EMS513U

A EROTHERMODYNAMICS OF F LOWS

2023/24

K SHITIJ S ABNIS

K . SABNIS @ QMUL . AC . UK

R OOM 311, S CHOOL OF E NGINEERING AND M ATERIALS S CIENCE

QUEEN M ARY U NIVERSITY OF L ONDON

M ANY THANKS TO FARIBORZ M OTALLEBI WHO DEVELOPED THIS COURSE

Module Description
The key people delivering the module are Kshitij Sabnis (module organiser), Nikos
Bempedelis (module deputy organiser) and Agamemnon Chaidos (teaching asso-
ciate).

The first part of the module deals with the derivation of the differential form of
the governing equations of fluid flows by using the fundamental conservation laws
(mass and linear momentum). Particular attention is given to the simplified form
of these equations at high Reynolds numbers. The concept of the boundary layer
is also introduced. Furthermore, the role of boundary layers in the production of
drag, stall and the performance of aeronautical/non-aeronautical vehicles and de-
vices is briefly discussed.

In the second part of the module the fundamentals of thermodynamics are re-
viewed. Basic concepts in compressible flows are also introduced, including: flow
regimes and Mach number, Mach waves, propagation of information, normal shock
waves, effects of area change/back pressure on gas flows, applications in propul-
sion systems and wind tunnels, flow measurement and flow visualisation in com-
pressible flows.

The module aims to:

• revisit the scope and applications of thermodynamics in compressible flows;

• review control volumes, revise mass/momentum conservation and introduce

energy conservation;

• introduce viscous flows and derive conservation laws with emphasis on high
Reynolds number flows;

• introduce simple boundary layer flows and drag estimation;

• describe transitional and turbulent flows, including their effects in aeronau-

tical applications;

• develop understanding the nature of compression and expansion waves;

• introduce reversible and irreversible waves, including the formation of shock

waves.

The objectives of the module are to enable you to:

• analyse the effect of boundary layer flows and compressibility on the perfor-
mance of aeronautical objects;

1
 • analyse and predict viscous drag;

• describe 1D compressible flow;

• compute the formation of waves and propagation of information in different

flow regimes;

• describe shock waves;

• perform fluid analysis for gas flows in variable area ducts and variable back
pressure;

• perform flow visualisation and measurement in compressible flows.

Alongside these lecture notes, the following textbooks are useful references:

• Mechanics of Fluids (F. White);

• Modern Compressible Flow (J. Anderson);

• Viscous Fluid Flow (F. White);

• Fundamentals of Fluid Mechanics (B. Munson, A. Rothmayer, T. Okiishi, W.

Huebsch);

• Fluid Mechanics (J. Douglas, J. Gasiorek, J. Swaffield);

• Engineering Fluid Mechanics (D. Elger, B. Williams, C. Crowe, J. Roberson).

Assessment is carried out through the following elements:

• CW1: lab report on “Flow around a cylinder” (15%)

• CW2: lab report on “Measurement in boundary-layer flows” (15%)

• final written exam in January assessment period (70%)

For students requiring resits, assessment is carried out through a reassessment

written exam in the LSR period (100%).

2
Table of Contents
1 Fundamentals of Viscous Flows 4
1.1 Basic Concepts in Fluid Mechanics . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Viscosity and Shear . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Governing Equations of Fluid Dynamics . . . . . . . . . . . . . . . . . 10
1.2.1 Conservation of Mass (Continuity Equation) . . . . . . . . . . 12
1.2.2 Conservation of Momentum 1: Control Volume Analysis . . . 15
1.2.3 Motion and Deformation of Fluid Elements . . . . . . . . . . 22
1.2.4 Conservation of Momentum 2: Navier–Stokes Equations . . 28
1.2.5 Exact Solutions of the Navier–Stokes Equations . . . . . . . . 33
1.2.6 Limiting Cases of the Navier–Stokes Equations . . . . . . . . 39
1.3 Boundary-Layer Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.3.1 Thin Boundary-Layer Equations . . . . . . . . . . . . . . . . . 48
1.3.2 Simple Solutions of Boundary-Layer Flows . . . . . . . . . . . 53
1.3.3 Applications of Boundary-Layer Theory . . . . . . . . . . . . . 64

2 Fundamentals of Compressible Flows 69

2.1 Basic Concepts in Compressible Flow . . . . . . . . . . . . . . . . . . 69
2.1.1 Constant-Area Duct with Friction: Fanno Flow . . . . . . . . 77
2.1.2 Constant-Area Duct with Heat Transfer: Rayleigh Flow . . . . 81
2.2 Normal Shock Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.2.1 Requirement for the Formation of Shock Waves . . . . . . . . 85
2.2.2 Working Equations of Normal Shock Waves . . . . . . . . . . 87
2.2.3 Pitot Probes in Supersonic Flow . . . . . . . . . . . . . . . . . 92
2.2.4 Moving Shock Waves . . . . . . . . . . . . . . . . . . . . . . . . 94
2.3 Flows through Variable-Area Ducts . . . . . . . . . . . . . . . . . . . . 95
2.3.1 Variable-Area Ducts and Sonic Speed . . . . . . . . . . . . . . 101
2.3.2 Effect of Back Pressure on Convergent Ducts . . . . . . . . . . 110
2.3.3 Convergent–Divergent Ducts . . . . . . . . . . . . . . . . . . . 115

3
 Fundamentals of Viscous

1.1 Basic Concepts in Fluid Mechanics

1 SECTION

Flows

Britannica Concise Encyclopaedia defines the mechanics of fluids as “the study of the
effects of forces and energy on liquids and gases. One branch of the field, hydrostatics,
deals with fluids at rest; the other, fluid dynamics, deals with fluids in motion and with
the motion of bodies through fluids. Liquids and gases are both treated as fluids because
they often have the same equations of motion and exhibit the same flow phenomena.
The subject has numerous applications in fields varying from aeronautics and marine
engineering to the study of blood flow and the dynamics of swimming.”

Continuous fluids In order to study and understand the principles of aerodynamics,

we must have a good knowledge of the three principal components: (1) the fluid, (2) the
body (i.e. the shape) and (3) the relative velocity between the body and the fluid. We will
start by considering the first of these components, the fluid.

Like solids, fluids are made of billions of molecules which are separated by empty space.
For example, for standard conditions at sea level, the number of molecules in a cubic
centimetre of air is approximately 2.5 × 1019 . The average distance a molecule moves
between successive collisions (also called the mean free path, λ) is about 6.63 × 10−8 m,
a very small distance. This implies that, at sea level when you move your hand through
the air, it cannot distinguish individual air molecules but instead feels like a continuous
medium. In contrast if we are at an altitude of say 100 km, the mean free path is about
0.32 m. Now, when you move your hand through the air, (if your hand were sensitive

4
enough) it would be able to feel the impacts of individual molecules. The air no longer
feels like a continuous medium. To capture the extent to which a fluid can be regarded
as continuous, the non-dimensional parameter used is called the Knusden number (Kn
= λ/L). Here, L is a characteristic dimension of the body, e.g. the width of your hand,
the thickness of an airfoil, the diameter of a cylinder, ...

The continuum postulate (which enables the fluid to be treated as a continuous medium)
requires that the Knudsen number should be less than approximately 0.01. Under such
conditions we can ignore the molecular structure of the fluid and replace this with a
physical model. This physical model has neither molecules nor structure, instead it
is continuous. By adopting such a model we can consider certain average properties
which are the representative of the statistical behaviour of the countless molecules of
the real fluid.

B
δn
A δs

Figure 1: Closely-neighbouring points in a fluid flow. The continuous nature of flow

properties allow us to relate these quantities at point A to those at point B (in the stream-
wise direction) and point C (in the flow-normal direction).

The fact that the fluid properties are continuous is significant. Say we know the proper-
ties of the fluid (velocity, temperature, pressure, density, entropy, ...) at point A in Fig-
ure 1. If we wish to know the equivalent properties a short distance away (either δs along
a streamline at point B or δn normal to a streamline at point C) we can use the math-
ematical framework of Taylor expansions to calculate the required information. In this
way, we can build up our knowledge of the properties around the entire flow field (using
the equations governing fluid mechanics which we will derive), in a way that would be
impossible if these quantities were discontinuous.

Fluid density Consider a very small element of fluid, which still contains a large enough
number of molecules to satisfy to continuum postulate. Define δV as the volume of this
fluid element. If the total mass of all the molecules within the element is δm then the
density of the fluid at the point represented by the fluid element is:

δm
ρ= (1)
δV

5
From a dynamical point of view, the density is an important property of the fluid be-
cause it is a measure of the fluid’s “inertial resistance” to being accelerated.

Fluid velocity By definition the velocity of a fluid element is equal to the ratio of the
total momentum of all the molecules making up the fluid element to the total mass of
all these molecules: Pn
m i v⃗i
⃗ = Pi =1
V (2)
n
i =1 m i

where the sum from i = 1 ... n is across all molecules within the fluid element. Of course,
more intuitively, the velocity of the fluid element is the vector quantity which defines
how fast it is moving through three-dimensional space.

1.1.1 Viscosity and Shear

Each and every particle in a fluid obeys the Newton’s law of motion. Newton’s law of
motion states that the sum of external forces acting on a particle is equal to its inertial
force (i.e. mass times acceleration). Whilst there are many different kinds of forces, they
can generally be put into two categories, body forces and surface forces. Body forces
act at a distance (e.g. gravitational, electrical, and electromagnetic). The only body force
to apply in most aerodynamic applications is gravity, which depends only on the weight
of the object and not on its speed. On the other hand, surface forces act by direct con-
tact, as shown for a single surface on a cuboidal fluid element in Figure 2. These surface
forces for a fluid can act normal to the contact surface (normal stresses, i.e. pressure) or
they can act tangential to the contact surface (shear stresses, i.e. viscous friction). The
aerodynamic lift on an object comes from the vertical component when the pressure
is integrated around all surfaces. Meanwhile, the aerodynamic drag typically has two
components, namely (1) the pressure drag from the horizontal component of the in-
tegrated pressure and (2) the friction drag from the integrated effect of shear stresses
around all surfaces of the object.

overall
normal stress surface
component stress

shear stress
fluid component
element

Figure 2: Decomposition of stress components on the upper surface of a cuboidal fluid

element. Similar stresses will also exist on the other faces.

6
The pressure acting on the walls of a swimming pool is an example of normal stress.
With regards to shear stresses, the fluid property associated with these forces is its dy-
namic (or absolute) viscosity. The nature of viscosity can be understood by considering
the following situation. There are two very long parallel plates a distance h apart, with
the space between them filled by a fluid (Figure 3). The lower plate is at rest and the up-
per plate moves to the right with a constant velocity U . Within the channel, there will be
a one-dimensional flow moving to the right with velocity (u) dependent on the vertical
coordinate (y).

plate moving at speed U

U

ut = U

y
h u(y) = U
h

ub = 0

plate at rest

Figure 3: Velocity profile in a viscous fluid between two parallel flat plates, with the
upper plate moving to the right at speed U . y is the spatial coordinate in the direction
normal to the plates.

We know from experience that the fluid adheres to both walls, also known as the “no-
slip condition”. Therefore, the velocity of the fluid particles at the lower plate is zero
(u b = 0) and, for those at the upper plate, the velocity u t = U . If the pressure is being
kept constant across the height of the channel, we can consider a simple case where the
velocity of fluid particles within the channel varies linearly between two plates:

y
u(y) = U (3)
h

In order to support this motion a tangential force must be applied to the upper plate.
This tangential force is in equilibrium with the frictional forces in the fluid. Experiment
shows that for many fluids (termed Newtonian fluids) this frictional force or its value
per unit area (i.e. shear stress, τ) is proportional to the plate velocity (U ) and inversely
is proportional to the distance between the two plates (h). Therefore, the shear stress
is proportional to U /h. Since U is the velocity difference between the two plates and h
is the vertical separation between the two plates, we note that the shear stress is there-
fore also proportional to the velocity gradient, d u/d y, i.e. the rate at which the velocity

7
changes in the direction perpendicular to the flow.

du
τ=µ (4)
dy

The constant of proportionality (µ) is a physical property of the fluid, called the absolute
viscosity or dynamic viscosity.

Equation 4 is known as Newton’s law of friction. Fluids that obey this law (such as
those in the situation described above) are termed Newtonian; others are termed non-
Newtonian. In non-Newtonian fluids, such as polymers or blood, the shear stress de-
pends non-linearly on the rate of the shear strain. However, the fluids considered in
the majority of aerospace applications are typically Newtonian in nature, and so this
module focuses primarily on Newtonian fluids.

Fluid viscosity By considering Equation 4, we can deduce the dimensions of dynamic

viscosity. Since the shear stress is expressed in N/m2 or Pa, the units of viscosity are
N·s/m2 or kg/m·s. In all fluid motion where frictional and inertia forces interact it is
important to consider the ratio of the viscosity (which governs friction) to the density
(which governs inertia). This ratio is called kinematic viscosity:

µ
ν= (5)
ρ

Dimensional analysis tells us that the units of m2 /s. It is important to not get confused
between dynamic and kinematic viscosities when evaluating numerical values, dimen-
sional analysis can be a useful check. As well as depending on the precise type of fluid
itself, this quantity is also a function of fluid temperature (it is worth noting that the
influence of pressure is negligible in most cases). The viscosity typically increases with
temperature for gases (molecules have more kinetic energy and collide with greater en-
ergy so the intermolecular forces get stronger) and decreases with temperature for liq-
uids (molecules have greater energy to overcome intermolecular forces). When trying
to determine how viscosity changes at different temperatures for air and dilute gases,
there are two widely-used relations. The Sutherland (1893) formula states:
 3/2
µ T Tref + S
=  (6)
µref Tref T +S

where µref and Tref are the viscosity and temperature at reference conditions, where their
values are known. S is an effective temperature called the Sutherland constant, which is
dependent on the fluid under consideration. The other common relation for calculating

8
viscosity from known reference conditions is the power-law formulation:
 n
µ T
=  (7)
µref Tref

where the exponent, n, depends on the fluid. For illustration, the values of these quan-
tities for some common gases are provided in Table 1.

Table 1: Power-law and Sutherland-law viscosity parameters for gases

Gas Tref µref n S
(K) (N·s/m2 ) (K)
Air 273 1.716 × 10−5 0.67 111
Argon 273 2.125 × 10−5 0.72 144
CO2 273 1.370 × 10−5 0.79 222
CO 273 1.657 × 10−5 0.71 136
N2 273 1.663 × 10−5 0.67 107
O2 273 1.919 × 10−5 0.69 139
H2 273 8.411 × 10−5 0.68 97
Steam 350 1.120 × 10−5 1.15 1064

Note that, while the values may be larger in some cases than others, all fluids contain
viscosity. So, why do we refer to some flow fields as viscous and other as inviscid? The
answer is that a viscous flow is one which transfers the effects of the fluid viscosity onto
the forces experienced by an object within it. Considering Equation 4 once again, we
see that (whilst µ is always non-zero) for the shear stress to be affected by this viscosity,
it is necessary for the velocity gradient to also be non-zero. Therefore, we generally
define viscous flows as those where there is not only viscosity in the fluid itself but also
where the flow field contains viscosity gradients. In such viscous flows, the integral of
the shear stress over the surface of the body gives one component of the overall drag,
namely friction drag.

Boundary layers In the frame of reference of an object moving through air, the air is
moving relative to the stationary body (as experienced in a wind tunnel). Very far from
the object, in the “free stream”, the air is all moving at uniform speed. Therefore, there is
no velocity gradient and thus no viscous effects, i.e. the flow is inviscid. However, near
the object the flow encounters a boundary – in Fig. 4 we consider the simple case of a
flat plate.

The fluid element immediately adjacent to the surface must be stationary due to the no-
slip condition. Considering the second fluid element outwards, in a real viscous fluid,
friction from the stationary fluid element slows it down considerably so it has a velocity
that is not quite zero but is very slow. Similarly for the next fluid element outwards,
which will be slow (but not quite as slow) and so on, until we reach the free-stream
velocity. Thus, the viscosity results in a continuous transition from zero velocity at the
surface to the free-stream velocity at some distance away. In practice, it is observed that

9
 u(y)
free stream
inviscid flow,
no velocity gradient so no shear stress

boundary layer
du
viscous flow, τ = µ
no slip, u = 0 dy

flat plate at rest

Figure 4: Velocity profile with flow moving above a flat plate, showing the boundary layer
as a viscous region. y is the spatial coordinate in the direction normal to the plate.

this transition in velocity tends to be confined to a relatively narrow region of flow close
to a surface. This narrow region, termed the “boundary layer”, is often an important
factor in the aerodynamic performance of bodies moving through a fluid.

The magnitude of the shear stress at the surface (which depends on both the fluid vis-
cosity and the local velocity gradient, as per Equation 4) is called the “wall shear stress”.
The integral of this wall shear stress over the surface of the body determines the compo-
nent of drag experienced by the object due to fluid friction at the surface, namely “skin-
friction drag”. Note that the fact that there is a change in velocity across the boundary
layer implies that there is a velocity gradient here and thus viscous effects are important.
Meanwhile, outside the boundary layer, the velocity is uniform, so there is no velocity
gradient and the flow is inviscid.

1.2 Governing Equations of Fluid Dynamics

The fundamental equations that govern how a fluid behaves have been known for more
than a century. However, in their complete form, these equations are quite formidable
and very difficult to solve for real applications even on modern, powerful computers.
Fluids, like solids, obey the principles of fundamental laws for physical systems. These
laws are:

• conservation of mass (continuity);

• conservation of momentum (Newton’s second law);

• conservation of energy (first law of thermodynamics);

• second law of thermodynamics.

In order to determine the flow behaviour in a given application, it is necessary to use

these equations to obtain three unknown quantities: the velocity vector (V⃗ ), the ther-
modynamic pressure (P ), and the absolute temperature (T ). In order to do so, it is nec-
essary to consider three elements:

10
 • the thermodynamic variables of the fluid which appear in the governing equa-
tions;

• the governing equations themselves (as listed above);

⃗ , P and T at every point around
• “boundary conditions”, i.e. specified values for V
the boundary of the flow region.

As we shall see, the final forms of the conservation equations contain four thermody-
namic variables, namely, the density (ρ), the enthalpy (h), and the two transport prop-
erties: viscosity (µ) and thermal conductivity (k). It is usual to assume that the system
under consideration is in a state of thermal equilibrium and therefore these four ther-
modynamic variables are uniquely determined by the values of P and T . Thus, the sys-
tem is completed by knowing the following relations:

ρ = ρ(P, T ) h = h(P, T ) µ = µ(P, T ) k = k(P, T ) (8)

which can be in the form of tables or empirical equations. The values of viscosity and
thermal conductivity are typically obtained from tables or relations such as Sutherland’s
law. From thermodynamics we already know that, for a perfect gas (including air), the
relationships for density and enthalpy are as follows:

P
ρ= h = CP T (9)
RT

where R is the gas constant, itself defined by R = R/m with the universal gas constant,
R = 8313 J/(kg·mol·K) and m being the molecular weight of the gas. C P is the specific
heat capacity of the gas at constant pressure and γ is the specific heat ratio. For illustra-
tion, Table 2 gives these quantities for some common fluids.

Table 2: Thermodynamic constants for some common fluids

Fluid m γ R CP
at 298 K (J/(kg·K) at 298 K
Air 28.964 1.400 287.06 1004.0
Argon 39.994 1.658 208.15 524.6
CO2 44.010 1.288 188.92 845.7
CO 28.010 1.398 296.83 1042.5
N2 28.013 1.400 296.80 1038.3
O2 32.000 1.395 259.82 916.9
H2 2.016 1.405 4124.2 14,315.0
Water 18.016 1.329 461.50 1863.1

Having established the thermodynamic properties, it is necessary to establish the gov-

erning equations themselves. To do so, we consider an arbitrary flow field whose prop-
erties are defined as functions of space and time. We assume that as fluid particles move
through the flow field they take on the properties of the field point that they occupy.
⃗ ) are known. We
Consider a point in space where the properties of the flow (ρ, P , T , V
now construct a cuboidal control volume (CV) around this point, which is at the centre

11
of the element (Figure 5. By using the continuum postulate, we can assert that the values
of the flow quantities on the faces of the cuboidal fluid element are adequately repre-
sented by a first-order Taylor series expansion about the centre of the element. Note
that if the fluid were not continuous and homogeneous (the composition is the same
everywhere) we would not be able to make such a statement.

a) velocity at centre b) mass flow at centre

v
ρv
d y/2
w
ρw

d y/2 u
ρu

d x/2 d z/2
d x/2 d z/2

c) mass flow at faces ∂ρv d y

ρv +
∂y 2

∂ρu d x
ρu −
∂x 2
∂ρw d z
ρw +
∂z 2
∂ρw d z
ρw −
∂z 2
∂ρu d x
y ρu +
∂x 2
z
x ∂ρv d y
ρv −
∂y 2
Figure 5: Cuboidal control volume for derivation of conservation laws.

1.2.1 Conservation of Mass (Continuity Equation)

Figure 5 allows us to consider the mass flowing into and out of the control volume.
Neglecting the possibility of mass destruction or production inside the control volume
(i.e. the existence of sources or sinks) the principle of conservation of mass is based on
the following rate equation:

rate of change mass flow rate mass flow rate

= − (10)
of mass within CV into CV out of CV

12
To formulate this equation algebraically, we need to evaluate the relevant variables at
the parallel faces of the CV. We know that the x-component of velocity at the centre of
the element is u (Figure 5a) and so the x-component of mass-flow rate (per unit area)
is ρu (Figure 5b). To determine the mass flowing out of the control volume in the x-
direction, we need to consider the right-hand face of the CV, which is marked with a
star in Figure 5c. We expand the mass-flow rate in the x direction using a Taylor series
in powers of the distance from the centre of the CV. Noting that the control volume has
width d x, so the distance from the centre is d x/2, the mass-flow rate (per unit area) out
of this right-hand face is:

mass flow rate

∂ρu ∂(ρu)2 (d x/2)2
per unit area out = ρu + (d x/2) + + ... (11)
∂x ∂2 x 2!
of right-hand face

According to the usual rules of differential calculus, we assume the CV to be so small

that terms containing the squares and higher powers of d x, d y and d z can be neglected.
Therefore the expression in Equation 11 can be written simply as:

mass flow rate

∂ρu
per unit area out = ρu + (d x/2) (12)
∂x
of right-hand face

To calculate the mass flow rate leaving the right-hand face, we now need to multiply by
the area of this face (= d y × d z):
  
mass flow rate out ∂ρu d x
= ρu +   d yd z (13)
of right-hand face ∂x 2

Note that there is also mass leaving the control volume from the top face and the far face
in Figure 5. Therefore, evaluating the rate of mass flow out of the CV from Equation 10,
we obtain:
        
mass flow rate  ∂ρu d x ∂ρv d y ∂ρw dz
= ρu +   d yd z + ρv +   d xd z + ρw +   d xd y
out of CV ∂x 2 ∂y 2 ∂z 2
 
∂ρu ∂ρv ∂ρw d xd yd z
= ρud yd z + ρvd xd z + ρwd xd y +  + + 
∂x ∂y ∂z 2
(14)

By considering the mass entering the control volume via the left, bottom and near faces
in Figure 5. The equivalent analysis gives the following expression for mass flow into the

13
control volume:
        
mass flow rate ∂ρu d x ∂ρv d y ∂ρw d z
= ρu −   d yd z + ρv −   d xd z + ρw −   d xd y
into CV ∂x 2 ∂y 2 ∂z 2
 
∂ρu ∂ρv ∂ρw d xd yd z
= ρud yd z + ρvd xd z + ρwd xd y −  + + 
∂x ∂y ∂z 2
(15)

The rate of change of mass within the control volume is defined as the time-derivative
of the mass contained within the CV:

rate of change ∂¡ ¢ ∂ρ
= ρd xd yd z = d xd yd z (16)
of mass within CV ∂t ∂t

since d x, d y and d z are constant with time. We now insert the expressions from Equa-
tions 14, 15 and 16 into the equation of conservation of mass (Equation 10):
     
∂ρ ∂ρu
∂ρw d xd yd z ∂ρv
 d xd yd z  = ρud yd z + ρvd xd z + ρwd xd y −  + +  
∂t ∂x ∂y ∂z 2
   
∂ρu ∂ρv ∂ρw d xd yd z
− ρud yd z + ρvd xd z + ρwd xd y +  + +  
∂x ∂y ∂z 2
(17)

We can then simply this equation by combining the two square brackets on the right-
hand side, dividing through by d xd yd z and moving all terms to the left-hand side. This
simplification results in the continuity equation:

∂ρ ∂ρu ∂ρv ∂ρw

+ + + =0 (18)
∂t ∂x ∂y ∂z

The above equation is true for all fluids with no assumptions made. We can now con-
sider what happens to the equation when we consider an incompressible flow, i.e. a flow
where the density is constant. As a result, the time-derivative of density is zero and the
density can be taken out of the partial spatial derivatives. We thus obtain the incom-
pressible continuity equation:

∂u ∂v ∂w
+ + =0 (19)
∂x ∂y ∂z

If the flow is steady, the time-derivatives of all properties (including density) are zero
and thus we obtain the steady continuity equation:

∂ρu ∂ρv ∂ρw

+ + =0 (20)
∂x ∂y ∂z

14
1.2.2 Conservation of Momentum 1: Control Volume Analysis

For fluid flows, Newton’s second law of motion (the principle of conservation of momen-
tum) is called the Navier–Stokes equation. This equation states that the rate of change
of momentum of a body is equal to the forces applied to that body. Momentum (like
force) is a vector quantity but we can resolve both into scalar quantities by considering
the components in the x-, y- and z-directions. We can therefore say, for example:

rate of increase rate of flux rate of flux sum of

of x-component of x-component of x-component x-component
= − +
of flow momentum of flow momentum of flow momentum of forces
within CV into CV out of CV acting on CV
(21)
The y- and z-directions are treated in a similar way. Again, referring to Figure 5, each
term in the above statement can be expressed in terms of mass flow rate and velocity.
Noting that the flow momentum per unit volume is the product of the density and the
velocity:

rate of increase of x-component ∂¡ ¢ ∂ρu

= ρud xd yd z = d xd yd z (22)
of flow momentum within CV ∂t ∂t

Similarly, the momentum flux per unit area is the product of the mass flow per unit area
and the velocity. We already have the mass flow rates at each of the faces in Figure 5c.
In order to determine the rate of flux of x-component of flow momentum, we therefore
need expressions for the x-component of flow velocity (u) at each of the faces, as shown
in Figure 6.

∂u d y
u+
∂y 2
∂u d x
u−
∂x 2 top
∂u d z
front u+
right ∂z 2
∂u d z
u−
∂z 2

∂u d x
y u+
∂x 2
z ∂u d y
u−
∂y 2
x

Figure 6: Variation of the x-component of velocity (u) on the faces of a cuboidal control
volume for derivation of the momentum equation.

We can now evaluate the rate of flux of x-component of flow momentum out of the

15
control volume (per unit area) for the right, starred face:

rate of flux of x-component   

  
∂ρu d x ∂u d x
of flow momentum out of CV = ρu +   u +   (23)
∂x 2 ∂x 2
per unit area (right face)

Multiplying by the area of the face (d yd z) to obtain the overall rate of flux of x-component
of flow momentum for this face, which we label as O R :

rate of flux of x-component 

    
∂ρu d x ∂u d x
of flow momentum out of CV = O R = ρu +   u +   d yd z (24)
∂x 2 ∂x 2
(right face)

We can also obtain similar expressions for the top face (O T ) and the far face (O F ), i.e. the
other faces where the mass flow is out of the control volume:

rate of flux of x-component 

    
∂ρv d y ∂u d y
of flow momentum out of CV = O T = ρv +   u +   d xd z
∂y 2 ∂y 2
(top face)
(25)
rate of flux of x-component      
∂ρw d z ∂u d z
of flow momentum out of CV = O F = ρw +   u +   d xd y
∂z 2 ∂z 2
(far face)

In exactly the same manner, we evaluate the rate of flux of the x-component of flow
momentum into the CV for the left face (I L ), the bottom face (I B ) and the near face (I N ):

rate of flux of x-component      

∂ρu d x ∂u d x
of flow momentum into CV = I L = ρu −   u −   d yd z
∂x 2 ∂x 2
(left face)

rate of flux of x-component      

∂ρv d y ∂u d y
of flow momentum into CV = I B = ρv −   u −   d xd z (26)
∂y 2 ∂y 2
(bottom face)

rate of flux of x-component      

∂ρw d z ∂u d z
of flow momentum into CV = I N = ρw −   u −   d xd y
∂z 2 ∂z 2
(near face)

Substituting these terms into Equation 21 and defining the sum of the x-component of
P
forces acting on the CV as F x , we obtain:

∂ρu X
d xd yd z = (I L + I B + I N ) − (O R + O T + O F ) + Fx (27)
∂t

We now evaluate the products in each of the momentum flux terms (Equations 24, 25
and 26). We recall that the control volume is small enough that we neglect any squared

16
and higher powers of d x, d y and d z. Simplification of the resulting equation leads to:
 
∂ρu ∂ρu 2 ∂ρuv ∂ρuw X
 + + +  d xd yd z = Fx (28)
∂t ∂x ∂y ∂z

Equivalent equations for the y- and z-components of momentum can also be obtained
using exactly the same method:
 
∂ρv ∂ρuv ∂ρv 2 ∂ρv w X
 + + +  d xd yd z = Fy
∂t ∂x ∂y ∂z

(29)
 
∂ρw ∂ρuw ∂ρv w ∂ρw 2 X
 + + +  d xd yd z = Fz
∂t ∂x ∂y ∂z

In order to continue developing these equations, it is necessary to formulate expressions

for the external forces acting upon the control volume, namely F x , F y and F z .

Forces acting on a fluid element The forces acting on the fluid in the control volume
are of two types: body forces and surface forces. As stated before, the only body force
acting in most aerospace applications is gravity, so we will use g x , g y and g z to represent
the components of body force per unit mass. Therefore, the total body force acting on
the fluid inside the CV is:

F x body⃗ F y body ⃗ F z body ⃗

¡X ¢ ¡X ¢ ¡X ¢ ¡X ¢
F body = i+ j+ k
(30)
= ρg x⃗ j + ρg z⃗
i + ρg y ⃗
³ ´
k d xd yd z

A surface force acting on each face of the control volume has two associated directions:
(1) the direction in which the force acts in a tangential direction and (2) direction in
which the force acts perpendicular (normal) to the face. Figure 2 shows an example of
these forces acting on one of the faces of a cuboidal fluid element.

We consider mathematical forms of these stresses by denoting the normal stresses by

σ and shear stresses by τ. By convention, the normal stress acting on the CV fluid is
defined to be positive in the outwards direction. Each shear stress component takes
two subscripts, the first subscript represents the axis to which the face is perpendicular,
and the second subscript represents the direction of the shearing stress. For example,
τx y corresponds to a stress in the y-direction which acts in the y −z plane. These stresses
are defined Figure 7 for an infinitely control volume, i.e. with d x = d y = d z = 0.

The surface stress distribution is therefore described by nine scalar quantities, which

17
 a) stresses on left and right faces

∂τx y d x
τx y +
∂σxx d x ∂x 2
σxx −
∂x 2

∂τxz d x ∂τxz d x
τxz − τxz +
∂x 2 ∂x 2

∂σxx d x
σxx +
∂x 2
∂τx y d x
τx y −
∂x 2

∂σ y y d y
c) stresses on top and bottom faces σy y +
∂y 2

∂τ y z d y
τy z +
∂y 2

∂τ y x d y
τy x +
∂y 2
∂τ y x d y
τy x −
∂y 2

∂τ y z d y
τy z −
∂y 2
y
∂σ y y d y z
σy y −
∂y 2 x

b) stresses on near and far faces

∂τz y d z
τz y +
∂z 2

∂τzx d z ∂σzz d z
τzx − σzz +
∂z 2 ∂z 2

∂σzz d z ∂τzx d z
σzz − τzx +
∂z 2 ∂z 2

∂τz y d z
τz y −
∂z 2
Figure 7: The normal and shear stress distributions on the faces of a cuboidal control
volume. For d x = d y = d z = 0, these schematics define the normal and shear stresses at
a point.

18
form the stress tensor:  
σ τx y τxz
 xx
σ=

 τy x σy y τy z 
 (31)
τzx τz y σzz

It turns out that, provided certain conditions are met, the stress matrix is symmetric,
i.e. τx y = τ y x , τxz = τzx and τ y z = τz y . We will briefly take a short detour to show that
this is the case, before returning to the effect of these stresses on the control volume
itself.

Aside: Symmetry of the stress tensor Note that the following analysis requires the
absence of “couple stresses”, i.e. there can be no moments per unit volume. From the
conservation of angular momentum, we know that the torque acting on a fluid element
must be equal to the time rate-of-change of the element’s angular momentum. For the
angular momentum in the z-direction:

dω
M= I zz (32)
dt

where I zz is the z-moment of inertia of the fluid element. Consider one of the faces of
the control volume in the x − y plane, which is shown in Figure 8.

τy x

y
x
τx y dy τx y

dx

τy x

Figure 8: Shear stress distributions in the x − y plane.

By summing the contribution from the stress on each side, the torque in the z-direction
is:

dy dx dy dx
M = τ y x d xd z − τx y d yd z + τ y x d xd z − τx y d yd z
2 2 2 2 (33)
= τ y x − τx y d xd yd z
¡ ¢¡ ¢

However, we also know that the z-moment of inertia of the element is:

1 ¢h ¡ ¢2 i
ρ d xd yd z (d x)2 + d y
¡
I zz = (34)
12

19
Substituting Equations 33 and 34 into Equation 32, we find that the rate of change of
angular velocity due to the shear stress:

dω τ y x − τx y d xd yd z
¡ ¢¡ ¢
12
= h (35)
ρ (d x)2 + ¡d y ¢2 ¡d xd yd z ¢
i
dt

As the volume of the element approaches zero, i.e. d xd yd z → 0, we note that the de-
nominator is proportional to the fifth power of the linear dimension whilst the numer-
ator is proportional to the third power of the linear dimension. If the term τ y x − τx y
¡ ¢

were non-zero, the denominator would vanish faster than the numerator and thus d ω/d t
would become infinitely large. Since this is not possible on physical grounds, it follows
that the term τ y x − τx y must be zero. Therefore,
¡ ¢

τ y x = τx y (36)

Using equivalent arguments, it is clear that τz y = τ y z and τzx = τxz . Thus the stress
tensor is symmetric.

Back to forces on the fluid element With a better understanding of the components
of the stress tensor, we return to the stress distributions on the faces normal to the x-,
and y- and z-directions, which are shown in Figure 7. For example, in the x-direction
we have:
¡X ¢ ¡X ¢ ¡X ¢
Fx surface
= Fx normal
+ Fx shear
(37)

We can use Figure 7 to evaluate these two terms. The forces in the x-direction come
from normal stresses on the left and right faces, as well as shear stresses on the other
faces. Remembering to multiply by the area over which each force acts:
   
∂σxx d x ∂σxx d x
F normal = σxx +  − σxx −
¡X ¢
 d yd z
∂x 2 ∂x 2

∂σxx
= d xd yd z
∂x

   
∂τ y x d y ∂τ y x d y (38)
F shear = τ y x +  − τ y x −
¡X ¢
 d xd z
∂y 2 ∂y 2
   
∂τzx d z ∂τzx d z
+ τzx +  − τzx −  d xd y
∂z 2 ∂z 2

∂τ y x ∂τzx
= d xd yd z + d xd yd z
∂y ∂z

20
Substituting these components into the x-component of the overall surface force,
 
¡X ¢ ∂σxx ∂τ y x ∂τzx
Fx = + +  d xd yd z (39)
surface ∂x ∂y ∂z

Similarly, for the other components:

 
¡X ¢ ∂τx y ∂σ y y ∂τz y
Fy = + +  d xd yd z
surface ∂x ∂y ∂z

(40)
 
¡X ¢ ∂τxz ∂τ y z ∂σzz
Fz = + +  d xd yd z
surface ∂x ∂y ∂z

Combining these surface forces with the body forces from Equation 30, we obtain the
total forces acting on the fluid inside the control volume:
 
∂σxx ∂τ y x ∂τzx
= ρg x d xd yd z + 
¡X ¢ ¡X ¢ ¡X ¢
Fx = Fx + Fx + +  d xd yd z
total body surface ∂x ∂y ∂z

 
∂τx y ∂σ y y ∂τz y
= ρg y d xd yd z + 
¡X ¢ ¡X ¢ ¡X ¢
Fy = Fy + Fy + +  d xd yd z
total body surface ∂x ∂y ∂z

 
∂τxz ∂τ y z ∂σzz
= ρg z d xd yd z + 
¡X ¢ ¡X ¢ ¡X ¢
Fz = Fz + Fz + +  d xd yd z
total body surface ∂x ∂y ∂z
(41)

These relations are general and therefore also apply when the fluid is at rest. But when
a fluid is at rest, all the shearing stresses are zero and the normal stresses σxx , σ y y and
σzz are equal to the hydrostatic pressure (−P ). The hydrostatic pressure is a scalar quan-
tity and hence is independent of direction. It is therefore common to separate out the
normal stress components as follows:

σxx = σ′xx − P

σ y y = σ′y y − P (42)

σzz = σ′zz − P

21
Here, σ′xx , σ′y y and σ′zz are called the “deviatoric stresses”. Decomposing the normal
stresses in this way, the overall forces are expressed as:
 
∂P ∂σ′xx ∂τ y x ∂τzx
= ρg x d xd yd z −
¡X ¢
Fx d xd yd z +  + +  d xd yd z
total ∂x ∂x ∂y ∂z

 
∂P ∂τx y ∂σ′y y ∂τz y
= ρg y d xd yd z −
¡X ¢
Fy d xd yd z +  + +  d xd yd z (43)
total ∂y ∂x ∂y ∂z

 
∂P ∂τxz ∂τ y z ∂σ′zz
= ρg z d xd yd z −
¡X ¢
Fz d xd yd z +  + +  d xd yd z
total ∂z ∂x ∂y ∂z

The next step in evaluating the momentum conservation equations is to relate the stresses
to the velocity field. In order to do so, it is necessary to first understand different types
of motion and deformation that fluid elements can undergo.

1.2.3 Motion and Deformation of Fluid Elements

In simple solids, stress is proportional to strain (Hooke’s law). In Newtonian fluids, the
stress is proportional to the rate of strain. Like solid matters, a fluid can undergo differ-
ent types of transformation:

• pure translation;
• extensional (or linear) deformation;
• angular deformation;
• angular rotation.

In general, any transformation of a fluid element can be decomposed into a combina-

tion of these four modes, as illustrated in Figure 9.

general pure extensional angular angular

transformation = translation + deformation + deformation + rotation

= + + +

Figure 9: Decomposition of the general transformation of a fluid element.

Pure translation Pure translation corresponds to all the velocity gradients being zero.
Therefore, the fluid element moves at a constant speed. As a result, there is no deforma-
tion and the fluid element retains its shape (Figure 10).

22
 uδt

A
v
vδt
A
u

Figure 10: Pure translation of a fluid element.

However, when there are changes in velocity components, there are non-zero velocity
gradients and so the other types of deformations can occur.

Extensional (or linear) deformation The diagram in Figure 11 represents the field of
relative velocities when all terms except ∂u/∂x are zero. In this example, we assume also
that ∂u/∂x > 0.

u u + (∂u/∂x) d x
D D′ C C′

A A′ B B′

δx

Figure 11: Extensional deformation in the x-direction of a fluid element.

The relative velocity of point B with respect to A is:

∂u
du = dx (44)
∂x

Therefore in time, δt , point A moves to A′ a distance A A ′ = uδt away. Meanwhile, point

B moves to B′ which is a distance:
 
∂u
B B ′ = (u + d u) δt = u + d x  δt (45)
∂x

The rate of elongation in the x-direction is the proportional change in x-length of the

23
element per unit time:
 
1 A ′ B ′ − AB
ϵ̇xx = lim  
δt →0 δt AB
 
1 AB − A A ′ + B B ′ − AB
= lim  
δt →0 δt AB
 
∂u
 
 d x − uδt + u + d x  δt − d x  (46)
1
 ∂x 

= lim  
δt →0 δt  dx
 

 

∂u
=
∂x

Similarly, the rate of elongation in y- and z-directions is:

∂v
ϵ̇ y y =
∂y
(47)
∂w
ϵ̇zz =
∂z

It is easy to visualise the deformation imparted to the fluid element by the simultaneous
action of all three elongations. As the element expands, there is a change in its volume.
The rate of change of volume, called the volumetric dilatation (θ̇) can be expressed as:
   
∂u ∂v ∂w

 d x + d xδt  d y + d yδt  d z + d zδt  − d xd yd z 
1
 ∂x ∂y ∂z 
θ̇ = lim

 
δt →0 δt  d xd yd z
 
(48)

 

∂u ∂v ∂w
= + +
∂x ∂y ∂z

∂u
For incompressible flows, ∂x + ∂v ∂w
∂y + ∂z = 0 and therefore the volumetric dilatation is
zero, as there is no change in volume. But, for a general fluid, if we define the convective
derivative of density (Dρ/D t ) such that:

Dρ ∂ρ ∂ρ ∂ρ ∂ρ
= +u +v +w (49)
Dt ∂t ∂x ∂y ∂z

then the continuity equation can be written as:

 
Dρ ∂u ∂v ∂w
+ρ + + =0 (50)
Dt ∂x ∂y ∂z

24
Therefore, by rearranging, we get:

∂u ∂v ∂w 1 Dρ
θ̇ = + + =− (51)
∂x ∂y ∂z ρ Dt

This equation suggests that, in a compressible flow, the volumetric dilatation is equal to
the negative of the convective rate of change of the local density.

Angular deformation Now assume that the cross-derivatives of the velocity compo-
nents, such as ∂u/∂y and ∂v/∂x are not zero. In order to quantify this type of deforma-
tion, consider an initially square fluid element, shown in Figure 12 at time, t , and at a
later time, t + δt .

C′

∂u
d yδt
∂y
D′
B′
∂v δβ
dy + d yδt ∂v
∂y d xδt
∂x
uδt δα
A′ ∂u
dx + d xδt
D C ∂x

dy

A B
dx

Figure 12: Deformation of a fluid element from time, t , to time, t + δt .

The fluid element ABCD has been subjected to rotation, extensional deformation, and
angular deformation. In a fluid, the rate of angular deformation (or rate of shear defor-
mation) is defined as the rate of change of the angle between two orthogonal fluid lines.
Therefore, for the two-dimensional element in Figure 12, the rate of angular deforma-
tion, φ̇x y , is given by how quickly the angle at point A (at the intersection of the two fluid
lines) decreases. This angle is initially π/2 but ends up at π/2 − δα − δβ. Thus:

π/2 − π/2 − δα − δβ
¡ ¢
φ̇x y = lim
δt →0 δt
δα + δβ
= lim (52)
δt →0 δt
dα dβ
= +
dt dt

From Figure 12, we can relate δα and δβ to the translational velocity gradients in the

25
flow:

∂v ∂v
d xδt δt ∂v
tan (δα) = lim ∂x = lim ∂x ≈ δt
δt →0 ∂u δt →0 ∂u ∂x
dx + d xδt 1+ δt
∂x ∂x
(53)
∂u ∂u
d yδt δt
∂y ∂y ∂u
tan δβ = lim δt
¡ ¢
= lim ≈
δt →0 ∂v δt →0 ∂v ∂y
dy + d yδt 1+ δt
∂y ∂y

For small angles, tan (δα) ≈ δα and tan δβ ≈ δβ. Therefore,

¡ ¢

dα ∂v
=
dt ∂x
(54)
dβ ∂u
=
dt ∂y

Inserting these expressions into Equation 52, we obtain:

∂u ∂v
φ̇x y = + (55)
∂y ∂x

In a similar way, it can be shown that:

∂v ∂w
φ̇ y z = +
∂z ∂y
(56)
∂w ∂u
φ̇zx = +
∂x ∂z

Note that, for mathematical reasons, the rate of shearing strain is defined as half the
rate of angular deformation:

1
ϵ̇x y = φ̇x y
2
1
ϵ̇ y z = φ̇ y z (57)
2
1
ϵ̇zx = φ̇zx
2

Similar to the stress tensor, there are nine quantities which form the rate of strain tensor:
 
ϵ̇xx ϵ̇x y ϵ̇xz
ϵ̇ = 
 
ϵ̇ y x ϵ̇ y y ϵ̇ y z 
 (58)
ϵ̇zx ϵ̇z y ϵ̇zz

26
This tensor is also symmetric so we can rewrite the elements in the bottom left-hand
corner as:  
ϵ̇xx ϵ̇x y ϵ̇xz
ϵ̇ = 
 
ϵ̇x y ϵ̇ y y ϵ̇ y z 
 (59)
ϵ̇xz ϵ̇ y z ϵ̇zz

In terms of the velocity gradients we have:

    
∂u 1 ∂u ∂v 1 ∂u ∂w
  +   + 

 ∂x 2 ∂y ∂x 2 ∂z ∂x 
    
 1 ∂u ∂v ∂v 1 ∂v ∂w 
 
ϵ̇ = 
 2 ∂y + ∂x
   +  (60)
 ∂y 2 ∂z ∂y 
     
 1 ∂u ∂w 1 ∂v ∂w ∂w
 

  +   +  
2 ∂z ∂x 2 ∂z ∂y ∂z

Note that the on-diagonal terms refer to the extensional deformation of the fluid ele-
ment along each axis. Meanwhile, the off-diagonal terms relate to the angular distor-
tion. In particular, these terms describe the rate of distortion of a right angle located
in a plane defined by the two axes in the subscript, e.g. ϵ̇x y defines how quickly a right
angle in the x − y plane gets distorted.

Angular rotation The final transformation to consider is angular rotation of the fluid
element. The z-component of angular velocity is defined as the average rate of rotation
of the two originally orthogonal lines, i.e. the x- and y-aligned sides in Figure 13.

C′

D′

B′
∂v δβ
dy + d yδt ∂v
∂y d xδt
∂x
δα
∂u
− d yδt A′
∂y
D C
∂u
dx + d xδt
∂x
dy

A B
dx

Figure 13: Deformation of a fluid element from time, t , to time, t + δt , showing the
effects of angular rotation.

By convention a counter-clockwise angular rotation is defined to be in the positive di-

rection. Therefore, in time δt , side AB rotates by δα and side AD rotates by δβ, both in

27
the positive direction. Therefore, the z-component of angular velocity, ωz , is:
   
1 δα + δβ 1 dα dβ
ωz = lim  =  +  (61)
δt →0 δt 2 2 dt dt

Following similar arguments to the angular deformation derivation, we find that:

dα ∂v
=
dt ∂x
(62)
dβ ∂u
=−
dt ∂y

Therefore,  
1 ∂v ∂u
ωz =  −  (63)
2 ∂x ∂y

The x- and y-components of angular velocity are obtained in a similar manner:

 
1 ∂w ∂v
ωx =  − 
2 ∂y ∂z
  (64)
1 ∂u ∂w
ωy =  − 
2 ∂z ∂x

The angular velocity vector is therefore written as:

      
1 ∂w ∂v ∂u ∂w ∂v ∂u
⃗
ω =  − ⃗i + − j +  − ⃗
⃗ k (65)
2 ∂y ∂z ∂z ∂x ∂x ∂y

For mathematical reasons, it is conventional to use the vorticity vector, ⃗

Ω, which is
defined as:       
∂w ∂v ∂u ∂w ∂v ∂u
⃗
Ω = 2⃗
ω =  − i +
⃗ − j +
⃗ − k
⃗ (66)
∂y ∂z ∂z ∂x ∂x ∂y

This analysis completes our introduction to the motion and deformation of fluid ele-
ments, so we can return to our consideration of the forces on a fluid element in analysing
the momentum equations.

1.2.4 Conservation of Momentum 2: Navier–Stokes Equations

The equations for conservation of momentum were derived in Equations 28 and 29.
These equations contain a term for the sum of forces on the fluid element, which was
evaluated in Equation 43. Now that we understand about the transformation of fluid
elements, we can consider the forms of the stress terms in this equation.

By analogy with Hooke elasticity for solids (which states that stress is proportional to
strain), the equivalent assumption for fluids was proposed by Stokes, i.e. that the varia-

28
tion of viscous stress is linearly related with strain rate. In other words, the relationship
between the stress tensor (σ, defined in Equation 31) and the strain rate tensor (ϵ̇, de-
fined in Equation 60) is linear. This assumption leads to the following relations:
 
∂u ∂v ∂w ∂u
σ′xx = λ  + +  + 2µ
∂x ∂y ∂z ∂x
 
∂u ∂v ∂w ∂v
σ′y y = λ  + +  + 2µ
∂x ∂y ∂z ∂y
 
∂u ∂v ∂w ∂w
σ′zz = λ  + +  + 2µ
∂x ∂y ∂z ∂z
  (67)
∂v ∂u
τx y = τ y x = µ  + 
∂x ∂y
 
∂w ∂v
τ y z = τz y = µ  + 
∂y ∂z
 
∂u ∂w
τzx = τxz = µ  + 
∂z ∂x

In these equations, µ is the dynamic viscosity which we are very familiar with. Mean-
while, λ is the second coefficient of viscosity - note that this quantity is only relevant
∂u
for compressible flows (because incompressible flows have ∂x
+ ∂v
∂y
+ ∂w
∂z
= 0 and so all
terms with λ are zero anyway). Note that, in contrast to µ, λ is not a physical property of
the fluid. Nevertheless, the two quantities are mathematically related by the expression,
3λ + 2µ = 0. Substituting this expression into Equations 67:
 
2 ∂u ∂v∂u ∂w
σ′xx = − µ  + +  + 2µ
3 ∂x ∂y ∂z ∂x
 
2 ∂u ∂v ∂w ∂v
σ′y y = − µ  + +  + 2µ (68)
3 ∂x ∂y ∂z ∂y
 
2 ∂u ∂v ∂w ∂w
σ′zz = − µ  + +  + 2µ
3 ∂x ∂y ∂z ∂z

Reintroducing the pressure terms, to go from the deviatoric normal stresses to the phys-

29
ical normal stresses, we obtain the following set of relations for all stress components:
 
2 ∂u ∂u ∂v ∂w
σxx = −P − µ  + +  + 2µ
3 ∂x ∂y ∂z ∂x
 
2 ∂u ∂v ∂w ∂v
σ y y = −P − µ  + +  + 2µ
3 ∂x ∂y ∂z ∂y
 
2 ∂u ∂v ∂w ∂w
σzz = −P − µ  + +  + 2µ
3 ∂x ∂y ∂z ∂z
  (69)
∂v ∂u
τx y = τ y x = µ  + 
∂x ∂y
 
∂w ∂v
τ y z = τz y = µ  + 
∂y ∂z
 
∂u ∂w
τzx = τxz = µ  + 
∂z ∂x

We can now substitute these stress terms into the equations for force (Equation 43)
and then, in turn, substitute the expression for force into the momentum conservation
equations (Equations 28 and 29) to obtain the final form of the momentum conservation
equations (in the x-, y- and z-directions, respectively), i.e. the Navier–Stokes equations:
  
∂ρu ∂ρu 2 ∂ρuv ∂ρuw ∂P ∂ ∂u 2 ∂u ∂v ∂w
+ + + =ρg x − + 2µ − µ  + + 
∂t ∂x ∂y ∂z ∂x ∂x ∂x 3 ∂x ∂y ∂z
     
∂ ∂v ∂u ∂ ∂u ∂w
+ µ  +  + µ  + 
∂y ∂x ∂y ∂z ∂z ∂x

  
∂ρv ∂ρuv ∂ρv 2 ∂ρv w ∂P ∂ ∂v ∂u
+ + + =ρg y − + µ  + 
∂t ∂x ∂y ∂z ∂y ∂x ∂x ∂y
     
∂ ∂v 2 ∂u ∂v ∂w ∂ ∂w ∂v
+ 2µ − µ + +  + µ  + 
∂y ∂y 3 ∂x ∂y ∂z ∂z ∂y ∂z

  
∂ρw ∂ρuw ∂ρv w ∂ρw 2 ∂P ∂ ∂u ∂w
+ + + =ρg z − + µ  + 
∂t ∂x ∂y ∂z ∂z ∂x ∂z ∂x
     
∂ ∂w ∂v ∂ ∂w 2 ∂u ∂v ∂w
+ µ  +  + 2µ − µ + + 
∂y ∂y ∂z ∂z ∂z 3 ∂x ∂y ∂z
(70)

∂ρ ∂ρu ∂ρv ∂ρw

Combined with the continuity equation (Equation 18: ∂t
+ ∂x
+ ∂y
+ ∂z
= 0), these

30
form the fundamental equations of motion for a viscous compressible flow.

Note that these equations can sometimes be simplified depending on the properties of
the flow field under consideration. For example, consider a case where the flow is steady
(i.e. all time-derivatives are zero) and incompressible (i.e. density is constant). Here, the
∂u
continuity equation becomes ∂x
+ ∂v
∂y
+ ∂w
∂z
= 0 (Equation 19) and Equation 70 reduces to
the steady, incompressible Navier–Stokes equations:
 
∂u ∂u ∂u 1 ∂P
µ ∂2 u ∂2 u ∂2 u
u +v +w = gx − +  + + 
∂x ∂y ∂z ρ ∂x ρ ∂x 2 ∂y 2 ∂z 2

 
∂v ∂v ∂v 1 ∂P
µ ∂2 v ∂2 v ∂2 v
u +v +w = gy − +  + +  (71)
∂x ∂y ∂z ρ ∂y ρ ∂x 2 ∂y 2 ∂z 2

 
∂w ∂w ∂w 1 ∂P
µ ∂2 w ∂2 w ∂2 w
u +v +w = gz − +  + + 
∂x ∂y ∂z ρ ∂z ρ ∂x 2 ∂y 2 ∂z 2

Conservation equations in cylindrical coordinates In many engineering problems,

the solution to conservation equations can sometimes be simplified by choosing a dif-
ferent coordinate system instead of the Cartesian one we have used so far. In particular,
the coordinate system should be chosen to take advantage of the symmetry properties
intrinsic to the particular problem under consideration. For example, if the problem
contains cylindrical symmetry, it makes sense to use the cylindrical coordinate system
shown in Figure 14.

x θ

Figure 14: Definition of cylindrical coordinate system.

Here z is the vertical coordinate, r is the distance in the x − y plane away from the z axis,
and θ is the counter-clockwise azimuthal angle in the x −y plane from the x-axis. There-
fore the relationship between the Cartesian and cylindrical coordinates is as follows:

x = r cos θ
y = r sin θ (72)
z=z

31
 ⃗ = vr ⃗
In this coordinate system, the velocity vector is defined as V e r +v θ⃗
e θ +v z⃗
e z . There-
fore, the continuity equation becomes:

∂v r vr 1 ∂v θ ∂v z
+ + + =0 (73)
∂r r r ∂θ ∂z

The momentum equation in the r -direction is:

 
∂v r ∂v r v θ ∂v r v θ2 ∂v r ∂P
ρ + vr + − + vz  =ρg r −
∂t ∂r r ∂θ r ∂z ∂r
 
∂2 v r 1 ∂v r vr 1 ∂2 v r 2 ∂v θ ∂2 v r
+µ + − + − + 
∂r 2 r ∂r r2 r 2 ∂θ 2 r 2 ∂θ ∂z 2
(74)

The momentum equation in the θ-direction is:

 
∂v θ ∂v θ v θ ∂v θ vr vθ ∂v θ 1 ∂P
ρ + vr + + + vz  =ρg θ −
∂t ∂r r ∂θ r ∂z r ∂θ
 
∂2 v θ 1 ∂v θ vθ 1 ∂2 v θ 2 ∂v r ∂2 v θ
+µ + − + + + 
∂r 2 r ∂r r2 r 2 ∂θ 2 r 2 ∂θ ∂z 2
(75)

The momentum equation in the z-direction is:

   
∂v z ∂v z v θ ∂v z ∂v z ∂P ∂2 v z 1 ∂v z 1 ∂2 v z ∂2 v z
ρ + vr + + vz  = ρg z − +µ + + +  (76)
∂t ∂r r ∂θ ∂z ∂z ∂r 2 r ∂r r 2 ∂θ 2 ∂z 2

For reference, these equations were derived using the following strain rate components:

∂v r
ϵ̇r r =
∂r
1 ∂v θ vr
ϵ̇θθ = +
r ∂θ r
∂v z
ϵ̇zz =
∂z
 
1 1 ∂v z ∂v θ (77)
ϵ̇θz =  + 
2 r ∂θ ∂z
 
1 ∂v r ∂v z
ϵ̇r z =  + 
2 ∂z ∂r
 
1 1 ∂v r ∂v θ v θ
ϵ̇r θ =  + − 
2 r ∂θ ∂r r

32
Similarly, the vorticity components are as follows:

1 ∂v z ∂v θ
Ωr = −
r ∂θ ∂z
∂v r ∂v z
Ωθ = − (78)
∂z ∂r
1 ∂ (r v θ ) 1 ∂v r
Ωz = −
r ∂r r ∂θ

1.2.5 Exact Solutions of the Navier–Stokes Equations

The Navier–Stokes equations in their complete form are nonlinear and no known an-
alytical solution exists for the general case. In certain particular cases, however, it is
possible to simplify the equations by making assumptions about the fluid, the flow or
the geometry in order to obtain analytical solutions. We will consider some of these
cases where such analytical solutions exist.

Parallel, steady, two-dimensional, incompressible flow through a straight channel

Consider a steady, two-dimensional, incompressible flow between two infinitely-large
parallel plates (Figure 15). Since flow is incompressible (ρ constant) and steady (time-
derivatives zero), we can use the incompressible continuity equation (Equation 19) and
the steady, incompressible Navier–Stokes equations (Equation 71). Since, the flow is also
two-dimensional, w = 0, and thus the continuity and momentum equations become:

∂u ∂v
continuity: + =0
∂x ∂y
 
∂u ∂u ∂P ∂2 u ∂2 u
x-momentum: ρu + ρv =−  +µ + (79)
∂x ∂y ∂x
∂y 2 ∂x 2
 
∂v ∂v ∂P ∂2 v ∂2 v
y-momentum: ρu + ρv =− + µ  2 + 2
∂x ∂y ∂y ∂x ∂y

y = +h

y
u(y) 2h
x

y = −h

Figure 15: Parallel, steady, two-dimensional, incompressible flow in a straight channel.

Note that there is no z-momentum equation since there is no flow in this direction.
Since the flow is symmetric and constrained by two parallel walls, no vertical compo-

33
nent of velocity is possible, and thus v = 0. In turn, this implies that the gradients of v
are also equal to zero:
∂v ∂v ∂2 v ∂2 v
= = = =0 (80)
∂x ∂y ∂x 2 ∂y 2
Setting v and its gradients to zero in Equation 79, we obtain:

∂u ∂v
continuity: =− =0
∂x ∂y
 
∂u ∂P ∂2 u ∂2 u
x-momentum: ρu =− +µ +  (81)
∂x ∂x ∂x 2 ∂y 2

∂P
y-momentum: =0
∂y

Substituting the relation from the continuity equation (∂u/∂x = 0) into the x-momentum
equation, we further obtain:
∂2 u ∂P
µ = (82)
∂y 2 ∂x
The continuity equation in Equation 81 tells us that u does not depend on x and is only
a function of y. Therefore, ∂2 u/∂y 2 is only a function of y. Considering Equation 82, it
must therefore be true that ∂P /∂x is also only a function of y. However, since ∂P /∂y = 0
(Equation 81), P is not a function of y and so the pressure gradient, ∂P /∂x, must be a
constant. With this knowledge it is possible to integrate Equation 82 twice with respect
to y:
1 ∂P
u(y) = y 2 + Ay + B (83)
2µ ∂x
where A and B are constants of integration. To evaluate these constants, it is necessary
to apply the necessary boundary conditions, namely the no-slip boundary conditions
at the upper and lower walls (u = 0 at y = ±h). We therefore obtain:

1 ∂P
h 2 + Ah + B = 0
2µ ∂x
(84)
1 ∂P
(−h)2 + A(−h) + B = 0
2µ ∂x

and thus:

A=0
h 2 ∂P (85)
B =−
2µ ∂x

34
Substituting these constants into Equation 83 gives the velocity distribution:
 2 
h 2 ∂Py
u(y) =   − 1  (86)
2µ ∂x h

From this velocity distribution, we can also calculate the shear stress at the lower wall
using Newton’s law of friction:
¯
∂u ¯
¯
τ y=−h = µ ¯
∂y ¯¯
y=−h
  
h 2 ∂P 2y
= µ   (87)
2µ ∂x h 2
y=−h

∂P
= −h
∂x

Couette flow The flow scenario is analogous to the previous example (Figure 15) but
the top wall is moving to the right at speed U . Since this is still a parallel, steady, incom-
pressible, two-dimensional flow, the governing equations again reduce to:

∂2 u ∂P
µ =
∂y 2 ∂x
(88)
∂P
=0
∂y

and thus, the velocity profile again takes the form:

1 ∂P
u(y) = y 2 + Ay + B (89)
2µ ∂x

In this case, however, the no-slip conditions at both walls produce a different set of
boundary conditions:

u =U at y = +h
(90)
u=0 at y = −h

By implementing these boundary conditions to determine the constants A and B :

1 ∂P
h 2 + Ah + B = U
2µ ∂x
(91)
1 ∂P
(−h)2 + A(−h) + B = 0
2µ ∂x

35
and thus:

U
A=
2h
(92)
U h 2 ∂P
B= −
2 2µ ∂x

Therefore, a different velocity profile emerges:

 2   
h 2 ∂P y U y
u(y) =   − 1  +  + 1  (93)
2µ ∂x h 2 h

Note that, in the case where there is no streamwise pressure gradient (∂P /∂x = 0), this
is commonly referred to as Couette flow and the velocity distribution reduces to:
 
U y
u(y) =  + 1 (94)
2 h

In this case, the shear stress at the lower wall is:

¯
∂u ¯ µU
¯
τ y=−h = µ ¯ = (95)
∂y ¯¯ 2h
y=−h

Flow through a straight, circular pipe Consider the flow through a straight, circular
pipe which is shown in Figure 16. Due to the symmetry properties of this flow, it makes
sense to use cylindrical coordinates. Here, z is the axial direction along the pipe and r is
the radial coordinate which goes from 0 at the centre to R at the pipe walls. We neglect
body forces and consider the case where the flow is steady.

r =R

r
2R
z

v z (r )

Figure 16: Flow through a straight, circular pipe.

Since the flow is only in the axial direction, we know that the other velocity components
are zero, v r = v θ = 0. Setting these velocity terms to zero in the cylindrical conservation

36
equations (Equations 73–76), we obtain:

∂v z
continuity: =0
∂z
∂P
r -momentum: =0
∂r
(96)
∂P
θ-momentum: =0
∂θ  
∂v z ∂P ∂2 v z 1 ∂v z 1 ∂2 v z ∂2 v z
z-momentum: ρv z =− +µ + + + 
∂z ∂z ∂r 2 r ∂r r 2 ∂θ 2 ∂z 2

Since v z is only a function of r , the z-momentum equation reduces to:

 
∂P ∂2 v z 1 ∂v z
− +µ + =0
∂z ∂r 2 r ∂r

∂2 v z
1 ∂v z 1 ∂P
=⇒ + = (97)
∂r 2
r ∂r µ ∂z
 
1d d vz 1 ∂P
=⇒ r =
r dr dr µ ∂z

The left-hand side of the equation is only a function of r , and therefore the right-hand
side must also be a function only of r . However, the r -momentum equation tells us that
∂P /∂r = 0, so P is not a function of r and therefore ∂P /∂z cannot be a function of r .
As a result, the right-hand side of the equation must be a constant. Therefore we can
integrate to find the velocity profile:

d vz r 2 ∂P
r = +A
dr 2µ ∂z
(98)
r 2 ∂P
=⇒ v z = + A ln r + B
4µ ∂z

where A and B are constants of integration. Since the velocity on the centre line (r = 0)
cannot be infinite, it must be true that A = 0. At r = R, the no-slip boundary condition
requires v z = 0 and so:
R 2 ∂P
B =− (99)
4µ ∂z
Therefore, the velocity profile, which is commonly known as Hagen–Poiseuille flow is
given by:
1 ∂P ¡
r 2 − R2
¢
vz = (100)
4µ ∂z

Flow between two concentric long circular pipes Consider the flow between two con-
centric, long cylinders, as shown in Figure 17. We define the inner and outer radii of

37
cylinders to be R i and R o , respectively. We also neglect body forces.

r v z (r )
z

r = Ri

r = Ro

Figure 17: Flow between two concentric cylinders.

Similar to the previous example, the only non-zero component of flow velocity is v z (r ).
Therefore, the Navier–Stokes equations in cylindrical coordinates reduce to:

∂2 v z 1 ∂v z 1 ∂P
+ = (101)
∂r 2 r ∂r µ ∂z

Integration, leads to the same general velocity profile:

r 2 ∂P
vz = + A ln r + B (102)
4µ ∂z

but now, the no-slip boundary conditions are: v z = 0 at r = R i and v z = 0 at r = R o .

These conditions lead to the following two equations to determine the constants:

R i2 ∂P
+ A ln R i + B = 0
4µ ∂z
(103)
R o2 ∂P
+ A ln R o + B = 0
4µ ∂z

Solving for A and B , we obtain:

 
1 ∂P R o2 − R i2
A=−  
4µ ∂z ln (R o /R i )
¡  (104)
1 ∂P R o2 − R i2 ln R i
¢
B=  − R i2 
4µ ∂z ln (R o /R i )

Therefore, we obtain the following relation for the velocity profile in the region from
r = R i to r = R o :  
1 ∂P ¡ R o2 − R i2
 r 2 − R2 −
¢
vz = i ln (r /R i ) (105)
4µ ∂z ln (R o /R i )

38
1.2.6 Limiting Cases of the Navier–Stokes Equations

It has already been mentioned that general analytical solutions for the Navier–Stokes
equations only exist in very special simple cases, such as the previous examples. How-
ever, useful approximations can be obtained by noting that the Navier–Stokes equations
comprise of two parts, a viscous part (which depends on the fluid viscosity) and invis-
cid part (which is viscosity-independent). Therefore, a natural approach to tackling the
equations is to consider two limiting cases: (1) very small viscosity values, and (2) very
large viscosity values. For case 1 (very low viscosity, e.g. air or water), when we ignore the
viscous terms, we will see that the equations reduce to the inviscid or “Euler” equations.
In case 2 (very viscous flows), we only need to worry about the viscous terms and can
neglect all other terms. By retaining only viscous terms, we will see that the equations
reduce to a set of linear second-order partial differential equations.

Suppose that, for a given flow condition, we want to study the importance of different
terms in the Navier–Stokes equations for a series of geometrically-similar objects which
have different scales. Similarly, for a given object, we might want to study the behaviour
of the equations for flows which have different speeds but are otherwise dynamically
similar, i.e. the flow field “looks the same” apart from the scaled velocity vectors. In order
to achieve these goals, it is necessary to make each flow quantity dimensionless with
respect to some reference condition. Let us define our reference parameters to be: L ref
for length, Uref for velocity, P ref for pressure, µref for viscosity, g ref for body force per unit
mass, t ref for time, and ρ ref for density. We can therefore define our non-dimensional
quantities as follows:

x y z
x= , y= , z=
L ref L ref L ref
u v w
u= , v= , w=
Uref Uref Uref
P µ t g ρ
P= , µ= , t= , g= , ρ=
P ref µref t ref g ref ρ ref

Now, let us consider the x-component of the unsteady, incompressible Navier–Stokes

equations:
 
∂u ∂u ∂u ∂u 1 ∂P
µ ∂2 u ∂2 u ∂2 u
+u +v +w = gx − +  + +  (106)
∂t ∂x ∂y ∂z ρ ∂x ρ ∂x 2 ∂y 2 ∂z 2

In order to “non-dimensionalise” the equation, it is necessary to replace each flow quan-

39
tity (u, t , x, etc.) with the corresponding non-dimensional form (e.g. replace u by uUref ):
     
Uref ∂u 2
Uref ∂u ∂u ∂u £ ¤ P ref 1 ∂P
 + u +v + w  =g ref g x −  
t ref ∂t L ref ∂x ∂y ∂z ρ ref L ref ρ ∂x
  
µrefUref µ ∂2 u ∂2 u ∂2 u
+   + + 
ρ ref L 2ref ρ ∂x 2 ∂y 2 ∂z 2

     
L ref ∂u ∂u ∂u ∂u g ref L ref £ ¤ P ref 1 ∂P
=⇒   + u +v +w = gx −  
t refUref ∂t ∂x ∂y ∂z 2
Uref ρ refUref
2 ρ ∂x
  
µref µ ∂2 u ∂2 u ∂2 u
+   + + 
ρ ref L refUref ρ ∂x 2 ∂y 2 ∂z 2
(107)

Consider two different flow scenarios where the geometry is the same and all the co-
efficients outside the square brackets (which are now non-dimensional) hold the same
values. In such a case, we would be solving the same partial differential equation in the
same domain with the same boundary conditions, and so the resulting flow fields would
be dynamically similar. Let us consider each of the coefficients in turn.

(1) The first coefficient on the left=hand side (L ref /t ref ·Uref ) can be expressed in terms of
a reference frequency, f ref , instead of the reference time as L ref f ref /Uref . This coefficient
is called the Strouhal number, Sr = L ref f ref /Uref , which is a measure of the characteristic
non-dimensional frequency of the unsteady flow.

2
(2) The first coefficient on the right-hand side (g ref L ref /Uref ) is the reciprocal of the square
p
of the Froude number, Fr = Uref / g ref L ref , which is a measure of how important inertial
effects are compared to gravitational effects.

2
(3) The second coefficient on the right-hand side (P ref /ρ ref ·Uref ) is known as the Euler
2
number, Eu = P ref /ρ ref ·Uref , which is a measure of how important pressure effects are
compared to inertial effects.

(4) The final coefficient on the right-hand side (µref /ρ ref · L ref ·Uref ) is the reciprocal of
the Reynolds number, Re = ρ ref L refUref /µref , which is a measure of how important iner-
tial effects are compared to viscous effects.

Therefore, we can write the unsteady, incompressible Navier–Stokes equation as:

     
  
∂u ∂u ∂u ∂u 1 £ ¤ 1 ∂P µ ∂2 u ∂2 u ∂2 u
1
Sr   + u +v +w = gx − Eu  +   + + 
∂t ∂x ∂y ∂z Fr2 ρ ∂x Re ρ ∂x 2 ∂y 2 ∂z 2
(108)

40
where

L ref f ref Uref P ref ρ ref L refUref

Sr = , Fr = p , Eu = , Re =
Uref g ref L ref ρ refUref
2 µref

Whilst it might be almost impossible to produce dynamically similar flows for the gen-
eral case (where all four parameters are at play), it is useful to consider those non-
dimensional numbers that have the most influence on the flow. For example, in a steady
flow the Strouhal number is not relevant. Similarly, in flows with no free surface, gravity
does not play a role and so the Froude number is not a significant parameter. For in-
compressible flows, density is constant and so a change in pressure does not change the
volume; therefore, the Euler number (which is related to Mach number) does not play
a role. On the other hand, the Reynolds number (the ratio of inertial force to viscous
force) can almost never be ignored – if we want to obtain two dynamically similar in-
compressible flows, the Reynolds number must be equal. Consider the x-component
of non-dimensional Navier–Stokes equation for an incompressible, steady scenario with
no free surface:  
∂u ∂u 1 µ ∂2 u ∂2 u ∂2 u
∂u
u +v +w =  + +  (109)
∂x ∂y ∂z Re ρ ∂x 2 ∂y 2 ∂z 2

At the start of the section, we discussed that it is common to examine the properties
of the Navier–Stokes equations in two extreme cases: case 1 where the fluid viscosity
is very large and case 2 where the viscosity is very small. We are now able to use non-
dimensional numbers to specify what precisely we mean by this: case 1 corresponds
to flows with very small Reynolds number and case 2 represents flows with very large
Reynolds number.

Flows with very small Reynolds number In cases with very large viscosity or very slow
flow, Re ≪ 1 which means that the viscous forces are considerably greater than the in-
ertial forces. Therefore, 1/Re → ∞ and to first order, it is possible to neglect the inertia
term (i.e. the left-hand side of Equation 109) as they are much smaller than the corre-
sponding viscous terms. We therefore obtain:

∂2 u ∂2 u ∂2 u
+ + =0 (110)
∂x 2 ∂y 2 ∂z 2

This is a linear second-order equation, which is considerably more amenable to math-

ematical treatment than the complete equation. Cases which satisfy this equation are
called creeping flows. Note that the omission of inertia terms is permissible from the
mathematical point of view because the order of the equation is not reduced, so that
the simplified equation (Equation 110) is able to satisfy as many boundary conditions
as the full equation (Equation 109). Figure 18 shows some experimental examples of
flows around objects at very low Reynolds number. Note that all cases faithfully rep-
resent the unseparated “creeping flow”, with the exception of small regions around the
stagnation points.

41
a) cylinder b) airfoil

c) rectangular block

Figure 18: Examples of low Reynolds number flows around different objects, visualised
using dye oil-flow visualisation. Source unknown.

Flows with very large Reynolds number The two most common fluids, air and water,
have relatively low viscosity. In addition, many flows of interest in engineering have
reasonably high velocities. The combination of very low viscosity and relatively high
velocity occurs when the Reynolds number is very high (Re ≫ 1, or 1/Re → 0). As a result,
for many practical applications, it is worth considering the Navier–Stokes equations in
their limit of high Reynolds number. Under such conditions, the viscous terms on the
right-hand side of Equation 109 can be neglected and we retain only the intertial terms
on the left-hand side of the equation. This results in the Euler equations:

∂u ∂u ∂u
u +v +w =0 (111)
∂x ∂y ∂z

Similar equations also exist which include the unsteady, pressure, or gravity terms from
the complete Navier–Stokes equation. However, it has been observed that the Euler
equations correspond to very poor prediction of an object’s drag. Mathematically, it is
worth noting that the omission of the viscous terms reduces the differential equation
from second order to first order, and so the solution of the simplified equation cannot
be made to satisfy the full boundary conditions of the original equation.

In order to illustrate the behaviour of the equation’s solutions at high Reynolds number,
it is useful to consider the temperature distribution around a hot body in a stream of
fluid (Figure 19). In the limiting case of zero velocity, where the fluid is at rest, the influ-
ence of the heated body will be approximately uniform on all sides. When a very small
velocity is applied from left to right, the fluid around the body still experiences an in-
creased temperature in all directions, although there is a bias towards the right (region
a). When the flow velocity is increased further, however, it is clear that the region af-
fected by the hot body shrinks into a narrow zone in the immediate vicinity of the body
and into a tail of heated fluid behind it (region b).

42
 a) temperature increase at low velocity

b) temperature increase at high velocity

Figure 19: Regions of increased temperature around a hot body at different velocities.

In summary, at small velocities the effect of the heated body, driven by the temperature
gradient, is felt by the whole region of flow around the body. On the other hand, for
large velocities the effect of the heated body is confined to regions near the body and
in its wake, whilst the rest of the flow field experiences zero temperature gradient and
remains at its original temperature.

An analogous argument holds for viscous effects. At low Reynolds number the effect of
friction at the wall surface can be felt in the whole region of the flow. As the Reynolds
number is increased, the effect of friction confined to a small (but still finite) layer along
the body surface, as well as the wake behind the object. Meanwhile, the remainder of
the flow field remains practically frictionless or inviscid. Therefore, in the high Reynolds
number limiting case, the flow field can be subdivided to two regions: (1) an external
region which is free from friction, and where the flow satisfies the Euler equations for
inviscid flow, and (2) an inner region very close to the body where the viscous terms
cannot be ignored. In this inner region, called the boundary layer, the complete Navier–
Stokes equations must therefore be used in order to correctly capture the flow.

Note that the reason that the Euler equations (on their own) do not accurately predict
drag is precisely because they are purely inviscid and so do not account for the impor-
tant effects of the boundary layers. In short, although the viscosity is very small at high
Reynolds numbers, this does not mean that we can completely ignore the viscous terms
in the Navier–Stokes equations. The subdivision of the flow field into the inviscid ex-
ternal flow and the viscous near-wall flow significantly reduces the mathematical diffi-
culties inherent to the Navier–Stokes equations. In particular, for cases where the thick-
ness of the inner region is small, the Navier-Stokes equations can be simplified to the
boundary-layer equations which are more amenable to mathematical treatment.

1.3 Boundary-Layer Flows

We have already seen that many practical problems in engineering involve high Reynolds
number flows (Re ≫ 1), where the flow field can be divided into two regions:

43
 • an outer, inviscid region where the velocity gradient is very small and the influ-
ence of viscosity is unimportant, so we can assume frictionless (or potential) flow;

• the boundary layer, which is a very thin inner region in the immediate vicinity of
the solid boundary where the velocity gradient normal to the wall is so large that
the shear stress, τ = µ ∂u/∂n, cannot be ignored.

Figure 20 shows flow over a flat plate. Due to the no-slip condition, the flow sticks to
the plate at y = 0 and viscosity causes a boundary layer to develop along the plate. The
velocity within the boundary layer varies from zero at wall, to free-stream velocity at
large distances from the wall. One common definition of the physical thickness of the
boundary layer, δ, is the distance where the velocity becomes 0.99 (or 0.995) of the local
free-stream velocity.

u u
boundary-layer edge

y
x

flat plate

Figure 20: Boundary-layer development over a flat plate.

The presence of the boundary layer affects the wider flow in a number of ways:

• friction at the wall caused by the velocity gradient in a viscous flow;

• displacement of external streamlines due to the reduction in mass flow in the

low-velocity boundary layer;

• reduction in momentum due to the lower velocity in the boundary layer;

We will now consider each of these effects in turn

Friction at the wall Inside the boundary layer, the product µ ∂u/∂y is non-zero and so
there is shear stress in this region. The shear stress is distributed such that its maximum
is at the solid boundary (where the velocity gradient is large) and it is zero at the edge
of the boundary layer (where there is no velocity gradient). At the solid boundary, the
shear stress is called the “skin friction” or wall shear stress, τw :
 
∂u
τw = µ   (112)
∂y
y=0

44
This physical quantity is converted into a non-dimensional form by dividing by the dy-
namic pressure to define the skin-friction coefficient:

τw
Cf = 1
(113)
ρ U2
2 e e

Here, ρ e and Ue are the density and the velocity, respectively, at the edge of the boundary
layer in the free stream.

Displacement of external streamlines Due to the formation of the boundary layer,

the mass flow rate per unit area is reduced. Therefore, the external streamlines in the
free stream are displaced outwards to compensate for the reduction in mass. In order
to estimate the extent of this displacement, consider the control volume ABCD in Fig-
ure 21.

u
external streamline C

H
B boundary-layer edge

h y δ
x drag

A D
L flat plate

Figure 21: Control volume for analysis of boundary layer on a flat plate.

As the boundary layer grows to thickness, δ, over distance, L, the height of the control
volume increases from h to H . Side AD is a solid wall and side BC is a displaced stream-
line, so no mass flows through these sections. The mass therefore enters through sides
AB and leaves through side CD. For steady flows, the conservation of mass can be writ-
ten as: Z B Z D
ρ y u y dy = ρ y u y dy
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
(114)
A C

Recalling that ρ e and Ue are the density and the velocity in the free stream (all of AB and
the section of CD outside the boundary layer), the equation reduces to:
Z h Z δ Z H
ρ e Ue d y = ρu d y + ρ e Ue d y (115)
0 0 δ

Therefore, Z δ
hρ e Ue = ρu d y + (H − δ) ρ e Ue (116)
0
Z δ Z δ¡
=⇒ (H − h) ρ e Ue = δρ e Ue − ρu d y = ρ e Ue − ρu d y
¢
(117)
0 0

45
We define the boundary-layer “displacement thickness”, δ∗ , to be the distance by which
an external streamline is displaced in the y-direction due to the presence of the bound-
ary layer. If the boundary layer did not exist, the external streamline would simply be at
height, h, and so it has been displaced by H − h.
Rδ¡  
0 ρ e Ue − ρu d y ρu
¢ Z δ
δ∗ = H − h = = 1 −  dy (118)
ρ e Ue 0 ρ e Ue

Note that for an incompressible flow, ρ = ρ e and so:

 
Z δ u
δ∗i = 1 −  dy (119)
0 Ue

The displacement thickness, δ∗ , is used as an alternative measure of the boundary-

layer thickness, although note that it is typically an order of magnitude smaller than
the boundary-layer thickness, δ.

Reduction in momentum Due to the formation of the boundary layer, the velocity in
this region is reduced and so the momentum flow rate decreases. The rate of momen-
tum per unit width of a fluid element with thickness δy is:

ρ y u y δy × u y
¡ ¢ ¡ ¢ ¡ ¢
(120)

If there had been no boundary layer, the momentum per unit width of the same element
of fluid (with mass ρu δy) would be:

ρu δy ×Ue (121)

Therefore, the reduction (or deficit) in momentum per unit width of this fluid element
due to the presence of the boundary layer is:

ρu δy × (Ue − u) (122)

Therefore, the total reduction (or deficit) in momentum per unit width due to the pres-
ence of the boundary layer is the integral over all such fluid elements in the y-direction:
Z ∞
ρu × (Ue − u) d y (123)
0

Outside the boundary layer (y = δ → ∞), u = Ue and so the term in brackets evaluates to
zero. Therefore, there is only a contribution from y = 0 → δ. The momentum deficit per
unit width is: Z δ
ρu × (Ue − u) d y (124)
0

We now define the “momentum thickness”, θ, as the distance which (if it contained flow
at ρ e all travelling at the free-stream velocity, Ue ) would account for the momentum

46
deficit. The momentum per unit width through such a region is:

ρ e ×Ue2 × θ (125)

Equating this expression to the momentum deficit:

Z δ
ρ e Ue2 θ = ρu (Ue − u) d y (126)
0
 
1 Z δ Z δ ρu u
=⇒ θ = ρu (Ue − u) d y = 1 −  dy (127)
ρ e Ue2 0 0 ρ e Ue Ue

This distance, θ, is equivalent to the distance a wall would have to be move such that
the same total momentum exists between this wall and an external reference plane for
an inviscid fluid. Note that for an incompressible flow, ρ = ρ e and so:
 
Z δ u u
θi = 1 −  dy (128)
0 Ue Ue

The momentum thickness can also be related to the drag of a flat plate with zero ex-
ternal pressure gradient. This drag is all due to wall friction, so the total drag per unit
width on one side of a plate of length L is:
Z L
drag = τw (x) d x (129)
0

By considering the momentum equation in the x-direction for the control volume in
Figure 21:
Z h Z H Z H
ρ e Ue2 d y − 2
ρu d y = ρ e Ue2 h − ρu 2 d y
X
drag = − Fx = (130)
0 0 0

Noting that mass conservation requires

Z H Z H ρu
hρ e Ue = ρu d y =⇒ h = dy (131)
0 0 ρ e Ue

we obtain
Z H ρu Z H
drag = ρ e Ue2 dy − ρu 2 d y (132)
0 ρ e Ue 0
 
drag Z H ρu Z H ρu 2 Z H ρu u
=⇒ = dy − dy = 1 −  dy =θ (133)
ρ e Ue2 0 ρ e Ue 0 ρ e Ue2 0 ρ e Ue Ue

We therefore have three measures of boundary-layer thickness, the physical thickness

(δ) where the velocity is 0.99 or 0.995 of the free-stream value, as well as the displace-
ment thickness (δ∗ ) and the momentum thickness (θ) . It is also common to define the

47
ratios of these thicknesses by defining η = y/δ:
 
δ∗ Z 1 ρu
= 1 −  dη (134)
δ 0 ρ e Ue

 
θ Z 1 ρu u
= 1 −  dη (135)
δ 0 ρ e Ue Ue

A particularly important quantity in boundary-layer theory is the ratio between dis-

placement thickness and momentum thickness, which is called the shape factor, H :

δ∗
H= (136)
θ

This quantity, which is always greater than 1, is important when considering the transi-
tion and separation of boundary layers, which we shall discuss in due course. For ref-
erence, the value of the shape factor is typically around 2.5 for laminar boundary layers
and between 1.2 and 1.6 for turbulent boundary layers.

1.3.1 Thin Boundary-Layer Equations

Consider the simple case of an incompressible, steady, two-dimensional flow in the ab-
sence of body forces. The conservation equations (continuity, x-momentum, y-momentum)
then become:

∂u ∂v
+ =0
∂x ∂y
 
∂u ∂u µ ∂2 u ∂2 u 1 ∂P
u +v =− +  + 
(137)
∂x ∂y ρ ∂x ρ ∂x 2 ∂y 2
 
∂v ∂v 1 ∂P µ ∂2 v ∂2 v
u +v =− +  + 
∂x ∂y ρ ∂y ρ ∂x 2 ∂y 2

Outside the boundary layer, where the flow is treated as inviscid, the continuity equa-
tion remains the same but the x- and y-momentum equations reduce to the Euler equa-
tions:

∂u ∂v
+ =0
∂x ∂y
∂u ∂u 1 ∂P
u +v =− (138)
∂x ∂y ρ ∂x
∂v ∂v 1 ∂P
u +v =−
∂x ∂y ρ ∂y

Inside the boundary layer, we cannot ignore the viscous terms and thus it is necessary
to solve the conservation equations in their complete form. However, at high Reynolds

48
number, the thickness of the boundary layer is relatively small and at large distances,
L, from the leading edge of the plate (where the boundary layer starts to grow), we can
assume that boundary layer is thin (i.e. δ/L ≪ 1). Under these assumptions, the con-
servation equations can be further simplified by ignoring some of the terms in the mo-
mentum equations. This approach uses an “order of magnitude” argument in order to
compare different terms in momentum equations. For example, the streamwise veloc-
ity is much larger than any wall-normal velocity component. In addition, the velocity
gradients in the wall-normal y-direction are much larger than the equivalent gradients
in the x- and z-directions, so the y-gradient terms dominate. These assumptions result
in the following thin boundary-layer equations:

∂u ∂v
+ =0
∂x ∂y
∂u ∂u 1 ∂P µ ∂2 u
u +v =− + (139)
∂x ∂y ρ ∂x ρ ∂y 2
∂P
=0
∂y

These equations are valid for thin boundary layers at locations significantly downstream
of the boundary-layer starting point. The simplifications have achieved the following:

• out of the three conservation equations, the y-momentum equation has effec-
tively disappeared;
• the pressure gradient in the normal (y-)direction is negligible, and can be evalu-
ated from the inviscid flow outside the boundary layer, so it is no longer an un-
known quantity;
• the number of unknowns has been reduced from three (P , u and v) to two (u and
v);
• the order of the Navier–Stokes equations has been preserved such that all the
physical boundary conditions can be satisfied.

The fact that the pressure is constant in the wall-normal direction is a key result. This
observation means that the boundary-layer development is influenced by the external,
inviscid pressure gradient. The external pressure gradient has a very important effect on
the boundary layer development. Let us consider how different pressure gradients affect
the development of the boundary layer. On an airfoil, for example, Figure 22 shows
that the streamline curvature imposed by the geometry causes the boundary layer to be
subjected to a favourable pressure gradient followed by an adverse pressure gradient.

In order to understand the behaviour of the velocity profile when the flow is subjected
to different pressure gradients, it is useful to examine the x-momentum equation inside
the boundary layer:
∂u ∂u 1 ∂P µ ∂2 u
u +v =− + (140)
∂x ∂y ρ ∂x ρ ∂y 2

49
 ∂P /∂x = 0
∂P /∂x > 0
∂P /∂x < 0

Figure 22: Pressure gradients experienced by the suction surface of an airfoil.

For any pressure gradient, this x-momentum equation must satisfy the boundary con-
ditions at the wall (y = 0), where both components of velocity are zero: u is zero due to
no slip, and v is zero since there is no flow through the wall. Therefore, at the wall, the
left-hand side of the equation becomes zero and (recalling that the pressure is indepen-
dent of y) we have:  
∂2 u ∂P
µ  = (141)
∂y 2 ∂x
y=0

Let us first consider the case where the pressure gradient is negative, such as around the
leading edge of the aerofoil in Figure 22. This means that, at the wall, µ ∂2 u/∂y 2 < 0.
¡ ¢

Since the velocity gradient in the direction normal to the wall (∂u/∂y) decreases towards
the edge of the boundary layer (Figure 23), the term µ ∂2 u/∂y 2 is negative everywhere
¡ ¢

inside the boundary layer.

y y y

u ∂u ∂2 u
∂y ∂y 2
Figure 23: Boundary-layer profile and gradients in a favourable pressure gradient.

If, on the other hand, the pressure gradient in the direction of the flow is positive (such
as towards the rear of an aerofoil), this is described as an adverse pressure gradient. In
this case, Equation 141 tells us that µ ∂2 u/∂y 2 > 0 near the wall. From the free stream,
¡ ¢

as one moves towards the wall at the edge of the boundary layer, the velocity always
starts by decreasing from the free stream velocity, i.e. the velocity gradient goes from
zero to positive. Therefore, at large distances from the wall, the term µ ∂2 u/∂y 2 must
¡ ¢

be negative, as shown in Figure 24. As a result, there must exist a point where the term
goes from positive near the wall to negative far away – at this point, µ ∂2 u/∂y 2 = 0.
¡ ¢

Such a point corresponds to an inflection of the velocity profile in the boundary layer.
Therefore, in a region of positive pressure gradient the velocity profile always displays

50
 y y y

inflection
point

u ∂u ∂2 u
∂y ∂y 2
Figure 24: Boundary-layer profile and gradients in a favourable pressure gradient.

an inflection point. When an adverse pressure gradient is applied, the fluid particles in-
side the boundary layer feel this pressure rise and, in order to move forward, they must
overcome this pressure increase. At some point, for sufficiently large adverse pressure
gradient, the particles near to the wall (which have the least momentum) cannot over-
come the pressure rise. Figure 25 shows how the boundary layer profile changes when
subjected to such a strong adverse pressure gradient.

pressure increasing (∂P /∂x > 0)

y
x

separation point (τw = 0) reversed flow (separated)

Figure 25: Effect of adverse pressure gradient on boundary-layer flows.

At some point, labelled the separation point, the boundary layer is deflected away from
the wall and moves into the free stream, which is called boundary-layer separation
or flow separation. At the point of separation, where the flow direction near the wall
changes, the friction at wall is zero:
 
∂u
τw, separation = µ   =0 (142)
∂y
y=0

Note that the boundary-layer equations are only valid as far as the point of separation.
After this point, the separated flow becomes so thick that the assumptions made earlier
(to derive the thin boundary-layer equations) no longer apply.
If an airfoil is at a low angle of attack, there is a mild adverse pressure gradient which the
boundary layer can typically negotiate and stay attached. However, as the angle of attack

51
a) open separation on airfoil b) closed separation bubble

separation point
separation bubble

c) laminar separation on flat plate d) fully-separated flat plate

reattachment point flow separation

closed laminar separation bubble attached flow on pressure surface

e) laminar separation on curved wall f ) separation over rectangular block

recirculation turbulent separation
separation point

g) circular cylinder (laminar) h) circular cylinder (turbulent)

separation point on cylinder turbulent separation and wake

recirculating eddies laminar separation

Figure 26: Examples of flow separation observed in experiment. Source unknown.

is increased, the pressure gradient becomes too severe and flow separation occurs. Note
that, compared to laminar cases, turbulent boundary layers tend to be more resistant
to separation as the turbulent kinetic energy associated with the velocity fluctuations
allows the boundary layer to negotiate more severe adverse pressure gradients.

Figure 26 shows several examples of flow separation around different geometries, which
typically causes a number of adverse effects. For example, if the flow separates from an
aircraft wing, it results in greater drag, lower lift and possible structural damage caused
by separation-induced vibration. In many practical applications, it is therefore neces-
sary to delay or prevent separation using passive or active flow control methods. Whilst
the topic of boundary-layer development and control on real geometries is an area of

52
significant research activity, for the remainder of this module, we will focus our atten-
tion on simpler cases, which permit a mathematical treatment.

1.3.2 Simple Solutions of Boundary-Layer Flows

In many practical situations the streamwise pressure gradient is not negligible and must
be considered in order to have a correct estimation of the boundary layer flow. We have
already seen, for example, that the flow around an airfoil wing section shows regions of
negative, almost zero and positive pressure gradient (Figure 22). One commonly-used
method to calculate boundary-layer flows under arbitrary pressure gradients is based on
the integration of the x-momentum equation. By integrating this equation, it is possible
to obtain a differential equation in terms of the average properties of the boundary layer:
momentum thickness, displacement thickness and skin friction. In such a way, we can
use approximate velocity profiles for the boundary layer and obtain information about
its development.

The von Karman momentum integral equation The x-momentum equation for a
thin boundary layer, derived in Equation 139, is:

∂u ∂u 1 ∂P µ ∂2 u
u +v =− + (143)
∂x ∂y ρ ∂x ρ ∂y 2

From our analysis of thin boundary-layer theory, we know that, at any arbitrary stream-
wise location (x) the pressure is nearly constant across the boundary layer in the direc-
tion normal to the surface (y). As a result, the pressure anywhere inside the boundary
layer at this location is equal to the local inviscid pressure outside the boundary layer.

u∞
B P∞

h
boundary-layer edge

u
y δ
x
A
x flat plate

Figure 27: Boundary layer on flat plate.

Therefore, using the flat-plate boundary layer described in Figure 27, the pressure inside

53
the boundary layer can be replaced by the local free-stream pressure, P ∞ :

∂u ∂u 1 ∂P ∞ µ ∂2 u
u +v =− + (144)
∂x ∂y ρ ∂x ρ ∂y 2

For the case of zero pressure gradient, ∂P ∞ /∂x = 0. Therefore:

∂u ∂u µ ∂2 u
u +v = (145)
∂x ∂y ρ ∂y 2

We now integrate this equation in the y-direction along AB, between the wall (y = 0) and
a point outside the boundary layer (y = h where h > δ):
   
Z h ∂u ∂u Z h µ ∂2 u
u +v  dy  dy = 
0 ∂x ∂y
ρ ∂y 2 0
   
Z h ∂u ∂u 1Z h ∂ ∂u
=⇒ u +v  dy = µ  d y
0 ∂x ∂y ρ 0 ∂y ∂y
  (146)
Z h ∂u ∂u 1Z h ∂ 1¡ ¡
(τ) d y = τ y = h − τ y = 0
¢ ¡ ¢¢
=⇒ u +v  dy =
0 ∂x ∂y ρ 0 ∂y ρ
 
Z h ∂u ∂u 1 τw
=⇒ u + v  d y = (0 − τw ) = −
0 ∂x ∂y ρ ρ

Using the two-dimensional, incompressible continuity equation (∂u/∂x + ∂v/∂y = 0),

we obtain an expression for the vertical component of velocity (v) in terms of u:

Z y ∂u
v =− dy (147)
0 ∂x

We can substitute this into the left-hand side of the boundary-layer equation:
 
Z h ∂u ∂u Z y ∂u τw
u − d y d y = −
0 ∂x ∂y 0 ∂x ρ
    (148)
Z h ∂u Z h ∂u Z y ∂u τw
=⇒ u  dy −  d y d y = −
0 ∂x 0 ∂y 0 ∂x ρ

Ry
(∂u/∂x) d y and d B = ∂u/∂y d y (such that B = u), we recall that A d B =
¡ ¢ R
For A = 0

54
 R
AB − B d A and thus can rewrite the second integral on the left-hand side:
    
Z h ∂u Z y ∂u Z h Z y ∂u ∂u
 d y d y =  d y  d y
0 ∂y 0 ∂x 0 ∂y 0 ∂x
 h  
Z y ∂u Z h ∂u
= u d y − u d y  (149)
0 ∂x 0 ∂x
0
 
Z h ∂u Z h ∂u
= u∞ dy − u d y 
0 ∂x 0 ∂x

Substituting this into the left-hand side of Equation 148, we obtain

   
Z h ∂u Z h ∂u Z h ∂u τw
u  d y − u∞ dy + u d y = −
0 ∂x ∂x 00 ∂x ρ
 
Z h ∂u Z h ∂u τw
=⇒ 2u  d y − u∞ dy =− (150)
0 ∂x 0 ∂x ρ
 
Z h ∂u 2 Z h ∂u τw
=⇒   d y − u∞ dy =−
0 ∂x 0 ∂x ρ

Recalling that we have a zero pressure gradient scenario (∂P ∞ /∂x = 0), this means that
∂u ∞ /∂x = 0 and so U∞ is a constant. Therefore, we can rewrite the left-hand side of the
equation:
 
Z h ∂u 2 Z h ∂ (uu ∞ ) τw
  dy − dy =−
0 ∂x 0 ∂x ρ
Z h ∂ £ ¤ Z h ∂ τw (151)
=⇒ u2 d y − [uu ∞ ] d y = −
0 ∂x 0 ∂x ρ
Z h ∂ τw
£ 2 ¤
=⇒ u − uu ∞ d y = −
0 ∂x ρ

Since the limits of the integration (0 and h) are independent of x, this allows us to take
the x-partial derivative outside the integral:

∂ ·Z τw h¡ ¸
2
¢
u − uu ∞ d y = −
∂x 0 ρ
    (152)
∂ Z h u u τw
2 1 −  d y = −
=⇒ −u ∞ 
∂x 0 u ∞ u∞ ρ

We note that, outside the boundary layer, for y = δ → h, the streamwise velocity, u = u ∞ ,
and thus the term inside the brackets evaluates to zero. Therefore, we only need to

55
consider the integral from y = 0 → δ:
   
∂ Z δ u u τw
2 1 −  d y = −
−u ∞  (153)
∂x 0 u∞ u∞ ρ

We now recall the definition of momentum thickness for an incompressible boundary

layer, θi , from Equation 128:
 
Z δ u u
θi = 1 −  dy (154)
0 Ue Ue

The velocity at the edge of the boundary layer, Ue , is roughly equal to the free-stream
velocity, u ∞ , so:  
Z δ u u
θi = 1 −  dy (155)
0 u∞ u∞

This is exactly the integral which appears in Equation 153 and so:

∂ τw
2
−u ∞ (θi ) = −
∂x ρ
∂θi τw
2 (156)
=⇒ u ∞ =
∂x ρ
∂θi τw
=⇒ =
∂x ρu ∞
2

For reference, in this equation τw = µ ∂u/∂y y=0 . We also know that the local skin-
¡ ¢

friction coefficient, C f is defined in Equation 113 as:

τw
Cf = 1
(157)
2
ρu ∞
2

Therefore,
∂θi Cf
= (158)
∂x 2
This equation is known as the von Karman momentum integral equation. It allows
us to calculate the development of the boundary-layer parameters if the velocity pro-
file inside the boundary layer is known. In order to perform such calculations, we first
need to know the boundary-layer velocity profile itself. As we shall see, this velocity pro-
file is typically defined in parametric form and then boundary conditions are applied to
determine the coefficients in the parametric equation. In particular, any physical ve-
locity profile must satisfy certain boundary conditions which can be divided into two
categories:

• primary boundary conditions;

• auxiliary (or secondary) boundary conditions.

56
Primary boundary conditions Consider the distributions in velocity and velocity gra-
dient for a typical boundary layer on a stationary surface, as shown in Figure 28. Obser-
vations reveal that the velocity profile must satisfy two primary boundary conditions.
Firstly, at the wall, the flow velocity must be equal to zero since the no-slip condition
applies:
u=0 at y = 0 (159)

Secondly, as the flow approaches the free stream the velocity tends to u ∞ . In other
words, at wall-normal distances greater than the boundary-layer thickness (i.e. for y > δ)
the changes in flow velocity can be ignored:

u = uδ at y = δ (160)

In this equation, u δ = 0.995u ∞ . Any velocity profile for a boundary-layer flow must nec-
essarily satisfy these two primary boundary conditions.

y y

u ∂u
∂y
Figure 28: Velocity and velocity gradient distribution for a typical boundary layer.

Auxiliary boundary conditions An inspection of the shear stress (velocity gradient)

distribution across the boundary layer in Figure 28 shows that the shear stress approaches
zero as the flow approaches the boundary-layer edge. This can be expressed mathemat-
ically as:  
∂u
τ y = δ = µ
¡ ¢
 =0 (161)
∂y
y=δ

The viscosity is not zero, result in the auxiliary boundary condition:

∂u
=0 at y = δ (162)
∂y

In some cases, it may be necessary to apply more boundary conditions in order to spec-
ify a velocity profile for a boundary layer. In these situations, the higher derivatives of
∂u/∂y can be used as additional boundary conditions. For example:

∂2 u
=0 at y = δ (163)
∂y 2

57
In general, such auxiliary boundary conditions can be expressed as:

∂n u
=0 at y = δ (164)
∂y n

Another useful boundary condition relates the pressure gradient to the second deriva-
tive of velocity at the wall. Equation 139 gives the x-momentum equation for a thin
boundary layer:
∂u ∂u 1 ∂P µ ∂2 u
u +v =− + (165)
∂x ∂y ρ ∂x ρ ∂y 2
We recall that the pressure gradient everywhere inside the boundary layer takes the free-
stream value, P = P ∞ . Also, this equation must be valid everywhere inside the boundary
layer, including at y = 0. But, at y = 0, the no-slip condition requires that u = v = 0, and
so the equation reduces to:
1 ∂P ∞ µ ∂2 u
0=− + (166)
ρ ∂x ρ ∂y 2
This therefore provides another auxiliary boundary condition:

∂2 u 1 ∂P ∞
= at y = 0 (167)
∂y 2 µ ∂x

In order to understand the application of these boundary conditions, it is useful to con-

sider a number of examples.

Example 1: Linear velocity profile Consider a velocity profile for a laminar boundary
layer given by the linear relation:
u = Ay + B (168)

Say we want to find the coefficients, A and B in terms of the boundary-layer thickness
(δ) and the flow velocity at the boundary-layer edge (u δ ). Since we have two unknowns
(A and B ) we require two boundary conditions. These can be satisfied by simply using
the two primary boundary conditions:

u=0 at y = 0
(169)
u = uδ at y = δ

We apply these boundary conditions to the given linear velocity profile:

0 = A ×0+B
(170)
uδ = A × δ + B

From these equations, we can determine the values of A and B :

uδ
A=
δ (171)
B =0

58
Substituting these values gives the linear boundary-layer velocity profile:

y
u = uδ (172)
δ

Defining the non-dimensional wall-normal distance, η = y/δ, we obtain:

u y
= =η (173)
uδ δ

Example 2: Cubic velocity profile Let us consider a cubic velocity profile for a zero
pressure-gradient boundary-layer flow:

u = Ay 3 + B y 2 +C y + D (174)

In this case, there are four unknowns so four boundary conditions are required. There-
fore, in addition to the two primary boundary conditions, it is necessary to use two aux-
iliary boundary conditions:

u = Ay 3 + B y 2 +C y + D = 0 at y = 0
u = Ay 3 + B y 2 +C y + D = u δ at y = δ
∂u
= 3Ay 2 + 2B y +C = 0 at y = δ (175)
∂y
∂2 u 1 ∂P ∞
= 6Ay + 2B = at y = 0
∂y 2 µ ∂x

The final boundary condition can be simplified by noting that the pressure gradient in
this case is zero:
∂2 u
= 6Ay + 2B = 0 at y = 0 (176)
∂y 2
We apply these four boundary conditions:

A × 03 + B × 02 +C × 0 + D = 0
A × δ3 + B × δ2 +C × δ + D = u δ
(177)
3A × δ2 + 2B × δ +C = 0
6A × 0 + 2B = 0

We can now obtain B and D from the first and last equations:

B =0
(178)
D =0

59
These then give:

A × δ3 +C × δ = u δ
(179)
3A × δ2 +C = 0

Multiplying the bottom equation by δ and subtracting the top equation gives:

2A × δ3 = −u δ
1
=⇒ A = − uδ (180)
2δ3
3
=⇒ C = u δ
2δ

We therefore obtain the velocity profile:

   
1 3
u = − uδ  y 3 +  uδ  y
2δ3 2δ
 3 (181)
u 1 y 3y
=⇒ = −  
uδ 2 δ 2 δ

Defining the non-dimensional wall-normal distance, η = y/δ, we obtain:

u 3 1
= η − η3 (182)
uδ 2 2

Now that we have obtained the velocity profile, we are able to calculate the boundary-
layer integral parameters (displacement and momentum thicknesses) and the wall shear
stress. Starting with the displacement thickness (Equation 119):
   
Z δ u Z 1 u
δ∗ =  dy =
1 − 1 −  δd η
0 uδ 0 uδ
 
δ∗ Z 1 u
=⇒ = 1 −  d η
δ 0 uδ
  
Z 1 3 1
= 1 −  η − η3  d η
0 2 2
(183)
 
Z 1 3 1
= 1 − η + η 3  d η
0 2 2
 1
3 1
= η − η2 + η4 
4 8
0
3 1 3
= 1− + =
4 8 8

60
Similarly, for the momentum thickness (Equation 128):
   
Z δ u u Z 1 u u
θ=  dy =1 − 1 −  δd η
0 uδ uδ 0 uδ uδ
 
θ Z 1 u u
=⇒ = 1 −  d η
δ 0 uδ uδ
  
Z 1 3 1 3 1
=  η − η3  1 − η + η3  d η
0 2 2 2 2
  (184)
Z 1 3 9 1 3 1
=  η − η2 − η3 + η4 − η6  d η
0 2 4 2 2 4
 1
3 3 1 3 1
=  η 2 − η3 − η4 + η5 − η7 
4 4 8 10 28
0

3 3 1 3 1 39
= − − + − =
4 4 8 10 28 280

Finally, the wall shear stress, τw , is:

     
∂u ∂u ∂η
τw = µ   = µ   
∂y ∂η ∂y
y=0 y=η=0 y=0
  
∂ 3 1 1
= µ u δ  η − η3 
∂η 2 2 δ
η=0
  (185)
µu δ ∂ 3 1
=  η − η3 
δ ∂η 2 2
η=0
 
µu δ 3 3 3 µu δ
=  − η2  =
δ 2 2 2 δ
η=0

Importantly, now that we have expressions for the momentum thickness and the wall
shear stress, we are able to use the von Karman momentum integral equation to evalu-
ate the development of the boundary layer with streamwise distance. We start with the

61
momentum integral equation provided in Equation 158:

dθ Cf τw
= =
dx 2 ρu 2
 δ 
d · 39 ¸ 1 3 µu δ
=⇒ δ =  
d x 280 ρu δ2 2 δ
(186)
13 d δ µ
=⇒ =
140 d x ρu δ δ
Z µ
13
Z
=⇒ δdδ = dx
140 ρu δ

Since the flow has zero pressure gradient, u δ , is a constant and does not vary with x.
Therefore,
13
1 µ
× δ2 = x +C 1 (187)
140 2 ρu δ
We now note that x = 0 is the start of the flat plate, where the boundary layer starts
growing, and thus δ = 0 at this point. This condition tells us that C 1 = 0, and so:

13 1 µ
× δ2 = x
140 2 ρu δ
(188)
280 µ
=⇒ δ2 = x
13 ρu δ

It is useful to recast this equation in terms of non-dimensional quantities, such as the

Reynolds number based on streamwise distance, x, which is given by Rex = u δ ρx/µ.
Therefore, we can write:

280 µx 2 280 x 2
δ2 = =
13 u δ ρx 13 Rex
v
u (189)
u 280 x 2 4.641x
=⇒ δ =
t
= p
13 Rex Rex

Therefore, we have obtained an equation for the boundary-layer thickness, δ, as a func-

tion of the streamwise distance, x. From this, it is possible to evaluate the development
of the other boundary-layer parameters.

3 1.740x
δ∗ = δ = p
8 Rex
39 0.646x
θ= δ= p (190)
280 Rex
 
3 µu δ 3 µu δ µu δ p µ ρu δ2
Rex = 0.323ρu δ2 
p
τw = = = 0.323  Rex = 0.323 p
2 δ 2 4.641x
p x u δ ρx Rex
Rex

62
Blasius solution for laminar boundary layers It is also possible to derive an exact so-
lution for steady, two-dimensional, incompressible, laminar boundary layers. Figure 29
shows a comparison of this solution (known as a Blasius profile) against experimental
data for laminar boundary layers at several different Reynolds numbers: Rex = 1.08×105 ,
1.82 × 105 , 3.64 × 105 , 5.46 × 105 and 7.28 × 105 .

7
exact Blasius solution
6 experimental data
5
s
u∞ ρ 4
y
µx 3

0
0 0.2 0.4 0.6 0.8 1
u/u ∞

Figure 29: Comparison of Blasius solution against experimental data. Source unknown.

The exact Blasius solution takes a form that is somewhat complicated. However, it is
possible to use this solution to derive the development of the associated boundary-layer
parameters:

5.2x
δ= p
Rex
1.718x
δ∗ = p
Rex
(191)
0.664x
θ= p
Rex
ρu δ2
τw = 0.332 p
Rex

It is interesting to note how close the coefficients in these expressions are very close
to those obtained in Example 2, using a “simple” cubic velocity profile. Note that all
boundary-layer thicknesses and wall shear stresses are functions of streamwise posi-
tion, x. The derivations also show that for two-dimensional, zero pressure-gradient,
laminar boundary layers, the wall shear stress is inversely proportional to the square
root of streamwise distance:
τw = K x −1/2 (192)

The constant, K , depends on the properties of the fluid (density and viscosity), free-
stream velocity and the shape of the velocity profile within the boundary layer.

63
1.3.3 Applications of Boundary-Layer Theory

The theory describing boundary layers can be used to explain various aspects of prac-
tical flow scenarios, such as the differences between laminar vs. turbulent boundary
layers, the effect of boundary-layer displacement, and the drag associated with the wall
shear stress. We shall now consider each of these aspects in turn.

Laminar and turbulent boundary layers When a boundary layer first begins to de-
velop, it is often laminar in nature, with only very small velocity fluctuations. How-
ever, these small fluctuations can give rise to instabilities, which grow as the boundary
layer develops and eventually causes a “transition” to a turbulent state. The topic of
boundary-layer transition is an area of active research, but some facts are well estab-
lished.

Higher Reynolds numbers tend to cause faster growth of the instabilities, and thus pro-
mote transition to a turbulent boundary-layer state. The precise point at which tran-
sition occurs depends on the geometry under consideration. For example, in a zero
pressure-gradient flat-plate flow, it is observed that transition occurs for Reynolds num-
ber (based on distance from boundary-layer starting point) of roughly Rex = 500, 000.
Meanwhile, in a fully-developed pipe flow, the critical Reynolds number based on pipe
diameter (D) is in the range, ReD = 2300 − 3500. Boundary-layer transition can also be
affected by the surface roughness, the pressure gradient the boundary layer is subjected
to, and any “tripping” devices installed to cause a small-scale separation which pro-
motes transition. Note that many practical aerospace applications feature sufficiently
high Reynolds number that transition occurs extremely close to the stagnation point
and so the boundary layers can be considered to be turbulent over the entire object’s
surface.

As well as having greater velocity fluctuations, the mean velocity profile for a turbulent
boundary layer is also quite different to a laminar boundary layer. The shape factor is
in the range 1.2 − 1.5 for turbulent boundary layers, significantly smaller than the value
of 2.5 for laminar boundary layers. This difference is associated with a much greater ve-
locity gradient (∂u/∂y) near the wall and thus a significantly larger wall shear stress, τw
(by roughly an order of magnitude). Recalling that the friction drag is related to the wall
shear stress, the enhanced wall shear stress means that turbulent boundary layers fea-
ture greater associated drag. On the other hand, the greater wall shear stress means that
turbulent boundary layers are more resistant to separation in the presence of adverse
pressure gradients, since they can withstand a greater reduction in wall shear stress be-
fore separation occurs (at τw = 0).

Effect of boundary-layer displacement The displacement thickness of the boundary

layer, which describes the distance by which an external streamline is displaced away
from the wall, has important consequences for the flow internal to a duct or channel.

64
Figure 30 shows the flow through a square duct of side, h, with entry velocity U0 . If the
flow was inviscid, there would be no boundary layer and so the flow everywhere would
have velocity U0 and there would be a slip condition at the walls, as shown in Figure 30a.

a) inviscid flow

h
U0 U0

b) viscous flow

U0 Ue Ue
boundary
layer

Figure 30: Effect of boundary layers on the flow through a square channel.

In reality, however, we know that there will be viscous effects and boundary layers will
develop on the tunnel walls, as described by Figure 30b. The boundary layer will have
zero thickness at the start of the duct but will get thicker as it develops in the down-
stream direction. The mass flow rate must be the same as the inviscid case, with the
same area, density and velocity at the entrance to the duct. For an incompressible flow,
the density is also the same, as is the overall flow area, which is defined by the geometry.
However, as a result of the formation of the boundary layer, the velocity close to the wall
decreases and so there is a reduction in mass flow rate in this region compared to the
inviscid case. Because the overall mass flow rate (across the entire duct) must be the
same as the inviscid case, this must mean that the free-stream velocity, Ue , in the region
outside the boundary layers must be increased to compensate for the reduction in mass
flow rate within the boundary layers. In other words, Ue > U0 .

In order to assess the effect that these boundary layers have on the free-stream velocity,
we consider the mass flow rate through the channel. At the entrance of the duct we have:

ṁ = ρU0 h 2 (193)

65
At point, x, in the duct, the mass flow can be split into two components: ṁ free stream for
the constant velocity core and ṁ boundary layer for the boundary layers:

ṁ (x) = ṁ free stream + ṁ boundary layer (194)

In the free stream, the effective duct side length is h − 2δ as there are boundary layers,
with thickness δ on each wall. Therefore, the effective area is (h − 2δ)2 and so the mass-
flow rate is:
ṁ free stream = ρUe (h − 2δ)2 (195)

There are four boundary layers (floor, ceiling and two sidewalls) which each extend to
a distance δ away from the wall and span a width h across the tunnel width/height.
Considering the floor boundary layer as an example, the mass flow within this is:
Z δ
ṁ floor boundary layer = ρuh d y (196)
0

All four boudnary layers must contain the same mass flow by symmetry. Therefore, the
total mass flow rate included within all the boundary layers is:
Z δ
ṁ boundary layer = 4 ρuh d y (197)
0

Combining the mass-flow rates together and equating them with the inlet mass-flow
rate, we obtain:
Z δ
2 2
ρU0 h = ρUe (h − 2δ) + 4 ρuh d y
0
Z δ
(198)
2 2 2
=⇒ U0 h = Ue h − 4Ue hδ + 4Ue δ + 4h udy
0

Since we are considering a duct with boundary-layer thickness much smaller than the
duct width, δ ≪ h. we can neglect the term quadratic in δ:
Z δ
2 2
U0 h = Ue h − 4Ue hδ + 4h udy
0
µ Z δ ¶
2
= Ue h − 4h Ue δ − udy
0
 
Z δ Z δ u
= Ue h 2 − 4h Ue d y −Ue d y (199)
0 0 Ue
 
Z δ u
= Ue h 2 − 4hUe  dy1 −
0Ue

= Ue h 2 − 4hUe δ∗ = Ue h 2 − 4hδ∗
¡ ¢

Therefore, from a mass-flow rate perspective, the effective area of the duct has been
reduced by 4hδ∗ . This corresponds to the boundary layer on four walls, each spanning
a distance h. Thus, the effective wall-normal distance that the boundary layer occupies

66
is 4hδ∗ /4h = δ∗ . This quantity is just the displacement thickness, i.e. the distance by
which an external streamline is displaced due to the presence of the boundary layer.

Typically, wind tunnel designers want the flow velocity to remain constant at the in-
let velocity, U0 . To achieve this, wind tunnels are designed with a test section which
diverges to account for the development and growth of the boundary layers. The diver-
gence is chosen such that the cross-sectional area of the inviscid flow, once the bound-
ary layers with displacement thickness, δ∗ (x) have been accounted for. Thus, to main-
tain a constant free-stream velocity, the physical cross-sectional area must be such that
such that A (x) − 4h (x) δ∗ (x) remains constant.

Drag caused by wall shear stress The relationship between the friction drag and the
momentum thickness has already been demonstrated. However, in general the drag
on the upper surface of an object shown in Figure 31a with width, W , and length, L, is
simply given by the integral of the wall shear stress, τw (x), over the distance where the
boundary layer develops (i.e. from x = 0 to L):
Z L
D =W τw (x) d x (200)
0

The distribution of wall shear stress depends on the development of the boundary layer.
However, it is clear from the exact Blasius solution and the approximations that one
common distribution has τw ∝ p1 and thus τw = kx −0.5 . It is thus useful to consider
Rex
the drag produced by this distribution:
Z L
D =W kx −0.5 d x
0
¤L (201)
= W k 2x 0.5 0
£
p
= 2W k L

It is interesting to note that despite having the same area, W × L, an object placed with
p
its length aligned with the flow (Figure 31a) has drag, D = 2W k L, whilst if the width is
p
aligned with the flow (Figure 31b) then D = 2Lk W . In the case that L > W , this means
that there is a higher drag if the flow passes over the narrow width of the object than if
it were to pass along its length. The mathematical explanation for this observation is
that τw ∝ x −0.5 and so the shear stress is largest right at the start of the boundary layer.
Where the width is aligned with the flow (Figure 31b), there is “more flow” (i.e. span L
rather than W ) which experiences the very high shear stress at the start of the boundary
layer, and thus a greater drag.

Physically the reason for the higher shear stress at the start of the boundary layer is
related to the fact that the shear stress is proportional to the velocity gradient ∂u/∂y.
Where the boundary layer is very thin, the velocity needs to go from the free-stream
value to zero at the wall (no slip) over a much smaller distance, hence the velocity gra-

67
 a) scenario A a) scenario B

W L

Figure 31: Flow scenarios for drag calculations.

dients are higher and the shear stress is greater. This reasoning is independent of the
exact wall shear stress distribution and thus similar arguments apply to the boundary-
layer development in many situations. The observation also explains why the majority
of the drag takes place at the front of a plate, where the boundary layer is thinnest. The
proportion of drag coming from the front half of the plate in the scenario described is:
R L/2
D front half W 0 kx −0.5 d x
= RL
D overall W −0.5 d x
0 kx
£ 0.5 ¤L/2
W k 2x 0
= £ ¤L (202)
W k 2x 0.5 0
p
L/2 1
= p = p ≈ 0.71
L 2

Therefore 71% of the drag is associated with the front half of the area of the plate, where
the boundary layer is thinner and thus the velocity gradients/shear stresses are largest.
The friction drag on all aeronautical objects of increasing complexity can be calculated
using analogous methods, by integrating the wall shear stress over the exposed surface
area of the object. Thus, it is generally useful to think about the viscous drag on object in
terms of the wall shear stress caused by the boundary layers developing on its surface.

68
2.1 Basic Concepts in Compressible Flow
2
Fundamentals of
Compressible Flows
SECTION

The term “compressible” means that the density varies throughout the flow field. The
change in density is often the result of a change in pressure from one point to another.
Whilst the classical way to define compressible flows relates to density variations, it is
also true that compressible flows are related to the speed of sound, although the link
between the two is not obvious. We will therefore start by discussing why the speed of
sound is relevant to compressible flows.

By definition, small disturbances in fluids travel at the speed of sound, a. Therefore,

one can think of the speed of sound, a, as the speed at which information propagates
in a fluid. Let us therefore consider the speed of propagation of a plane sound wave,
which takes the form of a weak pressure pulse or disturbance. Figure 32 shows a long
pipe which is filled with a stationary gas. A plane weak pressure pulse is produced by the
inward movement of a piston at speed, dV . This movement produces a pressure wave
which travels to the right at the speed of sound, a. The pressure and density of the fluid
on the left, once the wave has past, become P + d P and ρ + d ρ. This fluid moves at the
same speed as the piston and therefore has velocity, dV . The fluid on the right of the
wave front has not yet been affected by the wave, so is stationary with pressure, P , and
density, ρ.

In order to apply a control volume analysis, it is necessary to use a control volume which
remains fixed and, thus, it makes sense to change to a reference frame in which the

69
 physical reference frame wave reference frame
wave front stationary wave front
stationary control
dV air volume
dV a a − dV a

pressure P + dP pressure P + dP

P P

speed speed
a
dV a − dV
0

density ρ + dρ density ρ + dρ
ρ ρ

Figure 32: Propagation of a weak pressure wave generated by an accelerating piston.

pressure wave front is stationary. Therefore, flow enters the control volume with speed,
a, and leaves with speed, a − dV . For the cross-sectional area of the pipe given by A,
the conservation of mass gives:

ρ Aa = ρ + d ρ (A) (a − dV )
¡ ¢

d ρ dV (203)
=⇒ =
ρ a

Similarly, the conservation of momentum gives:

A [(P + d P ) − P ] = ρ Aa [a − (a − dV )]
(204)
=⇒ d P = ρadV

Combining the two equations gives:

dP
a2 =
dρ
s (205)
dP
=⇒ a =
dρ

This equation highlights why the speed of sound is important when considering com-
pressibility, as it shows how the speed of sound is a measure of how fluctuations in
pressure correlate with changes in density. For example, where an object travels much

70
slower than the speed of sound (and thus the speed of sound is relatively very large)
this corresponds to a large value of d P /d ρ and so a small value of d ρ/d P . Therefore,
changes in pressure cause negligible change in the density and so the density is con-
stant, i.e. the flow is incompressible. Conversely, when an object travels close to or
exceeds the speed of sound (and thus the speed of sound is small relative to the ob-
ject velocity), d P /d ρ is small and d ρ/d P is large, so pressure changes cause significant
density variations, i.e. the flow is compressible.

As a result, the definition of compressible flows as those where density varies is con-
sistent with defining compressible flows to be where flow velocities are comparable or
exceed the speed of sound. In other words, the study of compressible flows relies on
understanding how information (in the form of pressure disturbances caused by the
relative movement of a body through the fluid) travels and affects the other parts of the
flow field.

Speed of sound In order to evaluate the speed of sound, we need to calculate d P /d ρ.

To do so, we note that, in arriving at this equation, we have neglected friction and
heat transfer, so we are considering an adiabatic flow. Since the wave is of infinitesi-
mal strength, the process is also reversible. A process which is both adiabatic and re-
versible must also be isentropic, and so the isentropic relations apply. For perfect gases
(P = ρRT ), the isentropic relation between pressure and density is:

P
= constant
ργ
=⇒ ln P − γ ln ρ = constant
dP dρ (206)
=⇒ =γ
P ρ
dP P
=⇒ =γ
dρ ρ

By combining this relation with the equation of state (P = ρRT ), we obtain an expression
for the speed of sound:

a 2 = γRT
(207)
=⇒ a = γRT
p

where γ is the specific heat ratio of the fluid (1.4 for diatomic gases) and R is the specific
gas constant (287.05 J/kg K for air). Note that this can also be rewritten as:
s
R (208)
a= γ T
m

where R is the universal gas constant (8314.3 J/kmol K) and m is the gas molecular
weight (28.964 g/mol for air). In general, the ratio of specific heats (γ) varies only over
a small range, for example γ = 1.67 for monatomic gases and γ = 1.33 for polyatomic

71
gases. In contrast, the molecular weight of different gases can vary significantly and so
the gas’s molecular weight is the principal factor in determining the speed of sound.
q For
28.964
example, at a given temperature, the speed of sound in hydrogen is roughly 2.016
=
3.79 times the speed of sound in air.

Note that, for liquids, the speed at which sound waves propagate is very large. Therefore,
the pressure pulses produced at one point can be felt almost simultaneously at other
locations. For water and many common liquids, the speed of sound is about 1500 to
1700 m/s. Since the speed of fastest boats is not even close to this speed of sound, for
foreseeable future, vehicles in liquid can generally be treated as incompressible.

Mach number Consider the disturbances produced by a small object travelling at, say
680 m/s, in air. The pressure disturbances produced by this object propagate at the
speed of sound in air (340 m/s) which is about half the speed of the object itself. That
means the object is always ahead of its disturbances. In other words, any pressure in-
formation relating to disturbances are always behind of the creator of the disturbances.
If the object travels at the same speed (680 m/s) through hydrogen instead, its distur-
bances travel at a speed of about 1300 m/s, and are therefore ahead of their creator by
about 620 m/s. Therefore, it is the relative speed between the object and the distur-
bances which determines the way in which information travels around the flow field
and dictates the flow behaviour. It is common to use the ratio between the object speed
and the speed of sound as an indicator of the relative speed between the object and its
disturbances. This ratio, which is called the Mach number (M ), characterises the rela-
tive importance of compressibility:
V
M= (209)
a
For a perfect gas, we can use the derived expression for the speed of sound:

V
M=p (210)
γRT

This temperature-dependence of Mach number has consequences for an object travel-

ling with the same velocity at different altitudes. Since the air temperature depends on
altitude, the Mach number of the object is also altitude dependent. For example, Ta-
ble 3 shows how the Mach number of an object travelling at 350 m/s varies for different
altitudes.

Compressible flow regimes The definition of Mach number enables a classification of

different regimes of compressible flows. As the flow velocity increases, the Mach num-
ber increases through two different effects: (1) the velocity, V , increases directly and (2)
the reduction in temperature (as thermal energy is converted to kinetic energy) causes
the speed of sound ( γRT ) to reduce. In order to separate out these two effects, we can
p

72
 Table 3: Mach number of an object at 350 m/s at different altitudes
Altitude above Temperature Local speed Mach
sea level (m) (K) of sound (m/s) number
(a = γRT )
p
(M = V /a)
p 350
0 288.2 1.4 × 287 × 288.2 = 340.3 = 1.03
340.3
p 350
5000 255.7 1.4 × 287 × 255.7 = 320.5 = 1.09
320.5
p 350
10 000 223.2 1.4 × 287 × 223.2 = 299.5 = 1.17
299.5
p 350
20 000 216.7 1.4 × 287 × 216.7 = 295.0 = 1.19
295.0

consider the steady, adiabatic flow in a stream tube. The steady-flow energy equation is:

V2
h+ = constant
2 (211)
=⇒ V 2 + 2h = constant

In this equation, V is the velocity and h is the enthalpy. Noting that h = C P T , T =

a 2 /(γR) and γR/C P = γ − 1, we obtain:

2
V2+ a 2 = constant (212)
γ−1

Eventually, as the speed continues to increase, the temperature continue to decrease

until, in theory, it eventually ends up at T = 0 K (although, of course, this state is impos-
sible to reach in reality). At this point, a = 0 m/s, and so the velocity takes its maximum
possible value, V = Vmax . Therefore, we have:

2
V2+ a 2 = Vmax
2
(213)
γ−1

Inspection of this equation reveals that it simply describes a section of an ellipse, as

shown in Figure 33. As the velocity goes from 0 to Vmax , the speed of sound goes from a 0
(the speed of sound at zero velocity) to 0.
From Figure 33, it is possible to identify five different flow conditions:

• (1) incompressible flow – the flow velocity, V , is very small compared to the speed
of sound, a, and so M ≪ 1. The speed of sound does not change very much in this
regime, so any changes in Mach number are due to changes in flow velocity.
• (2) subsonic flow – the flow velocity, V , is close to the speed of sound, a, but still
always smaller than it and so M < 1. The speed of sound still does not change
very much in this regime, so any changes in Mach number are primarily due to
changes in flow velocity.
• (3) transonic flow – the flow velocity, V , is comparable with the speed of sound,
a, and so M ≈ 1. The speed of sound does change in this regime, so any changes
in Mach number are due to changes in both flow velocity and speed of sound.

73
 speed
of sound,
(1) incompressible flow
a
(2) subsonic flow
(3) transonic flow

a0

(4) supersonic flow

V =a
(M = 1)
V <a

V >a

(5) hypersonic flow

flow velocity, V Vmax

Figure 33: Speed of sound versus flow velocity for a steady, adiabatic, compressible flow.

• (4) supersonic flow – the flow velocity, V , is larger than the speed of sound, a, and
so M > 1. Any changes in Mach number are due to changes in both flow velocity
and speed of sound.

• (5) hypersonic flow – the flow velocity, V , is much larger than the speed of sound,
a, and so M ≫ 1. The speed of sound changes significantly in this regime, so any
changes in Mach number are largely due to changes in the speed of sound.

Classifying the flow regimes in this way highlights the importance of the relative velocity
compared to the speed of sound. Another important aspect of the relative magnitude of
the flow velocity to the speed of sound concerns the propagation of information from
a source through a flow field. Figure 34 describes the propagation of a spherical sound
wave emanating from a point source (P). There are four possible scenarios:

• stationary source (Figure 34a) – the sound waves (i.e. the pressure information)
propagate in all directions at the speed of sound, producing spherical wave fronts.

• source moving at subsonic speed (Figure 34b) – the pressure information trav-
els faster than the source, P. Therefore, the particles in the surrounding flow field
know in advance about the disturbance and so they gradually adjust themselves
to the new boundary conditions.

• source moving at sonic speed (Figure 34c) – the pressure information travels at
the same speed as the source, and so the surrounding particles have no time to go

74
 a) stationary source (V = 0) b) subsonic source (V < a)

P P′ P
a dt
3V d t
2a d t
3a d t 3a d t

c) sonic source (V = a) d) supersonic source (V > a)

P P′ P
′
P
3V d t = 3a d t 3a d t
µ

3V d t > 3a d t

Figure 34: Propagation of pressure information produced by a point source.

through gradual change to adjust themselves.

• source moving faster than sonic speed (Figure 34d) – the pressure information
lags behind the source, P. Therefore, the pressure waves cannot overtake the mov-
ing source and so, in all successive positions, the moving source is ahead of the
pressure waves.

In the final, supersonic case, Figure 34d shows that the different pressure wave fronts are
enveloped by a conical surface (or a wedge for two-dimensional flows) which is called a
Mach cone. The half-angle of the cone defines the Mach angle, µ. From the geometry
in Figure 34d, it is clear that:
1
sin µ = (214)
M
Therefore, the propagation of pressure information through supersonic flows is intrin-
sically related to the Mach number.

Nature of pressure waves As a fluid particle moves through a pressure wave, it phys-
ical and kinematic properties change. For example, its pressure can either increase or
decrease. In other words, the wave can be of compressive or expansive nature. The
former is called a “compression wave” and the latter an “expansion wave”. The type of

75
a) compressive waves
wave front 4 wave front 1
P + 4d P a4 a1 P
T + 4d T T

T + 3d T T + d T a1 < a2 < a3 < a4

T + 2d T
pressure
P + 4d P
pressure gradient direction of wave
becomes steeper propagation
P + dP
P

b) expansive waves
wave front 4 wave front 1
P − 4d P a4 a1 P
T − 4d T T

T − 3d T T − dT a1 > a2 > a3 > a4

T − 2d T
pressure
P
pressure gradient
becomes less steep direction of wave
propagation
P − 4d P

Figure 35: Wave formation by successive acceleration of a piston.

wave that occurs in a physical flow field depends on the nature of the boundary condi-
tions that the flow is subjected to. In order to understand the effect of these two types of
flow field, we consider the wave propagation in a long pipe containing a stationary gas
(Figure 35).

Figure 35a shows a piston accelerating towards the right into the gas, which is initially
stationary. The first right-ward movement of the piston (from velocity 0 to dV ) pro-
duces a pressure wave which moves to the right at the local speed of sound, a 1 = γRT .
Since this wave is considered to have infinitesimally small strength, we assume that the
process is isentropic. As a result, the temperature and the pressure of the fluid through
which this isentropic wave has passed increase from P and T to P +d P and T +d T . Since
the pressure increases, this wave is a compression wave. If the piston speed is subse-
quently changed from dV to, say, 2dV then a second pressure wave forms. This second
pressure wave moves to the right at the new local speed of sound, a 2 = γR(T + d T ).
Therefore this second pressure wave begins to catch up with the first wave.

76
As the piston continues to accelerate, fresh waves are produced. Since each new wave
is produced in an environment with a higher temperature, the speed of propagation
of each new wave is larger than the previous one. Therefore, each new wave starts to
catch up with the previous wave, and the distance over which the pressure rise occurs
becomes smaller and smaller. The net result of this process is that the pressure gradient
becomes increasingly steep. Eventually, the pressure gradient becomes infinitely steep,
with an associated large gradient in the temperature. This infinitely steep wave is called
a shock wave. Since the changes in temperature and pressure are happening over a very
short distance, the process cannot be reversible and therefore, the isentropic hypothesis
ceases to exist.

Now let us consider what happens if the piston accelerates towards the left instead (Fig-
ure 35b). In this case, the pressure decreases through each wave and so these are expan-
sion waves. As well as decreasing the static pressure, each expansion wave also reduces
the temperature. Therefore, each new wave forms in an environment with lower pres-
sure and temperature than the previous wave. As a result, the speed of propagation of
each new wave ( γRT ) is slower than the previous wave. Therefore, the net result of
p

the piston accelerating in this way is that the expansion waves spread out and cause a
gradual change in pressure and temperature, which corresponds to a reversible process.
Since there is no heat transfer involved, the expansion process is also adiabatic and so
can be treated as isentropic.

In this module, we do not discuss expansion waves in great detail but, instead, focus on
the behaviour of shock waves. A good starting point for understanding normal shock
waves is to examine the flow through ducts with constant area. In general, it is possible
for compressible flow through a constant-area duct to be subjected to both friction and
heat transfer. For simplicity, we will consider two cases in turn: (1) the flow through
a constant-area duct with friction but no heat transfer (also called “Fanno flow”), and
(2) the flow through a constant-area duct with heat transfer but no friction (also called
“Rayleigh flow”).

2.1.1 Constant-Area Duct with Friction: Fanno Flow

In the case of Fanno flow, the changes in flow conditions are due to friction, i.e. the wall
shear stress. In order to analyse the flow, we will consider the control volume shown
in Figure 36. We can apply a number of conservation equations to this control volume:
conservation of mass; the energy equation; the momentum equation; the second law of
thermodynamics. Let us consider each of these equations in turn.

Conservation of mass The conservation of mass states that the mass flow rate, ṁ is
constant. Since ṁ = ρ AV , this means that ρ AV must be constant. However, noting that

77
 friction force
P1 P2
V1 V2
ρ1 ρ2
T1 T2
S1 control S2
volume
friction force

Figure 36: Fanno flow: adiabatic, constant-area duct flow with friction.

A does not change for a constant-area duct, we have:

ρV = constant
(215)
=⇒ ρ 1V1 = ρ 2V2

The energy equation The energy equation applied to the control volume in Figure 36
is given by:
" Ã ! #
V22 − V12
ṁ (h 2 − h 1 ) + + g (z 2 − z 1 ) = Q̇ + Ẇ (216)
2

For a horizontal duct, there are no changes in potential energy (z 2 − z 1 = 0). Also, in the
absence of heat transfer and mechanical work, we obtain:
à !
V22 − V12
(h 2 − h 1 ) + =0
2
V22 V12 (217)
=⇒ h 2 + = h1 +
2 2
V2
=⇒ h + = constant = h 0
2

In this equation, h 0 is the stagnation enthalpy, which corresponds to the enthalpy value
when the flow is at rest (V = 0). We now want to derive an expression for temperature,
T , at any point along the duct in terms of the pressure, P , at that point and the flow
properties at the entrance to the control volume (station 1). Recalling that h = C P T :

ρ 2V 2
CP T + = h0 (218)
2ρ 2

78
Noting that mass conservation tells us that ρV is constant so, anywhere along the duct,
ρV = ρ 1V1 (the values at station 1):

ρ 21V12
CP T + = h0
2ρ 2
ρ 21V12
=⇒ C P T + ³ 2 ´ = h 0
(219)
2 RP2 T 2
 
ρ 21V12 R 2
=⇒   T 2 +C P T − h 0 = 0
2P 2

Solving for the temperature, T , and taking the positive solution:

s
2ρ 21V12 R 2
−C P + C P2 + h0
P2 (220)
T=
ρ 21V12 R 2
P2

The momentum equation Applying the momentum equation in the flow direction to
the control volume in Figure 36, from station 1 to station 2, we have:

P 1 A 1 + ṁV1 = P 2 A 2 + ṁV2 + friction force (221)

Since this is a constant-area duct, A 1 = A 2 = A, and so:

P 1 A + ṁV1 = P 2 A + ṁV2 + friction force (222)

The second law of thermodynamics The second law of thermodynamics states that:

1
T dS = dh − dP (223)
ρ

P
We write the density in terms of pressure and temperature (ρ = RT
) and express the
γR
enthalpy as a function of temperature (h = C P T where C P = γ−1 ) to obtain:

dT 1
d S = CP
− R dP
T P
(224)
dS dT γ − 1 dP
=⇒ = −
CP T γ P

Integrating this expression from station 1 to station 2:

S2 − S1 T2 γ−1 P2
µ ¶ µ ¶
= ln − ln (225)
CP T1 γ P1

79
Therefore, at any point along the duct:

S − S1 T γ−1 P
µ ¶ µ ¶
= ln − ln (226)
CP T1 γ P1

The Fanno line Station 1 represents the reference values of the flow quantities at the
entrance to the control volume. For known values of these parameters at station 1, it is
possible to calculate equivalent properties at the exit of the control volume (station 2)
as a function of the exit pressure, P 2 . Therefore, for any value of P 2 , it is possible to:
v
u 2 2 2
t 2 2ρ 1V1 R
u
−C P + C P + h0
P 22
1. calculate T2 from the energy equation (Equation 220): T2 =
ρ 21V12 R 2
P 22
γ−1
³ ³ ´ ³ ´´
2. calculate S 2 from the second law (Equation 225): S 2 = C P ln TT21 − γ ln PP 21 +S 1

P2
3. calculate ρ 2 from the ideal gas law: ρ 2 = RT 2

ρ 1 V1
4. calculate V2 from the conservation of mass (Equation 215): V2 = ρ2

5. calculate M 2 from the definition of Mach number: M 2 = p V2

γRT2

Therefore, for any given exit pressure, P 2 , the other flow properties can be determined.
When the above procedure is performed for a range of values for P 2 , we obtain the locus
of states with the same value of total enthalpy (h 0 ) and mass flux as the entry condi-
tions. This locus of flow properties is called the “Fanno line” from the initial condition
(Figure 37). Essentially, for a given known initial condition, the changes in flow prop-
erties as a result of friction through a constant area duct must lie along the Fanno line
through this initial point.

h M Fanno line
Fanno line

initial state

initial
state

S S

Figure 37: Enthalpy–entropy and enthalpy–Mach number diagram of Fanno flow.

80
2.1.2 Constant-Area Duct with Heat Transfer: Rayleigh Flow

The case of Rayleigh flow also considers compressible, one-dimensional flow through
a duct, but in this case we neglect the effects of friction. Instead, the changes in flow
conditions are the result of heat transfer, i.e. heating or cooling the flow. To analyse the
flow, we consider the control volume shown in Figure 38. Again, we apply the conser-
vation equations to this control volume: conservation of mass; the energy equation; the
momentum equation; the second law of thermodynamics.

heat transfer

P1 P2
V1 V2
ρ1 ρ2
T1 T2
S1 control S2
volume

heat transfer

Figure 38: Rayleigh flow: frictionless, constant-area duct flow with heat transfer.

Conservation of mass The conservation of mass states that the mass flow rate, ṁ =
ρ AV , must be constant. Since A does not change for a constant-area duct, we have:

ρV = constant
(227)
=⇒ ρ 1V1 = ρ 2V2

The energy equation The energy equation applied to the control volume in Figure 38
is given by:
" Ã ! #
V22 − V12
ṁ (h 2 − h 1 ) + + g (z 2 − z 1 ) = Q̇ + Ẇ (228)
2

For a horizontal duct, there are no changes in potential energy (z 2 − z 1 = 0). Also, in the
absence of mechanical work, we obtain:
à !
V22 − V12
(h 2 − h 1 ) + =q
2
V22 V12 (229)
=⇒ h 2 + = h1 + +q
2 2
=⇒ h 02 = h 01 + q

In this equation, q is the heat transferred per unit mass, q = Q/m. As before, h 0 is the
stagnation enthalpy.

81
The momentum equation Applying the momentum equation in the flow direction to
the control volume in Figure 38, there is no frictional force so we have:

P 1 A 1 + ṁV1 = P 2 A 2 + ṁV2 (230)

Since this is a constant-area duct, A 1 = A 2 = A, and so:

P 1 A + ṁV1 = P 2 A + ṁV2 (231)

Recalling that ṁ = ρ 1 AV1 = ρ 2 AV2 , we have:

P 1 A + ρ 1 AV12 = P 2 A + ρ 2 AV22
=⇒ P 1 + ρ 1V12 = P 2 + ρ 2V22
¢2 ¢2 (232)
ρ 1 V1 ρ 2 V2
¡ ¡
=⇒ P 1 + = P2 +
ρ1 ρ2

Using the ideal gas law to rewrite the density in terms of pressure and temperature, we
obtain:
¢2 ¢2
ρ 1 V1 ρ 2 V2
¡ ¡
P1 + P1
= P2 + P2
RT1 RT2 (233)
¢2 ¢2
ρ 1 V1 ρ 2 V2
¡ ¡
=⇒ P 1 + RT1 = P 2 + RT2
P1 P2

Therefore, along the duct:

¡ ¢2
ρV
P+ RT = constant (234)
P

The second law of thermodynamics Using the same approach as Fanno flow, the sec-
ond law of thermodynamics (T d S = d h − ρ1 d P ) leads to the following relation:

S2 − S1 T2 γ−1 P2
µ ¶ µ ¶
= ln − ln (235)
CP T1 γ P1

Therefore, at any point along the duct:

S − S1 T γ−1 P
µ ¶ µ ¶
= ln − ln (236)
CP T1 γ P1

The Rayleigh line Station 1 represents the reference values of the flow quantities at
the entrance to the control volume. For known values of these parameters at station 1,
it is possible to calculate equivalent properties at the exit of the control volume (station
2) as a function of the exit pressure, P 2 . Therefore, for any value of P 2 , it is possible to:

1. calculate T2 from the momentum equation (Equation 233), recalling also that ρV

82
 · ¸ · ¸
(ρ 1 V1 )2 (ρ 1 V1 )2
P1 + P1 RT1 −P 2 P 2 P1 + P1 RT1 −P 2 P 2
is constant (Equation 227): T2 = 2 = 2
R (ρ 2 V 2 ) R (ρ 1 V 1 )

γ−1
³ ³ ´ ³ ´´
2. calculate S 2 from the second law (Equation 235): S 2 = C P ln TT21 − γ ln PP 21 +S 1

P2
3. calculate ρ 2 from the ideal gas law: ρ 2 = RT 2

ρ 1 V1
4. calculate V2 from the conservation of mass (Equation 227): V2 = ρ2

5. calculate M 2 from the definition of Mach number: M 2 = p V2

γRT2

Therefore, for any given exit pressure, P 2 , the other flow properties can be determined.
When the above procedure is performed for a range of values for P 2 , we obtain the locus
of states with the same momentum flux and mass flux as the entry conditions. This lo-
cus of flow properties is called the “Rayleigh line” from the initial condition (Figure 39).
Essentially, for a given known initial condition, the changes in flow properties as a result
of heat transfer through a constant area duct must lie along the Rayleigh line through
this initial point.

h M Fanno line
Fanno line

initial state
Rayleigh
line Rayleigh line

initial
state

S S

Figure 39: Enthalpy–entropy and enthalpy–Mach number diagram of Rayleigh flow.

2.2 Normal Shock Waves

We can use a similar analysis to consider the flow through a normal shock wave (Fig-
ure 40) and to therefore establish the governing equations. In deriving such equations,
it is reasonable to assume that shock wave has a very small thickness, thus acting as a
discontinuity across which the flow properties experience an abrupt change. We can
also assume that the flow is steady and one-dimensional, i.e. the flow upstream and
downstream of the shock wave is uniform. We further assume that there is neither fric-
tion nor heat transfer at the shock because the abrupt changes occur over a relatively
small surface. Finally, any chemical reactions, including evaporation and condensation,
are ignored. It is important to note that, since changes to flow properties are assumed

83
to occur abruptly, the transition from upstream conditions to downstream conditions
cannot be isentropic, and thus must involve irreversibility.

P1 normal P2
V1 shock V2
ρ1 ρ2
T1 T2
S1 control S2
volume

Figure 40: Control volume around a stationary normal shock wave.

The control volume around the normal shock wave in Figure 40 is such that the flow
condition before the shock is denoted by subscript 1 and the properties after the shock
are represented by subscript 2. For a one-dimensional, adiabatic flow in the absence of
friction we consider each governing equation as before.

Conservation of mass The conservation of mass states that the mass flow rate, ṁ =
ρ AV , must be constant. Since A does not change for a constant-area duct, we have:

ρV = constant
(237)
=⇒ ρ 1V1 = ρ 2V2

The energy equation The flow is adiabatic and so the stagnation enthalpy remains
unchanged, analogous to the energy equation for Fanno flow. Thus, the energy equation
applied to the control volume in Figure 40 is given by:

V12 V22
h1 + = h2 + = h0 (238)
2 2

In this equation, h is static enthalpy and h 0 is the stagnation enthalpy.

The momentum equation Applying the momentum equation in the flow direction
to the control volume in Figure 40, we note that there is no frictional force and so the
momentum equation takes the same form as for Rayleigh flow:

P 1 A 1 + ṁV1 = P 2 A 2 + ṁV2
ṁ (239)
=⇒ P 2 − P 1 = (V2 − V1 )
A

The second law of thermodynamics Using the same approach as Fanno flow and
Rayleigh flow, the second law of thermodynamics (T d S = d h − ρ1 d P ) leads to the fol-

84
lowing relation:

S2 − S1 T2 γ−1 P2
µ ¶ µ ¶
= ln − ln (240)
CP T1 γ P1

Physical interpretation An inspection of the governing equations shows that, for known
initial conditions (state 1), the post-shock properties (state 2) must satisfy:

• the same continuity equation as Fanno and Rayleigh flows;

• the same energy equation as Fanno flow;

• the same momentum equation as Rayleigh flow;

• the same second law of thermodynamics as Fanno and Rayleigh flows.

Therefore, for any given initial condition, the flow condition downstream of the shock
must be on the Fanno line through the initial state, which is defined by the continuity
equation, energy equation and second law of thermodynamics for adiabatic Fanno flow.
Similarly, the flow condition downstream of the shock must also be on the Rayleigh line
through the initial state, which is defined by the continuity equation, momentum equa-
tion and second law of thermodynamics for frictionless Rayleigh flow. Therefore, the
downstream condition must be at the intersection of the Fanno line and the Rayleigh
line that passes through the initial state, as shown in Figure 41.

h M Fanno line
Fanno line final state
(Mach 0.58)
initial state
(Mach 2)

Rayleigh line
normal Rayleigh
shock line
normal shock

initial state
(Mach 2)
final state
(Mach 0.58)

S S

Figure 41: Enthalpy–entropy and enthalpy–Mach number diagram for a normal shock.

2.2.1 Requirement for the Formation of Shock Waves

It is possible to use these findings to establish when a shock wave is able to form. A
shock wave results in movement from one point on a Fanno line (or Rayleigh line) to
another, and must also result in an entropy increase (second law of thermodynamics).
It is therefore worth evaluating the point of maximum entropy along a Fanno line (or

85
Rayleigh line). Considering a Fanno line, every point along this curve must have con-
stant total enthalpy:
V2
h+ = constant =⇒ d h + V dV = 0 (241)
2
Similarly, every point along the Fanno line must also satisfy continuity:

ρV = constant =⇒ ρdV + V d ρ = 0 (242)

However, we also know from thermodynamics that:

dP
T ds = dh − (243)
ρ

At the point of maximum entropy along the Fanno line, we know that d s = 0 and so:

dP
dh = (244)
ρ

We can combine Equation 241 and Equation 244 to obtain:

d P + ρV dV = 0 (245)

At the same time, we know from Equation 242 (by multiplying both sides by V ) that:

V 2 d ρ + ρV dV = 0 (246)

Therefore, by subtracting Equation 245 from Equation 246, we have, at the point of max-
imum entropy: s
dP dP
V2= =⇒ V = =a (247)
dρ dρ
where a is the speed of sound. Therefore, the point of maximum entropy on a Fanno line
corresponds to the sonic velocity. Similarly, it can be shown that the point of maximum
entropy on a Rayleigh line also corresponds to sonic velocity. In order to establish the
necessary condition for the formation of a shock wave, we recall that the flow through a
shock wave is irreversible and so, according to the second law of thermodynamics, the
entropy must increase, d s > 0. Therefore,

dP
T d s > 0 =⇒ d h − >0
ρ
dP
=⇒ d h > (248)
ρ
dh
=⇒ ³ ´ > 1
dP
ρ

86
Let us call this ratio, d h/ d P /ρ = α. Therefore the requirement that the entropy must
¡ ¢

increase through a shock wave is equivalent to saying that:

α>1 (249)

From conservation of energy, we know that:

V2
h+ = constant =⇒ d h + V dV = 0
2
dP (250)
=⇒ α + V dV = 0
ρ
=⇒ αd P + ρV dV = 0

From conservation of mass, we have:

ρV = constant =⇒ ρdV + V d ρ = 0
(251)
=⇒ ρV dV + V 2 d ρ = 0

Combining Equation 250 and Equation 251, we obtain:

dP
αd P − V 2 d ρ = 0 =⇒ V 2 = α
dρ
=⇒ V 2 = αa 2 (252)
V p
=⇒ = α
a

Given that we know that α > 1 (from Equation 249, this means that:

V
>1 (253)
a

It is therefore clear that shock waves can only form if the velocity is greater than the
speed of sound or, in other words, if the flow is supersonic. Therefore, as shown in
Figure 41, a normal shock wave takes the flow from supersonic to subsonic with an as-
sociated increase in entropy.

2.2.2 Working Equations of Normal Shock Waves

Flow properties change across the normal shock wave and so it is useful to derive ex-
pressions to describe these changes in conditions such as temperature, pressure and
Mach number.

Stagnation temperature The shock wave is an adiabatic process and thus the stagna-
tion enthalpy (or total enthalpy) does not change across it:

h 01 = h 02 (254)

87
If the gas has a specific heat capacity at constant pressure (C P ) which remains constant
through the process, then the face that the stagnation enthalpy remains the same means
that the stagnation temperature also does not change across a shock wave:

T01 = T02 (255)

Physically, stagnation temperature can be thought of as a measure of the overall energy

in the system (kinetic as well as thermal). A shock wave converts kinetic energy (in the
form of flow velocity) to thermal energy but does not change the overall energy of the
flow. Note that the assumption of constant C P does not hold for hypersonic flows and
thus, whilst the shock wave is still adiabatic, the stagnation temperature does not re-
main the same.

Static temperature We recall the relationship between stagnation temperature and

static temperature:

T01 γ−1 2
= 1+ M1
T1 2
(256)
T02 γ−1 2
= 1+ M2
T2 2

Since stagnation temperature remains constant across the shock wave (T01 = T02 ) we
can combine the two equations (dividing one by the other) to obtain the relation for
static temperature change across the normal shock:

γ−1
2
T2 1 + 2 M 1
= (257)
T1 1 + γ−1 M 2
2 2

The static temperature, which is a measure of the thermal energy of the flow, increases
through a normal shock since some kinetic energy has been converted to thermal en-
ergy.

Static pressure The relationship for the static pressure ratio can be derived by starting
from the momentum equation:

P 1 + ρ 1V12 = P 2 + ρ 2V22 (258)

By converting from velocity (V ) to Mach number (M = V / γRT ), we obtain:

p

P 1 + ρ 1 γRT1 M 12 = P 2 + ρ 2 γRT2 M 22 (259)

88
We now use the ideal gas law (P = ρRT ) to write:

P 1 + γP 1 M 12 = P 2 + γP 2 M 22
=⇒ P 1 1 + γM 12 = P 2 1 + γM 22
¡ ¢ ¡ ¢
(260)
P 2 1 + γM 12
=⇒ =
P 1 1 + γM 22

The static pressure increases through a normal shock because it represents a collection
of infinitesimal compression waves (each of which cause a pressure rise) that have coa-
lesced to form a discontinuity.

Mach number Let us start by determining the relationship between the Mach number
and the static pressure/static temperature ratios across the shock wave. From the ideal
gas law, we have:

P 1 = ρ 1 RT1
(261)
P 2 = ρ 2 RT2

Dividing one equation by the other gives:

T2 P 2 ρ 1
= (262)
T1 P 1 ρ 2

From continuity (ρ 1V1 = ρ 2V2 ) we obtain:

s
T2 P 2V2 P 2 M 2 γRT2 P 2 M 2 T2
p
= = = (263)
T1 P 1V1 P 1 M 1 γRT1 P 1 M 1 T1
p

Therefore, the ratio of Mach numbers is:

s
M2 P 1 T2
= (264)
M1 P 2 T1

We already know the static pressure ratio (Equation 260) and the static temperature ratio
(Equation 257):

P 2 1 + γM 12
=
P 1 1 + γM 22
2γ−1 (265)
T2 1 + 2 M 1
=
T1 1 + γ−1 M 2
2 2

89
Inserting these expressions into Equation 264, we obtain:
v
γ−1
u
2
M 2 1 + γM 12 u 1 + 2 M 1
u
=
M 1 1 + γM 22 1 + γ−1 M 2
t
2 2 (266)
!2 γ−1
1 + 2 M 12
Ã
M 22 1 + γM 22
=⇒ = γ−1
M 12 1 + γM 12 1 + 2 M 22

We can now solve this equation for M 22 , which gives the Mach number downstream of a
normal shock as a function of the upstream Mach number (M 1 ):

γ − 1 M 12 + 2
¡ ¢
M 22 = (267)
2γM 12 − γ − 1
¡ ¢

Stagnation Pressure In order to determine the variation of stagnation pressure across

a normal shock wave, we start by noting that:

P 02 P 02 P 2 P 1
= × × (268)
P 01 P 2 P 1 P 01

We already have expressions for all three terms. We have determined the static pressure
ratio across the shock (Equation 260) and also know how the stagnation pressures are
related to the static pressures:

¶ γ ¶ γ
P 02 T02 γ−1 γ − 1 2 γ−1
µ µ
= = 1+ M2
P2 T2 2
P 2 1 + γM 12
= (269)
P 1 1 + γM 22
¶ γ ¶ γ
P 01 T01 γ−1 γ − 1 2 γ−1
µ µ
= = 1+ M1
P1 T1 2

Inserting these relations into Equation 268, we obtain:

¶ γ ¶ γ
P 02 γ − 1 2 γ−1 1 + γM 12 γ − 1 2 − γ−1
µ µ
= 1+ M2 × × 1+ M1 (270)
P 01 2 1 + γM 22 2

To ensure that the right-hand side is only a function of M 1 , we can use Equation 267 to
express every instance of M 2 in terms of M 1 . Following some rearrangement, we then
obtain the stagnation pressure ratio:

"µ 1
!γ #− γ−1
¶Ã
P 02 2γ γ−1 γ−1 2
= M 12 − +¡ (271)
γ+1 γ+1 γ+1 γ + 1 M 12
¢
P 01

The stagnation pressure, which is a measure of the kinetic energy of the flow, decreases
through a normal shock since some kinetic energy has been converted to thermal en-
ergy.

90
 Summary of working equations The working equations for normal shock waves de-
scribe how each of the flow properties vary through the shock. Now that we have an
expression for M 2 (M 1 ), we can write all these relations in terms of M 1 , the Mach num-
ber ahead of the normal shock.

γ − 1 M 12 + 2
¡ ¢
Mach number: M 22 =
2γM 12 − γ − 1
¡ ¢

P2 2γ γ−1
Static pressure: = M 12 −
P1 γ+1 γ+1

ρ 2 V1 γ + 1 M 12
¡ ¢
Density and velocity: = =
ρ 1 V2 2 + γ − 1 M 12
¡ ¢
  
T2 2γ γ−1 γ−1 2
Static temperature: = M2 −  +¡ 
γ + 1 1 γ + 1 γ + 1 γ + 1 M 12
¢
T1

1
  γ −
P 02 2γ γ−1 γ−1 2 γ−1
Stagnation pressure: =  M 12 −  +¡ ¢ 2 
P 01 γ+1 γ+1 γ + 1 γ + 1 M1

A useful way to visualise these changes in properties is to consider the temperature–

entropy diagram for a normal shock wave, which is shown in Figure 42.

stagnation states P 01

P 02 < P 01
T02 = T01
P2 > P1
T2 > T1 post-shock
(state 2)
temperature, T
P1

T1
pre-shock
(state 1)

entropy, s s1 s2 > s1

Figure 42: Temperature–entropy diagram for a normal shock wave.

Entropy production Figure 42 shows that a shock wave is associated with an entropy
rise, which is due to the fact it is an irreversible process. The increase in entropy for
a perfect gas is related to the changes in stagnation pressure and temperature by the

91
following equation:    
T02 P 02
∆s = C P ln   − R ln   (272)
T01 P 01

Since shock waves are adiabatic, the stagnation temperature does not change across a
shock wave (T02 = T01 ). Therefore, the production in entropy is only due to the reduction
in stagnation pressure:  
P 02
∆s = −R ln   (273)
P 01

Shock waves versus an isentropic process It is also useful to consider the relation-
ship between the pressure ratio and the density ratio across the shock wave, also known
as the Rankine–Hugoniot relation. From the previous definitions of how the density
and static pressure vary through the shock, it can be shown that the Rankine–Hugoniot
relation takes the form:

γ + 1 ρ2 γ + 1 P2
+1 +1
P2 γ − 1 ρ1 ρ2 γ − 1 P1
= =⇒ = (274)
P1 ρ2 γ+1 ρ1 P2 γ+1
− + +
ρ1 γ−1 P1 γ−1

Considering the limit of an infinitely strong shock, where M 1 → ∞, we have:

P2 2γ γ−1
= M 12 − →∞ (275)
P1 γ+1 γ+1

Considering this limit in Equation 274, the density ratio approaches the limit:

ρ2 γ+1
→ (276)
ρ1 γ−1

For air (γ = 1.4), this limit is ρ 2 /ρ 1 → 6.0. Figure 43 shows the variation of pressure
ratio (P 2 /P 1 ) against density ratio (ρ 2 /ρ 1 ) for a shock wave in air. On the same graph,
the relation between pressure and density for an isentropic process (P /ρ γ = constant)
is also shown. As the pressure ratio decreases, the shock wave becomes weaker and
the shock equation approaches the equation for an isentropic process. This means that
very weak shock waves can be treated as isentropic and, in this case, they are simply
compression waves.

2.2.3 Pitot Probes in Supersonic Flow

Pitot probes are commonly used in wind tunnel experiments to obtain pressure mea-
surements. When a Pitot tube exposed to a subsonic flow, the fluid decelerates isen-
tropically to rest, and thus the stagnation pressure is measured at the entrance to the
Pitot tube. In a wind tunnel, this measurement will therefore be equal to the stagnation
pressure in the settling chamber of the wind tunnel (where the flow can be assumed to

92
 102

shock wave
P 2 /P 1

101

isentropic

100
1 2 3 4 5 6
ρ 2 /ρ 1

Figure 43: The Rankine–Hugoniot curve compared to an isentropic process.

state 1
(at P 01 )

stagnation
streamline Pitot probe

supersonic
flow

state 2
bow shock wave (at P 02 < P 01 )
(almost normal at probe tip)

Figure 44: Pitot probe measurements in supersonic flow.

be at rest). At supersonic speed, however, the flow cannot go from a supersonic velocity
to rest isentropically. Instead, a shock wave is formed ahead of the entrance to the Pitot
tube, as shown in Figure 44.

If the Pitot probe is aligned with the flow, the portion of the shock wave directly in front
of the Pitot probe tip is almost normal to the flow and can thus be treated as a normal
shock wave. This means that the total pressure sensed by the Pitot tube is not the stag-
nation pressure of the flow (P 01 in Figure 44) but is actually the stagnation pressure after
the normal shock wave (P 02 ). This pressure can be related to the upstream stagnation
pressure using the normal shock wave equations:

"µ 1
!γ #− γ−1
¶Ã
P 02 2γ γ−1 γ−1 2
= M 12 − +¡ (277)
γ+1 γ+1 γ+1 γ + 1 M 12
¢
P 01

The flow stagnation pressure, P 01 , is typically be the same as the stagnation pressure

93
from the settling chamber. The Pitot probe measurement will provide the post-shock
stagnation pressure, P 02 . It is therefore possible to use this information to solve for M 1
in Equation 277 and thus obtain the Mach number of the flow ahead of the Pitot probe.

2.2.4 Moving Shock Waves

There are many practical applications in which the shock waves are not stationary, such
as the flow in a shock-tube, shock waves generated by explosions and the sudden clo-
sure of a valve in a gas pipeline. If the speed of the moving shock wave is constant, it is
possible to use the working equations derived for the stationary shocks to solve prob-
lems involving non-stationary shock waves. To do so, it is necessary to convert the flow
field to a frame of reference in which the shock wave is still to apply the standard equa-
tions, as shown in Figure 45.

moving shock wave

P1 Vs P2
stationary V1 = 0 V2
flow ρ1 ρ2

stationary shock wave

P1 P2
Vs V s − V2
ρ1 ρ2

control volume

Figure 45: Conversion of moving shock waves to a stationary reference frame.

In this example of a moving shock wave, the shock is moving to the left at speed, Vs .
Therefore, the frame of reference in which the shock is stationary involves adding a
speed, Vs to the right. Therefore the flow at the left of the shock is now travelling at
0+Vs = Vs . The shock itself is travelling at −Vs +Vs = 0. The flow at the right of the shock
was previously travelling at −V2 (since it was flowing towards the left) so, in the new ref-
erence frame, has velocity −V2 + Vs = Vs − V2 . The normal shock equations can now be
applied as usual to the velocities in this new reference frame, where the shock wave is

94
stationary. For example, we know that the velocity ratio across a shock wave is:

γ + 1 M 12 γ + 1 V12
¢¡ ¡ ¢
Vpre-shock
= = (278)
Vpost-shock 2 + γ − 1 M 12 2a 12 + γ − 1 V12
¡ ¢ ¡ ¢

Here, a 1 is the speed of sound in region 1, to the left of the shock wave. Applying this
relation to the reference frame with a stationary shock in Figure 45, we obtain:

γ + 1 Vs2
¡ ¢
Vs
= (279)
Vs − V2 2a 12 + γ − 1 Vs2
¡ ¢

Thus, if we knew the temperature in region 1 (and thus we knew a 1 ) and the veloc-
ity of air in region 2 (V2 ), we would be able to solve for Vs to determine the speed of
the shock wave. Note that the static pressure, static temperature and density are not
affected by this change of reference frame (although the stagnation quantities are af-
fected by changing reference frame). Therefore, in the case of a moving shock, the gas
over which the shock has passed is at a higher pressure than the undisturbed gas.

2.3 Flows through Variable-Area Ducts

In this module, we restrict ourselves to considering steady, one-dimensional flows which
are isentropic in all regions apart from shock waves. However, we do consider the ef-
fect of area changes to the duct in which these one-dimensional flows are contained.
For example, consider a large reservoir storing a gas at pressure P 0 , which is connected
to the external atmosphere through a variable-area duct. The duct can take a number
of shapes: convergent, divergent, convergent–divergent and divergent–convergent, as
shown in Figure 46. In this figure, the supply pressure, P 0 , is usually the stagnation pres-
sure at the inlet of the duct which, in jet engines, corresponds to the stagnation pressure
at the exit of the turbine or afterburner. Meanwhile, P b is the local back pressure which,
for jet engines and rockets, is the local atmospheric pressure at flight altitude.

Regardless of the shape, the flow through the duct depends on a number of parameters,
including the pressure ratio (P 0 /P b ), the area change, surface friction, heat transfer, etc.
The combined effect of these parameters determines the mass flow rate as well as the
distribution of velocity, pressure, temperature and density along the duct. At this stage,
we ignore other effects and consider only the effects of (1) pressure ratio (P 0 /P b ) and
(2) area change on the flow field. (Note that, although the viscous effects in the form of
wall friction are always present, their effect is usually secondary to the one-dimensional
flow behaviour and so, for now, we assume that the flow is frictionless.) The flow in
such variable-area ducts impacts the performance of practical applications which fea-
ture these geometries, such as the nozzles and diffusers of jet engines and rockets.

First, let us consider the effect of area change for one-dimensional, steady, isentropic
flow. Since the flow is isentropic (d s = 0), we can say that it is adiabatic (δQ = 0) and

95
convergent duct divergent duct

P0 Pb P0 Pb

convergent–divergent duct divergent–convergent duct

P0 Pb P0 Pb

Figure 46: Examples of variable-area duct driven by a reservoir at P 0 .

A +d A

control
volume
A

V V + dV
P P + dP
T T + dT
ρ ρ + dρ
M M +dM

dx

Figure 47: Control volume in a variable area duct.

96
reversible (δW = 0). We also assume that the drag is horizontal such that g d z = 0. Now
let us consider each of the conservation equations in turn, using the control volume in
Figure 47.

Conservation of mass Conservation of mass tells us that the mass-flow rate through
the duct is constant, and thus:

ṁ = constant
(280)
=⇒ ρ AV = constant

Taking the differential of this mass flow rate from the start to the end of the control
volume, we have:

d ṁ = d ρ AV = AV d ρ + ρV d A + ρ AdV = 0
¡ ¢
(281)

Dividing through by ρ AV , we obtain:

d ρ d A dV
+ + =0 (282)
ρ A V

The energy equation The energy equation is given by the fact that the stagnation en-
thalpy, h 0 (= C P T0 ) is constant:

C P T0 = constant
µ
V2
¶ (283)
=⇒ C P T + = constant
2C P

Taking the differential of this equation from the start to the end of the control volume,
(recalling that C P is a constant) we have:

V2
· µ ¶¸
d CP T + =0
2C P
V2
µ ¶
=⇒ C P d T + =0
2C P (284)
µ 2¶
V
=⇒ C P d T + d =0
2
=⇒ C P d T + V dV = 0

We can rewrite this equation as:

dT V
+ dV = 0
T CP T
d T γR V2 dV
=⇒ + × × =0 (285)
T C P γRT V
d T γR M 2 dV
=⇒ + =0
T CP V

97
The momentum equation Applying the momentum equation in the flow direction to
the control volume in Figure 47, we have:

P A + ṁV + force = (P + d P ) (A + d A) + ṁ (V + dV ) (286)

Ignoring the d Ad P term, which is second-order small, we obtain:

force = Ad P + P d A + ṁdV (287)

This forms the basis of calculating the force on the flow due to the tunnel walls. In turn,
Newton’s third law allows us to relate this quantity to the force in the streamwise direc-
tion that the flow exerts on the duct walls. Such calculations are essential to determine
the thrust produced by the nozzle in propulsion applications but, for now, we will leave
the momentum equation at this point.

Equation of state The ideal gas law allows us to relate the flow properties at the start
and the end of the control volume:

P = ρRT
=⇒ d P = d ρRT
¡ ¢

=⇒ d P = RT d ρ + Rρd T
(288)
dP dρ dT
=⇒ = +
ρRT ρ T
dP dρ dT
=⇒ = +
P ρ T

Isentropic flow Since we know that the flow within the variable-area duct is constant,
we can use the isentropic relation within our control volume:

P
= constant
ργ
=⇒ P ρ −γ = constant
=⇒ d P ρ −γ = 0
¡ ¢

=⇒ ρ −γ d P + P d ρ −γ = 0
¡ ¢
(289)
=⇒ ρ −γ d P + P −γρ −γ−1 d ρ = 0
¡ ¢

=⇒ d P = P γρ −1 d ρ
dP dρ
=⇒ =γ
P ρ

98
Combining equations We start by substituting the expression for d P /P from Equa-
tion 289 into Equation 288:

dρ dρ dT
γ= +
ρ ρ T
(290)
dT ¡ ¢ dρ
=⇒ = γ−1
T ρ

We already have an expression for d T /T in Equation 285, which gives:

γR M 2 dV ¡ ¢ dρ
− = γ−1
CP V ρ
(291)
dρ γR M 2 dV
=⇒ +¡ =0
ρ γ − 1 CP V
¢

γR
Recalling that C P = γ−1 , we therefore have:

dρ dV
= −M 2 (292)
ρ V

In order to obtain an expression for the change in flow velocity as a function of duct area
change, we now substitute d ρ/ρ for the expression obtained in Equation 282:

d A dV dV
µ ¶
− + = −M 2
A V V
¡ 2 ¢ dV dA
=⇒ M − 1 = (293)
V A
dV 1 dA
=⇒ = 2
V M −1 A

We can also obtain similar equations for the other flow properties as a function of area
change. For example, Equation 292 gives the density variation:

dρ M2 d A
=− 2 (294)
ρ M −1 A

In turn, Equation 290 provides the variation in temperature:

γ−1 M2 d A
¡ ¢
dT
=− (295)
T M2 −1 A

Finally, we can use Equation 289 to obtain the pressure variation:

dP γM 2 d A
=− 2 (296)
P M −1 A

99
As these flow properties change, the speed of sound, a, is also affected. Recalling that
¢1/2
(1) a = γRT and that (2) γ and R are both constants, we have:
¡

1¡ ¢−1/2 ¡
γRT γRd T
¢
da =
2
1¡ ¢−1/2 ¡ ¢ dT
= γRT γRT
2 T
(297)
1¡ ¢1/2 d T
= γRT
2 T
1 dT
= a
2 T

Dividing both sides by a, we obtain:

da 1 dT
= (298)
a 2 T

Substituting in the temperature-variation relation (Equation 295), we can obtain an ex-

pression for how the speed of sound changes with area:

da γ−1 M2 d A
=− (299)
a 2 M2 −1 A

We can therefore obtain the change in Mach number (M = V /a) with area:

M = V a −1
=⇒ d M = a −1 dV + V −a −2 d a
¡ ¢

V dV V d a
=⇒ d M = −
a µV a a¶ (300)
dV d a
=⇒ d M = M −
V a
d M dV d a
=⇒ = −
M V a

Substituting in the change in velocity with area (Equation 293) and the change in sound
speed with area (Equation 299), we get:

dM 1 d A γ−1 M2 d A
= 2 +
M M −1 A 2 M2 −1 A
γ−1 (301)
dM 1+ 2 M2 d A
=⇒ =
M M2 −1 A

This is a key equation, which lets us determine how the Mach number through a duct
varies in response to any changes in area. In particular we can see that:

• a subsonic flow (M < 1) in a convergent duct (d A < 0) accelerates: M 2 −1 < 0 and

d A/A < 0 so the Mach number increases (d M /M > 0);

• a subsonic flow (M < 1) in a divergent duct (d A > 0) decelerates: M 2 − 1 < 0 and

d A/A > 0 so the Mach number decreases (d M /M < 0);

• a supersonic flow (M > 1) in a convergent duct (d A < 0) decelerates: M 2 − 1 > 0

100
 and d A/A < 0 so the Mach number decreases (d M /M < 0);

• a supersonic flow (M > 1) in a divergent duct (d A > 0) accelerates: M 2 −1 > 0 and

d A/A > 0 so the Mach number increases (d M /M > 0).

Figure 48 shows a summary of how all the flow properties change with area: velocity
(Equation 293), Mach number (Equation 301), pressure (Equation 296), temperature
(Equation 295) and density (Equation 294).

subsonic flow (M < 1) through subsonic flow (M < 1) through

convergent duct (d A < 0) divergent duct (d A > 0)
A A +d A
A +d A A

dV > 0 dV < 0
M <1 dM > 0 M <1 dM < 0
dP < 0 dP > 0
dT < 0 dT > 0
dρ < 0 dρ > 0

supersonic flow (M > 1) through supersonic flow (M > 1) through

convergent duct (d A < 0) divergent duct (d A > 0)
A A +d A
A +d A A

dV < 0 dV > 0
M >1 dM < 0 M >1 dM > 0
dP > 0 dP < 0
dT > 0 dT < 0
dρ > 0 dρ < 0

Figure 48: Change in flow properties through variable-area ducts.

2.3.1 Variable-Area Ducts and Sonic Speed

Consider a convergent duct with subsonic flow at the inlet. We know that any decrease
in area accelerates the flow, but it is worth considering whether there are any limits to

101
the acceleration. Let us start with Equation 301:

γ−1 2
dM 1+ 2 M dA
= (302)
M M2 −1 A

We have a subsonic flow (M < 1) in a converging duct, so as the area changes from A
to A + d A (d A < 0) then flow Mach number will change by d M (where d M > 0). If we
continue to decrease the area of the duct, the flow continues to accelerate (as long as
the applied pressure difference allows it) until it reaches sonic speed (M = 1). Now let us
consider what happens when there is any further area change beyond the point at which
we attain sonic speed. For this we will consider two cases: (1) a convergent section after
reaching sonic speed, and (2) a divergent section after reaching sonic speed.

Convergent section after reaching sonic speed After reaching M = 1 at section 1 in

Figure 49, let us see what happens when we further decrease the area. As a result of this
change the Mach number changes from M 1 = 1 to M 2 = 1+d M at section 2. For a subse-
quent decrease in area to section 3, the Mach number further changes to M 3 = 1+2d M .
There are three possibilities: (1) d M > 0, (2) d M < 0, or (3) d M = 0. In other words, after
reaching sonic velocity, any additional reduction in area can cause the Mach number
to increase (flow becomes supersonic), to decrease (flow becomes subsonic again) or to
stay the same (flow remains sonic).

1 2 3

M<1 M=1
(subsonic) (sonic)

M1 = 1
M2 = 1 + d M
M 3 = 1 + 2d M

Figure 49: Convergent section after reaching sonic speed.

If option (1) were true, then we would have d M > 0 so M 3 > M 2 > M 1 = 1. In other words
a convergent duct would be accelerating a supersonic flow. We have seen that this can-
not be true, since a supersonic flow decelerates in a convergent duct. If option (2) were
true, then we would have d M < 0 so M 3 < M 2 < M 1 = 1. In other words a convergent
duct would be decelerating a subsonic flow. We have seen that this cannot be true, since
a subsonic flow accelerates in a convergent duct. Therefore, the only remaining option
is case (3) where d M = 0. Effectively this means that if the area further decreases beyond
M = 1 then the sonic condition moves and positions itself at the point of lower area. For

102
a convergent nozzle, this will simply be at the exit of the nozzle.

This argument suggests that, in convergent ducts, if the flow reaches sonic speed at the
exit then any further reduction in area does not change the exit Mach number. This
condition is referred to as the flow being choked. For a fixed exit area, supply pressure
and supply temperature the choked condition corresponds to the maximum mass flow
rate that can go through the nozzle.

Divergent section after reaching sonic speed Now let us consider the opposite situ-
ation, where the area increases after reaching sonic speed at section 1 (Figure 50. The
increase in area causes the Mach number to change from M 1 = 1 to M 2 = 1 + d M at
section 2. For a further increase in area to section 3, the resultant change in Mach num-
ber results in M 3 = 1 + 2d M . There are two possibilities: (1) d M > 0 or (2) d M < 0. In
other words, after reaching sonic velocity, any subsequent increase in area can cause
the Mach number to increase (flow becomes supersonic) or to decrease (flow becomes
subsonic again).

1 2 3

M<1 M=1
(subsonic) (sonic)

M1 = 1
M2 = 1 + d M
M 3 = 1 + 2d M

Figure 50: Divergent section after reaching sonic speed.

If option (1) were true, then we would have d M > 0 so M 3 > M 2 > M 1 = 1. In other words
a divergent duct would be accelerating a supersonic flow, which is possible. If option (2)
were true, then we would have d M < 0 so M 3 < M 2 < M 1 = 1. In other words a divergent
duct would be decelerating a subsonic flow, which is also possible. Note that we do
not consider the case d M = 0 since a change in area always causes a change in Mach
number – we only discussed this possibility for the convergent section because neither
option (1) nor option (2) were possible in that case.

We refer to the point of minimum area (section 1 in Figure 50) as the throat. The mag-
nitude of Mach number variation depends on how much the area increases after sec-

103
tion 1, according to Equation 301. It turns out that, for a given area ratio (A exit /A throat ),
whether we get accelerating supersonic flow (option (1)) or decelerating subsonic flow
(option (2)) in the divergent section of the nozzle depends on the pressure ratio across
the nozzle. This is essentially the ratio between the supply pressure and the pressure
at the exit, which is often the local atmospheric pressure outside the nozzle (the back
pressure).

These arguments imply that after reaching M = 1, an increase in area can either accel-
erate or decelerate the flow, with both solutions physically possible. Therefore, we can
conclude that, to accelerate or decelerate flow from the sonic condition, the area must
be increased. Furthermore, for a fixed area ratio, the operation of the nozzle depends
on the pressure ratio across the nozzle. Finally, we now know that, in order to accelerate
the flow from subsonic to supersonic speeds, the flow must go through a convergent–
divergent duct. Given that we know the effect of area change on different flow param-
eters, we can now quantify what area change is required to accelerate or decelerate the
flow to another Mach number. We will consider a number of different cases in turn
which illustrate these processes.

Case 1: Acceleration from subsonic to sonic In order for a subsonic flow to become
sonic, it needs to accelerate and therefore we need to use a convergent duct. Figure 51
describes such a process. At the inlet (station 1), the Mach number is M 1 < 1. We con-
ventionally denote a station where the velocity is sonic by ∗ – in this case, this station is
at the exit, M ∗ = 1.

1 *

M<1 M=1
(subsonic) (sonic)

M1 M∗ = 1
A1 A∗

Figure 51: Convergent duct accelerating a flow from subsonic to sonic.

For one-dimensional flow the mass flow rate at any point is ṁ = ρ AV . We can write
density in terms of pressure and temperature (using the ideal gas equation) and we can
write the velocity in terms of Mach number and temperature:

γ
r
P P
AM γRT =
p
ṁ = AM p (303)
RT R T

104
The local static pressure is related to the corresponding local stagnation pressure by the
isentropic relation:
  γ
γ−1
P0 γ−1
= 1 + M 2 (304)
P 2

Similarly, the local static temperature is related to the corresponding local stagnation
temperature:
T0 γ−1
= 1+ M2 (305)
T 2
Inserting these relations into Equation 303, we obtain:
  − γ   −1 −1/2
γ−1
γ γ−1 γ−1
r
ṁ = AM P 0 1 + M 2  T0 1 + M 2 
  
R 2 2
− γ 1
γ−1 + 2

γ γ−1
r
P0 (306)
= AM p 1 + M 2
R T0 2
 − γ+1
2(γ−1)
γ γ−1
r
P0
=p AM 1 + M 2
T0 R 2

We apply conservation of mass from station 1 to station ∗ (ṁ 1 = ṁ ∗ ) and apply this
expression for the mass-flow rate to obtain:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ γ−1 P 0∗ γ ∗ ∗ γ−1
r r
P 01
p A 1 M 1 1 + M 12  =q A M 1+ M ∗2 
T01 R 2 T0∗ R 2
 − γ+1  − γ+1 (307)
2(γ−1) 2(γ−1)
P 01 γ−1 P 0∗ γ−1
=⇒ p A 1 M 1 1 + M 12  = q A ∗ M ∗ 1 + M ∗2 
T01 2 T0∗ 2

We are considering isentropic flows and thus we have P 01 = P 0∗ and T01 = T0∗ :

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 γ−1
A 1 M 1 1 + M 12  = A ∗ M ∗ 1 + M ∗2  (308)
2 2

Furthermore, we know that the exit (station ∗) has M ∗ = 1, and so:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 γ+1
A 1 M 1 1 + M 12  = A∗   (309)
2 2

105
We can rearrange this to obtain:

 − γ+1
2(γ−1)
γ+1
A1 1 
 
=  2 
 (310)
A ∗ M1 
 γ−1


1+ M 12
2

This is a key equation which allows us to calculate the area ratio required to accelerate
a subsonic flow at a given Mach number, M 1 , to the sonic condition. Note that, if we
wanted to perform the opposite process to decelerate a sonic flow at M ∗ = 1 to a sub-
sonic flow with Mach number, M 2 , we would need to use a divergent duct (Figure 52)
with an area ratio given by the same equation:

 − γ+1
2(γ−1)
γ+1
A2 1 
 
=  2 
 (311)
A ∗ M2 
 γ−1


1+ M 22
2

* 2

M=1 M<1
(sonic) (subsonic)

M∗ = 1 M2
A∗ A2

Figure 52: Divergent duct accelerating a flow from sonic to subsonic.

Case 2: Deceleration from supersonic to sonic In order for a supersonic flow to be-
come sonic, it needs to decelerate and therefore we need to use a convergent duct. Fig-
ure 53 describes such a process. At the inlet (station 1), the Mach number is M 1 < 1 and
at the outlet, the velocity is sonic (M = 1) so we denote this station as ∗. In order to cal-
culate the area ratio required to decelerate the supersonic from M 1 to M ∗ = 1, we simply
use the area-ratio equation that we derived for Case 1 (Equation 310):

 − γ+1
2(γ−1)
γ+1
A1 1 
 
=  2 
 (312)
A ∗ M1 
 γ−1


1+ M 12
2

106
 1 *

M>1 M=1
(supersonic) (sonic)

M1 M∗ = 1
A1 A∗

Figure 53: Convergent duct decelerating a flow from supersonic to sonic.

Again the converse is also true – in order to accelerate a flow from sonic (M ∗ = 1) to a
supersonic Mach number, M 2 , we would need to use a divergent duct (Figure 54) with
an area ratio given by the same equation:

 − γ+1
2(γ−1)
γ+1
A2 1 
 
=  2 
 (313)
A ∗ M2 
 γ−1


1+ M 22
2

* 2

M=1 M>1
(sonic) (supersonic)

M∗ = 1 M2
A∗ A2

Figure 54: Divergent duct accelerating a flow from sonic to supersonic.

Case 3: Acceleration of a subsonic flow In order to accelerate a subsonic flow, we need

to use a convergent duct, as shown in Figure 55. At the inlet (station 1), the Mach num-
ber is M 1 and at the outlet (station 2), the Mach number is M 2 . Both stations are sub-
sonic (M 1 < 1 and M 2 < 1) and the flow is accelerating so M 2 > M 1 . We also include

107
station ∗ as a reference, because we know that the Mach number here is 1, so it acts as a
useful benchmark.

1 2 *

M<1 M<1 M=1

(subsonic) (subsonic) (sonic)

M1 M2 M∗ = 1
A1 A2 A∗

Figure 55: Convergent duct accelerating a subsonic flow.

We know that mass-flow considerations mean that:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 γ−1
A 1 M 1 1 + M 12  = A 2 M 2 1 + M 22 
2 2
 − γ+1

γ−1 2(γ−1) (314)

2
A 1 M2  1 + 2 M2 

=⇒ =  
A 2 M1 
 γ−1 

2
1+ M1
2

We can indeed use this equation directly to figure out the required area ratio. However,
it is often easier to consider the ratios A 1 /A ∗ and A 2 /A ∗ since these are often tabulated
in data books against the relevant Mach numbers, M 1 and M 2 , respectively. Therefore,
we can simply take these values from a data book and use:
³ ´
A1
A1 A∗
=³ ´ (315)
A2 A2
A∗

Again, the converse is also true. In order to decelerate a subsonic flow (M 1 < 1, M 2 < 1
and M 2 < M 1 ), we would require a divergent duct (Figure 56). In this case the benchmark
sonic station comes before station 1. We again use the same equation to determine the
necessary area ratio:
³ ´
A2
A2 A∗
=³ ´ (316)
A1 A1
A∗

108
 * 1 2

M=1 M<1 M<1

(sonic) (subsonic) (subsonic)

M∗ = 1 M1 M2
A∗ A1 A2

Figure 56: Divergent duct decelerating a subsonic flow.

Case 4: Deceleration of a supersonic flow In order to decelerate a supersonic flow,

we need to use a a convergent duct, as shown in Figure 57. At the inlet (station 1), the
Mach number is M 1 and at the outlet (station 2), the Mach number is M 2 . Both stations
are supersonic (M 1 > 1 and M 2 > 1) and the flow is decelerating so M 2 < M 1 . We also
include station ∗ as a reference, because we know that the Mach number here is 1, so it
acts as a useful benchmark.

1 2 *

M>1 M>1 M=1

(supersonic) (supersonic) (sonic)

M1 M2 M∗ = 1
A1 A2 A∗

Figure 57: Convergent duct decelerating a supersonic flow.

As before, we know that mass-flow considerations give Equation 314:

 − γ+1
2(γ−1)
γ−1
2
A1 M2  1 + 2 M2 

=   (317)
A 2 M1  γ−1 

2
1+ M1
2

Again, it is often easier to consider the ratios A 1 /A ∗ and A 2 /A ∗ which are tabulated in

109
data books against the relevant Mach numbers, M 1 and M 2 , respectively. Therefore, we
can simply take these values from a data book and use:
³ ´
A1
A1 A∗
=³ ´ (318)
A2 A2
A∗

Similarly, the opposite argument still holds. In order to accelerate a supersonic flow
(M 1 > 1, M 2 > 1 and M 2 > M 1 ), we would require a divergent duct (Figure 58) with the
benchmark sonic station before station 1. We again use the same equation to determine
the necessary area ratio:
³ ´
A2
A2 A∗
=³ ´ (319)
A1 A1
A∗

* 1 2

M=1 M>1 M>1

(sonic) (supersonic) (supersonic)

M∗ = 1 M1 M2
A∗ A1 A2

Figure 58: Divergent duct accelerating a supersonic flow.

2.3.2 Effect of Back Pressure on Convergent Ducts

The effect of back pressure on the flow through convergent ducts forms a key part on
the operation of such geometries in aircraft propulsion applications. Figure 59 shows a
convergent duct with fixed area ratio, as featured, for example, on the jet engines of civil
aircraft. A gas (such as air) enters the duct with a stagnation pressure, P 0 , and a stagna-
tion temperature, T0 . These stagnation quantities could, for example, correspond to the
stagnation conditions of the gases at the exit of a turbine or the stagnation conditions of
the air at the fan exit of a turbofan engine. The pressure outside the nozzle is the back
pressure, P b , which corresponds to the local atmospheric pressure, P atm . We want to
ascertain the effect of the pressure ratio across the nozzle (P 0 /P atm ) on the mass flow
rate going through the nozzle.

110
 j

P0
T0

P b = P atm

Mj
Aj
Pj

Figure 59: Effect of back pressure on flow through a convergent duct.

We can start by considering the mass flow rate at the exit of the nozzle, which is labelled
as section j in Figure 59:

Pj γ Pj
q r
ṁ = ρ j A j V j = Aj Mj γRT j = Aj Mj p (320)
RT j R Tj

The local static pressure is related to the corresponding local stagnation pressure by the
isentropic relation (Equation 304):

  γ
γ−1
P0 j γ−1
= 1 + M 2j  (321)
Pj 2

Similarly, the local static temperature is related to the corresponding local stagnation
temperature (Equation 305):
T0 j γ−1
= 1+ M 2j (322)
Tj 2
Inserting these relations into the expression for mass flow rate, we obtain:

 − γ+1
2(γ−1)
P0 j γ γ−1
r
ṁ = p A j M j 1 + M 2j  (323)
T0 j R 2

If the flow between the reservoir and the nozzle exit is isentropic, there should be no
change in stagnation quantities so P 0 j = P 0 and T0 j = T0 . Therefore, we have:

 − γ+1
2(γ−1)
γ γ−1
r
P0
ṁ = p A j M j 1 + M 2j  (324)
T0 R 2

111
It is common to define a non-dimensional version of the mass flow rate:
p
ṁ C P T0
Q= (325)
P0 A j

Inserting our expression for mass-flow rate, we therefore obtain:

− γ+1
p

2(γ−1)
γ γ−1
r
C P T0 P 0 2
Q= p A j M j
1 + M j
P0 A j T0 R 2
 − γ+1 (326)
2(γ−1)
CP γ γ−1
r
= M j 1 + M 2j 
R 2

Since R
CP = C PC−C
P
V
= 1 − γ1 , we have:

 − γ+1
2(γ−1)
γ γ−1
s
Q= M j 1 + M 2j 
1 − γ1 2
 − γ+1
2(γ−1)
γ−1
s
γ2
= M j 1 + M 2j  (327)
γ−1 2
 − γ+1
2(γ−1)
γ γ−1
=p M j 1 + M 2j 
γ−1 2

Therefore, this non-dimensional mass flow rate, Q is a function of local Mach number
and is commonly tabulated in data books. We can also write Q in terms of the local
stagnation pressure–static pressure ratio. From Equation 304, we have:
γ
P0 j γ−1 2
µ ¶
γ−1
= 1+ Mj (328)
Pj 2

Recalling that the flow through the nozzle is isentropic (P 0 j = P 0 ), we rearrange for M j :
v
u "µ ¶ γ−1 #
u 2 P 0 γ
Mj = t −1 (329)
γ−1 Pj

Inserting this into the definition for Q, we obtain:

v
u "µ ¶ γ−1 # µ ¶ γ+1
γ u 2 P0 γ P 0 − 2γ
Q=p t −1
γ−1 γ−1 Pj Pj
v
u "µ ¶ γ−1 # µ ¶ γ+1
γ u P0 γ P0 − γ (330)
= t2 −1
γ−1 Pj Pj
v
u "µ ¶ 2 µ ¶ γ+1 #
− − γ
γ ut2 P 0 γ − P 0
=
γ−1 Pj Pj

112
Now let us consider the behaviour of the mass flow rate as a function of the pressure
ratio across the nozzle, P 0 /P b .

No flow through the nozzle At a pressure ratio P 0 /P b = 1, there is no pressure differ-

ence across the nozzle. Since there is no pressure gradient, there is nothing to drive the
flow and so the mass flow rate is zero. In this situation, the pressure everywhere is equal
to P 0 , and thus P j = P b . Inserting the value, P 0 /P j = 1 into the expression for Q, we
simply obtain Q = 0 which matches out expectation for zero mass flow.

Subsonic flow through the nozzle As the pressure ratio, P 0 /P b , increases above 1, the
pressure differences induces a flow from left to right through the nozzle, and thus the
mass-flow rate becomes larger than zero. To start with, the flow is subsonic everywhere
and the flow adjusts itself such that the exit pressure (P j ) is equal to the back pressure
(P b ). In this scenario, the non-dimensional mass flow rate is given by:
v
u "µ ¶ 2 µ ¶ γ+1 #
− − γ
γ ut2 P 0 γ − P 0 (331)
Q=
γ−1 Pb Pb

Within this regime, we have P j = P b and so the Mach number at the exit is given by:
v
u "µ ¶ γ−1 #
u 2 P0 γ
Mj = t −1 (332)
γ − 1 Pb

Therefore, as the pressure ratio, P 0 /P b , continues to increase, the exit Mach number,
M j , also increases.

Sonic condition at nozzle exit As the pressure ratio increases, at some point the exit
Mach number, M j , reaches 1 (the sonic condition). At this point, we still have P j = P b
and so the critical pressure-ratio value at which the exit Mach number, M j , becomes 1
is:
¶ γ ¶ γ
P0 P0 γ − 1 2 γ−1 γ − 1 γ−1
µ ¶ µ ¶ µ µ
= = 1+ Mj = 1+ (333)
Pb critical Pj M =1
2 2
For air, γ = 1.4 and so this critical pressure ratio is:

1.4
P0
µ ¶ µ ¶
1.4 − 1 1.4−1
= 1+ = 1.89293 (334)
Pb critical 2

For typical exhaust gases, a more appropriate value for γ is 1.333, in which case:

¶ 1.333
P0
µ ¶ µ
1.333 − 1 1.333−1
= 1+ = 1.85242 (335)
Pb critical 2

113
At this critical pressure ratio, the nozzle is said to be choked. We can use Equation 327
to determine the non-dimensional mass-flow rate in this scenario, where M j = 1:

 − γ+1
2(γ−1)
γ γ−1
Q=p × 1 × 1 + × 12 
γ−1 2
 − γ+1 (336)
2(γ−1)
γ γ−1
=p 1 + 
γ−1 2

Substituting in values of γ = 1.4 and 1.333, we obtain Q = 1.281 for air and Q = 1.347 for
exhaust gases.

Underexpanded condition Now we consider what happens when the pressure ratio,
P 0 /P b , increases beyond the critical value. Equation 332 shows that, if it was still true
that P j = P b then:
v
u "µ ¶ γ−1 #
u 2 P0 γ
Mj = t −1 (337)
γ − 1 Pb

In this case, for P 0 /P b greater than the critical value, M j would exceed 1 and the flow
would become supersonic. However, our analysis of incremental area changes in con-
vergent ducts has shown that flow cannot go from subsonic to supersonic in a conver-
gent duct. Therefore, instead the flow remains sonic at the exit of the nozzle, M j = 1 and
thus P j ̸= P b . A consequence of this observation is that the pressure ratio to the exit,
P 0 /P j , remains at the value given by the critical sonic condition:

P0 P0
µ ¶
= (338)
Pj P b critical

Therefore, the value of P 0 /P b must be greater than P 0 /P j which means that the exit
pressure, P j , must be higher than the back pressure, P b . Thus the flow exiting the noz-
zle requires a further pressure drop to match the pressure of the surrounding air. The
required pressure drop is achieved through an expansion process outside the nozzle exit
which gradually reduces the pressure from P j to P b . Since further expansion is required
to meet equilibrium conditions, the flow in the nozzle is described as underexpanded.
Note that, since the Mach number remains at 1 in the underexpanded condition, the
non-dimensional flow rate remains at the value obtained for the critical condition:
 − γ+1
2(γ−1)
γ  γ−1
Q=p 1+  (339)
γ−1 2

This value of Q is therefore the maximum (non-dimensional) mass flow rate that can be
achieved by the nozzle, and we therefore denote the value by Q max . Thus, Q max = 1.281
for air and Q max = 1.347 for exhaust gases. As the pressure ratio P 0 /P b continues to
increase, the Mach number remains 1, the non-dimensional mass flow rate remains at

114
Q max and the difference between P j and P b gets larger.

Figure 60 describes the variation in Mach number and non-dimensional mass flow rate
for a convergent nozzle defined by Figure 59.

critical critical
M condition Q condition
subsonic under- subsonic under-
flow expanded flow expanded

Q max

1 pressure ratio, P 0 /P b 1 pressure ratio, P 0 /P b

Figure 60: Exit Mach number and non-dimensional mass flow versus pressure ratio.

2.3.3 Convergent–Divergent Ducts

We have seen that it is necessary to use a convergent–divergent nozzle in order to ac-

celerate a flow from subsonic to supersonic. Similar to convergent nozzles, there are a
series of different regimes as the pressure ratio from the reservoir (P 0 ) to the external
atmosphere (P b ) is increased. In this case, shown in Figure 61, there are two relevant
locations: (1) the point of minimum area, or throat, denoted by “t” and (2) the exit, de-
noted by “j”. Again, we can start by considering the mass flow rate. We can start by
considering the mass flow rate at an arbitrary station, “i”:

γ
r
Pi Pi
ṁ = ρ i A i Vi = A i M i γRTi =
p
A i Mi p (340)
RTi R Ti

The local static pressure is related to the corresponding local stagnation pressure by the
isentropic relation (Equation 304):

  γ
γ−1
P 0i γ−1
= 1 + M i2  (341)
Pi 2

Similarly, the local static temperature is related to the corresponding local stagnation
temperature (Equation 305):
T0i γ−1
= 1+ Mi j 2 (342)
Ti 2

115
Inserting these relations into the expression for mass flow rate, we obtain:

 − γ+1
2(γ−1)
γ γ−1
r
P 0i
ṁ = p A i M i 1 + M i2  (343)
T0i R 2

Now let us consider the behaviour of the mass flow rate as a function of the pressure
ratio across the nozzle, P 0 /P b .

t j
Pb

P0
T0

Mt Mj
At Aj
Pt Pj

Figure 61: Effect of back pressure on flow through a convergent–divergent duct.

No flow through the nozzle At a pressure ratio P 0 /P b = 1, there is no pressure differ-

ence across the nozzle. Since there is no pressure gradient, there is nothing to drive the
flow and so the mass flow rate is zero. In this situation, shown in Figure 62, the pressure
everywhere is equal to P 0 , and thus P j = P b .

Subsonic flow through the nozzle As the pressure ratio, P 0 /P b , increases above 1, the
pressure differences induces a flow from left to right through the nozzle, and thus the
mass-flow rate becomes larger than zero. To start with, the flow is subsonic everywhere
and the flow adjusts itself such that the exit pressure (P j ) is equal to the back pressure
(P b ). In this scenario, shown in Figure 63, the subsonic flow accelerates in the conver-
gent section up to the throat Mach number, M t , which must be less than 1 since the flow
is subsonic everywhere. The subsequent flow in the divergent section is also subsonic
and therefore decelerates through this region of the duct up to the exit Mach number,
M j . In this scenario, there are no shock waves (since there is no supersonic flow) and
thus the flow is isentropic everywhere and the stagnation pressure remains constant.

116
 t j
Pb
P0
=1

P0 M =0 M =0 M =0
T0

0
distance along nozzle

Figure 62: No flow through the nozzle.

Therefore, we have P 0 j = P 0t = P 0 . Since we also know that P j = P b , we obtain:

  γ
γ−1
P0 P0 j γ−1
= = 1 + M 2j  (344)
Pb Pj 2

Thus, the Mach number at the exit is given by:

v
u "µ ¶ γ−1 #
u 2 P0 γ
Mj = t −1 (345)
γ − 1 Pb

Therefore, as the pressure ratio, P 0 /P b , continues to increase, the exit Mach number,
M j , also increases.

Sonic flow at throat In the subsonic regime, the flow accelerates in the convergent
section and decelerates in the divergent section, so the peak Mach number must occur

117
 t j
h i
Pb Pb
1< P0
< P 0 choked

P0 M <1 M <1 M <1

T0

P j = Pb

0
distance along nozzle

Figure 63: Subsonic flow through the nozzle.

at the throat. As the pressure ratio, P 0 /P b , increases the Mach number everywhere in-
creases until the peak Mach number, at the throat, reaches 1 (the sonic condition), as
shown in Figure 64. At the throat, where M t = 1, the mass flow rate is given by:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ γ−1 γ  γ−1
r r
P 0t P 0t
ṁ t = p A t M t 1 + M t2  =p At 1 +  (346)
T0t R 2 T0t R 2

At the nozzle exit, the mass flow rate is given by:

 − γ+1
2(γ−1)
P0 j γ γ−1
r
ṁ j = p A j M j 1 + M 2j  (347)
T0 j R 2

118
 t j
h i
Pb Pb
P0
= P 0 choked

P0 M <1 M =1 M <1
T0

P j = Pb

0
distance along nozzle

Figure 64: Sonic flow at throat.

Equating the two mass flow rates, which must be the same, we have ṁ j = ṁ t , and so:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
P0 j γ γ−1 γ  γ−1
r r
P 0t
p A j M j 1 + M 2j  =p At 1 +  (348)
T0 j R 2 T0t R 2

Assuming isentropic flow (no shock waves) throughout the duct, we have T0 j = T0t = T0
and P 0 j = P 0t = P 0 . Therefore:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
P0 γ−1 P0 γ−1
p A j M j 1 + M 2j  = p A t 1 + 
T0 2 T0 2
 − γ+1  − γ+1 (349)
2(γ−1) 2(γ−1)
γ−1 At  γ−1
=⇒ M j 1 + M 2j  = 1+ 
2 Aj 2

A key parameter determining the exit Mach number is therefore the area ratio, A j /A t ,

119
between the exit of the nozzle at the throat. This equation has two solutions for M j , one
which is subsonic and the other which is supersonic. To find the solutions in practice,
we note that, since the throat is at Mach 1, A t = A ∗ . Therefore, the Mach numbers that
are listed against the area ratio A j /A t = A/A ∗ in a data table provide the values of exit
Mach number, M j . For example, when γ = 1.4 and A j /A t = 1.82, the two solutions are
M j , subsonic = 0.341 and M j , supersonic = 2.089.

At the limit where the nozzle is only just choked, the flow previously was subsonic ev-
erywhere and so it is the subsonic solution which applies here too (the flow in the diver-
gent section cannot go from all subsonic to all supersonic in one instant). In this case,
the back pressure is the same as the exit pressure (P b = P j ) and thus:

    γ
γ−1
P0 P0 γ−1
  = = 1 + M 2j , subsonic  (350)
Pb Pj 2
choked

Unlike the case of the convergent nozzle (where the exit Mach number is 1), in this case
the Mach number is 1 at the throat instead. As a result, the “critical” pressure ratio is
dependent on the area ratio, A j /A t , which defines the exit Mach number, M j in the
equation. In the example above, data tables for M = 0.341 tell us that [P 0 /P b ]choked =
1.084.

Normal shock in divergent section If the pressure ratio, P 0 /P b continues to increase,

the throat Mach number cannot increase further so remains at 1. Therefore, the flow
remains choked and the mass flow rate is still given by:

 − γ+1
2(γ−1)
γ  γ−1
r
P 0t (351)
ṁ = p At 1 + 
T0t R 2

The flow between the reservoir and the throat is isentropic (there cannot be any shock
waves in subsonic flow), which means that T0t = T0 and P 0t = P 0 . Therefore:

 − γ+1
2(γ−1)
γ  γ−1
r
P0 (352)
ṁ = ṁ choked = p At 1 + 
T0 R 2

The flow in the convergent part of the nozzle remains the same as the case where the
flow was just choked. However now, the pressure ratio, P 0 /P b , is larger which has the
consequence that P b /P 0 is smaller and so the exit Mach number needs to increase as
a result. In order to achieve this, the flow immediately downstream of the throat now
becomes supersonic. However, for small increases in P 0 /P b above the critical choking
value, the entire divergent section cannot become supersonic at once. Instead, the di-
vergent section immediately downstream of the throat has supersonic flow, which ac-
celerates due to the area increase. At some point in the divergent section, there is a
normal shock wave, which causes the flow to go from supersonic to subsonic. After this

120
shock wave, the subsonic flow in the diverging duct section decelerates up to the exit.
The scenario is described by Figure 65.

t x j
h i h i
Pb Pb Pb
P 0 choked < P0 < P 0 exit shock

P0 M <1 M =1 M >1 M <1

T0

P j = Pb

choked case
0
distance along nozzle

Figure 65: Normal shock in divergent section.

If the shock wave is at station “x”, the mass flow rate in the supersonic flow just ahead
of the shock wave at this station is:
 − γ+1
2(γ−1)
P 0x, pre-shock γ γ−1
r
2 (353)
ṁ x, pre-shock = p A x M x, pre-shock 1 + M x, pre-shock

T0x, pre-shock R 2

There are no shock waves between the reservoir and “x” so T0x, pre-shock = T0 and P 0x, pre-shock =
P 0 . Therefore:

 − γ+1
2(γ−1)
γ γ−1
r
P0
ṁ x, pre-shock = p A x M x, pre-shock 1 + M x2  (354)
T0 R 2

Equating this mass flow to the mass flow rate defined by the choked flow at the throat,

121
we obtain:
 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 At  γ−1
2 (355)
M x, pre-shock 1 + M x, pre-shock
 = 1+ 
2 Ax 2

As before, this equation has two solutions for M x , one which is subsonic and the other
which is supersonic which we find by noting that, since the throat is at Mach 1, A t = A ∗ .
Therefore, the Mach numbers that are listed against the area ratio A x /A t = A/A ∗ in a
data table provide the values of the shock Mach number, M x . In this case, however, we
know that the flow immediately ahead of the shock wave must be supersonic, so we
select the supersonic value out of the two possible solutions.

Note that the stagnation pressure drops through the shock wave so that P 0 j ̸= P 0 . In
other words the flow is isentropic before and after the shock wave, but the stagnation
pressure changes at the shock wave itself. As the pressure ratio, P 0 /P b is increased, the
shock wave moves further downstream. As it does so, since the supersonic flow in the
divergent section ahead of the shock is isentropic, it must (at any point with area, A)
satisfy the equation:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 At  γ−1
M 1 + M 2 = 1+  (356)
2 A 2

Therefore, the Mach number at any point in the supersonic region is defined by the area
ratio of that point to the throat A/A t = A/A ∗ . Therefore, as the shock moves down-
stream , the supersonic flow upstream of it must follow exactly the same Mach number
profile in each case (Figure 66) up to the point of the shock wave itself.

t j
M

0
distance along nozzle

Figure 66: Normal shock at different positions in divergent section.

As the shock wave moves downstream, the supersonic flow accelerates further along
the divergent section, and thus the shock Mach number is greater. The stronger shock
has a lower post-shock, subsonic Mach number. However, there is a shorter section of

122
the divergent duct in which the flow decelerates and thus the exit Mach number, M j , is
higher. As a result, P 0 j /P j is higher. Since the back pressure equals the exit pressure in
this case, P j = P b . Although the stagnation pressure drop through the shock means that
P 0 j ̸= P 0 and thus P 0 j /P j = P 0 j /P b ̸= P 0 /P b , it is still true that P 0 /P b increases as the
shock moves downstream. Therefore the pressure ratio is directly related to the shock
position and, importantly, increasing nozzle pressure ratios cause the shock wave to
move further downstream until such time that the shock wave sits at the nozzle exit.

t j
h i
Pb Pb
P0
= P 0 exit shock

P0 M <1 M =1 M >1 M <1

T0

P j = Pb

0
distance along nozzle

Figure 67: Normal shock wave at nozzle exit.

Normal shock wave at nozzle exit When the normal shock wave is sitting at the nozzle
exit, the flow is still choked at the throat where M t = 1 and thus the mass flow rate is:

 − γ+1
2(γ−1)
γ  γ−1
r
P0 (357)
ṁ = ṁ choked = p At 1 + 
T0 R 2

123
We still have the same subsonic flow accelerating upstream of the throat, and acceler-
ating supersonic flow downstream of the throat. However, now the entire duct is su-
personic up until the normal shock wave, as shown in Figure 67. The flow up to the
point of the shock wave is isentropic and so the pre-shock Mach number, M j , pre-shock is
determined by solving:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 At  γ−1
M j , pre-shock 1 + M 2j , pre-shock  = 1+  (358)
2 Aj 2

Note that this is the same equation we solved to determine the subsonic exit Mach num-
ber (M j , subsonic ) when the nozzle was just choked (subsonic flow in the divergent sec-
tion) but now we take the supersonic solution, M j , supersonic . There is isentropic flow
between the reservoir and the shock wave, so that P 0 j , pre-shock = P 0 . The static pressure
ahead of the shock wave is given by the local supersonic Mach number:

  γ
γ−1
P 0 j , pre-shock γ−1
= 1 + M 2j , supersonic  (359)
P j , pre-shock 2

The static pressure ratio through the shock is:

P j , post-shock 2γ γ−1
= M 2j , supersonic − (360)
P j , pre-shock γ+1 γ+1

Since the flow immediately downstream of the shock wave is subsonic, the pressure
here (P j , post-shock ) must be equal to the back pressure (P b ). Since we already know that
P 0 j , pre-shock = P 0 , the condition with a normal shock at the exit of the nozzle corre-
sponds to a nozzle pressure ratio:
   −1
P0 P 0 j , pre-shock P 0 j , pre-shock P j , post-shock
  = = × 
Pb P j , post-shock P j , pre-shock P j , pre-shock
exit shock
  γ  −1 (361)
γ−1
γ−1 2γ γ−1
= 1 + M 2j , supersonic   M 2j , supersonic − 
2 γ+1 γ+1

Here, the value of M j , supersonic comes from the solution to Equation 358. In practice,
it is common to use tabulated data to make calculations simpler. For example, in the
case of the area ratio A j /A t = 1.82 from earlier, we already saw that data tables give
us M j , supersonic = 2.089. The same data tables show that P /P 0 at this Mach number is
0.111 and so P 0 j , pre-shock /P j , pre-shock = 1/0.111 = 8.992. Similarly, the data table pro-
vides P s /P 1 = P j , post-shock /P j , pre-shock = 4.925. Therefore, this example has a critical
pressure ratio, [P 0 /P b ]exit shock = 8.992/4.925 = 1.825.

Overexpanded flow: shock waves outside nozzle If the pressure ratio, P 0 /P b , is in-
creased even further, then any pressure waves move outside the nozzle. Note that the

124
 t j
h i h i
Pb Pb Pb
P 0 exit shock
< P0
< P 0 perfect expansion

external shocks
P0 M <1 M =1 M >1
T0
P j < Pb

edge of jet

0
distance along nozzle

Figure 68: Overexpanded flow, with shock waves outside nozzle.

flow is still choked and so the mass flow rate is given by ṁ = ṁ choked . In this regime,
the flow in the diverging section is supersonic and the flow throughout the nozzle is
isentropic. The Mach number at the exit of the nozzle is given by:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 At  γ−1
M j 1 + M 2j  = 1+  (362)
2 Aj 2

This is the equation which we have solved previously, and it is clear that we take the
supersonic solution for M j . Since the flow throughout the nozzle is isentropic (there are
no shock waves), we have P 0 j = P 0 . The static pressure at the exit is given by:

  γ
γ−1
P0 j γ−1
= 1 + M 2j , supersonic  (363)
Pj 2

125
Therefore, we have:
  γ
γ−1
P0 P0 j γ−1
= = 1 + M 2j , supersonic  (364)
Pj Pj 2

For the example used previously, where A j /A t = 1.82 and M j , supersonic = 2.089, we have
P 0 /P j = 8.992. Therefore, in the regime where 1.825 < P 0 /P b < 8.992, we must have
P j ̸= P b . Specifically, P 0 /P b < P 0 /P j and so the exit pressure must be smaller than
the back pressure, which is why the condition is known as overexpanded (the flow in
the nozzle has been expanded too much). It is therefore necessary to have a pressure
rise outside the nozzle to eventually attain the external back-pressure value, P b . This
pressure rise is achieved by a series of shock waves (and expansion waves) outside the
nozzle, as shown in Figure 68. Since the pressure in the jet is lower than the external
atmospheric pressure, the edges of the jet are not horizontal but converge towards one
another. Note that the precise wave pattern is beyond the scope of the current module.

t j
h i
Pb Pb
P0
= P 0 perfect expansion

P0 M <1 M =1 M >1
T0
P j = Pb

edge of jet

0
distance along nozzle

Figure 69: Perfectly expanded flow, with no waves inside or outside the nozzle.

126
Perfectly expanded flow: no waves inside or outside the nozzle As the pressure ratio,
P 0 /P b is further increased, there will eventually be a point where P 0 /P b = P 0 /P j . This
condition means that the exit pressure is equal to the back pressure (P j = P b ) and so
the flow is said to be perfectly expanded. The flow is still choked (ṁ = ṁ choked ) and
there is accelerating supersonic flow throughout the divergent section. Since the flow is
isentropic throughout the nozzle, P 0 j = P 0 , and so this scenario occurs when:

    γ
γ−1
P0 P0 j γ−1
  = = 1 + M 2j , supersonic  (365)
Pb Pj 2
perfect expansion

Since the exit pressure is equal to the back pressure, there is no need for any waves
outside the nozzle, as shown in Figure 69.

t j
h i
Pb Pb
P0
> P 0 perfect expansion

expansion waves
P0 M <1 M =1 M >1
T0
P j > Pb

edge of jet

0
distance along nozzle

Figure 70: Underexpanded flow, with expansion waves outside the nozzle.

Underexpanded flow: expansion waves outside the nozzle If the pressure ratio, P 0 /P b
is increased even further, we still have choked flow (ṁ = ṁ choked ) and the flow in the di-

127
verging section is still isentropic and supersonic. The Mach number at the exit of the
nozzle is still given by the supersonic solution to the equation:

 − γ+1  − γ+1
2(γ−1) 2(γ−1)
γ−1 At  γ−1
M j 1 + M 2j  = 1+  (366)
2 Aj 2

The exit pressure is still the same static pressure, P j , which satisfies:

  γ
γ−1
P0 P0 j γ−1
= = 1 + M 2j , supersonic  (367)
Pj Pj 2

We recall that P 0 /P j = [P 0 /P b ]perfect expansion but now we have P 0 /P b > [P 0 /P b ]perfect expansion .
Therefore, P 0 /P j < P 0 /P b which means that P j > P b . In other words, the exit pressure
must be greater than the back pressure, which is why the condition is known as under-
expanded (the flow in the nozzle has still requires further expansion to reach external
atmospheric conditions). It is therefore necessary to have a pressure drop outside the
nozzle to eventually attain the external back-pressure value, P b . This pressure drop is
achieved by a series of expansion waves outside the nozzle, as shown in Figure 70. Since
the pressure in the jet is greater than the external atmospheric pressure, the edges of
the jet are not horizontal but diverge away from one another. Again, the precise wave
pattern is beyond the scope of the current module.

Summary of regimes in a convergent–divergent nozzle Therefore there are eight dif-

ferent regimes which can exit in a convergent–divergent nozzle, which are governed by
the pressure ratio across the nozzle, P 0 /P b , and the area ratio between the exit and the
throat, A j /A t . A summary of these different regimes is as follows:

1. P 0 /P b = 1: no flow through the nozzle, with ṁ = 0, M j = 0 and P j = P b ;

2. 1 < P 0 /P b < [P 0 /P b ]choked : complete subsonic operation with a peak subsonic

Mach number at the throat, such that ṁ < ṁ choked , M j < 1 and P j = P b ;

3. P 0 /P b = [P 0 /P b ]choked : choked subsonic operation with a peak Mach number of 1

at the throat such that ṁ = ṁ choked , M j = M j , subsonic and P j = P b ;

4. [P 0 /P b ]choked < P 0 /P b < [P 0 /P b ]exit shock : choked subsonic/supersonic operation

with a normal shock wave inside the divergent section of the nozzle (accelerating
supersonic flow pre-shock and decelerating subsonic flow post-shock), such that
ṁ = ṁ choked , M j , subsonic < M j < 1 and P j = P b ;

5. P 0 /P b = [P 0 /P b ]exit shock : choked supersonic operation with a supersonic flow through-

out the divergent section and a normal shock wave at the exit of the nozzle, such
that ṁ = ṁ choked , M j , subsonic < M j < 1 and P j = P b ;

6. [P 0 /P b ]exit shock < P 0 /P b < [P 0 /P b ]perfect expansion : choked overexpanded condition

with a supersonic flow throughout the divergent section and shock waves outside
the nozzle, such that ṁ = ṁ choked , M j = M j , supersonic and P j < P b ;

128
 7. P 0 /P b = [P 0 /P b ]perfect expansion : choked perfectly expanded condition with a su-
personic flow throughout the divergent section and no waves outside the nozzle,
such that ṁ = ṁ choked , M j = M j , supersonic and P j = P b ;

8. P 0 /P b > [P 0 /P b ]perfect expansion : choked underexpanded condition with a super-

sonic flow throughout the divergent section and expansion waves outside the noz-
zle, such that ṁ = ṁ choked , M j = M j , supersonic and P j > P b .

Representative Mach number distributions for these regimes are also summarised in
Figure 71.

t j

P0
T0 Pb

M 8

M j , supersonic 7

1
6

5
4
M j , subsonic 3
2
0 1
distance along nozzle

Figure 71: Summary of flow regimes in a convergent–divergent nozzle.

129