)y 4 + O(y 6 ). (8.103)

8.4. TWO-DIMENSIONAL FLOW IN THE THROAT OF A LAVAL NOZZLE 187

To make φ(x, y) an even function of y is just a consequence of the symmetry of the flow
problem. The functions f (x), g(x) and h(x) have to be determined so that equation (8.102)
is satisfied to a certain order. A possible solution, valid in the neighbourhood of x = 0, is:
1 2
f (x) = kx + O(x3 ),
2
γ+1 2
g(x) = k x + O(x2 ) (8.104)
2
(γ + 1)2 3
h(x) = k + O(x)
24
where k is a free parameter which will be determined later when the contour of the nozzle
wall is discussed. Combining equations (8.103) and (8.104), the solution for φ(x, y) valid near
the sonic point reads:
   
1 2 3 γ+1 2
φ(x, y) = kx + O(x ) + k x + O(x ) y 2 +
2
2 2
(8.105)
(γ + 1)2 3
 
4 6
k + O(x) y + O(y ).
24

When omitting the higher order terms O(x3 ), O(x2 )y 2 , O(x)y 4 and O(y 6 ) this solution satisfies
exactly the P.D.E.:

−(γ + 1)φx φxx + φyy = 0. (8.106)

The term φx φxy in equation (8.102) appears to be a higher order term as well!
From equation (8.105) the velocity field near x = 0, y = 0 may be obtained as
γ+1 2 2
u = a∗ (1 + kx k y ) + O(x2 , xy 2 , y 4 ), (8.107)
2
(γ + 1)2 3 3
 
∗ 2
v = a (γ + 1)k xy + k y + O(x2 y, xy 3 , y 5 ). (8.108)
6

Observe that along the center line, where y = 0, we have v = 0 and u = a∗ · (1 + kx + O(x2 )).
If k > 0 then u < a∗ for x < 0 and u > a∗ for x > 0. Obviously the origin is sonic and k > 0
implies transition from the subsonic to the supersonic region. Using the velocity field just
obtained in equations (8.107) and (8.108), the sonic line where M = 1 (or u2 + v 2 = (a∗ )2 )
may be found as:
γ+1 2
xsonic line = − ky + O(y 4 ). (8.109)
2
The sonic line has a parabolic shape; for k > 0 it runs upstream (x < 0) if y 6= 0, see figure
8.6.
Observe the difference with one-dimensional theory which predicts the sonic line as a line
x = constant. Another aspect of interest is the curve where the flow is parallel to the x-axis,
so where v = 0. From equation (8.108) we obtain two solutions: the trivial solution y = 0 and
the non-trivial one:
γ+1 2
v=0: x=− ky + O(y 4 ). (8.110)
6
188 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

Figure 8.6: Two-dimensional flow in the throat of a Laval nozzle (k > 0).

Similar to the sonic line, the v = 0 curve is also a parabola running upstream (x < 0) if y 6= 0,
see figure 8.6. Across the v = 0 curve the v-component changes sign: upstream of this curve
v is negative, downstream v is positive.
This implies a streamline behaviour as sketched qualtitatively in figure 8.6. In inviscid
flow theory every streamline mau be viewed as a solid wall. So we observe now a converging-
diverging channel with a throat at v = 0! Now take a particular streamline and call the point
in the throat as point T . For a prescribed height of the throat yT the x location of the throat
follows from equation (8.110) as
γ+1 2
xT = − kyT . (8.111)
6
At this x-location the sonic line has the height y ∗ ; with equation (8.109) it can be calculated
from
γ+1
xT = − k(y ∗ )2 . (8.112)
6
The ratio between the height of the throat yT and the location of the sonic point in the throat
is now found to be
y∗ 1√
= 3 ≈ 0.577. (8.113)
yT 3
In a two-dimensional planar nozzle the throat appears to be partially subsonic (near and on
the center line) and partially supersonic (near the wall contour). There appears only one
sonic point in the throat; this is in contrast with the one-dimensional theory, which predicts
an entirely sonic throat!

Shape of the wall contour

The transonic throat problem just solved in equation (8.107) and (8.108) still contains the
unknown parameter k. When considering the velocity distribution along the center line:
u(x, 0) = a∗ · (1 + kx + O(x2 )), (8.114)
8.4. TWO-DIMENSIONAL FLOW IN THE THROAT OF A LAVAL NOZZLE 189

k can be interpreted physically as the first derivative in streamline direction of the dimension-
less velocity u/a∗ :
 
∂u
= a∗ k, (8.115)
∂x u=a∗

taken at the sonic point: u = a = a∗ . This is just an interpretation in terms of a local variable
and not yet in terms of global variables such as the geometry of the Laval nozzle. To find such
a relationship let us calculate the shape of an arbitrary streamline.
Streamlines in the throat area of the nozzle are solutions of the ordinary differential equa-
tion
dy v(x, y)
=
dx u(x, y)
(γ+1)2 3 3 (8.116)
(γ + 1)k 2 xy + 6 k y + O(x2 y, xy 3 , y 5 )
= γ+1 2 2 .
2 2 5
1 + kx + 2 k y + O(x , xy , y )

Now we attempt to find the shape of that particular streamline that forms a throat at
T (xT , yT ), see figure 8.6. Since u(x, y) and v(x, y) are atmost known approximately it is not
possible to find the exact shape of the streamline. Only an approximation that is consistent
with the order estimation of u(x, y) and v(x, y) can be obtained.
The streamline going through yT is a perturbation of the streamline y = yT being the
unperturbed streamline of the uniform sonic flow (u = a∗ , v = 0) going through T . Obviously
in this unperturbed flow the flow direction is zero everywhere. In the perturbed situation
the streamline going through T attains its minimum y-value at T and therefore it may be
approximated by the expression

y − yT = α(x − xT )2 + H.O.T. (8.117)

The coefficient α can be determined by requiring that (8.117) is an approximate solution of

the streamline equation (8.116). This yields:

k 2 kyT
y − yT = (γ + 1) (x − xT )2 + H.O.T. (8.118)
2
Since every inviscid streamline mey be replaced by a solid wall, (8.118) gives an expression
for the wall contour of a converging-diverging nozzle with a throat height yT = H. The local
parameter k is related to the radius of curvature R of the wall contour at the throat.
From (8.118) the radius of curvature of the nozzle contour follows as

1
= (γ + 1)k 2 H. (8.119)
R

Remember that the parameter k is a measure for the streamline velocity gradient ∂ ∗
∂x (u/a )
at the throat. Combining (8.115) and (8.119) we get the result
s
∂ u 1
∗
=± . (8.120)
∂x a RH(γ + 1)
190 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

The streamwise velocity gradient in the throat depends on the radius of the curvature and on
the height of the throat; two parameters very typical for the throat geometry! A small throat
height and a highly curved nozzle contour result into a high velocity gradient in the throat.
The flow properties and especially the velocity distribution in the Laval nozzle may now be
obtained in a non-dimensional form if x/H and y/H are used as spatial variables. Furthermore,
if k is expressed in terms of the throat height H and the radius of curvature R then the
expressions, originally presented by Oswatitch and Rothstein are obtained:
r    
u H 1 H x 1 H  y 2
= 1+ √ + (8.121)
a∗ R γ+1 R H 2 R H
    √  1  
v H x y y+1 H 2 y 3
= + (8.122)
u∗ R H H 6 R H

Observe the appearance of the dimensionless parameter H/R in this representation of the
velocity field.

8.4.1 Mass flow through a two-dimensional nozzle

Consider the mass flow through the upper half of a symmetric nozzle having a throat height
H and a nozzle wall curvature R, see figure 8.7.

Figure 8.7: Two-dimensional nozzle

The mass flow through the throat at x = xT is defined as

Z H Z H
ρ u
ṁ = ρu dy = ρ∗ a∗ dy (8.123)
0 0 ρ∗ a∗

where ρu(x, y) = ρu(xT , y). From equation (8.107):

u u0 γ+1 2 2
= 1 + = 1 + kx + k y + H.O.T.
a∗ a∗ 2
8.4. TWO-DIMENSIONAL FLOW IN THE THROAT OF A LAVAL NOZZLE 191

The ratio ρ/ρ∗ follows from

a 2
ρ/ρ∗ =
γ−1

a ∗
( 1
2 2 !) γ−1
u0 u0 v0
 
γ−1
= 1− 2 ∗+ + .
2 a a∗ a∗
0 v0
Since au∗ and a∗ are small with respect to one, a binomial expression 1 can be applied to get
the result:
2 2
u0 u0 v0
 
ρ 1 1 2−γ 0 2
∗
= 1 − ∗
− − + (u ) + H.O.T.
ρ a 2 a∗ 2 a∗ 2

The mass flux ρu may be approximated as

!
u0 u0 γ − 1 u0 2 1 v 0 2
     
ρu
= 1+ ∗ 1− ∗ − − + ...
ρ∗ u∗ a a 2 a∗ 2 a∗
γ + 1 u0 2 1 v 0 2
   
=1− − + ...
2 a∗ 2 a∗
0 v0
Inserting au∗ and a∗ from (8.107) and (8.108) and taking into account the proper order esti-
mation, we find:
(  2 )
ρu γ+1 2 2 3 2 γ+1 4 4
=1− k x + k (γ + 1)xy + k y . (8.124)
ρ∗ u∗ 2 2

The mass flow through the nozzle follows by integrating the mass flux ρu along the nozzle
height at the throat where x = xT ; xT itself is a function of the throat height H and the
radius of curvature R:
r
γ+1 2 Hp H
xT = − kH = − γ+1 . (8.125)
6 6 R
The mass flow through a two-dimensional nozzle with throat height H and radius of
curvature R may now be written as

H2 γ + 1
 
∗ ∗
ṁ2D = ρ a H 1 − 2 . (8.126)
R 90

Notice the difference with the mass flow according to the one-dimensional theory which pre-
dicts:

ṁ1D = ρ∗ a∗ H. (8.127)

There is a two-dimensional effect which reduces the mass flow with an amount that increases
with the ratio H/R, see figure 8.8. For example: at H/R = 1 the mass flow reduction is
2.66 % but as H/R = 4 a mass flow reduction of 42.66 % has to be accepted.
1 2
Binomial expansion of (1 + ε)α = 1 + αε + α(α − 1) ε2! . . .
192 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS

Figure 8.8: Mass flow reduction in a two-dimensional nozzle

The reduction of the mass flow is a direct consequence of the variable mass flux ρu in the
throat. The mass flux distribution in the throat can be obtained from equation (8.124) by
taking x = xT = − γ+1 2
6 H ; this yields:

(γ + 1)2 4 4
 
ρu
= 1 − k (H − 6H 2 y + 9y 4 )
ρ∗ u∗ throat 72
(8.128)
γ + 1 (H 2 − 3y 2 )2
 
=1− .
72 H 2 R2
A sketch of the mass flux distribution (according to equation (8.128)) is given in figure
8.9.
Observe that the mass flux√attains its maximum one-dimensional value ρu = ρ∗ a∗ just at
the sonic point (y = y ∗ = H/ 3). At y = 0 (at the center line of the nozzle) the mass flux
takes the value
γ + 1 H2
 
∗ ∗
(ρu)center line = ρ a 1 − (8.129)
72 R2
and at the nozzle wall y = H we find a mass flux
H2
 
∗ ∗ 4
(ρu)nozzle wall = ρ a 1 − (γ + 1) 2 (8.130)
72 R
The mass flux defect at the nozzle wall appears four times larger than that at the center line.
Moreover, the subsonic part of the nozzle throat transports a higher amount of mass flow than
the supersonic part does. This can be verified by calculating the mass ratio
γ+1
2
ṁsubsonic 1 1 − 135 H R
=√ (8.131)
ṁtotal 3 1 − γ+1 H 2


90 R
8.4. TWO-DIMENSIONAL FLOW IN THE THROAT OF A LAVAL NOZZLE 193

Figure 8.9: Mass flux distribution in the throat.

For example, if H/R = 1 the mass ratio takes a value 0.583 and if H/R = 3 this ratio becomes
0.639. Indeed the subsonic part of the throat transports a higher mass flow than the supersonic
part does.

8.4.2 Literature on the subject

Relevant information on this subject can be found in the following references:

1. Oswatitsch & Rothstein: “Das Strömungsfield in einer Laval Düse”, Deutsche Lufo I,
91-102, 1942.

2. Hall, I.M.: “Transonic Flow in Two-dimensional and Axi-Symmetric Nozzles”, Quarterly

Journal on Mechanics and Applied Mathematics 15, 487-508, 1962.

3. Oswatitsch, K.:“Spezialgebiete der Gasdynamik”, Springer Verlag; pp. 94

194 CHAPTER 8. TWO-DIMENSIONAL STEADY POTENTIAL FLOWS
Chapter 9

The ’first non linear’ description of

steady two dimensional supersonic
flow

9.1 Inviscid Burgers equation

It is possible to derive an inviscid Burgers equation for simple waves in steady 2D supersonic
flow. Let us assume that simple waves are created in the upper half plane by a bottom wall.
This wall may have a more or less arbitrary shape but far upstream the wall is plane and
parallel to the oncoming free stream with Mach number M∞ > 1, see figure 9.1. We assume a
homentropic and isenthalpic flow meaning that s and H are uniform. Above the wall a simple
wave will be formed.
This created simple wave field is characterized by the condition that the invariant V − is
constant in the whole flow domain. Due to the uniformity of s and H this yields that

V − = ν + ϕ = constant = V∞
−
, (9.1)

hold in the whole field. This simple-wave field is still governed by the characteristic equation

∂ ∂
(ν − ϕ) + tan (ϕ + µ) (ν − ϕ) = 0. (9.2)
∂x ∂y

Figure 9.1: Simple waves created by a bottom wall in a supersonic flow

195
196CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

dy
The invariant V + = ν − ϕ is constant along the Γ+ characteristics having the slope dx =
tan (ϕ + µ). But if this result is combined with the fact that ν + ϕ = constant everywhere
it is clear that each variable ν or ϕ at its own is constant along Γ+ ; and furthermore ,
ν = constant implies µ = constant and thus also the slope tan (ϕ + µ) is constant along Γ+ :
the Γ+ characteristics are straight lines.
From equation (9.1) it is obvious that in the simple wave domain the flow angle ϕ is only a
function of the Mach number, formally written as
−
ϕ = V∞ − ν(M ). (9.3)

The function is monotonic in M because ν(M ) is monotonic in M . But from equation (9.3) we
must conclude that also the variables (ν − ϕ) and tan(µ + ϕ) are purely monotonic functions
of the Mach number M , thus formally written:
−
ν − ϕ = 2ν(M ) − V∞ = F (M ), (9.4)

and
−


tan (ϕ + µ(M )) = tan V∞ − ν(M ) + µ(M ) = V (M ). (9.5)

In principal it is now possible to eliminate the Mach number in (9.4) and (9.5) to obtain a
direct expression for (ν − ϕ) in terms of v = tan(ϕ + µ) yielding

(ν − ϕ) = F ∗ (v) = F ∗ (tan(ϕ + µ)) .

Equation (9.2) may be rewritten as:

∂ ∂
(F ∗ (v)) + v (F ∗ (v)) = 0,
∂x ∂y
or
dF ∗ ∂v dF ∗ ∂v
+v = 0. (9.6)
dv ∂x dv ∂y
dF ∗
Now get rid of dv in both terms of (9.6), yielding
∂v ∂v
+v = 0, (9.7)
∂x ∂y
with v = tan(ϕ + µ). Equation (9.7) is the desired form, it is the inviscid Burgers equation
valid for simple waves (pointing upwards) in a two dimensional steady supersonic flow.
The characteristics of the Burgers equation
dy
= v = tan(ϕ + µ), (9.8)
dx
are just the Γ+ characteristics in the simple wave.
In conservative form the inviscid Burgers equation becomes
 
∂v ∂ 1 2
+ v = 0. (9.9)
∂x ∂y 2
This form is suitable in finding weak solutions to model discontinuities (see next section).
9.2. DISCONTINUITIES OF THE INVISCID BURGERS EQUATION 197

Figure 9.2: Shock formed at ramp with angle δ

9.2 Discontinuities of the inviscid Burgers equation

Equation (9.7) is applicable in the continuous part of the flow; once a shock is formed the
solutions break down because the assumption that both entropy (s) and the invariant V − are
uniform is violated. For weak shocks however the error in the solution is small because shock
jumps are of the order (∆p)3 .
Weak solutions of the Burgers equation in conservative form:
 
∂v ∂ 1 2
+ v = 0,
∂x ∂y 2
are given by
1/2v 2
   
dy 1
= = (vpost + vpre } , (9.10)
dx shock [v] 2
or in terms of ϕ + µ:
 
dy 1
= {tan(ϕ + µ)post + tan(ϕ + µ)pre ) . (9.11)
dx shock 2
In words: the slope of the shock is the average of the slopes of the characteristics merging into
the shock, see figure 9.2.
In this figure the supersonic flow along a ramp with ramp angle δ is considered. Upstream
of the ramp there is a uniform supersonic flow with Mach number M1 and corresponding
Prandtl-Meyer angle ν1 . Let the (x,y)-frame be chosen such that the flow angle ϕ1 = 0.
Referring to the simple wave condition (9.1) the flow properties downstream on the ramp are
again uniform with Mach number M2 , flow angle ϕ2 = δ and Prandtl-Meyer angle ν2 = ν1 − δ.
Apparently ν2 < ν1 and therefore M2 < M1 and µ2 > µ1 . The slope of the characteristics
obey the inequality
tan(ϕ2 + µ2 ) > tan(ϕ1 + µ1 ),
or
vpost > vpre .
Obviously the characteristics merge and terminate at the shock having the slope
 
dy 1
= {tan(ϕ2 + µ2 ) + tan(ϕ1 + µ1 )} .
dx shock 2
198CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Shock formation Shock formation appears if characteristics start to intersect. In order

to study the shock formation process, consider an arbitrary non-linear distribution of v(x, y)
prescribed as boundary data along y = 0. Let the distribution be given by

v(x, 0) = vB (x).

To find the location where the shock formation starts, consider an arbitrary point xj and its
neighbour xj + ∆x on the line y = 0. The characteristics going through xj and xj + ∆x are
given by:

y = vB (xj )(x − xj ),

and

y = vB (xj + ∆x)(x − xj − ∆x),

respectively. Both characteristics intersect at (xI , yI ), for ∆x → 0 we find:

   2 
vB (xj ) vB vB
xI − xj = = 0
, yI = . (9.12)
v 0 B x=xj
 
dvB v B x=xj
dx x=xj

Obviously, there are only intersecting characteristics in the upper half plane (y > 0) if v 0 B > 0.
Shock formation may be expected where vB (x) increases in streamwise direction. Equation
(9.12) contains xj being the x-location on the boundary value line (y = 0) of the characteristic
running to (xI ,yI ) where neighbouring characteristics intersect.
Intersecting characteristics announce the possibility of shock development which really hap-
pens where yI is minimal. This value of yI is called ys , the corresponding value of xI is similarly
called xs . Let xjmin be the value of xj where vB 2 /v 0 is minimal, then shock formation starts
B
at (xs ,ys ):
   2 
vB vB
xs − xjmin = 0
, ys = , (9.13)
vB v0B
where xjmin has to be solved from
2
2v 0 B (xjmin ) − vB v 00 B (xjmin ) = 0, (9.14)

(primes indicate differentiation).

Example of a shock development Assume a boundary-value distribution vB (x) at y = 0

as follows
x < 0 : vB = 1,
0 < x < 1 : vB = 1 + x,
x > 1 : vB = 2.

Along interval 0 < x < 1, vB (x) is increasing so that shock formation may be expected. Using
equation (9.14) yields xjmin = 0; shock formation starts at (xs , ys ) = (1, 1), see figure 9.3.
From the boundary value line (y = 0), x < 0, characteristics emanate and run into the
dy
upper half plane. They are parallel and have a slope dx = 1; a uniform domain with v(x, y) =
9.2. DISCONTINUITIES OF THE INVISCID BURGERS EQUATION 199

Figure 9.3: Shock formation process

constant(= 1) results. If we further assume in domain Àa parallel √ flow with ϕ1 = 0, then

v = 1 implies a uniform supersonic flow with Mach number M1 = 2 (tan µ1 = 1).
dy
The characteristics originating from x > 1, y = 0 are also parallel, they have a slope dx =
o
2. A uniform domain Áis formed with M2 = 1.19 and ϕ2 = 5.9 . In the shaded area
between domains Àand Ácharacteristics converge, the flow variables are constant along a
characteristic but they vary if one crosses a Γ+ characteristic, so a real simple wave is formed.
All characteristics terminate at the shock, the characteristics in domain Àterminate at the
front part of the shock, those in the shaded area and in domain Áterminate at the rear part
of the shock.
Let us try to find the shock shape from the O.D.E.
 
dy 1
= (vpost + vpre ) .
dx shock 2
Take an arbitrary point (x, y) on the shock, then
vpre = 1,  
y
vpost = vB x − vpost .

For the curved part of the shock the expression for vpost becomes
y
vpost = 1 + x − ,
vpost
or

vpost 2 − vpost (1 + x) + y = 0. (9.15)

200CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

The O.D.E. that has to be solved becomes

dy 1
= (1 + vpost ) , (9.16)
dx 2
where vpost is determined by equation (9.15).
The boundary condition that has to be satisfied is: vpost = 1 at (x, y) = (1, 1); this condition
guarantees that the shock starts to develop at the point (xs , ys ) = (1, 1) with the correct slope
dy
dx = 1.
The solution procedure will be accomplished in two steps; in the first step vpost will be deter-
mined as a function of x, in the second step the R.H.S. of equation (9.16) is then a known
function of x and integral calculus enables us to find the shock shape ys = ys (x).
In order to set up the first step, equation (9.15) has to be differentiated w.r.t. x, this yields
(using vp = vpost )
dvp dvp dy
2vp − (1 + x) − vp + = 0.
dx dx dx
dy
The shock slope dx can be eliminated from equation (9.16) and expressed in terms of the
variable vp . Then the post shock value of v (i.e. vp ) is found as the solution of the O.D.E
dvp vp − 1
= , (9.17)
dx 2 (2vp − 1 − x)
with boundary condition: x = 1, vp = 1.
The solution for x > 1 is the linear graph
3x + 1
vp = . (9.18)
4
Shock development is accomplished when vp = 2. This happens at x = 7/3. The shape of the
shock follows from the O.D.E. (equation(9.16))
 
dy 1 3x + 1 5 + 3x
= 1+ = .
dx 2 4 8
Taking into account the boundary conditions xs = 1, ys = 1 the shock shape is found as
1
3x2 + 10x + 3 .


yshock = (9.19)
16
It is a parabolic curve that steepens when moving downstream until x = 37 , y = 83 . Here the
shock formation is accomplished; point T in figure 9.3. For x > 73 the shock maintains a
dy
constant slope dx = 2.
As in the unsteady case the shock strength is measured as the difference ∆v = vpost − vpre .
The shock strength varies along the shock curve as
3
∆v = (x − 1) ,
4
which shows that the shock strengthens linearly from ∆v = 0 at x = 1 to ∆v = 1 at x ≥ 73 .
Streamlines in the steady two dimensional flow field are governed by the equation
 
dy
= tan ϕ.
dx streamline
9.2. DISCONTINUITIES OF THE INVISCID BURGERS EQUATION 201

Writing tan ϕ = tan(ϕ + µ − µ), the streamline equation can be arranged into
 
dy βv − 1
= , (9.20)
dx streamline β+v
√
where β = M 2 − 1 = cot µ.
Equation (9.20) is hard to solve, at least analytically because β is a complicated expression
in terms of the variable v and v itself depends on x and y according to:
 y y
v(x, y) = vB x − =1+x− ,
v v
leading to

v 2 − v(1 + x) + y = 0, (9.21)

valid in the simple wave region.

On a particular streamline, y depends on x and with equation (9.21) also v depends on x.
The v distribution along a streamline satisfies the O.D.E. (differentiate equation (9.21) w.r.t.
x!)

dv dv dy
2v − (1 + x) −v = 0.
dx dx dx
dy
Now dx can be substituted from equation (9.20) yielding

1 + v2
 
dv
= . (9.22)
dx streamline (β + v)(2v − 1 − x)

This expression, when solved, yields the v-distribution along a streamline.

In order to detect the course of a particular streamline, e.g. the extension of the streamline
that starts in domain Àat level y = h, into the simple wave, we have to apply one of the
following boundary conditions:

− underneath s : v = 1(β = 1) , x = h,
(9.23)
− behind curved part of shock : v = 3x+1 4 , 16h = 3x2 + 10x + 3.

Once v(x) is determined from equation (9.22) then essentially β(x) is known along the stream-
line and equation (9.20) can be integrated yielding the shape of the streamline in the simple
wave domain.
For the particular problem, defined in figure 9.3, with oncoming flow properties: β = 1, v = 1,
the characteristic variable v lies in the interval 1 ≤ v ≤ 2. The flow variable (β + v) in
equation (9.22) may be approximated as

(v − 1) (v − 1)2
(β + v) ≈ (β + v)a = 2 + + .
2 10
Although, the comparison with the exact value (see figure 9.4) shows a fair agreement, its
use in equation (9.22) does not help that much than an analytical solution of equation (9.22)
could be found. Consequently, we have to resort to numerical integration of the equations
(9.22) and (9.20) finally.
202CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Figure 9.4: Approximation of β + v in the neighbourhood β = 1, v = 1

9.3 Shock pattern created by a biconvex airfoil

A biconvex airfoil with symmetrical shape, a sharp leading edge (L.E.) and a sharp trailing
edge (T.E.) is placed in a uniform supersonic flow (Mach number M∞ ) at zero incidence
(α = 0), as shown in figure 9.5. Only the upper half of the flow is shown.

The strength of the shocks are assumed sufficiently weak, they are attached to the edges
of the airfoil so that simple wave theory, e.g. presented by the inviscid Burgers equation is
applicable. Using the simple wave conditions (see equation (9.1)):
−
V∞ = ν + ϕ,

the Prandtl-Meyer angle ν(x) along the airfoil contour y = f (x) results. Once the distribution
ν(x) is known also other variables as the Mach number M (x), Mach angle µ(x) and character-
istic direction v(x) can be determined without much effort. Hence for a given airfoil contour
the v-distribution on the contour: v(x, f (x)) is known and the (straight) characteristics ema-
nating from the contour y = f (x) can be drawn. A characteristic pattern results and shock
appear if characteristics intersect.
To study the shock pattern let us consider the inverse problem where v(x) is prescribed and
the corresponding airfoil contour has to be derived. Furthermore we assume that v(x) is
prescribed on the x-axis (y = 0) rather than on the contour (yet unknown) itself. Since char-
acteristics are straight lines and v is constant along a characteristic a simple extrapolation of
the straight characteristic from airfoil contour onto the x-axis allows us to do so; see figure
9.6.
9.3. SHOCK PATTERN CREATED BY A BICONVEX AIRFOIL 203

Figure 9.5: Shock pattern about a pointed airfoil

Figure 9.6: Projection of boundary data on x-axis

204CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Figure 9.7: N-wave boundary value distribution at y = 0

By this procedure v(x, f (x)) on the contour is projected onto y = 0 resulting into a
corresponding and equivalent distribution: vB (x) on y = 0. Thus vB (x) generates exactly the
same wave pattern as the actual airfoil contour: y = f (x), with prescribed v(x, f (x)) does.
Consider the following boundary value distribution vB (x) at y = 0:

x<0 vB = v∞ > 0, 
0<x<1 vB = vL − (vL − vT ) x, (9.24)
x>1 vB = v∞ > 0,


shown in figure 9.7 and being a typical N-wave.

In the oncoming uniform flow field all characteristics have the slope v∞ = tan (µ∞ ) > 0.
Across the leading edge shock the Mach number drops and the characteristic slope jumps to
vL > v∞ . Over the airfoil the flow expands and v decreases linearly to vT < vL being the
pre-shock value of the trailing edge shock. Across the trailing edge shock v jumps to the free
stream value v∞ and keeps this value further downstream. The main objective is twofold;
first find the shock wave pattern and its behaviour (e.g. shock delay) in the farfield (y → ∞);
secondly find the airfoil shape (y = f (x)) that generates that wave pattern.

Bow shock The bow shock shape is governed by the O.D.E.

 
dy 1
= (vpre + vpost ) . (9.25)
dx shock 2

Take an arbitrary point (x, y) on the bow shock then,

)
vpre = v∞ ,
y
 (9.26)
vpost = vB x − vpost .
9.3. SHOCK PATTERN CREATED BY A BICONVEX AIRFOIL 205

Using equation (9.24) and the shorter notation vp = vpost there results
 
y
vp (x, y) = vL − (vL − vT ) x − ,
vp (x, y)
or

vp2 − {vL − (vL − vT )x} vp = y(vL − vT ). (9.27)

The O.D.E. that has to be solved becomes

 
dy 1
= (v∞ + vp ) , (9.28)
dx shock 2

where vp follows from equation (9.27).

The boundary conditions that must be fulfilled care that the bow shock starts at the leading
edge, so at (x, y, ) = (0, 0), vp = vL . The solution procedure that results into the shock shape
contains two steps; first vp will be determined as a function of x, in the second step the R.H.S.
of equation (9.28) is a known function of x and integral calculus enables us to get the shock
shape ys = ys (x). To set up the first step equation (9.27) is differentiated w.r.t. x, this yields

dvp dvp dy
2vp − {vL − ∆ · x} + ∆ · vp = ∆ · ,
dx dx dx
here ∆ = vL −vT being the total drop of v along the airfoil contour from leading edge to trailing
edge. The shock slope may be substituted from equation (9.28). Then the vp -distribution along
the bow shock is governed by the O.D.E.:

dvp ∆ (v∞ − vp )
= , (9.29)
dx 2 (2vp − vL + ∆ · x)

supplemented with the boundary conditions: x = 0, vp = vL . Standard solution techniques

for O.D.E.’s teach us that the solution of equation (9.29) becomes

∆ 2vp3 vp2 v3
(vp − v∞ )2 x + − (2v∞ + vL ) + vp vL v∞ = L , (9.30)
2 3 2 6
vp (x) is a monotonic decreasing function of x, see figure 9.8. For large x (x → ∞) vp approaches
the free stream value v∞ .
Let us define the strength of the bow shock by ∆B = vp − v∞ ; figure 9.8 shows that the
strength of the shock goes to zero for large x.

The asymptotic behaviour of the shock strength for x → ∞ may be obtained from (9.30)
as (take vp → v∞ ):

∆ v3 v3 2
vL v∞
(∆B )2 x ∝ L + ∞ − .
2 6 3 2
Introducing the non-dimensional variables

e = ∆,
∆ e B = ∆B ,
∆ veL =
vL
,
v∞ v∞ v∞
206CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Figure 9.8: Shock strength vs x

a suitable form follows

s r
e B ∝ (e 2 + veL 1
∆ vL − 1) .
3∆
e x

Since veL − 1 can be interpreted as the non-dimensional shock strength at the leading edge the
asymptotic behaviour of the shock strength gets the form:
s
e B (0) 3 + ∆B (0) √1 .
e
∆e B,x→∞ ∝ ∆ (9.31)
3∆e x

So we observe shock decay with increasing x as √1x .

The precise shock shape in the physical plane can be determined, at least in principal, by
integrating equation (9.28) with vp (x) solved from equation (9.30). This is left as a homework
exercise in numerical analysis.
Since vp (x) decreases monotonically to the limiting value vp = v∞ at x = ∞, it is obvious
that the shock has a curved shape with decreasing slop for increasing x. For x → ∞ the shock
attains the slope of the characteristics of the oncoming flow.

Tail shock The tail shock is determined by the O.D.E.

 
dy 1
= (vpre + vpost ) . (9.32)
dx shock 2

If (x, y) is an arbitrary point on the tail shock then

  )
y
vpre = vB x − vpre ,
(9.33)
vpost = v∞ .
9.4. WAVE INTERACTIONS 207

Let us now compare this set of equations with those derived in (9.25) and (9.26) determining
the bow shock. Evidently they are exactly the same! Both bow shock and tail shock are
governed by the same O.D.E. namely equation (9.29). In case of the tail shock vp refers to
vpre . Although bow and tail shock are governed by the same O.D.E. they have different shapes
because their boundary conditions differ.
For the tail shock the boundary condition warrants that this shock starts at the trailing edge
of the airfoil: (x, y) = (1, 0), vp = vT . The solution of equation (9.29) that fulfills these
boundary conditions reads:

∆ 2vp3 vp2 v3 ∆ 2
(vp − v∞ )2 x + − (2v∞ + vL ) + vp vL v∞ = T + v∞ , (9.34)
2 3 2 6 2
where vp (x) is the pre-shock v-distribution along the tail shock.
vp (x) is monotonic increasing with x, for large x (x → ∞) it approaches the free stream value
v∞ , see figure 9.8. The monotonic increase from vp = vT to the limiting value vp = v∞ (at
x → ∞) indicates a curved trailing edge shock with increasing slope for increasing x. For
x → ∞ the shock gets the slope of the characteristics of the free stream. Let the shock
strength of the tail shock be defined as ∆T = v∞ − vp ; from figure 9.8 we observe shock decay
at larger distances from the airfoil.
Physically, shock decay is caused by the interaction of expansion characteristics emanating
from the aft part of the airfoil with the trailing edge shock.
The asymptotic behaviour of the shock strength for x → ∞ may be obtained from equation
(9.34) as (take vp → v∞ ):

∆ v3 v3 2
vT v∞
(∆T )2 x ∝ T + ∞ − .
2 6 3 2
Introducing the non-dimensional variables:

e = ∆,
∆ e T = ∆T ,
∆ veT =
vT
,
v∞ v∞ v∞
we find the form:
s
2 + veT 1
e T ∝ (1 − veT )
∆ √ . (9.35)
3∆
e x

We observe shock decay with increasing x as √1 ; a result very similar to what happens with
x
the bow shock (see equation (9.31)).

9.4 Wave interactions

As we have done in the unsteady 1-D case, the Burgers model for simple waves enables us to
study non-linear wave interaction.
The biconvex airfoil, treated in section 9.3 is an example of this phenomenon. The bow wave
is weakened due to the interaction with expansion waves that come from the front part of the
airfoil; similarly the trailing edge shock wave is weakened by expansion waves coming from the
aft part of the airfoil. In this section two other examples of wave interaction will be presented:
shock-expansion interaction and the backward facing slope.
208CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Figure 9.9: Shock-expansion interaction

9.4.1 Shock-expansion interaction

Assume boundary conditions on the x-axis: vB (x) given as:

x < − 21 vB = 1,
− 12 <x<0 vB = 3,
x>0 vB = 2.

This distribution represents a shock at x = − 12 and a centered expansion at x = 0, see figure

dy
9.9. The shock emanating from x = − 12 has a slope dx = 2. The first characteristic of the
expansion fan hits the shock at (1, 3). Then, for larger y, the shock gets curved. The shock
shape is governed by the O.D.E.
 
dy 1 y
= 1+ ,
dx shock 2 x

supplemented with the boundary condition that it runs through (x, y) = (1, 3). The solution
that fulfills this boundary condition reads
√
ys = 2 x + x.
dy
The last characteristic of the fan (v = 2) hits the shock when the shock has the slope dx = 23 .
dy
This happens at the point (4, 8). For x > 4 the shock is straight keeping the slope dx = 23 .
The shock strength in the interval 1 ≤ x ≤ 4 can be calculated as
√
2 x+x 2
∆ = vpost − vpre = −1= √ ,
x x
9.4. WAVE INTERACTIONS 209

Figure 9.10: v distribution at different y levels

indicating shock decay from ∆ = 2 at x = 1 to ∆ = 1 at x = 4. The v-distribution at different

y levels is drawn in figure 9.10.
The stronger shock with ∆ = 2 present at y = 0, is partially ’eaten’ by the expansion, and a
less stronger shock having ∆ = 1 remains at y ≥ 8.
Notice the non-linear v decrease from v = 3 to v = 2. Its graph is the orthogonal hyperbolae:
vx = y.

9.4.2 Backward facing slope

A uniform stream with Mach number 1.4 flows downhill of a 30o backward facing slope, then it
is compressed back into the original direction, see figure 9.11. At the shoulder (O) a centered
expansion appears which further downstream interacts with the shock wave generated at the
compression corner (P ). The free stream (domain À) conditions with ϕ1 = 0 are: M1 = 1.4,
µ1 = 45.58o and v1 = 1.020. The flow conditions in the uniform domain Áalong the backward
facing slope may be calculated as M2 = 2.494, µ2 = 23.64o and v2 = −0.111. Downstream
of the shock wave the flow is turned back to the original flow direction (ϕ3 = 0o ), the flow
conditions here (domain Â) are the same as those in domain À(explain why?). To calculate the
shock shape take the shoulder (O) as the origin of (x, y) coordinate system. The v distribution
in the expansion fan then reads v = xy .
dy
The shock separating domains Áand Âis straight and has a slope dx = 21 (v2 + v3 ) = 0.454.
It hits the last characteristic (v = −0.111) in the point Q (3.161, −0.351). From this point a
curved shock runs downstream. It is governed by the O.D.E.
   
dy 1 x 1 x
= + v3 = + 1.020 ,
dx 2 y 2 y
and it has to fulfill the boundary condition that it runs through the point Q.
The shock shape that satisfies all these requirements has the form:
√
ys = A x + Bx, (x > xQ ),
with A = −2.011, B = v3 = 1.020.
The shock strength becomes
y
∆ = vpost − vpre = v3 − ,
x
210CHAPTER 9. THE ’FIRST NON LINEAR’ DESCRIPTION OF STEADY TWO DIMENSIONAL SUPER

Figure 9.11: Flow downhill of a backward facing slope

or
A
∆ = −√ .
x

The shock decays asymptotically to zero strength as √1 .

x
Chapter 10

Steady one-dimensional viscous flow

10.1 Channel flow: assumptions and basic equations

If a gas travels with high speed through a channel of sufficient length the effects of viscosity and
associated entropy change can not be neglected. We attempt to study these flows and suppose
that the channel has a variable area A(x), that the fluid is a perfect gas with constant specific
heats and that the walls of the channel are perfectly insulated (adiabatic wall conditions). The
flow will be treated as one-dimensional and friction in the boundary layer will be accounted
for by assuming a tangential stress τ acting on the circumference, and which is modelled by
1
τ = f ρv 2 .
2
Here f is a friction coefficient, depending mainly on the Reynolds number Re. Experiments
from Keenan and Neumann 1 teach us that for pipes having a length/diameter ratio 10 <
L/D < 50 and 25.000 < ReD < 700.000 that the friction coefficient takes the values:

faverage ≈ 0.003 − 0.0065 incompressible flow

faverage ≈ 0.002 − 0.003 compressible flow onto M = 3

Regarding turbulent incompressible flows Nikuradse has presented the implicit relation:
1 p
√ = −0.8 + 2 log(Re 4f ).
4f

In this chapter f is taken constant along the channel. Finally it is assumed that internal heat
sources (combustion, chemical reaction, condensation etc.) are absent.
Gas flows satisfying the conditions just spelled out are governed by the following set of
equations that appear when the conservation laws for mass, momentum and energy are applied
onto the control volume V as shown in figure 10.1.

Continuity:
∂ y {
ρ dV + ρv ds = 0. (10.1)
∂t
V ∂V
1
Keenan, J.H. and Neumann: Friction in pipes at supersonic velocities, NACA TN 963, 1945.

211
212 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Figure 10.1: Control volume in quasi one-dimensional flow

For steady one-dimensional flow this yields:

{
ρv ds = ρ2 v2 A2 − ρ1 v1 A1 = 0. (10.2)
∂V

The integral form of the continuity equation then becomes

ρvA = constant (10.3)

or in differential form
dρ dv dA
+ + = 0. (10.4)
ρ v A

Momentum:
The momentum equation without body force terms is:
∂ y { { {
ρv dV + v(ρv ds) = − p ds + τ̄¯ ds̄. (10.5)
∂t
V ∂V ∂V ∂V

For a steady one-dimensional flow between the stations 1 and 2 this reduces to
Z 2 Z 2
ρvA(v2 − v1 ) = p1 A1 − p2 A2 + p dA − τ πDh dx, (10.6)
1 1

where Dh is the hydraulic diameter of the channel (πDh is the circumference of the channel
cross-section). v
Apparently the one-dimensional form of the viscous term ∂V τ̄¯ ds̄ in equation (10.6) is
written as
{ Z 2
τ̄¯ ds̄ = − τ πDh dx. (10.7)
∂V 1

To understand this expression consider a small element with length dx of the control volume,
see figure 10.2. The fluid inside this control volume experiences a friction force from the wall
of the channel along its circumference.
10.1. CHANNEL FLOW: ASSUMPTIONS AND BASIC EQUATIONS 213

Figure 10.2: Friction force on a small fluid element

Let us write the circumference as πDh with Dh as the hydraulic diameter. The friction
force (in flow direction) on the fluid element is
dFvx = −τ πDh ds cos α
= −τ πDh dx.
Integrating from station 1 to station 2 yields equation (10.7).
The differential form of the momentum equation for one-dimensional viscous flow follows
from (10.6) by taking stations 1 and 2 a distance dx apart; this results into
ρvA du = −d(pA) + pdA − τ πDh dx,
or
ρvAdv + Adp − τ πDh dx = 0. (10.8)

Energy equation
Recall the energy equation without body forces:
∂ y { { {
ρE dV + ρE(v̄ · ds̄) = − p(v̄ · ds̄) + (τ̄¯ · v̄) · ds̄
∂t
V ∂V ∂V ∂V
| {z } | {z }
work done by pressure forces work done by viscous forces
y { (10.9)
+ ρc dV − q̄ · ds̄ .
|V {z } ∂V
| {z }
volumetric heating conduction

For a steady one-dimensional adiabatic flow between the stations 1 and 2 this reduces to
p2 p1
ρvA(E2 − E1 ) = −ρvA( − ),
ρ2 ρ1
or
p p v2 v2
E+ =e+ + =h+ = H = constant. (10.10)
ρ ρ 2 2
214 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

In differential form with dh = cp dT (perfect gas) the energy equation reads

cp dT + v dv = 0. (10.11)

Equations (10.4), (10.8) and (10.11) form the basic set of equations that govern one-dimensional
steady adiabatic flows.

Assume a variable area channel whose cross-sectional area A(x) is a known function of x.
So we have three equations for the unknowns v(x), ρ(x), p(x) and T (x). To close the system,
we will use the perfect gas law in differential form:

dp dρ dT
= + . (10.12)
p ρ T

To solve the system of equations the variables ρ(x), p(x) and T (x) will be eliminated so that
a single equation for the velocity v(x) or the Mach number M (x) results. First however some
algebra is performed to attain a more suitable form of the system. Execute the following
operations:

• Write cross-sectional area A(x) in the momentum

p equation and in the continuity equa-
tion in terms of the hydraulic diameter Dh = 4A(x)/π.

• Divide the momentum equation (10.8) by the cross-sectional momentum ρv 2 A and sub-
stitute the friction coefficient f from equation (10.1).

• Divide the energy equation (10.11) by the enthalpy cp T and remember that v 2 /cp T =
(γ − 1)M 2 .

Then the following system of equations results:

dρ dv 2dDh
continuity: + + = 0, (10.13a)
ρ v Dh
 
dp dv 2f
momentum: + γM 2 + dx = 0, (10.13b)
p v Dh
dT dv
energy: + (γ − 1)M 2 = 0, (10.13c)
T v
dρ dT dp
perfect gas law: + = (10.13d)
ρ T p

In order to solve this system a relation between the Mach number M and velocity v is still
needed. For the specific case of a perfect gas the energy equation (10.9) reduces to

a2 v2
+ = constant,
γ−1 2

which gives the following relation connecting the Mach number and velocity:

dv 1 dM
=  . (10.14)
v 1 + 2 M2 M
γ−1
10.2. CONSTANT-AREA CHANNEL FLOW; FANNO LINE 215

Using the set of equations (10.13a)–(10.13d), the variables dp/p, dρ/ρ and dT /T may be
eliminated and finally we obtain

dM 2M (1 + γ−1 2 0 2
2 M )(Dh − γf M )
= , (10.15)
dx Dh (M 2 − 1)

where Dh0 = dD dx . For a known channel shape Dh (x), (10.15) may be solved to get the
h

corresponding mach number distribution along the channel. Equation (10.15) will now be
analysed to understand the physics of one-dimensional viscous flow in a variable area channel.
First we address the special case of flow through a constant area channel.

10.2 Constant-area channel flow; Fanno line

Consider a constant-area channel having its entrance at x = 0 and having a certain length L.
The entrance conditions are specified as M0 , v0 and h0 . The Mach number distribution along
the channel is governed by the O.D.E.:

dM 2M 3 (1 + γ−1 2
2 M )γf
=− . (10.16)
dx Dh (M 2 − 1)

This equation tells already the important but also paradoxical message that friction (f 6= 0)
causes a Mach numbert distribution in a constant area channel.
Before analyzing equation (10.16) further some interesting properties of the flow can be
discovered already by considering the integral forms of the continuity equation and the energy
equation e.g.

J = ρv = constant,

and
v2
H =h+ = constant.
2
Combining both equations yields
 2
1 1
h + J2 = H, (10.17)
2 ρ

J being the mass flux through the channel. Equation (10.17) is an algebraic relation between
the thermodynamic variables h (enthalpy) and ρ1 (specific volume). This relation defines a
family of curves (a particular curve depends on the choice of J and H) in the (h, ρ1 )-plane, see
figure (10.3). These parabolic curves are well known as Fanno lines.
The entrance conditions v0 , h0 and rho0 fix the particular Fanno curve that describe all
possible states (h, ρ1 ) in the channel flow. Since we have assumed adiabatic flow the second law
states that the entropy cannot decrease following a particle; thus a gas that moves in positive
x-direction obeys the entropy condition:

ds
entropy condition: ≥ 0. (10.18)
dx
216 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Figure 10.3: Fanno line in the (h, ρ1 )-plane

Following a fluid particle there can oly be an increase of entropy. So it seems a good idea to
study the graph of entropy level curves in the (h, ρ1 )-plane. From the entropy formula:
 γ−1
T ρ0
= es−s0 /cv ,
T0 ρ

it follows that iso-entropy lines in the (h, ρ1 )-plane have the form
 γ−1
1
h = constant.
ρ
The qualitative behaviour of these lines are shown in figure 10.3. For a particular value of
the entropy s = s∗ the iso-entropy curve just touches the Fanno line in point c. An iso-
entropy curve having s < s∗ intersects the Fanno line twice, in fact in the the points a and b.
Iso-entropy curves with s > s∗ nowhere intersect the Fanno line.
The entropy condition (10.18) implies that in a constant area flow the Fanno line is followed
in a way corresponding to the arrows in figure 10.3 when following a particle. This means
that irrespective the particular choice of the entrance conditions the channel develops in such
a way that it tends to attain the flow conditions of point c (h = h∗ , ρ = ρ∗ ) where entropy
becomes as large as possible (s = s∗ ). It will be shown that in this point the Mach number
is unity (M = 1). To that end observe that at point c the Fanno curve and the iso-entropy
curve have the same slope.
The slope of the Fanno curve is
!  
dh 2 1
= −J ,
d ρ1 ρ
Fanno

the slope of an iso-entropy curve is

!
∂h
= −(γ − 1)ρh = −ρa2 ,
∂ ρ1
s
10.2. CONSTANT-AREA CHANNEL FLOW; FANNO LINE 217

equating both yields

J 2 = ρ2 a2
or
v 2 = a2
which shows that M = 1. Thus in point c the flow becomes just sonic. ‘Above’ c the flow
is subsonic (M < 1) because the density increases and consequently (J = constant) the
velocity decreases. In a similar way it may be concluded that ‘below’ c the flow is supersonic
(M > 1). The two distinct branches of the Fanno line thus represent subsonic flow (h > h∗ )
and supersonic flow (h < h∗ ). In the absence of discontinuities there is no possibility of passing
from one branch to another because that would imply crossing point c and the condition of
ds/dx > 0 would be violated. Thus an initially supersonic flow remains supersonic but slows
down to M = 1; an initially subsonic flow remains subsonic but accelerates to M = 1. The
effect of friction is that it drives the Mach number to unity. This is an interesting conclusion
having the consequence that an initially subsonic flow defines a particular channel length Lmax
at which exit the sonic conditions are just attained: this phenomenon is referred to as fricional
choking; it will be discussed later.
Let us return to the O.D.E. (10.16), which governs the Mach number distribution along
the channel and let us construct a qualitative sketch of the integral curves in the (M, x)-plane.
The friction coefficient f is a positive quantity, so from (10.16) we conclude:
dM
> 0, for M < 1,
dx
dM
= ∞, for M = 1,
dx
dM
< 0, for M > 1.
dx
The graph dMdx as a function of M , shown in figure 10.4, illustrates a monotonic increase
of dM
dx if the flow is subsonic. In supersonic flow dM
dx attains a negative maximum value.
The corresponding particular Mach number satisfies the relation:
3(γ − 1)M 4 − (5γ − 7)M 2 − 6 = 0;
√
for γ = 7/5 this particular Mach number takes the value M = 4 5.
Based on the graph dMdx as a function of M the integral curves in the (M, x)-plane have a
pattern as is shown in figure 10.5.
A particular curve depends on the choice of the entrance Mach number M0 . The arrows
indicate the direction in which an integral curve is followed when following a gas particle.
Obviously the Mach number goes to unity. With every entrance Mach number M0 there
corresponds a maximum tube length Lmax beyond which choking occurs. From figure 10.5 we
expect a zero tube length when M0 = 1.
Equation (10.16) will now be integrated from the entrance x = 0 with entrance Mach
number M0 to an arbitrary section x with Mach number M (x) in order to get an expression
for the maximum tube length. Integration yields:
 
2 2 M 2 1 + γ−1 M 2 (x)
4γf M (x) − M0 γ+1 0 2
x= 2 2 + ln  . (10.19)
Dh M (x)M0 2 M 2 (x) 1 + γ−1 M 2
2 0
218 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

dM
Figure 10.4: Graph of dx vs M .

Figure 10.5: Mach number distribution in a constant area tube.

The maximum tube length occurs at x = Lmax with M (x) = 1, so there is obtained
4γf 1 − M02 γ + 1 (γ + 1)M02
Lmax = + ln . (10.20)
Dh M02 2 2(1 + γ−1 2
2 M0 )

The maximum tube length depends on the entrance Mach number M0 . Evidently for M0 = 1
10.3. A CHANNEL WITH A VARIABLE CROSS-SECTIONAL AREA 219

this length is zero; for M0 → 0, Lmax → ∞ but M0 → ∞, Lmax tends to the limiting value
(see figure 10.6):

Lmax (γ + 1) ln γ+1
γ−1 − 2
= .
Dh 4γf

For γ = 1.4 and f = 0.003 we find Lmax ≈ 68.5Dh . The phenomena that a maximum tupe

Figure 10.6: Maximum tube length.

length exists for a specified entrance Mach number is usually referred to as frictional choking.
Friction reduces the effective cross-sectional area of the tube and drives the Mach number
towards unity. Let us now conclude now by inviting the reader to comment on the interesting
question: “What will happen if a uniform free stream with Mach number M0 enters a tube
with length L where L exceeds the maximum tube length Lmax (M0 ?” Distinguish the two
cases M0 > 1 and M0 < 1 and be aware of the possibility that in the supersonic case shocks
may appear.

10.3 A channel with a variable cross-sectional area

The Mach number distribution in a variable area channel is governed by the O.D.E.:

dM 2M (1 + γ−1 2 0 2
2 M )(Dh − γf M )
= , (10.15)
dx Dh (M 2 − 1)

where the channel shape is defined by the variation of its hydraulic diamter

Dh = F (x). (10.21)

To study the properties of the solution curves M (x), equation (10.15) will be viewed as a
nonlinear system of two ordinary differential equations having the formal shape:

dM
= P (M, x), (10.22a)
dλ
dx
= Q(M, x), (10.22b)
dλ
220 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

where functions P (M, x) and Q(M, x) are defined as

γ−1 2
P (M, x) = 2M (1 + M )(F 0 (x) − γf M 2 ),
2
Q(M, x) = F (x)(M 2 − 1),

and where λ is a parameter running along solution curves. System (10.22a),(10.22b) is nonlin-
ear because its right-hand side depends nonlinearly on the variables M and x. The variables
M and x are seen as depending on λ, the (M, x)-plane is referred to as the phase plane of the
system.
Standard theory regarding nonlinear systems2 will be applied to get the qualitative be-
haviour of solution curves (integral curves) in the phase plane.
Integral curves in the phase plane have a slope
dM P (M, x)
= . (10.23)
dx Q(M, x)

Obviously these slopes are uniquely defined if P (M, x) 6= 0 or Q(M, x) 6= 0. However if we

have P = 0 and Q = 0 at a common point this point becomes singular in the sense that the
slope of the integral curves is not defined there. System (10.22a)-(10.22b) features singular
points if:

M = 1, and F 0 (x) = γf. (10.24)

Let this happen in the phase plane at x = x∗ . Since γ and f are both positive, singular points
only appear in the diverging parts of a channel, where F 0 > 0. To analyse integral curves near
the singularity, it is standard to linearize the system near (M, x) = (1, x∗ ) where x∗ satisfies
equation (10.24). The linearized system that follows reads:
 dM   ∂P ∂P   
dλ ∂M ∂x M −1
= (10.25)
dx
dλ
∂Q
∂M
∂Q
∂x
x − x∗

where the matrix has to be evaluated at M = 1 and x = x∗ ; this yields:

−2γ(γ + 1)f (γ + 1)F 00

   
PM Px
= . (10.26)
QM Qx M =1 2F 0
x=x∗

Important parameters for further analysis are the trace (T ) and the Jacobian (J) of the
matrix. Here they take the form

T = −2γ(γ + 1)f,
(10.27)
J = −2(γ + 1)F F 00

Converging-diverging nozzle
Let us apply the theory of the specific situation of a converging-diverging nozzle. The shape
of the contour and the corresponding phase plane (with solution curves M = M (x)) is shown
in figure 10.7.
2
See for example: Jordan, D.W. & P. Smith, Nonlinear Ordinary Differential equations, Clarendon Press,
Oxford 1979
10.3. A CHANNEL WITH A VARIABLE CROSS-SECTIONAL AREA 221

Figure 10.7: Viscous effects near the throat of a nozzle

Behind the throat in the diverging part the condition F 0 (x) = γf is fulfilled at x = x∗ .
The area behind the throat is increasing with x so that at x = x∗ also F 00 (x∗ ) > 0. Thus at
x = x∗ the Jacobian is negative and the solution curves form a saddle-point there, the two
separatices of the saddle, labelled in figure 10.7 as (1) and (2), have slopes determined by
p
−γ(γ + 1)f ± γ 2 (γ + 1)2 f 2 − 2(γ + 1)F F 00
 
dM
= . (10.28)
dx 1,2 2F

Separatrix (1) has a positive slope and separatrix (2) has a negative slope.
The transition from subsonic to supersonic flow or vice versa is made possible by the
presence of the saddle point and takes place downstream of the throat in the divergent part
of the channel. Observe the contrast with inviscid flow where transitin always happens in
the throat. Due to viscous effects a subsonic flow in the throat section is accelerated and
a supersonic flow is slowed down. The positive slope dM to zero if F 00 (x∗ ) → 0


dx 1 tends
and the negative slope dM ∗


dx 2 tends to −γ(γ + 1)f /F (x ) which means that the transition
from subsonic to supersonic flow can be made as slow as we please but the transition from
supersonic to subsonic flow has a limiting value.
From figure 10.7 we observe that an increase of entrance Mach number in the subsonic
domain moves the position of the highest Mach number from the throat downstream towards
the sonic section. Similarly a decrease of a supersonic entrance Mach number moves the
position of the minimum Mach number upstream onto the sonic section. Entrance Mach
numbers defining integral curves in the shaded region will terminate at M = 1. A sonic section
222 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

upstream of x∗ is predicted but transition through sonic conditions is not possible. The nozzle
is choked and the corresponding range of entrance Mach numbers cannot be realized unless
the nozzle is terminated before the sonic section or a shock (only for a supersonic entrance)
appears.
Because x∗ lies downstream of the throat frictional choking is possible even in the diverging
part of the nozzle if the influence of friction dominates that of the area increase.

Converging-diverging nozzle with inflexion point

Usually a supersonic wind tunnel is equipped with a converging-diverging nozzle having an

inflexion point (F 00 = 0) downstream of the throat. The shape of such a contour together with
the corresponding phase plane portrait is shown in figure 10.8.

Figure 10.8: Typical nozzle contour with a corresponding phase plane

Assume a nozzle contour with a point of inflexion between locations x∗1 and x∗2 where the
condition F 0 (x) = γf holds. At x∗1 and x∗2 singular points occur in the phase plane on the
line M = 1. The presence of an inflexion point between x∗1 and x∗2 implies F (x∗1 ) > 0 and
F (x∗2 ) < 0. Using equation (10.27) the Jacobians in these points have opposite signs, i.e.
J(x∗1 ) < 0 and J(x∗2 ) > 0. So the singular point at x = x∗1 is a saddle point. Standard theory
about nonlinear systems of differential equations teaches us that J > 0 represents either a
node if 4J − T 2 < 0 or a spiral (focus) if 4J − T 2 > 0. Evaluating equation (10.27) in x = x∗2 ,
10.3. A CHANNEL WITH A VARIABLE CROSS-SECTIONAL AREA 223

M = 1 yields

(4J − T 2 )x∗2 ,M =1 = (γ + 1) 2F F 00 + (γ + 1)(F 0 )2 .


(10.29)

Let us discuss here in more detail the spiral case (the nodal case is left as an excercise), thus
we consider

−γ + 1 0 2
F F 00 < (F )
2

in x∗2 . Figure 10.8 already gives a qualitative picture of some integral curves. Observe that
all integral curves have a vertical slope at the line M = 1 and a horizontal slope on the curve
F 0 (x) = γf M 2 . This curve passes through the singularies (x∗1 , 1) and (x∗2 , 2) and must have
a maximum somewhere between x∗1 and x∗2 . Because the trace T in x∗2 , M = 1 is negative
the spiral singularity is asymptotically stable which means that trajectories spiral into the
singularity if the parameter λ increases. But if the spiral is stable then separatrix S1 runs into
the spiral whereas separatrix S2 bypasses the spiral singularity.
Set aside the local details near the singularites the trajectory pattern looks very similar
one that corresponds to a constant area channel (shown in figure 10.5). For every entrance
Mach number the flow in the variable area channel is driven to sonic conditions. So for every
entrance Mach numner there corresponds a maximum tube length Lmax . We do not attempt
to find an analytic relation between Lmax and M0 but we perform a qualitative research; the
result is shown in figure 10.9.

Figure 10.9: Maximum tube length for a converging-diverging nozzle with an inflexion point

For increasing subsonic entrance Mach number M0 < M1 the Lmax decreases gradually.
At M0 = M1 Lmax jumps from the large value L3 to the much smaller length L1 . For
M1 < M0 ≤ 1 the Lmax decreases to zero at M0 = 1. For supersonic entrance Mach numbers
there is a gradual increase of Lmax from zero to L2 if M0 increases to M2 . At M0 = M2 , Lmax
jumps from the short value L1 to a larger value L2 < L3 . Beyond M0 = M2 , Lmax increases
continuously with increasing M0 .
224 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

10.4 Internal structure of a shock wave

10.4.1 Introductory comments
The thickness of a shock wave refers to the spatial region in the flow field where the variation of
flow properties is irreversible and where entropy increases. In most practical cases this region
is very thin (comparable to the mean free path of molecules) causing high spatial gradients
of temperature, pressure, density and velocity. In many applications the shock thickness is
neglected and shocks are modelled as a true discontinuity. But doing so the velocity and
temperature gradients become infinite causing infinite viscous stress and infinite heat flux.
This ambiguity is resolved by introducing a shock or shock layer with a finite thickness.
Notice the analogue with the boundary layer, both feature thin regions of large viscous stress
and high heat flux.
The Navier-Stokes model will be applied to resolve the internal structure of a normal shock
wave. Therefore we assume a perfect one-dimensional flow with a flow direction normal to
the shock wave. Let this direction be the x-direction. Further let us assume the shock to be
stationary, so we expect that acceleration is not of prime importance; nevertheless a detailed
study on the structure of accelerating (or decelerating) shocks would be welcome. Finally,
thermal radiation and diffusion are not taken into account.

10.4.2 Navier-Stokes equations for one-dimensional flow

To derive the Navier-Stokes equations in differential form for one-dimensional flow let us
recall the conservation laws (in integral form) from chapter 1. The influence of body forces
and volumetric heating is not taken into account so the terms accounting for these influences
are dropped.

∂ y {
Continuity: ρ dV + ρv̄ · n̄ ds = 0,
∂t
V ∂V
∂ y { { {
Momentum: ρv̄ dV + ρv̄ v̄ · n̄ ds + pn̄ ds = τ̄¯ · n̄ ds,
∂t
V ∂V ∂V ∂V
∂ y { {
Energy: ρE dV + ρEv̄ · n̄ ds + pv̄ · n̄ ds =
∂t
{ V {∂V ∂V

(τ̄¯ · v̄) n̄ ds − q̄ · n̄ ds
∂V ∂V

Here τ̄¯ is the viscous stress tensor and q̄ is the heat flux due to conduction.
For a Newtonian fluid:
 
∂ui ∂uj
τij = λ∇ · v̄ + µ + .
∂xj ∂xi
Fourier’s law:

q̄ = −k∇T.

The divergence theorem and the gradient theorem are applied to tranfer surface integrals to
volume integrals. Combining the volume integrals and recalling that V is fixed in space but
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 225

otherwise arbitrarily chosen the integrand has to be zero for all points in space. This yields
the differential form:
∂ρ
+ ∇ · ρv̄ = 0, (10.30)
∂t
∂ρv̄
+ ∇ · ρv̄v̄ + ∇p = ∇τ̄¯, (10.31)
∂t
∂
(ρE) + ∇ · ρv̄E = −∇ · (pv̄) + ∇(τ̄¯ · v̄) − ∇ · q̄. (10.32)
∂t
For a one-dimensional steady flow this reduces to

∂
(ρu) = 0,
∂x
∂u ∂p ∂
ρu + = τxx ,
∂x ∂x ∂x  
∂E ∂ ∂T ∂T ∂
ρu + (pu) = k + (uτxx ),
∂x ∂x ∂x ∂x ∂x

with τxx = λ ∂u ∂u
∂x + 2µ ∂x the set of equations becomes

∂
(ρu) = 0, (10.33)
∂x  
∂u ∂p ∂ ∂u
ρu =− + (λ + 2µ) , (10.34)
∂x ∂x ∂x ∂x
   
∂H ∂ ∂u ∂ ∂T
ρu = (λ + 2µ)u + k . (10.35)
∂x ∂x ∂x ∂x ∂x

10.4.3 Equations of shock structure

Due to the one-dimensionality of the problem (there are only variations in x-direction), the
effects of viscosity come from the normal viscous stress τxx and from heat conduction in
x-direction. Shear stresses are absent here.
The viscous coefficients λ and µ appear in the combination û = λ + 2µ; taking the Stokes’
hypotheses for granted (λ = − 23 û) we assume that λ + 2µ = 43 µ. The set of equations can be
integrated once giving the shock structure:

J = ρu, (10.36)
du
K = p + ρu2 − (λ + 2µ) , (10.37)
dx
du dT
L = ρuH − (λ + 2µ)u −k , (10.38)
dx dx
where J, K and L are integration constants representing the mass flux, momentum flux and
energy respectively. Note that the viscosity coefficients λ and µ have not been assumed to be
constant. Assuming flow to the right the upstream and downstream conditions are:

x = −∞ : u = u1 , p = p1 , ρ = ρ1 , T = T1 , µ̂ = µ̂1 ,
x = +∞ : u = u2 , p = p2 , ρ = ρ2 , T = T2 , µ̂ = µ̂2 .
226 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Far upstream (x → −∞) and far downstream (x → +∞) the derivatives ∂u ∂T

∂x and ∂x vanish.
The system (10.36)–(10.38) reduces to the well-known Rankine-Hugoniot equations
ρ1 u1 = ρ2 u2
p1 + ρ1 u21 = p2 + ρ2 u22
1 1
H1 = h1 + u21 = h2 u22 = H2
2 2
describing the shock discontinuity. The Rankine-Hugoniot equations are invariant under the
transformation u1 = −u1 and u2 = −u2 ; the Rankine-Hugoniot equations are not decisive
regarding the direction of the flow. Expansion shocks and compression shocks are both allowed
by the Rankine-Hugoniot conditions.

10.4.4 An estimate for the shock-thickness

The momentum equation (10.37) contains only the derivative du dx and this fact enables us to
make an estimate of the shock thickness. In the interior of the shock the velocity u decreases
from u1 (pre-shock value) to u2 (post-shock value), see figure 10.10.

Figure 10.10: Internal velocity distribution

The differential equation that describes this behaviour reads

du
(µ + 2λ) = p + ρu2 − K.
dx
Let K be evaluated at x = −∞, this yields
K = p1 + ρ1 u21 ,
and the O.D.E. governing the velocity distribution in the shock reads
du
(µ2λ) = (p − p1 ) + J (u − u1 ) . (10.39)
dx
The O.D.E. cannot be solved yet because the pressure p(x) is still unknown. In order to get an
estimate of the shock thickness we evaluate the derivative du
dx at the midpoint of the velocity
distribution. At the midpoint (subscript m) the velocity has the value:
1 1
um = u1 + (u2 − u1 ) = u1 + ∆u. (10.40)
2 2
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 227

The shock thickness δ is defined as, see figure 10.10:

 
du (u2 − u1 ) ∆u
=+ =+ . (10.41)
dx m δ δ
The pressure at the midpoint is obtained by performing a Taylor expansion of p = p(ρ, s)

(ρ − ρ1 )2
   2   
∂p ∂ p ∂p
p = p1 + (ρ − ρ1 ) + 2
+ ··· + (s − s1 ) + . . .
∂ρ s ∂ρ s 2 ∂s ρ

For weak shocks (s − s1 ) is at least of the order (ρ − ρ1 )3 and if the analysis is restricted to
second order terms (ρ − ρ1 )2 , the pressure can be calculated as

a21 (ρ − ρ1 )2
p = p1 + a21 (ρ − ρ1 ) + (γ − 1) .
ρ1 2
Using equation (10.36) in the form ρ1 u1 = ρm um = J the density ρm at the midpoint is
related to the velocity um as
ρ1 ∆u
ρm − ρ1 = − . (10.42)
2u1 + ∆u
The pressure at the midpoint is
2
ρ1 a21

(γ + 1) ∆u
∆u + ρ1 a21 + O ∆u3 .


pm − p1 = − (10.43)
2u1 2 2u1
Evaluating equation (10.39) in the midpoint for the shock thickness δ:
(µ + 2λ) ρ1 a21 (γ + 1) (∆u) ρ1 u21
=− + ρ1 a21 + ,
δ 2u1 2 4u21 2u1
or
 
(µ + 2λ) 2 1 γ+1 ∆u
2 = ρ1 a1 M1 − + ρ1 a1 , (10.44)
δ M1 4 a1 M12
u1
where M1 is the Mach number M1 = a1 . The velocity jump ∆u depends on the shock Mach
number according to
2 M12 − 1


∆u
=− .
a1 (γ + 1)M1
If this relation is inserted into equation (10.44) the shock thickness δ follows from
M2 − 1
 
2 (µ + 2λ) 1
= ρ1 a1 M1 − − 1 3 .
δ M1 2M1
Taking the weak shock limit (M1 → 1), the estimate for the shock thickness becomes:
2 (µ + 2λ)
δ= .
ρ1 a1 (M1 − 1)
Notice that the shock thickness becomes large if the Mach number approaches M = 1; the
weaker the shock the larger its thickness.
228 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

10.4.5 Entropy production in the shock zone

From inviscid theory its known that passing a shock is a non-isentropic process; entropy
increases across a shock discontinuity.
Let us now take the viscous model of the shock to study the entropy production in the shock
zone. From the first law of thermodynamics

dp
T ds = dh − ,
ρ
we write
ds dH du dp
ρuT = ρu − ρu2 −u .
dx dx dx dx
dp
Inserting dx and ρu dH
dx from equations (10.34) and (10.35) and introducing µ̂ = µ + 2λ we
find
      
ds 1 d du d dT d du
ρu = µ̂u + k −u µ̂ . (10.45)
dx T dx dx dx dx dx dx

The first and second term in the right hand side may be worked out as follows

d du d du 2
µ̂ du




dx µ̂u dx
= u dx dx
+ µ̂ dx
,
1 d dT d k du k dT 2
T dx k dx = dx T dx + T 2 dx ,

and the expression for the entropy gradient becomes

 2  2  
ds µ̂ du k dT d k dT
ρu = + 2 + . (10.46)
dx T dx T dx dx T dx

The first two terms in the R.H.S. are always positive, they will increase the entropy when
moving downstream. The last term vanishes at the boundaries x = −∞ and x = +∞;
integration w.r.t. x yields
Z 2 "  2   #
µ̂ du k dT 2
ρu (s2 − s1 ) = + 2 dx, (10.47)
1 T dx T dx

which clearly shows that passing a shock in down stream direction increases the entropy.

10.4.6 More about the shock structure

The internal structure of the shock region will now be discussed in more detail.
The equations (10.36), (10.37) and (10.38):

J = ρu,

du
K = p + ρu2 − µ̂ ,
dx
du dT
L = ρuH − µ̂u −k ,
dx dx
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 229

are the starting point for such a discussion.

They have to be integrated to get the variation of u(x) and T (x) in the shock zone between the
upstream state À(u = u1 , T = T1 ) and the downstream state Á(u = u2 , T = T2 ) where du dx = 0
and dT
dx = 0.
Using ρu = J = constant, the equations (10.37) and (10.38) may be written in the form:
du JRT
µ̂ = −K + Ju + , (10.48)
dx u
 
dT 1 2
k = −L + Ku + J cv T − u , (10.49)
dx 2
a set of O.D.E.’s with u(x) and T (x) as the two unknowns and J, K and L as constants. The
ratios K/J and L/J may be interpreted in terms of the variables defining state À. From:
 
2 RT1
K = p1 + ρ1 u1 = ρ1 RT1 + Ju1 = J + u1 ,
u1

u21
 
L = ρ1 u1 H1 = J cp T1 + ,
2
is obtained that
RT1 + u21 u21
K/J = , L/J = cp T1 + .
u1 2
u T
Introduce the scaled variables ũ = u1 , θ= T1 , into equations (10.48) and (10.49), this results
into the system:
µ̂ dũ
ũ = ũ (ũ − 1) + α (θ − ũ) , (10.50)
J dx
and
kα dθ α 1
= (θ − 1) + α (ũ − 1) − (ũ − 1)2 , (10.51)
JR dx γ−1 2
where α is a parameter defined as:

−1
α = RT1 /u21 = γM12 . (10.52)

The equations (10.50) and (10.51) form a dynamical system for the variables ũ(x) and θ(x).
The system is autonomous because the independent variable x is not explicitly present. The
various solutions may be viewed as integral parths in the phase plane (ũ, θ). For the shock
structure problem a solution curve has to be sought that starts in the point (ũ, θ) = (1, 1)
where dũ dθ
dx and dx have to be zero and indeed this conditions is fulfilled by the equations (10.50)
and (10.51).
Our next question will be: starting from uniform state À: (ũ, θ)1 = (1, 1) is it possible that
the flow may change onto another uniform state Á: (ũ, θ)2 = (ũ2 , θ2 ) which is different from
state À?
To find such a uniform state we take dũ dθ
dx = 0 and dx = 0 in equations (10.50) and (10.51) and
have to solve:

ũ (ũ − 1) + α (θ − ũ) = 0, (10.53)

230 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

α 1
(θ − 1) + α (ũ − 1) − (ũ − 1)2 = 0. (10.54)
γ−1 2
There are two solutions: ũ = 1 and ũ = ũ2 ; ũ = 1 represents the original flow (state À) and
ũ = ũ2 (state Á) with
γ − 1 + 2γα
ũ2 = , (10.55)
γ+1
represents a normal shock. Equation (10.55) is equivalent with the well known normal shock
relation
u2 (γ − 1) M12 + 2
= .
u1 (γ + 1) M12
In the (ũ, θ) phase plane a solution curve has to be found which runs from state À: (ũ, θ) =
(1, 1) to state Á: (ũ, θ) = (ũ2 , θ2 ).
To see whether such a solution curve indeed exists let us do a plane analysis of equations
(10.50) and (10.51). To obtain a qualitative impression of the phase portrait the singular
points of equations (10.50) and (10.51) are considered.
They appear at:

ũ = 1, θ = 1 and ũ = ũ1 , θ = θ2 .

Notice that they represent the uniform states Àand Ábetween which the solution for the shock
structure problem has to be sought. Since state Áis a post-shock state we know that ũ2 < 1
and θ2 > 1, implying that point 2 lies above and left from point 1 in the (ũ, θ)-plane; see
figure 10.11.
The points 1 and 2 have Jacobians

Figure 10.11: Phase plane with some integral curves

kα2 µ̂R kα2 µ̂Rũ2

J1 = (1 − γα) , J2 = (γα − 1)
γ−1 γ−1
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 231

respectively.
Since 0 < γα < 1 it appears that the uniform state Àis represented by a node and that
uniform state Áis represented by a saddle in the phase plane.
dθ
The iso-lines dũ = 0, ∞ are parabolas:

2
dθ
dũ = 0 : θ = γ−1
2α (ũ − 1) − (γ − 1) (ũ − 1) + 1 ,
2
dθ
dũ = ∞ : θ = − ũα + 1+α
α ũ .

They are depicted in figure 10.11 which also shows some integral curves emanating from the
singularity at ũ = 1, θ = 1. The phase portrait as it is built up with integral curves reveals
that an infinite number of integral curves start from uniform state Àbut that only one of them
terminates at the uniform state Á. So there is just one unique solution describing the shock
structure between pre-state Àand post-state Á.
To find this unique solution analytically we have to solve equations (10.50) and (10.51) together
with the boundary conditions ũ = 1, θ = 1 and ũ = ũ2 , θ = θ2 . This problem is hard to solve
and a solution for the general problem where µ̂ and k are still functions of the temperature is
still not available. However various attempts have led to some interesting solutions valid for
special cases.Two examples of special solutions: the Taylor solution and the Becker solution
will be introduced (briefly) in the next section (10.4.7). More interesting details about the
shock structure problem may be found in the excellent text: Von Mises (1958)3 ch. III, art. 11.

When an attempt is made to construct a numerical approximation of the shock structure

it is very likely that the numerical iteration procedure does not converge if the pre-shock state
Àis taken as starting point of the numerical process.
The nodal character of state Àimplies that an infinite number of integral curves emanate from
the starting point and all these curves diverge when they leave the neighbourhood of state À.
This divergent behaviour causes a growing deviation of the numerical path with respect to
the exact path which connects states Àand Á.
The saddle character of state Álet the numerical path bypass state Áso that the boundary con-
dition which states that the integral path should terminate in state Áis not (approximately)
fulfilled by the numerical computation.
A careful look at figure 10.11 suggests that a smooth and converged solution is expected if
the post-shock state Áis taken as starting point in the numerical process. The integral path
followed by the numerical marching procedure converges to the exact integral curve because
the integral curves converge in marching direction so that small deviations between the com-
putational and exact path are surpressed naturally.
Figure 10.12 shows a typical curve for the shock structure region obtained by numerical cal-
µc
culation for M1 = 2.0, γ = 1.4 and P r = kp = 34 . The dependence of the viscosity with
temperature is modelled with the power-law assumption µ = µ1 (θ)0.76 .

10.4.7 Special solutions for the shock structure problem

Two special solutions of equations (10.50) and (10.51) may attract our attention. The first
one is discussed by Taylor who took the assumption of zero heat conduction.

3
Richard von Mises (1958), Mathematical Theory of Compressible Fluid Flow, Academic Press, Inc.
232 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Figure 10.12: Integral curves for Mach number M1 = 2.0, γ = 1.4 and Prandtl number
P r = 43 . The transition curve from 1 to 2 is a straight line (from von Mises (1958)).

Taylor’s solution (no heat conduction, k = 0)

G.I. Taylor has shown that equations (10.50) and (10.51) can be integrated in closed form
when the thermal conductivity k is set to zero. This is not a realistic assumption, since it is
known that the ratio µ/k varies with temperature. However the solution is of interest because
some principle flow features are already present under the assumption k = 0. Taking k = 0,
equations (10.50) and (10.51) become:

µ̂ dũ
ũ = ũ (ũ − 1) − αũ + αθ, (10.56)
J dx
with

γ − 1 (ũ − 1)2
θ =1+ − (γ − 1) (ũ − 1) . (10.57)
α 2
Eliminating θ there follows

µ̂ dũ (γ + 1) 2 1
ũ = ũ − γ (1 + α) ũ2 − γ (1 + α) ũ + (γ − 1) + γα. (10.58)
J dx 2 2
Except from a scale factor J and a translation of x, the solution of this equation only depends
on the parameter α.
From equation (10.53) and (10.54) it is obvious that the uniform states ũ = ũ1 = 1 and ũ = ũ2
are roots of the right-hand side of equation (10.58) inplying that it can be written as

µ̂ dũ (γ + 1)
ũ =− (1 − ũ) (ũ − ũ2 ) . (10.59)
J dx 2
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 233

When µ̂ can be taken constant, equation (10.59) can be integrated leading to the analytical
solution as introduced by Taylor:
2 µ̂ ln (1 − ũ) − ũ2 ln (ũ − ũ2 )
x= . (10.60)
γ+1J (1 − ũ2 )
Notice that for x → −∞, ũ → ũ1 = 1 and for x → ∞, ũ → ũ2 ; consequently equation (10.60)
describes transition from the pre-shock state onto the post-shock state which occurs in the
interval (−∞, ∞).
In studying the transition and in particular a shock thickness estimate, consider two inter-
mediate values u− + − +
 and u defined as u = 1 −  (1 − ũ2 ) and u = ũ2 +  (1 − ũ2 ).  is any
number satisfying 0 <  < 1/2.
The x-locations where u− + − +
 and u appear in the transition region are x and x respectively;
they follow from
2 µ̂ ln (1 − ũ± ±
 ) − ũ2 ln (ũ − ũ2 )
x± = ,
γ+1J (1 − ũ2 )
and differ by the amount
2 µ̂ 1 + ũ2 1 − 
∆x = x+ − x− = ln .
γ + 1 J 1 − ũ2 
For given values of the flux J and the shock strength as expressed in terms of ũ2 , the right-
hand side tends to zero as µ̂ decreases no matter how small  may be.
Let us for example calculate the shock thickness in air in a windtunnel experiment. Assume the
tunnel is driven with settling chamber conditions pt = 5 bar, Tt = 228 K. Assume a normal
shock appears at M = 2. The pre-shock conditions are: p1 = 0.639 bar, ρ1 = 1.39 kg/m3 and
T1 = 160 K; the post-shock conditions are: p2 = 2.88 bar, ρ2 = 3.71 kg/m3 and T1 = 270 K.
The velocities in front of and behind the shock are respectively u1 = 507.14 m/s and
u2 = 190.15 m/s.
If we take  = 0.05 then we find ∆x = 0.14 × 10−3 mm. Thus, since the total velocity drop
u1 − u2 = 317 m/s, then 90 percent of this drop is effected within a distance of 0.00014
mm! This is a significant result: the thickness of the layer within which the major part of
the transition from pre-state Àto post-state Áoccurs tends towards zero with µ̂ → 0 and is
actually extremely small in air under normal conditions.

Becker’s solution Becker has found an exact solution of the complete problem, equations
µc
(10.50) and (10.51), in the case that the Prandtl number: P r = kp = 34 . The complete
problem is described by the system of first order O.D.E.’s, given in equations (10.50) and
(10.51) given in the unknowns ũ and θ. One could eliminate either ũ or θ to obtain a second-
order differential equation for θ or ũ respectively. The resulting equation however would not be
easy to handle and a better procedure is to eliminate x by dividing the two O.D.E.’s obtaining

dθ µ̂R
α
γ−1 (θ − 1) + α (ũ − 1) − 21 (ũ − 1)2
= . (10.61)
ũdũ kα ũ (ũ − 1) + α (θ − ũ)
µ̂R
The coefficient k is a dimensionless quantity very close related to the Prandtl number:
µ̂R µ̂cp γ − 1 4 µcp γ − 1 4γ−1
= = = P r. (10.62)
k k γ 3 k γ 3 γ
234 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Observe that the Stokes hypothesis: λ = −2/3µ is used to write µ̂ = 2µ + λ = 43 µ. For dry air
under normal conditions the Prandtl number varies only slightly, roughly between 0.68 and
0.77.
Becker observed in 1922 that for P r = 3/4 equation (10.61) results in an exact solution:
γ−1
1 − ũ2 .


θ−1= (10.63)
2γα
When this solution is expressed in terms of the dimensional variables u and T one finds

u2 γ u2
+ RT = + cp T = constant. (10.64)
2 γ−1 2
The total enthalpy is constant through the whole shock zone. But if that is true then one
must conclude from equation (10.38) that du dT
dx and dx are related such that

du dT
(λ + 2µ) u +k = 0,
dx dx
or
3
dT µ̂ Pr
=− = 4 ,
udu k cp

which corresponds exactly with the value of T1 dT

du that is obtained from equation (10.63) with
a Prandtl number of 34 .
In the Becker solution the viscous dissipation and heat conduction are in balance resulting
into a constant total enthalpy in the entire shock zone.

10.4.8 Entropy behaviour in the shock zone

In this section we will investigate some properties of the behaviour of the entropy in the shock
zone. One reason to do this investigation is that we want to know how the total entropy rise
is built up in the shock zone. Are there parts in the shock zone that are almost isentropic and
are there other parts in the shock zone that are responsible for the main part of the entropy
rise? Regarding these questions let us take a particular example e.g. a shock wave in air
(γ = 1.4) at a Mach number M = 2. Let us study the Taylor shock and the Becker shock as
well.
In scaled variables the Taylor shock is governed by, (see equation (10.57))

γ − 1 (ũ − 1)2
θTaylor = 1 + − (γ − 1) (ũ − 1) ,
α 2
and the Becker solution gives (equation (10.62))
γ−1
θBecker = 1 + (1 − ũ)2 .
2γα
For both shocks the transport coefficients µ and k are assumed to be constant. In the particular
situation M1 = 2 the parameter α takes the value α = 5/28 and the post-shock conditions are
u2 = 0.375u1 (ũ2 = 0.375), T2 = 1.6875T1 (θ2 = 1.6875).
The velocity-temperature relation in the shock interior has graphs that may be viewed in figure
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 235

Figure 10.13: Taylor and Becker solution for the shock structure.

10.13. Obviously the Taylor shock and the Becker shock have states Àand Áin common but
otherwise their behaviour differs substantially. A striking difference appears when considering
the entropy. Figure 10.13 shows two entropy level curves; one carrying the pre-state value s1
and the other carrying the post-state value s2 . Taking the pre-state as reference, the entropy
can be calculated from
s − s1
= ln θ + (γ − 1) ln ũ. (10.65)
cv
From figure 10.13 one observes that the curve representing the Taylor shock intersects each
entropy level curve only once implying a monotonic increase of entropy when following a
particle downstream. Since the Taylor curve is tangent (the proof is left to the reader) to the
entropy level curves s1 and s2 the streamwise entropy gradient is zero far upstream (x → −∞)
and far downstream (x → +∞). A numerical calculation of the entropy distribution in the
shock zone of the Taylor shock (figure 10.15) reflects this observation. The most typical
characteristic of the Taylor shock is the absence of heat conduction in the shock zone. Of
course severe temperature gradients appear but due to the assumption k = 0 no heat will
be transferred between adjacent fluid particles. So the energy content of a fluid element can
only be changed by work; done by pressure forces and/or viscous (normal) forces. To get
some physical insight in this process let us study the breakdown of energy into its constituting
components like internal energy, kinetic energy, work done by viscous forces and work done
by pressure forces. Let us start by considering the energy equation for 1-dimensional steady
flow in the form of equation (10.38)
udu dT
L = ρuH − µ̂ −k = constant,
dx dx
p
or with k = 0 (Taylor shock) and inserting H = E + ρ

du
L = ρuE + pu − µ̂u ,
dx
236 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Figure 10.14: Velocity decrease in a Taylor shock at M1 = 2.

Figure 10.15: Entropy increase in a Taylor shock at M1 = 2.

or
pu µ̂u du L
E+ − = = constant. (10.66)
J J dx J
The first term represents the total energy (E), being the sum of the internal energy (e) and
the kinetic energy ( 12 u2 ). The second term is the energy needed to run against the adverse
pressure gradient. The third term is the energy needed to overcome viscous normal stresses.
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 237

In case of the Taylor shock these terms may be calculated as follows:

cv T αθ
internal energy: Ei = u21
= γ−1 ,
1 u2 1
kinetic energy: Ek = 2 u2 = 2 ũ,
1
pu
work done by pressure forces: Wp = Ju 2 = αθ,
1
µ̂u du
work done by viscous forces: Wv = − Ju 2 dx = −ũ (ũ − 1) + αũ − αθ.
1

Note that the kinetic energy of the pre-state: u21 is chosen as a reference.
The enery breakdown for a Taylor shock at M1 = 2 is calculated in table 10.1 and depicted
in figures 10.16 and 10.17. Due to the assumption of a Taylor shock, which gives a relation
between velocity and temperature, the internal energy Ei and kinetic energy Ek are coupled.
In figure 10.16 we see a monotonic increase of internal energy and mutually a monotonic de-
crease of kinetic energy when passing the shock in downstream direction. The energy needed
to run against the adverse pressure gradient is also increasing monotonically.
The energy needed to overcome viscous forces is growing in the low pressure part of the

Figure 10.16: Energy breakdown in a Taylor shock.

shock, it reaches a maximum at about ũ = 0.7, then it decreases to zero at the post state.
From figure 10.17 we conclude that a fluid element that passes a shock experiences a decrease
of total energy (E) in the front part of the shock caused by the fact that the strong decrease of
kinetic energy is not sufficiently compensated by the moderate increase of internal energy. In
238 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

ũ θ Ei Ek E Wp H Wv
1.0 1.0 0.4464 0.5000 0.9464 0.1786 1.1250 0
0.95 1.023 0.4567 0.4513 0.9080 0.1827 1.0907 0.0343
0.90 1.051 0.4692 0.4050 0.8742 0.1877 1.0619 0.0631
0.85 1.085 0.4844 0.3613 0.8457 0.1938 1.0395 0.0755
0.80 1.125 0.5022 0.3200 0.8222 0.2009 1.0231 0.1019
0.75 1.170 0.5223 0.2813 0.8036 0.2089 1.0125 0.1125
0.70 1.221 0.5450 0.2450 0.7900 0.2180 1.0080 0.1170
0.65 1.277 0.5701 0.2113 0.7814 0.2280 1.0094 0.1156
0.60 1.339 0.5978 0.1800 0.7778 0.2391 1.0169 0.1081
0.55 1.407 0.6281 0.1513 0.7794 0.2513 1.0307 0.0943
0.50 1.480 0.6607 0.1250 0.7857 0.2643 1.0500 0.0750
0.45 1.559 0.6960 0.1013 0.7973 0.2784 1.0757 0.0493
0.40 1.643 0.7335 0.0800 0.8135 0.2934 1.1069 0.0181
0.375 1.6875 0.7533 0.0703 0.8236 0.3014 1.1250 0

Table 10.1: Energy breakdown in Taylor shock

Figure 10.17: Energy breakdown in a Taylor shock.

this front part of the shock the loss of kinetic energy is used to overcome the adverse pressure
gradient and the work against viscous normal stresses. In the aft part of the shock zone the
total energy (E) starts recovering. Observe that due to viscous forces the total enthalpy is
not constant here; passing the shock we see first a decrease of H and further downstream an
increase of H.
Finally let us pay attention to the Becker shock and in particular to the entropy variation
through the shock. First we remember that the Becker shock is characterized by a perfect bal-
ance between viscous dissipation and heat conduction resulting into a constant total enthalpy
through the entire shock zone.
Concerning entropy it appears (figure 10.13) that this parameter attains a maximum value:
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 239

smax somewhere in the shock zone and being larger than its post-state value s2 . To understand
this entropy overshoot let us consider the entropy distribution as calculated for the Becker
shock having M1 = 2 and as shown in figure 10.18.
Obviously a maximum appears at ũ ≈ 0.6. This maximum is about 1.6 times larger than its

Figure 10.18: Breakdown of entropy in a Becker shock at M1 = 2.

post-state value!
As we know, the heat flux is related to temperature gradients. Consider a small one-dimensional
fluid element and in particular the heat flux through its upstream face and its downstream
face. As long as this fluid element resides in the front part of the shock, the temperature
gradient at the upstream face is higher then at the downstream face. So in the front part of
the shock there is a net influx of heat into the fluid element. Fluid elements residing in the aft
part of the shock experience higher temperature gradients at the downstream face than those
at their upstream face. Therefore in the aft part of the shock fluid elements experience a net
outflux of heat.
It is this changing behaviour of net heat flux that is responsible for the entropy overshoot we
have discovered in figure 10.18.
In order to understand this more clearly let us breakdown the entropy according to its defini-
tion:
Z Z
dQ
s= + dsirr , (10.67)
T
into a part that comes from the heat flux and into the irreversible part.
Since dQ changes from Rpositive to negative when moving downstream through the shock we
expect a maximum of dQ T somewhere inside the shock. Let dQ be calculated as follows:
consider a one-dimensional fluid element of unit mass and having a length `, see figure 10.19.
Due to mass conservation the length of the fluid element varies with the density as ρ` = 1.
The heat flux into the fluid element through its upstream face is
     2  
dT dT d T
dQ̇in = k =k + ` .
dx x+` dx x dx2 x
240 CHAPTER 10. STEADY ONE-DIMENSIONAL VISCOUS FLOW

Figure 10.19: Heat flux into a one dimensional fluid element.

The heat flux that leaves the fluid element through its downstream face is
 
dT
dQ̇out = k .
dx x
The net heat flux during a time lapse dt then becomes:
 
  dT
dQ = dQ̇in − dQ̇out = k `dt.
dx x
Since the fluid element itself moves with the velocity u the net heat influx during a traveling
distance dx is:
 2 
d T `dx
dQ = k .
dx2 u
Finally using ρ` = 1 (conservation of mass) and ρu = J is constant we find

dQ k d2 T k d2 T du
= dx = . (10.68)
T JT dx2 JT dx2 du
dx

This expression is now evaluated for the Becker shock by substituting:

dT γ+1 2 γ−1
k = −J u + Ku − L,
dx 2γ γ
and
du γ+1 γ−1L
µ̂ =J u−K + ,
dx 2γ γ u
which leads to the final expression
dQ γ−1
= {γ (1 + α) − (γ + 1) ũ} dũ. (10.69)
cv T γαθ
Inserting the Becker solution:
γ−1
1 − ũ2 ,


θ =1+
2γα
equation (10.69) can be integrated giving the result
γ (1 + α) (B + ũ) (B − 1)
Z
dQ
= ln + (γ + 1) ln θ, (10.70)
cv T B (B − ũ) (B + 1)
10.4. INTERNAL STRUCTURE OF A SHOCK WAVE 241

2γα
where B 2 = 1 + γ+1 and where the pre-state ũ = 1, θ = 1 is chosen as the reference state.
Figure 10.18 is now completed by showing the behaviour of cdQ
R
vT
through the shock zone.
R dQ
As was expected cv T attains a maximum value and causes the entropy overshoot which
is already mentioned. Furthermore we see in figure 10.18 that the irreversible part of the
entropy sirr increases continuously when moving downstream. This irreversible part of the
entropy can attributed to the presence of viscous normal stresses. To verify this statement
consider equation (10.45):
     
ds d du d dT d du
JT = µ̂u + k −u µ̂ .
dx dx dx dx dx dx dx
d
k dT


Equation (10.68) can be used to eliminate dx dx
   
ds d du dQ d du
JT = µ̂u +J −u µ̂ ,
dx dx dx dx dx dx
or
(   )
ds 1 du 2 JdQ
J = µ̂ + . (10.71)
dx T dx T dx

This equation may be compared to equation (10.67) where we have given the usual breakdown
of entropy into a part that comes from the heat flux and into an irreversible part. Comparison

2
shows that the viscous term µ̂ du
dx causes the irreversibly of the shock wave.