Main

F L U I D - S T R U C T U R E I N T E R A C T I O N F U N D A M E N TA L S
O N T H E M U L C A H Y O N E - D I M E N S I O N A L V I B R AT I O N
I N S TA B I L I T Y A N A LY S I S
a. riedinger
universidad tecnológica nacional - facultad regional de bahía blanca

september 19, 2023
CONTENTS
1 Nonlinear dynamics fundamentals 5

1.1 One dimmensional flows on the line 7
1.2 Bifurcations 12
1.3 Two-dimensional linear systems 27
1.4 Phase plane 33
2 Fluid mechanics fundamentals 41

2.1 Definition of liquids and gases 42
2.2 The continuum hypothesis 42
2.3 Properties of a fluid 43
2.4 Fluid statics 48
2.5 Descriptions of fluid motion 53
2.6 Kinematics of deformation and motion 58
2.7 Conservation of mass 62
2.8 Newton’s Law of Motion 69
2.9 General equations of motion 79
2.10 Turbulence 87
2.11 Unsteady flows 101
2.12 Internal flows 112
2.13 Entrance flow and developed flow 112
2.14 Laminar flow in a pipe 113
2.15 Laminar flow between parallel plates 115
2.16 Turbulent flow in a pipe 117
3 Solyd mechanics fundamentals 121

3.1 Coupled two degrees of freedom spring-mass system analysis 121
3.2 Free vibration 122
3.3 Forced vibration 129
4 Fluid-solid interaction fundamentals 131

4.1 Dimensionless fundamental equations 132
4.2 Solid with a still fluid 141
4.3 Coupling with a fast fluid 157
4.4 Coupling with a slow flow 169
4.5 Coupling with any type of flow 174
2
3
5 Mulcahy’s leakeage flow model for a single pipe with horizontal walls 191
5.1 Mathematical model 191
6 Mulcahy’s leakage flow model for a single pipe with rotating walls 199
6.1 Fluid fundamental equations 200
6.2 Approximation based upon dimensional analysis 209
6.3 Lubrication approximation 216
6.4 Solid fundamental equations 220
6.5 Linear solution 222
6.6 Nonlinear solution 224
N O N L I N E A R D Y N A M I C S F U N D A M E N TA L S
1
1.1 One dimmensional flows on the line 7
1.2 Bifurcations 12
1.3 Two-dimensional linear systems 27
1.4 Phase plane 33
There are two main types of dynamical systems: differential equations and iterated
maps (also known as difference equations). Differential equations describe the evolution
of systems in continuous time, whereas iterated maps arise in problems where time is
discrete. Differential equations are used much more widely in science and engineering,
and we shall therefore concentrate on them. Later in the book we will see that iterated
maps can also be very useful, both for providing simple examples of chaos, and also as
tools for analyzing periodic or chaotic solutions of differential equations. Now confining
our attention to differential equations, the main distinction is between ordinary and
partial differential equations. For instance, the equation for a damped harmonic oscillator
d2 x dx
m 2
+b +kx = 0 (1.1)
dt dt
is an ordinary differential equation, because it involves only ordinary derivatives dx/dt

and d2 x/dt2 . That is, there is only one independent variable, the time, t. In contrast, the
heat equation
∂u ∂2 u
= 2 (1.2)
∂t ∂x
is a partial differential equation - it has both time, t, and space, x, as independent

variables. Our concern in this book is with purely temporal behavior, and so we deal
with ordinary differential equations almost exclusively. A very general framework for
ordinary differential equations is provided by the system
x˙1 = f 1 ( x1 , . . . , xn )
.. (1.3)
.
x˙n = f n ( x1 , . . . , xn )
Here the overdots denote differentiation with respect to t. Thus ẋi ≡ dxi/dt. . The
variables x1 , . . . , xn might represent concentrations of chemicals in a reactor, populations
of different species in an ecosystem, or the positions and velocities of the planets in the
solar system. The functions f 1 , . . . , f n are determined by the problem at hand.
5
6 nonlinear dynamics fundamentals
For example, the damped oscillator, Equation 1.1, can be rewritten in the form of Eq.
1.3 thanks to the following trick: we introduce new variables x1 = x and x2 = x. Then
x˙1 = x2 , from the definitions, and
b k
x˙2 = ẍ = −
ẋ − x
m m (1.4)
b k
= − x2 − x1
m m
from the definitions and the governing equation, Eq. 1.1. Hence, the equivalent system
may be written as
x˙1 = x2
b k (1.5)
x˙2 = − x2 − x1
m m
This system is said to be linear, because all the x, on the right-hand side appear to the
first power only. Otherwise the system would be nonlinear. Typical nonlinear terms are
products, powers, and functions of the xi , such as x1 x2 , ( x1 )3 or cos x2 .
For example, the swinging pendulum is governed by the equation
g
ẍ + sin x = 0 (1.6)
L
where x is the angle of the pendulum from vertical, g is the acceleration due to gravity,
and L is the length of the pendulum. The equivalent system is nonlinear
x˙1 = x2
g (1.7)
x˙2 = − sin x1
L
Nonlinearity makes the pendulum equation very difficult to solve analytically. The
usual way around this is to fudge, by invoking the small angle approximation sin x ≃ x
for x << 1. This converts the problem to a linear one, which can then be solved easily.
But by restricting to small x, we are throwing out some of the physics, like motions
where the pendulum whirls over the top. Is it really necessary to make such drastic
approximations?
It turns out that the pendulum equation can be solved analytically, in terms of elliptic
functions. But there ought to be an easier way. After all, the motion of the pendulum is
simple: at low energy, it swings back and forth, and at high energy it whirls over the top.
There should be some way of extracting this information from the system directly.
Here is the rough idea. Suppose we happen to know a solution to the pendulum
system, for a particular initial condition. This solution would be a pair of functions x1 (t)
and x2 (t), representing the position and velocity of the pendulum. If we construct an
abstract space with coordinates ( x1 , x2 ) then , the solution ( x1 (t), x2 (t)) corresponds to
a point moving along a curve in this space (Figure 1.1).
This curve is called a trajectory, and the space is called the phase space for the system.
The phase space is completely filled with trajectories, since each point can serve as an
initial condition.
Our goal is to run this construction in reverse: given the system, we want to draw
the trajectories, and thereby extract information about the solutions. In many cases,
1.1 one dimmensional flows on the line 7
Figure 1.1: Pendulum trajectory
geometric reasoning will allow us to draw the trajectories without actually solving the
system.
Some terminology: the phase space for the general system Equation 1.3 is the space
with coordinates x1 , . . . , xn . Because this space is n-dimensional, we will refer to Eq. 1.3
as an n-dimensional system or an nth-order system. Thus n represents the dimension of
the phase space.
1.1 one dimmensional flows on the line
Given the general system in Equation 1.3, for n = 1 we get a single equation
ẋ = f ( x ) (1.8)
which denotes a one dimmensional system. Here, x (t) is a real-valued function of

time and f ( x ) is a smooth real-valued function of x. Being f just a function of x, and not
of x and t, i.e. f ( x, t), defines the system as nonautonomous. Autonomous systems shall
be discussed later.
1.1.1 A geometric way of thinking
Pictures are often more helpful than formulas for analyzing nonlinear systems. Here
we illustrate this point by a simple example. Along the way we will introduce one of
the most basic techniques of dynamics: interpreting a differential equation as a vector field.
Consider the following nonlinear differential equation
ẋ = sin x (1.9)
To emphasize our point about formulas versus pictures, we have chosen one of the
few nonlinear equations that can be solved in closed form. We separate the variables
and then integrate
dx
dt = (1.10)
sin x
which implies
ˆ
t = csc x dx
(1.11)
= − ln | csc x + cot x | + C
To evaluate the constant C, suppose that x = x0 at t = 0. Then, C = ln | csc x0 + cot x0 |,

and the solution is
csc x0 + cot x0
t = ln (1.12)
csc x + cot x
This result is exact, but a headache to interpret. In contrast, a graphical analysis of

1.9 is clear and simple, as shown in Fig. 1.2. We think of t as time, x as the position of
an imaginary particle moving along the real line, and ẋ as the velocity of that particle.
Then the differential equation ẋ = sin x represents a vectorfield on the line: it dictates
the velocity vector ẋ at each x. To sketch the vector field, it is convenient to plot ẋ versus
x, and then draw arrows on the x − axis to indicate the corresponding velocity vector at
each x. The arrows point to the right when ẋ > 0 and to the left when ẋ < 0.
Figure 1.2: Graphical representation of Equation 1.9
Here is a more physical way to think about the vector field: imagine that fluid is
flowing steadily along the x-axis with a velocity that varies from place to place, according
to the rule ẋ = sin x. As shown in Figure 1.2, the flow is to the right when ẋ > 0 and to
the left when ẋ < 0. At points where x = 0, there is no flow; such points are therefore
called fixed points. You can see that there are two kinds of fixed points in Figure 1.2: solid
black dots represent stable fixed points (often called attractors or sinks, because the flow is
toward them) and open circles represent unstable fixed points (also known as repellers or
sources).
Armed with this picture, we can now easily understand the solutions to the differential
equation ẋ = sin x. We just start our imaginary particle at x0 , and watch how it is
carried along by the flow. We should admit though that a picture cannot tell us certain
quantitative things: for instance, we do not know the time at which the speed | ẋ | is
greatest. But in many cases qualitative information is what we care about, and then
pictures are fine.
And these ideas can be extended to any one-dimensional system ẋ = f ( x ). We just
need to draw the graph of f ( x ) and then use it to sketch the vector field on the real line
(the x-axis in Figure 1.3).
As before, we imagine that a fluid is flowing along the real line with a local velocity
f ( x ). This imaginary fluid is called the phase fluid, and the real line is the phase space.
The flow is to the right where f ( x ) > 0 and to the left where f ( x ) < 0. To find the
solution to x = f ( x ) starting from an arbitrary initial condition x0 , we place an imaginary
Figure 1.3: Fixed points and stability
particle (known as phase point) at x0 , and watch how it is carried along by the flow. As
time goes on, the phase point moves along the x-axis according to some function x (t).
This function is called the trajectory based at x0 , and it represents the solution of the
differential equation starting from the initial condition x0 . A picture like Figure 1.3,
which shows all the qualitatively different trajectories of the system, is called a phase
portrait.
The appearance of the phase portrait is controlled by the fixed points x ∗ , defined by
f ( x ∗ ) = 0; they correspond to stagnation points of the flow. In Figure 1.3, the solid black
dot is a stable fixed point (the local flow is toward it) and the open dot is an unstable
fixed point (the flow is away from it).
In terms of the original differential equation, fixed points represent equilibrium solu-
tions (sometimes called steady, constant, or rest solutions, since if x = x ∗ initially, then
x (t) = x ∗ for all time). An equilibrium is defined to be stable if all sufficiently small
disturbances away from it damp out in time. Thus stable equilibria are represented geo-
metrically by stable fixed points. Conversely, unstable equilibria, in which disturbances
grow in time, are represented by unstable fixed points.
1.1.2 Linear stability analysis
Frequently one would like to have a more quantitative measure of stability, such as
the rate of decay to a stable fixed point. This sort of information may be obtained by
linearizing about a fixed point, as we now explain.
Let x ∗ be a fixed point, and let η̇ = ẋ = f ( x ) = f ( x ∗ + η ) be a small perturbation
away from x ∗ . To see whether the perturbation grows or decays, we derive a differential
equation for η. Differentiation yields
d
η̇ = ( x − x ∗ ) = ẋ (1.13)
dt
since x ∗ is constant. Thus, η̇ = ẋ = f ( x ) = f ( x ∗ + η ). Now using Taylor’s expansion,

assumming f ( x ) as a smooth and continuous function, we obtain
f ( x ∗ + η ) = f ( x ∗ ) + η f ′ ( x ∗ ) + O(η 2 ) + . . . + O(η n ) (1.14)

where O(η 2 ) denotes cuadratically small terms in η, and so on until O(η n ). Finally,
note that f ( x ∗ ) = 0 since x ∗ is a fixed point. Hence
η̇ = η f ′ ( x ∗ ) + O(η 2 ) (1.15)
Now, if f ′ ( x ∗ ) ̸= 0, the O(η 2 ) and superior are negligible and we may write the
approximation
η̇ ≃ η f ′ ( x ∗ ) (1.16)
This is a linear equation in η, and is called the linearization about x ∗ . It shows that
the perturbation η (t) grows exponentially if f ′ ( x ∗ ) > 0 and decays if f ′ ( x ∗ ) < 0. If
f ′ ( x ∗ ) = 0, the O(η 2 ) terms and superior are not negligible and a nonlinear analysis is
needed to determine stability.
The upshot is that the slope f ′ ( x ∗ ) at the fixed point determines its stability. The
slope is always negative at a stable fixed point. The importance of the sign of f ′ ( x ∗ )
was clear from our graphical approach; the new feature is that now we have a measure
of how stable a fixed point is - that is determined by the magnitude of f ′ ( x ∗ ). This
magnitude plays the role of an exponential growth or decay rate. Its reciprocal 1/| f ′ (x∗ )|
is a characteristic time scale; it determines the time required for x (t) to vary significantly
in the neighborhood of x ∗ .
Fixed points dominate the dynamics of first-order systems. In all our examples so
far, all trajectories either approached a fixed point, or diverged to ±∞. In fact, those
are the only things that can happen for a vector field on the real line. The reason is that
trajectories are forced to increase or decrease monotonically, or remain constant (Figure
1.4). To put it more geometrically, the phase point never reverses direction.
Figure 1.4: Impossibility of oscillations
Thus, if a fixed point is regarded as an equilibrium solution, the approach to equi-

librium is always monotonic-overshoot and damped oscillations can never occur in a
first-order system. For the same reason, undamped oscillations are impossible. Hence
there are no periodic solutions to ẋ = f ( x ).
These general results are fundamentally topological in origin. They reflect the fact that
ẋ = f ( x ) corresponds to flow on a line. If you flow monotonically on a line, you will
never come back to your starting place - that is why periodic solutions are impossible.
(Of course, if we were dealing with a circle rather than a line, we could eventually return
to our starting place. Thus vector fields on the circle can exhibit periodic solutions).
It may seem surprising that solutions to ẋ = f ( x ) cannot oscillate. But this result
becomes obvious if we think in terms of a mechanical analog. We regard ẋ = f ( x ) as a
limiting case of Newton’s law, in the limit where the inertia term m ẍ is negligible.
For example, suppose a mass m is attached to a nonlinear spring whose restoring

force is F ( x ), where x is the displacement from the origin. Furthermore, suppose that
the mass is immersed in a vat of very viscous fluid, like honey or motor oil (Figure 1.5),
so that it is subject to a damping force b ẋ. Then Newton’s law is
m ẍ + b ẋ = F ( x ) (1.17)
Figure 1.5: Mechanical analog: overdamped systems
If the viscous damping is strong compared to the inertia term (b ẋ >> m ẍ ), the system
should behave like b ẋ = F ( x ), or equivalently ẋ = f ( x ), where f ( x ) = b−1 F ( x ).In this
overdamped limit, the behavior of the mechanical system is clear. The mass prefers to
sit at a stable equilibrium, where f ( x ) = 0 and f ′ ( x ) < 0.If displaced a bit, the mass
is slowly dragged back to equilibrium by the restoring force. No overshoot can occur,
because the damping is enormous. And undamped oscillations are out of the question.
These conclusions agree with those obtained earlier by geometric reasoning.
1.1.3 Euler’s method
Throughout this chapter we have used graphical and analytical methods to analyze
first-order systems. Every budding dynamicist should master a third tool: numerical
methods. In the old days, numerical methods were impractical because they required
enormous amounts of tedious hand-calculation. But all that has changed, thanks to the
computer. Computers enable us to approximate the solutions to analytically intractable
problems, and also to visualize those solutions. In this section we take our first look at
dynamics on the computer, in the context of numerical integration of ẋ = f ( x ).
The problem can be posed this way: given the differential equation ẋ = f ( x ), subject
to the condition x = x0 , at t = t0 , find a systematic way to approximate the solution x (t).
Suppose we use the vector field interpretation of ẋ = f ( x ). That is, we think of a fluid
flowing steadily on the x-axis, with velocity f ( x ) at the location x. Imagine we are riding
along with a phase point being carried downstream by the fluid. Initially we are at x0 ,
and the local velocity is f ( x0 ). If we flow for a short time ∆t, we will have moved a
distance f ( x0 ) ∆t. Of course, that is not quite right, because our velocity was changing a
little bit throughout the step. But over a sufficiently small step, the velocity will be nearly
constant and our approximation should be reasonably good. Hence our new position
x (t0 + ∆t) is approximately x0 + f ( x0 ) ∆t. Let us call this approximation x1 . Thus
x (t0 + ∆t) ≃ x1 = x0 + f ( x0 ) ∆t (1.18)
Now we iterate. Our approximation has taken us to a new location x1 ; our new velocity
is f ( x1 ); we step forward to x2 = x1 + f ( x1 ) ∆t; and so on. In general, the update rule is
xn+1 = xn + f ( xn ) ∆t (1.19)
This is the simplest possible numerical integration scheme. It is known as Euler’s

method.
Euler’s method can be visualized by plotting x versus t (Figure ??). The curve shows
the exact solution x (t), and the open dots show its values x (tn ) at the discrete times
tn = t0 + n ∆t. The black dots show the approximate values given by the Euler method.
As you can see, the approximation gets bad in a hurry unless ∆t is extremely small.
Hence Euler’s method is not recommended in practice, but it contains the conceptual
essence of the more accurate methods to be discussed later (such as the Finite Volume
Method).
Figure 1.6: Visualization of Euler’s method
1.2 bifurcations
As we have seen in Section 1.1, the dynamics of vector fields on the line is very limited:
all solutions either settle down to equilibrium or head out to ±∞. Given the triviality of
the dynamics, what is interesting about one-dimensional systems? Answer: Dependence
on parameters. The qualitative structure of the flow can change as parameters are varied.
In particular, fixed points can be created or destroyed, or their stability can change. These
qualitative changes in the dynamics are called bifurcations, and the parameter values at
which they occur are called bifurcation points.
Bifurcations are important scientifically - they provide models of transitions and
instabilities as some control parameter is varied. For example, consider the buckling of a
beam. If a small weight is placed on top of the beam in Figure 1.7, the beam can support
the load and remain vertical. But if the load is too heavy, the vertical position becomes
unstable, and the beam may buckle.
Figure 1.7: Buckle of a beam
Here the weight plays the role of the control parameter, and the deflection of the beam
from vertical plays the role of the dynamical variable x.
1.2 bifurcations 13
One of the main goals of this book is to help you develop a solid and practical under-
standing of bifurcations. This section introduces the simplest examples: bifurcations of
fixed points for flows on the line. We begin with the most fundamental bifurcation of all.
1.2.1 Saddle-node bifurcation
The saddle-node bifurcation is the basic mechanism by which fixed points are created
and destroyed. As a parameter is varied, two fixed points move toward each other,
collide, and mutually annihilate.
The prototypical example of a saddle-node bifurcation is given by the first order
system
ẋ = r + x2 (1.20)
where r is a parameter, which may be positive, negative, or zero. When r is negative,

there are two fixed points, one stable and one unstable (Figure 1.8a).
Figure 1.8: Prototypical example of a saddle-node bifurcation
As r approaches 0 from below, the parabola moves up and the two fixed points move
toward each other. When r = 0, the fixed points coalesce into a half-stable fixed point at
x ∗ ≃ 0(Figure 1.8b). This type of fixed point is extremely delicate - it vanishes as soon as
r > 0, and now there are no fixed points at all (Figure 1.8c).
In this example, we say that a bqurcation occurred at r = 0, since the vector fields for
r < 0 and r > 0 are qualitatively different.
Figure 1.9: Bifurcation diagram of a saddle-node bifurcation
Figure 1.9 shows the bifurcation diagram for the bifurcation. The parameter r is
regarded as the independent variable, and the fixed points are shown as dependent
variables.
1.2.2 Transcritical bifurcation
There are certain scientific situations where a fixed point must exist for all values of
a parameter and can never be destroyed. For example, in the logistic equation and
other simple models for the growth of a single species, there is a fixed point at zero
population, regardless of the value of the growth rate. However, such a fixed point
may change its stability as the parameter is varied. The transcritical bifurcation is the
standard mechanism for such changes in stability.
The normal form for a transcritical bifurcation is
ẋ = r x − x2 (1.21)
Figure 1.10 shows the vector field as r varies. Note that there is a fixed point at x ∗ = 0
for all values of r.
Figure 1.10: Prototypical example of a transcritical bifurcation
For r < 0, there is an unstable fixed point at x ∗ = r and a stable fixed point at x ∗ = 0.
As r increases, the unstable fixed point approaches the origin, and coalesces with it when
r = 0. Finally, when r > 0, the origin has become unstable, and x ∗ = r is now stable.
Some people say that an exchange of stabilities has taken place between the two fixed
points.
Note the important difference between the saddle-node and transcritical bifurcations:
in the transcritical case, the two fixed points do not disappear after the bifurcation -
instead they just switch their stability.
Figure 1.11: Bifurcation diagram of a transcritical bifurcation
1.2.3 Pitchfork bifurcation
We turn now to a third kind of bifurcation, the so-called pitchfork bifurcation. This
bifurcation is common in physical problems that have a symmetry. For example, many
problems have a spatial symmetry between left and right. In such cases, fixed points
1.2 bifurcations 15
tend to appear and disappear in symmetrical pairs. In the buckling example of Figure
1.7, the beam is stable in the vertical position if the load is small. In this case there is a
stable fixed point corresponding to zero deflection. But if the load exceeds the buckling
threshold, the beam may buckle to either the left or the right. The vertical position
has gone unstable, and two new symmetrical fixed points, corresponding to left and
right-buckled configurations, have been born.
There are two very different types of pitchfork bifurcation. The simple type is called
supercritical and will be discussed first.
1.2.3.1 Supercritical pitchfork bifurcation
The normal form of the supercritical pitchfork bifurcation is
ẋ = r x − x3 (1.22)
Note that this equation is invariant under the change of variables x → − x. That is,
if we replace x by − x and then cancel the resulting minus signs on both sides of the
equation, we get Eq. 1.22 back again. This invariance is the mathematical expression of
the left-right symmetry mentioned earlier. (More technically, one says that the vector
field is equivariant, but we will use the more familiar language.)
Figure 1.12: Supercritical pitchfork bifurcation vector fild for different values of r
When r < 0, the origin is the only fixed point, and it is stable. When r = 0, the origin
is still stable, but much more weakly so, since the linearization vanishes. Now solutions
no longer decay exponentially fast-instead the decay is a much slower algebraic function
of time. This lethargic decay is called critical slowing down in the physics literature.
Finally, when r > 0, the origin has become unstable. Two new stable fixed points appear
√
on either side of the origin, symmetrically located at x ∗ = ± r. The reason for the term
"pitchfork" becomes clear when we plot the bifurcation diagram (Figure 1.13).
Figure 1.13: Bifurcation diagram of a supercritical pitchfork bifurcation

1.2.3.2 Subcritical pitchfork bifurcation
In the supercritical case ẋ = r x − x3 discussed above, the cubic term is stabilizing: it acts
as a restoring force that pulls x (t) back toward x = 0. If instead the cubic term were
destabilizing, as
ẋ = r x + x3 (1.23)
then we would have a subcritical pitchfork bifurcation.
Figure 1.14: Bifurcation diagram of a subcritical pitchfork bifurcation
Comparing Fig. 1.13 to Fig. 1.14, the pitchfork is inverted. The nonzero fixed points
√
x∗ = ± −r are unstable, and exist only below the bifurcation (r > 0), which motivates
the term subcritical. More importantly, the origin is stable for r < 0 and unstable for r > 0,
as in the supercritical case, but now the instability for r > 0 is not opposed by the cubic
term - in fact the cubic term lends a helping hand in driving the trajectories out to infinity.
This effect leads to blow-up: one can show that x (t) → ±∞ in finite time, starting from
any initial condition x ̸= 0. In real physical systems, such an explosive instability is
usually opposed by the stabilizing influence of higher-order terms. Assuming that the
system is still symmetric under x → − x, the first stabilizing term must be x5 . Thus the
canonical example of a system with a subcritical pitchfork bifurcation is
ẋ = r x + x3 − x5 (1.24)
There is no loss in generality in assuming that the coefficients of x3 and x5 are 1.
Figure 1.15: Bifurcation diagram for Eq. 1.24
For small x , the bifurcation diagram (Fig. 1.15)looks just like Figure 1.14: the origin is
locally stable for r < 0, and two backward-bending branches of unstable fixed points
bifurcate from the origin when r = 0. The new feature, due to the x5 term, is that the
1.2 bifurcations 17
unstable branches turn around and become stable at r = rs , where rs < 0. These stable
large-amplitude branches exist for all r > rs .
There are several things to note about Figure 1.15:
• In the range rs < r < 0, two qualitatively different stable states coexist, namely the
origin and the large-amplitude fixed points. The initial condition x0 , determines
which fixed point is approached as t → ∞. One consequence is that the origin
is stable to small perturbations, but not to large ones - in this sense the origin is
locally stable, but not globally stable.
• The existence of different stable states allows for the possibility of jumps and
hysteresis as r is varied. Suppose we start the system in the state x ∗ = 0, and then
slowly increase the parameter r (indicated by an arrow along the r-axis of Figure
1.16). Then the state remains at the origin until r = 0, when the origin loses stability.
Figure 1.16: Jumps and hysteresis in the bifurcation diagram
Now the slightest nudge will cause the state to jump to one of the large-amplitude
branches. With further increases of r, the state moves out along the large-amplitude
branch. If r is now decreased, the state remains on the large-amplitude branch,
even when r is decreased below O. We have to lower r even further (down past rs )
to get the state to jump back to the origin. This lack of reversibility as a parameter
is varied is called hysteresis.
• The bifurcation at rs , is a saddle-node bifurcation, in which stable and unstable
fixed points are born out the clear blue sky as r is increased (see Section 1.2.1).
1.2.4 Overdamped bead on a rotating hoop
In this section we analyze a classic problem from first-year physics, the bead on a rotating
hoop. This problem provides an example of a bifurcation in a mechanical system. It also
illustrates the subtleties involved in replacing Newton’s law, which is a second-order
equation, by a simpler first-order equation.
A bead of mass m slides along a wire hoop of radius r. The hoop is constrained to
rotate at a constant angular velocity ω about its vertical axis, as depicted in Figure 1.17a.
The problem is to analyze the motion of the bead, given that it is acted on by both
gravitational and centrifugal forces. This is the usual statement of the problem, but now
we want to add a new twist: suppose that there is also a frictional force on the bead
that opposes its motion. To be specific, imagine that the whole system is immersed in a
vat of molasses or some other very viscous fluid, and that the friction is due to viscous
damping.
(b) Coordinates
(a) Bead on a rotating hoop
Figure 1.17: Depiction of the system
Let ϕ be the angle between the bead and the downward vertical direction. By conven-
tion, we restrict ϕ to the range −π < ϕ < π, so there is only one angle for each point
on the hoop. Also, let ρ = r sin ϕ denote the distance of the bead from the vertical axis.
Then the coordinates are as shown in Figure 1.17b.
Now we write Newton’s law for the bead. There is a downward gravitational force m g,
a sideways centrifugal force m ρ ω 2 , and a tangential damping force b ϕ̇. (The constants
g and b are taken to be positive; negative signs will be added later as needed.) The
hoop is assumed to be rigid, so we only have to resolve the forces along the tangential
direction, as shown in Figure 1.18. After substituting ρ = r sin ϕ in the centrifugal term,
and recalling that the tangential acceleration is r ϕ̈, we obtain the governing equation
m r ϕ̈ = −b ϕ̇ − m g sin ϕ + m r ω 2 sin ϕ cos ϕ (1.25)
Figure 1.18: Forces diagram
This is a second-order differential equation, since the second derivative ϕ̈ is the highest
one that appears. We are not yet equipped to analyze second-order equations, so we
would like to find some conditions under which we can safely neglect the m r ϕ̈ term.
Then Eq. 1.25 reduces to a first-order equation, and we can apply our machinery to it.
Of course, this is a dicey business: we cannot just neglect terms because we feel like
it. But we will for now, and then at the end of this section we will try to find a regime
where our approximation is valid.
So, our concern now is with the first-order system
b ϕ̇ = −m g sin ϕ + m r ω 2 sin ϕ cos ϕ

r ω2 (1.26)

= m g sin ϕ cos ϕ − 1
g
1.2 bifurcations 19
The fixed points of Equation 1.26 correspond to equilibrium positions for the bead.
Equation 1.26 shows that there are always fixed points where sin ϕ∗ = 0, namely ϕ∗ = 0
(the bottom of the hoop) and ϕ∗ = π (the top). The more interesting result is that there
are two additional fixed points if
r ω2
>1 (1.27)
g
that is, if the hoop is spinning fast enough. These fixed points satisfy ϕ∗ = ± cos−1 ( g/r ω2 ).
To visualize them, we introduce a parameter
r ω2
γ= (1.28)
g
and solve cos ϕ∗ = 1/γ graphically. We plot cos ϕ vs. ϕ, and look for intersections with
the constant function 1/γ, shown as a horizontal line in Figure 1.19.
Figure 1.19: Plot of cos ϕ vs ϕ
For γ < 1 there are no intersections, whereas for γ > 1 there is a symmetrical pair of
intersections to either side of ϕ∗ = 0. As γ → ∞, these intersections approach ±π/2.
Figure 1.20: Fixed points on the hoop for the cases γ < 1 and γ > 1
To summarize our results so far, let us plot all the fixed points as a function of the
parameter γ (1.21). As usual, solid lines denote stable fixed points and broken lines
denote unstable fixed points.
We now see that a supercritical pitchfork bifurcation occurs at γ = 1.
Here is the physical interpretation of the results: When γ < 1, the hoop is rotating
slowly and the centrifugal force is too weak to balance the force of gravity. Thus the
bead slides down to the bottom and stays there. But if γ > 1, the hoop is spinning fast
enough that the bottom becomes unstable. Since the centrifugal force grows as the bead
moves farther from the bottom, any slight displacement of the bead will be amplified.
The bead is therefore pushed up the hoop until gravity balances the centrifugal force;
Figure 1.21: Fixed points as a function of γ
this balance occurs at ϕ∗ = ± cos−1 ( g/r ω2 ). Which of these two fixed points is actually
selected depends on the initial disturbance. Even though the two fixed points are entirely
symmetrical, an asymmetry in the initial conditions will lead to one of them being chosen
- physicists sometimes refer to these as symmetry-broken solutions. In other words, the
solution has less symmetry than the governing equation.
What is the symmetry of the governing equation? Clearly the left and right halves
of the hoop are physically equivalent - this is reflected by the invariance of under the
change of variables ϕ → −ϕ. As we mentioned in Section 1.2.3, pitchfork bifurcations
are to be expected in situations where such a symmetry exists.
Now we need to address the question: When is it valid to neglect the inertia term
m r ϕ̈ in Eq. 1.25? At first sight the limit m → 0 looks promising, but then we notice that
we are throwing out the baby with the bathwater: the centrifugal and gravitational terms
vanish in this limit too. So we have to be more careful.
In problems like this, it is helpful to express the equation in dimensionless form (at
present, all the terms in Equation 1.25 have the dimensions of force.) The advantage of a
dimensionless formulation is that we know how to define small - it means much less than
1. Furthermore, nondimensionalizing the equation reduces the number of parameters by
lumping them together into dimensionless groups. This reduction always simplifies the
analysis.
There are often several ways to nondimensionalize an equation, and the best choice
might not be clear at first. Therefore we proceed in a flexible fashion. We define a
dimensionless time τ by
1
τ= (1.29)
T
where T is a characteristic time scale to be chosen later. When T is chosen correctly,

the new derivatives dϕ/dτ and d2 ϕ/dτ2 should be O(1), i.e., of order unity. To express these
new derivatives in terms of the old ones, we use the chain rule:
dϕ dϕ dτ 1 dϕ
ϕ̇ ≡ = = (1.30)
dt dτ dt T dτ
and similarly
1 d2 ϕ
ϕ̈ = (1.31)
T 2 dτ 2
1.2 bifurcations 21
Hence, Equation 1.25 becomes
m r d2 ϕ b dϕ
2 2
=− − m g sin ϕ + m r ω 2 sin ϕ cos ϕ (1.32)
T dτ T dτ
Now since this equation is a balance of forces, we nondimensionalize it by dividing by

a force, e.g. m g. This yields the dimensionless equation
2
r ω2

r d ϕ b dϕ
=− − sin ϕ + sin ϕ cos ϕ (1.33)
g T 2 dτ 2 m g T dτ g
Each of the terms in parentheses is a dimensionless group. We recognize the group

rω 2/g
in the last term - that is our old friend γ from earlier in the section.
We are interested in the regime where the left-hand side of Eq. 1.33 is negligible
compared to all the other terms, and where all the terms on the right-hand side are of
comparable size. Since the derivatives are O(1) by assumption, and sin ϕ ≃ O(1), we
see that we need
b r
≃ O(1) and << 1 (1.34)
mgT g T2
The first of these requirements sets the time scale T - a natural choice is
b
T= (1.35)
mg
Then the condition r/g T2 << 1 becomes

r m g 2
<< 1 (1.36)
g b
or equivalently,
b2 >> m2 g r (1.37)
This can be interpreted as saying that the damping is very strong, or that the mass is
very small, now in a precise sense.
The condition Eq. 1.36 motivates us to introduce a dimensionless group
m2 g r
ε= (1.38)
b2
Then Eq. 1.33 becomes
d2 ϕ dϕ
ε 2
=− − sin ϕ + γ sin ϕ cos ϕ (1.39)
dτ dτ
As advertised, the dimensionless Equation 1.39 is simpler than Eq. 1.25: the five
parameters m, g, r, ω and b have been replaced by two dimensionless groups γ and ε.
In summary, our dimensional analysis suggests that in the overdamped limit ε → 0, Eq.
1.39 should be well approximated by the first-order system
dϕ
= f (ϕ) (1.40)
dτ
where
f (ϕ) = − sin ϕ + γ sin ϕ cos ϕ
(1.41)
= sin ϕ (γ cos ϕ − 1)
Unfortunately, there is something fundamentally wrong with our idea of replacing

a second-order equation by afirst-order equation. The trouble is that a second-order
equation requires two initial conditions, whereas a first-order equation has only one. In
our case, the bead’s motion is determined by its initial position and velocity. These two
quantities can be chosen completely independent of each other. But that’s not true for
the first-order system: given the initial position, the initial velocity is dictated by the
equation dϕ/dτ = f (ϕ). Thus the solution to the first-order system will not, in general, be
able to satisfy both initial conditions.
We seem to have run into a paradox. Is Eq. 1.40 valid in the overdamped limit or not?
If it is valid, how can we satisfy the two arbitrary initial conditions demanded by Eq.
1.39? The resolution of the paradox requires us to analyze the second-order system. We
haven’t dealt with second-order systems before - that’s the subject of a section to come.
1.2.4.1 Phase plane analysis
Up until now, we have exploited the idea that a first-order system ẋ = f ( x ) can be
regarded as a vector field on a line. By analogy, the second-order system Eq. 1.25 can be
regarded as a vector field on a plane, the so-called phase plane.
The plane is spanned by two axes, one for the angle ϕ and one for the angular velocity
dϕ/dτ .
To simplify the notation, let
dϕ
Ω = ϕ′ ≡ (1.42)
dτ
where prime denotes differentiation with respect to τ. Then an initial condition for Eq.
1.39 corresponds to a point (ϕ0 , Ω0 ) in the phase plane (Figure 1.22). As time evolves, the
phase point (ϕ(t), Ω(t)) moves around in the phase plane along a trajectory determined
by the solution to Eq. 1.39.
Figure 1.22: Representation of the phase plane
Our goal now is to see what those trajectories actually look like. As before, the key
idea is that the diflerential equcrtion can be interpreted as a vector field on the phase
space. To convert Eq. 1.39 into a vector field, we first rewrite it
ϵ Ω′ = f (ϕ) − Ω (1.43)
1.2 bifurcations 23
Along with the definition ϕ′ = Ω, this yields the vector field
ϕ′ = Ω
1 (1.44)
Ω′ = ( f (ϕ) − Ω)
ε
We interpret the vector (ϕ′ , Ω′ ) at the point (ϕ, Ω) as the local velocity of a phase fluid
flowing steadily on the plane. Note that the velocity vector now has two components,
one in the Ω-direction and one in the ϕ-direction. To visualize the trajectories, we just
imagine how the phase point would move as it is carried along by the phase fluid.
In general, the pattern of trajectories would be difficult to picture, but the present case
is simple because we are only interested in the limit ε → 0. In this limit, all trajectories
slam straight up or down onto the curve C defined by f (ϕ) = Ω, and then slowly ooze
along this curve until they reach a fixed point (Figure 1.23).
Figure 1.23: Pattern of trajectories
To arrive at this striking conclusion, let us do an order-of-magnitude calculation.

Suppose that the phase point lies off the curve C. For instance, suppose (ϕ, Ω) lies an
O(1) distance below the curve C, i.e., Ω < f (ϕ) and f (Ω) − Ω ≃ O(1). Then Eq. 1.44
shows that Ω′ is enormously positive: Ω′ ≃ O(1/ε). Thus the phase point zaps like
lightning up to the region where f (ϕ) − Ω ≃ O(ε). In the limit ε → 0, this region is
indistinguishable from C. Once the phase point is on C, it evolves according to Ω ≃ f (ϕ);
that is, it approximately satisfies the first-order equation ϕ′ = f (ϕ).
Our conclusion is that a typical trajectory is made of two parts: a rapid initial transient,
during which the phase point zaps onto the curve where ϕ′ = f (ϕ) ,followed by a much
slower drift along this curve.
Now we see how the paradox is resolved: The second-order system does behave like
the first-order system, but only after a rapid initial transient. During this transient, it is
not correct to neglect the term ε d2 ϕ/dτ2 . The problem with our earlier approach is that
we used only a single time scale T = b/m g; this time scale is characteristic of the slow
drift process, but not of the rapid transient.
The difficulty we have encountered here occurs throughout science and engineering.
In some limit of interest (here, the limit of strong damping), the term containing the
highest order derivative drops out of the governing equation. Then the initial conditions
or boundary conditions cannot be satisfied. Such a limit is often called singular. For
example, in fluid mechanics, the limit of high Reynolds number is a singular limit; it
accounts for the presence of extremely thin boundary layers in the flow over airplane
wings. In our problem, the rapid transient played the role of a boundary layer - it is a
thin layer of time that occurs near the boundary t = 0. The branch of mathematics that
deals with singular limits is called singular perturbation theory.
1.2.5 Imperfect bifurcations and catastrophes
As we mentioned earlier, pitchfork bifurcations are common in problems that have a

symmetry. For example, in the problem of the bead on a rotating hoop (Section 1.2.4,
there was a perfect symmetry between the left and right sides of the hoop. But in many
real-world circumstances, the symmetry is only approximate - an imperfection leads to a
slight difference between left and right. We now want to see what happens when such
imperfections are present.
For example, consider the system
ẋ = h + r x − x3 (1.45)
If h = 0, we have the normal form for a supercritical pitchfork bifurcation, and there is
a perfect symmetry between x and − x. But this symmetry is broken when h ̸= 0; for
this reason we refer to h as an imperfection parameter.
Equation 1.45 is a bit harder to analyze than other bifurcation problems we have
considered previously, because we have two independent parameters to worry about
(h and r). To keep things straight, we will think of r as fixed, and then examine the
effects of varying h. The first step is to analyze the fixed points of Eq. 1.45. These can
be found explicitly, but we would have to invoke the messy formula for the roots of a
cubic equation. It is clearer to use a graphical approach. Thus, we plot the graphs of
y = r x − x3 and y = −h on the same axes, and look for intersections (Figure 1.24). These
intersections occur at the fixed points of Eq. 1.45. When r ≤ 0, the cubic is monotonically
decreasing, and so it intersects the horizontal line y = −h in exactly one point (Figure
1.24a). The more interesting case is r > 0; then one, two, or three intersections are
possible, depending on the value of h (Figure 1.24b).
Figure 1.24: Graphys of y = r x − x3 and y = −h
The critical case occurs when the horizontal line is just tangent to either the local
minimum or maximum of the cubic; then we have a saddle-node bifurcation. To find the
1.2 bifurcations 25
values of h at which this bifurcation occurs, note that the cubic has a local maximum
3
when /dx r x − x = r − 3 x = 0.
d 2
r
r
xmax = (1.46)
3
and the value of the cubic at the local maximum is

r
3 2r r
r xmax − ( xmax ) = (1.47)
3 3
Similarly, the value at the minimum is the negative of this quantity. Hence saddle-node
bifurcations occur when hc (r ) = 23r 3r , where
p
r
2r r
h c (r ) = (1.48)
3 3
Equation 1.45 has three fixed points for |h| < hc (r ) and one fixed point for |h| > hc (r ).
To summarize the results so far, we plot the bifurcation curves h = ±hc (r ), in the
(r, h) plane (Figure 1.25). Note that the two bifurcation curves meet tangentially at
(r, h) = (0, 0); such a point is called a cusp point. We also label the regions that correspond
to different numbers of fixed points. Saddle-node bifurcations occur all along the
boundary of the regions, except at the cusp point, where we have a codimension-2
bifurcation (this fancy terminology essentially means that we have had to tune two
parameters, h and r, to achieve this type of bifurcation. Until now, all our bifurcations
could be achieved by tuning a single parameter, and were therefore codimension-1
bifurcations).
Figure 1.25: Bifurcation curves in the (r, h) plane
Pictures like Figure 1.25 will prove very useful in our future work. We will refer to
such pictures as stability diagrams. They show the different types of behavior that occur
as we move around in parameter space (here, the (r, h) plane).
Figure 1.26: Bifurcation diagram of x ∗ vs r for fixed h

Now, Figure 1.26 shows the results in a more familiar way. When h = 0 we have
the usual pitchfork diagram (Figure 1.26a) but when h ̸= 0, the pitchfork disconnects
into two pieces (Figure 1.26b). The upper piece consists entirely of stable fixed points,
whereas the lower piece has both stable and unstable branches. As we increase r from
negative values, there is no longer a sharp transition at r = 0; the fixed point simply
glides smoothly along the upper branch. Furthermore, the lower branch of stable points
is not accessible unless we make a fairly large disturbance.
Figure 1.27: Bifurcation diagram of x ∗ vs h for fixed r
Alternatively, we could plot something like Figure 1.27 to interpret the results. When
r ≤ 0 there is one stable fixed point for each h (Figure 1.27a). However, when r > 0
there are three fixed points when |h| < hc (r ), and one otherwise (Figure 1.27b). In the
triple-valued region, the middle branch is unstable and the upper and lower branches
are stable. Note that these graphs look like Figure 1.24 rotated by 90◦ .
There is one last way to plot the results, which may be apealing in some cases where
it is useful to picture things in three dimensions. This method of presentation contains
all of the others as cross sections or projections. If we plot the fixed points x ∗ above the
(r, h) plane, we get the cusp catastrophe surface shown in Figure 1.28. The surface folds
over on itself in certain places. The projection of these folds onto the (r, h) plane yields
the bifurcation curves shown in Figure 1.25. A cross section at fixed h yields Figure 1.26,
and a cross section at fixed r yields Figure 1.27.
Figure 1.28: Cusp catastrophe surface
The term catastrophe is motivated by the fact that as parameters change, the state
of the system can be carried over the edge of the upper surface, after which it drops
discontinuously to the lower surface. This jump could be truly catastrophic for the
equilibrium of a bridge or a building.
1.3 two-dimensional linear systems 27
1.3 two-dimensional linear systems
A two-dimensional linear system is a system of the form
ẋ = a x + b y
(1.49)
ẏ = c x + d y
where a, b, c, d are parameters. This system can be written more compactly in matrix
form as
ẋ = A x (1.50)
where

x
x=
y
(1.51)
a b
A=
c d
Such a system is linear in the sense that if x1 and x2 , are solutions, then so is any linear
combination c1 x1 + c2 x2 . Notice that ẋ = 0 when x = 0, so x ∗ = 0 is always a fixed
point for any choice of A.
The solutions of ẋ = A x can be visualized as trajectories moving on the ( x, y) plane,
in this context called the phase plane. Our first example presents the phase plane analysis
of a familiar system.
1.3.1 Example of a simple harmonic oscillator
The vibrations of a mass hanging from a linear spring are governed by the linear
differential equation
m ẍ + k x = 0 (1.52)
where m is the mass, k is the spring constant, and x is the displacement of the mass
from equilibrium (Figure 1.29). We shall give a phase plane analysis of this simple
harmonic oscillator.
Figure 1.29: Diagram of the simple harmonic oscillator
It is easy to solve Eq. 1.52 analytically in terms of sines and cosines. But that is
precisely what makes linear equations so special - for the nonlinear equations of ultimate
interest to us, it is usually impossible to find an analytical solution. We want to develop

methods for deducing the behavior of equations like 1.52 without actually solving them.
The motion in the phase plane is determined by a vector field that comes from the
differential equation Eq. 1.52. To find this vector field, we note that the state of the
system is characterized by its current position x and velocity v; if we know Figure 1.29
the values of both x and v, then Eq. 1.52 uniquely determines the future states of the
system. Therefore we rewrite Eq. 1.52 in terms of x and v, as follows
ẋ = v
k (1.53)
v̇ = − x
m
Thus, we have rewritten the the differential equation Eq. 1.53 rewritten in terms of v.
To simplify the notation, let ω 2 = k/m. Then Eq. 1.53 becomes
ẋ = v
(1.54)
v̇ = −ω 2 x
The system from Eq. 1.54 assigns a vector ( ẋ, v̇) = (v, −ω 2 x ) at each point ( x, v), and
therefore represents a vector field on the phase plane.
For example, let us see what the vector field looks like when we are on the x-axis.
Then v = 0 and so ( ẋ, v̇) = (0, −ω 2 x ). Hence the vectors point vertically downward
for positive x and vertically upward for negative x (Figure 1.30). As x gets larger in
magnitude, the vectors (0, −ω 2 x ) get longer. Similarly, on the v-axis, the vector field is
( ẋ, v̇) = (v, 0), which points to the right when v > 0 and to the left when v < 0. As we
move around in phase space, the vectors change direction as shown in Figure 1.30.
Figure 1.30: Representation of vectors on the ( x, v) plane
it is helpful to visualize the vector field in terms of the motion of an imaginary fluid.
In the present case, we imagine that a fluid is flowing steadily on the phase plane with a
local velocity given by ( ẋ, v̇) = (v, −ω 2 x ). Then, to find the trajectory starting at ( x0 , v0 ),
we place an imaginary particle or phase point at ( x0 , v0 ) and watch how it is carried
around by the flow.
The flow in Figure 1.30 swirls about the origin. The origin is special, like the eye of a
hurricane: a phase point placed there would remain motionless, because ( ẋ, v̇) = (0, 0)
when ( x, v) = (0, 0); hence the origin is a fixed point. But a phase point starting
anywhere else would circulate around the origin and eventually return to its starting
point. Such trajectories form closed orbits, as shown in Figure 1.31. Figure 1.31 is called
the phase portrait of the system - it shows the overall picture of trajectories in phase space.
What do fixed points and closed orbits have to do with the original problem of a
mass on a spring? The answers are beautifully simple. The fixed point ( x, v) = (0, 0)
Figure 1.31: Representation of closed trajectories on the phase portrait
corresponds to static equilibrium of the system: the mass is at rest at its equilibrium
position and will remain there forever, since the spring is relaxed. The closed orbits have
a more interesting interpretation: they correspond to periodic motions, i.e., oscillations of
the mass. To see this, just look at some points on a closed orbit (Figure 1.32). When the
displacement x is most negative, the velocity v is zero; this corresponds to one extreme
of the oscillation, where the spring is most compressed (Fig. 1.32a).
Figure 1.32: Correspondence between the phase portrait and the dynamics of the system
In the next instant as the phase point flows along the orbit, it is carried to points
where x has increased and v is now positive; the mass is being pushed back toward
its equilibrium position. But by the time the mass has reached x = 0, it has a large
positive velocity (Figure 1.32b) and so it overshoots x = 0. The mass eventually comes to
rest at the other end of its swing, where x is most positive and v is zero again (Figure
1.32c).Then the mass gets pulled up again and eventually completes the cycle (Figure
1.32d).
1.3.2 Classification of linear systems
We want to study the general case of an arbitrary 2 × 2 matrix, with the aim of classifying
all the possible phase portraits that can occur. Recall that the x and y axes played a crucial
geometric role. They determined the direction of the trajectories as t → ∞. They also
contained special straight-line trajectories: a trajectory starting on one of the coordinate
axes stayed on that axis forever, and exhibited simple exponential growth or decay along
it. For the general case, we would like to find the analog of these straight-line trajectories.
That is, we seek trajectories of the form
x (t) = exp (λ t) v (1.55)
where v ̸= 0 is some fixed vector to be determined, and λ is a growth rate, also to be

determined. If such solutions exist, they correspond to exponential motion along the
line spanned by the vector v.
To find the conditions on v and λ, we substitute x (t) = exp (λ t) v into ẋ = A x, and
obtain λ exp(λ t) v = exp(λ t) A v. Canceling the nonzero scalar factor exp(λ t) yields
Av = λv (1.56)
which says that the desired straight line solutions exist if v is an eigenvector of A with
corresponding eigenvalue λ. In this case we call the solution Eq. 1.55 an eigen-solution.
Let us recall how to find eigenvalues and eigenvectors. In general, the eigenvalues of
a matrix A are given by the characteristic equation det( A − λ I ), where I is the identity
matrix. For a 2 × 2 matrix

a b
A= (1.57)
c d
the characteristic equation becomes
a−λ b
det =0 (1.58)
c d−λ
which expanding the equation shields
λ2 − τ λ + ∆ = 0 (1.59)
where
τ = trace( A) = a + d and ∆ = det( A) = a d − b c (1.60)
Then
√
τ+τ 2 − 4λ
λ1 =
√2 (1.61)
τ − τ 2 − 4λ
λ2 =
2
are the solutions of the quadratic equation, Eq. 1.59. In other words, the eigenvalues
depend only on the trace and determinant of the matrix A.
The typical situation is for the eigenvalues to be distinct: λ1 ̸= λ2 . In this case, a
theorem of linear algebra states that the corresponding eigenvectors v1 and v2 , are
linearly independent, and hence span the entire plane (Figure 1.33). In particular,
any initial condition x0 , can be written as a linear combination of eigenvectors, say
x0 = c1 v1 + c2 v.2 v2 .
This observation allows us to write down the general solution for x (t) - it is simply
x (t) = c1 exp(λ1 t) v1 + c2 exp(λ2 t) v2 (1.62)

Figure 1.33: Eigenvectors in the plane
Why is this the general solution? First of all, it is a linear combination of solutions to
ẋ = A x, and hence is itself a solution. Second, it satisfies the initial condition x (0) = x0 ,
and so by the existence and uniqueness theorem, it is the only solution.
Let us now analyze what would happen in the phase portraits given the different
eigenvectors and eigenvalues. We shall start with the case λ1 > λ2 . The first eigensolution
grows exponentially, and the second eigensolution decays. This means the origin is a
saddle point. Its stable manifold is the line spanned by the eigenvector v2 , corresponding
to the decaying eigensolution. Similarly, the unstable manifold is the line spanned by
v1 . As with all saddle points, a typical trajectory approaches the unstable manifold as
t → ∞, and the stable manifold as t → −∞. Figure 1.34 shows the phase portrait.
Figure 1.34: Phase portrait of a saddle point
Now, let us sketch the phase portrait for the case λ2 < λ1 < 0. Both eigensolutions
decay exponentially. The fixed point is a stable node, and the eigenvectors are not
mutually perpendicular, in general. Trajectories typically approach the origin tangent
to the slow eigendirection, defined as the direction spanned by the eigenvector with the
smaller |λ|. In backwards time (t → −∞), the trajectories become parallel to the fast
eigendirection. Figure 1.35 shows the phase portrait (if we reverse all the arrows in
Figure 1.35, we obtain a typical phase portrait for an unstable node).
Figure 1.35: Stable nodes

What happens if the eigenvalues are complex numbers? If the eigenvalues are complex,
the fixed point is either a center (Figure 1.36a) or a spiral (Figure 1.36b). We have already
seen an example of a center in the simple harmonic oscillator (Sec. 1.3.1). The origin
is surrounded by a family of closed orbits. Note that centers are neutrally stable, since
nearby trajectories are neither attracted to nor repelled from the fixed point. A spiral
would occur if the harmonic oscillator were lightly damped. Then the trajectory would
just fail to close, because the oscillator loses a bit of energy on each cycle.
Figure 1.36: Centers and spirals
We can show the type and stability of all the different fixed points on a single diagram
(Figure 1.37).
Figure 1.37: Classification of fixed points
The axes are the trace τ and the determinant ∆ of the matrix A. All of the information
in the diagram is implied by the following formulas
1 p
λ1,2 = τ ± τ2 − 4 ∆
2
∆ = λ1 λ2 (1.63)
τ = λ1 + λ2
The second and third equations can be obtained by writing the characteristic equation
in the form (λ − λ1 ) (λ − λ2 ) = λ2 − τ λ + ∆ = 0.
To arrive at Figure 1.37, we make the following observations:
1. If ∆ < 0, the eigenvalues are real and have opposite signs; hence the fixed point is
a saddle point.
2. If ∆ > 0, the eigenvalues are either real with the same sign (nodes), or complex
conjugate (spirals and centers). Nodes satisfy τ 2 − 4 ∆ > 0 and spirals satisfy
τ 2 − 4 ∆ < 0. The parabola τ 2 − 4 ∆ is the borderline between nodes and spirals;
star nodes and degenerate nodes live on this parabola. The stability of the nodes
1.4 phase plane 33
and spirals is determined by τ. When τ < 0, both eigenvalues have negative

real parts, so the fixed point is stable. Unstable spirals and nodes have τ > 0.
Neutrally stable centers live on the borderline τ = 0 where the eigenvalues are
purely imaginary.
3. If ∆ = 0, at least one of the eigenvalues is zero. Then the origin is not an isolated
fixed point. There is either a whole line of fixed points, or a plane of fixed points,
if A = 0.
Figure 1.37 shows that saddle points, nodes, and spirals are the major types of fixed
points; they occur in large open regions of the (∆, τ ) plane. Centers, stars, degenerate
nodes, and non-isolated fixed points are borderline cases that occur along curves in the
(∆, τ ) plane. Of these borderline cases, centers are by far the most important. They occur
very commonly in frictionless mechanical systems where energy is conserved.
1.4 phase plane
This section begins the study of two-dimensional nonlinear systems. First we consider
some of their general properties. Then we classify the kinds of fixed points that can
arise, building on our knowledge of linear systems. The theory is further developed
through a series of examples from biology (competition between two species) and physics
(conservative systems, reversible systems, and the pendulum). The section concludes
with a discussion of index theory, a topological method that provides global information
about the phase portrait. This chapter is mainly about fixed points.
1.4.1 Fixed points in two-dimensional nonlinear systems
The general form of a vector field on the phase plane is
x˙1 = f 1 ( x1 , x2 )
(1.64)
x˙2 = f 2 ( x1 , x2 )
where f 1 and f 2 are given functions. This system can be written more compactly in
vector notation as
ẋ = f ( x ) (1.65)
where x = ( x1 , x2 ) and f = f ( x1 , x2 ). Here x represents a point in the phase plane,

and ẋ is the velocity vector at that point. By flowing along the vector field, a phase point
traces out a solution x (t), corresponding to a trajectory winding through the phase plane
(Figure 1.38).
Figure 1.38: Trajectory of x(t) in the phase plane
Furthermore, the entire phase plane is filled with trajectories, since each point can play
the role of an initial condition.
For nonlinear systems, there is typically no hope of finding the trajectories analytically.
Even when explicit formulas are available, they are often too complicated to provide
much insight. Instead we will try to determine the qualitative behavior of the solutions.
Our goal is to find the system’s phase portrait directly from the properties of f ( x ).
So, we shall extend the linearization technique developed earlier for one-dimensional
systems (Section 1.1.2). The hope is that we can approximate the phase portrait near a
fixed point by that of a corresponding linear system.
Consider the system
ẋ = f ( x, y)
(1.66)
ẏ = g( x, y)
and suppose that ( x ∗ , y∗ ) is a fixed point; i.e.
f ( x ∗ , y∗ ) = 0 and g( x ∗ , y∗ ) = 0 (1.67)
Let
u = x − x∗ and v = y − y∗ (1.68)
denote the components of a small disturbance from the fixed point. To see whether
the disturbance grows or decays, we need to derive differential equations for u and v.
Let us do the u-equation first
u̇ = ẋ since x ∗ is constant
= f ( x ∗ + u, y∗ + v) by substitution
∂f ∂f
= f ( x ∗ , y∗ ) + u +v + O(u2 , v2 , u v) Taylor series expansion (1.69)
∂x ∂y
x∗ y∗
∂f ∂f
=u +v + O(u2 , v2 , u v) since f ( x ∗ , y∗ ) = 0
∂x ∂y
x∗ y∗
Note that altough the partial derivatives ∂ f/∂x and ∂ f/y may look like functions, in fact
they are not. They are just mere numbers, since they are evaluated at the fixed points
( x ∗ , y∗ ) respectively.
In the same fashion, we can find the v-equation to be
∂g ∂g
v̇ = u +v + O(u2 , v2 , u v) (1.70)
∂x ∂y
x∗ y∗
Hence, the disturbance (u, v) evolves according to

"∂f ∂f #
u̇ u
= ∂x∂g
∂y
∂g + O(u2 , v2 , u v) (1.71)
v̇ ∂x ∂y
v
The matrix
"∂f ∂f
#
∂x ∂y
A= ∂g ∂g (1.72)
∂x ∂y
1.4 phase plane 35
is called the Jacobian matrix at the fixed point ( x ∗ , y∗ ).

Now since the quadratic terms in Eq. 1.71 are tiny, because u and v in themselves are
quite tiny already, it is tempting to neglect them altogether. If we do that, we obtain the
linearized system
"∂f ∂f #
u̇ u
= ∂x∂g
∂y
∂g (1.73)
v̇ ∂x ∂y
v
Is it really safe to neglect the quadratic terms in Eq. 1.71? In other words, does the
linearized system give a qualitatively correct picture of the phase portrait near ( x ∗ , y∗ )?
The answer is yes, as long as the fixed point for the linearized system is not one of the borderline
cases. In other words, if the linearized system predicts a saddle, node, or a spiral, then
the fixed point really is a saddle, node, or spiral for the original nonlinear system.
1.4.2 Rabbits versus sheeps model of competition
We shall analyze the classic Lotka-Volterra model of competition between two species,
here imagined to be rabbits and sheep. Suppose that both species are competing for the
same food supply (grass) and the amount available is limited. Furthermore, ignore all
other complications, like predators, seasonal effects, and other sources of food. Then
there are two main effects we should consider:
1. Each species would grow to its carrying capacity in the absence of the other. This
can be modeled by assuming logistic growth for each species. Rabbits have a
legendary ability to reproduce, so perhaps we should assign them a higher intrinsic
growth rate.
2. When rabbits and sheep encounter each other, trouble starts. Sometimes the
rabbit gets to eat, but more usually the sheep nudges the rabbit aside and starts
nibbling (on the grass, that is). We shall assume that these conflicts occur at a rate
proportional to the size of each population. (If there were twice as many sheep, the
odds of a rabbit encountering a sheep would be twice as great.) Furthermore, we
assume that the conflicts reduce the growth rate for each species, but the effect is
more severe for the rabbits.
A specific model that incorporates these assumptions is
ẋ = x (3 − x − 2 y)
(1.74)
ẏ = y (2 − x − y)
where x (t) represents the population of rabbits and y(t) the population of sheeps.
The coefficients have been chosen to reflect this scenario, but are otherwise arbitrary.
To find the fixed points for the system, we solve ẋ = 0 and ẏ = 0 simultaneously. Four
fixed points are obtained: (0, 0), (0, 2), (3, 0) and (1, 1). To classify them, we compute
the Jacobian

3−2x−2y −2 x
A= (1.75)
−y 2−x−2y
Now we shall consider the four fixed points in turn. Let us start with the fixed point
(0, 0), the Jacobian results

3 0
A= (1.76)
0 2
The eigenvalues are λ1,2 = 3, 2, so (0, 0) is an unstable node. Trajectories leave the
origin parallel to the eigenvector for λ1 = 2, i.e. tangential to v = (0, 1), which spans
the y-axis. (Recall the general rule: at a node, trajectories are tangential to the slow
eigendirection, which is the eigendirection with the smallest |λ|.) Thus, the phase portrait
near (0, 0) looks like Figure 1.39.
Figure 1.39: Phase portrait near the fixed point (0, 0)
Let us move on now to the fixed point (0, 2). The Jacobian gives

−1 0
A= (1.77)
−2 −2
This matrix has eigenvalues λ1,2 = −1, −2, as can be seen from inspection, since Sthe
matrix is triangular. Hence the fixed point is a stable node. Trajectories approach along
the eigendirection associated with λ = −1;you can check that this direction is spanned
by v = (1, −2). Figure 1.40 shows the phase portrait near the fixed point.
Then, for the fixed point (3, 0) we have

−3 −6
A= (1.78)
0 −1
which results in λ1,2 = −3, −1. This is also a stable node. The trajectories approach
along the slow eigendirection spanned by v = (3, −1), as shown in Figure 1.41.
Lastly, the fixed point (1, 1) resuls in the Jacobian

−1 −2
A= (1.79)
−1 −1
√
which has τ = −2, ∆ = −1 and λ = −1 ± 2. Hence this is a saddle point. As you
can check, the phase portrait near (1, 1) is as shown in Figure 1.42.
1.4 phase plane 37
Combining Figures 1.39-1.42 we get Figure 1.43, which already conveys a good sense of
the entire phase portrait. Furthermore, notice that the x and y axes contain straight-line
trajectories, since ẋ = 0 when x = 0, and ẏ = 0 when y = 0.
Figure 1.43: Combined phase portraits of the fixed points
Now we use common sense to fill in the rest of the phase portrait (Figure 1.44). For
example, some of the trajectories starting near the origin must go to the stable node on
the x-axis, while others must go to the stable node on the y-axis. In between, there must
be a special trajectory that cannot decide which way to turn, and so it dives into the
saddle point. This trajectory is part of the stable manifold of the saddle, drawn with a
heavy line in Figure 1.44.
Figure 1.44: Phase portrait of the model
The other branch of the stable manihld consists of a trajectory coming in from infinity.
A computer-generated phase portrait (Figure 1.45) confirms our sketch. The phase
portrait has an interesting biological interpretation. It shows that one species generally
drives the other to extinction. Trajectories starting below the stable manifold lead to
eventual extinction of the sheep, while those starting above lead to eventual extinction
of the rabbits. This dibits chotomy occurs in other models of competition and has led
biologists to formulate the principle of competitive exclusion, which states that two
species competing for the same limited resource typically cannot coexist.
Figure 1.45: Computer generated phase portrait of the model
Our example also illustrates some general mathematical concepts. Given an attracting
fixed point x ∗ , we define its basin of attraction to be the set of initial conditions x0 , such
that x (t) → x ∗ as t → ∞. For instance, the basin of attraction for the node at (3, 0)
consists of all the points lying below the stable manifold of the saddle. This basin is
shown as the shaded region in Figure 1.46.
Figure 1.46: Basin of attraction for fixed point (3, 0)
Because the stable manifold separates the basins for the two nodes, it is called the basin
boundary. For the same reason, the two trajectories that comprise the stable manifold are
traditionally called separatrices. Basins and their boundaries are important because they
partition the phase space into regions of different long-term behavior.
1.4.3 Index theory
Linearization is a prime example of a local method: it gives us a detailed microscopic

view of the trajectories near a fixed point, but it cannot tell us what happens to the
trajectories after they leave that tiny neighborhood. Furthermore, if the vector field starts
with quadratic or higher-order terms, the linearization tells us nothing.
In this section we discuss index theory, a method that provides global information
about the phase portrait. It enables us to answer such questions as: must a closed trajec-
tory always encircle a fixed point? If so, what types of fixed points are permitted? What
types of fixed points csn coalesce in bifurcations? The method also yields information
about the trajectories near higher-order fixed points. Finally, we can sometimes use index
arguments to rule out the possibility of closed orbits in certain parts of the phase plane.
1.4 phase plane 39
The index of a closed curve C is an integer that measures the winding of the vector
field on C. The index also provides information about any fixed points that might
happen to lie inside the curve, as we shall see.
This idea may remind of a concept in electrostatics. In that subject, one often introduces
a hypothetical closed surface (a Gaussian surface) to probe a configuration of electric
charges. By studying the behavior of the electric field on the surface, one can determine
the total amount of charge inside the surface. Amazingly, the behavior on the surface
tells us what is happening far away inside the surface. In the present context, the electric
field is analogous to our vector field, the Gaussian surface is analogous to the curve C,
and the total charge is analogous to the index.
Figure 1.47: Representation of a closed curve C
Now let us make these notions precise. Suppose that ẋ = f ( x ) is a smooth vector field
on the phase plane. Consider a closed curve C (Figure 1.47). This curve is not necessarily
a trajectory - it is simply a loop that we are putting in the phase plane to probe the
behavior of the vector field. We also assume that C is a simple closed curve (i.e., it does not
intersect itself) and that it does not pass through any fixed points of the system. Then at
each point x on C, the vector field ẋ = ( ẋ, ẏ) makes a well-defined angle
ẏ
ϕ = tan−1 (1.80)
ẋ
wiht the x-axis.

As x moves counterclockwise around C, the angle ϕ changes continuously since the
vector field is smooth. Also, when x returns to its starting place, ϕ returns to its original
direction. Hence, over one circuit, ϕ has changed by an integer multiple of 2π. Let [ϕ]C
be the net change in ϕ over one circuit. Then the index of the closed curve C with respect
to the vector field f is defined as
1
IC = [ ϕ ]C (1.81)
2π
Thus, IC , is the net number of counterclockwise revolutions made by the vector field
as x moves once counterclockwise around C. To compute the index, we do not need to
know the vector field everywhere; we only need to know it along C.
Now we list some of the most important properties of the index:
1. Suppose that C can be continuously deformed into C ′ without passing through a

fixed point. Then IC = IC′ . This property has an elegant proof: Our assumptions
imply that as we deform C into C ′ , the index IC , varies continuously. But IC , is an
integer - hence it cannot change without jumping. (To put it more formally, if an
integer-valued function is continuous, it must be constant.)
2. If C does not enclose any fixed points, then IC = 0. Proof: By property (1), we can
shrink C to a tiny circle without changing the index. But ϕ is essentially constant
on such a circle, because all the vectors point in nearly the same direction, thanks
to the assumed smoothness of the vector field. Hence, [ϕ]C = 0, and therefore
IC = 0.
3. If we reverse all the arrows in the vector field by changing t → −t, the index is
unchanged - it stays the same. Proof: All angles change from ϕ to ϕ + π .Hence
[ϕ]C stays the same.
4. Suppose that the closed curve C is actually a trajectory for the system, i.e., C is a
closed orbit. Then IC = +1.
The properties above are useful in several ways. Perhaps most importantly, they allow
us to define the index of a fixed point, as follows:
• Suppose x ∗ is an isolated fixed point. Then the index I of x ∗ is defined as IC , where

C is any closed curve that encloses x ∗ and no other fixed points. By property (1)
above, IC , is independent of C and is therefore a property of x ∗ alone. Therefore
we may drop the subscript C and use the notation I for the index of a point.
And they also allow us to define some useful theorems:
• If a closed curve C surrounds n isolated fixed points x1 ∗ , . . . , xn∗ , then
IC = I1 + . . . + In (1.82)
where Ik is the index of x ∗k for k = 1, . . . , n.

• Any closed orbit in the phase plane must enclose fixed points whose indices sum
to +1.
n
∑ Ik = 1 (1.83)
k =1
F L U I D M E C H A N I C S F U N D A M E N TA L S
2
2.1 Definition of liquids and gases 42
2.2 The continuum hypothesis 42
2.3 Properties of a fluid 43
2.4 Fluid statics 48
2.5 Descriptions of fluid motion 53
2.6 Kinematics of deformation and motion 58
2.7 Conservation of mass 62
2.8 Newton’s Law of Motion 69
2.9 General equations of motion 79
2.10 Turbulence 87
2.11 Unsteady flows 101
2.12 Internal flows 112
2.13 Entrance flow and developed flow 112
2.14 Laminar flow in a pipe 113
2.15 Laminar flow between parallel plates 115
2.16 Turbulent flow in a pipe 117
In the context of fluid mechanics, a fluid is a substance that can flow and deform
under the influence of applied forces. Unlike solids, fluids do not possess a definite
shape and can easily change their shape in response to external forces.
Fluids can exist in two main states: liquid and gas. Liquids are relatively incompress-
ible and have a fixed volume, while gases are compressible and can expand or contract
to fill the available space. Both liquids and gases exhibit similar fluid behavior, but their
properties and characteristics may differ.
The fundamental properties of fluids include: density, viscosity, pressure and temper-
ature. So, fluid mechanics is the study of how fluids behave under various conditions
and the principles governing their motion and forces. It encompasses areas such as fluid
dynamics, which deals with the motion of fluids, and fluid statics, which focuses on
fluids at rest. Understanding the behavior of fluids is essential in many engineering and
scientific applications, including designing efficient transportation systems, analyzing
fluid flow in pipes, studying weather patterns, and developing aerospace technologies.
The primary objective of most such analyses in fluid dynamics is to determine the
velocity, u(r, t), a function of the position vector r and time t. Once u is determined, the
forces exerted by the fluid on various surfaces can be calculated in a straightforward
manner.
41
42 fluid mechanics fundamentals
2.1 definition of liquids and gases
The defining property of fluids, embracing both liquids and gases, lies in the ease
with which they may be deformed. A piece of solid material has a definite shape, and
that shape changes only when there is a chnage in the external conditions. A portion
of fluid, on the other hand, does not have a preferred shape, and different elements
of a homogeneous fluid may be rearranged freely without affecting the macroscopic
properties of the portion of fluid. The fact that relative motion of different elements of a
portion of fluid can, and in general cases does, occur when forces act on th efluid gives
rise to the science of fluid dynamics.
The distinction between solids and fluids is not a sharp one, since there are many
materials which in some respects behave like a solid and in other respects like a fluid. A
simple solid might be regarded as a material of which the shape, and the relative positions
of the contituent elements, change by a small amount only, when there is a small change
There is no one term in the forces acting on it. Correspondingly, a simple fluid might be defined as a material
in general use, so we such that the relative positions of the elements of the material change by an amount
define it as ’simple’
which is not small whne suitable chosen forces are applied to the material.
Whe shall suppose then that the fluid unders discussion cannot withstand any tendency
by applied florces to deform it in a way which leave sthe volume unchanged. It should
be noted that a simple fluid may offer resistance to attempts to deform it; what the
definition implies is that the resistance cannot prevent the deformation from occurring,
or, equivanlently, that the resisting force vanishes with the rate of deformation. Since we
shall be concerned then exclusively with the kind of idealized material described here as
a simple fluid, there is no need to use the term further; we shall therefore refer only to
fluids.
The distinction between fluids and gases is much less fundamental, so far as dynamical
studies are concerned. For reasons related to the nature of intermolecular forces, most
substances can exist in either of two stable phases which exhibit the property of fluidity,
or easy deformability. The density of a substance in the liquid phase is normally much
larger than that in th egaseous phase, but this is not in itself a significant basis for
distinction, since it leads mainly to a difference in the magnitudes of forces required
to produce given magnitudes of acceleration rahter than to a difference in the types of
motion. The most important difference between the mechanical properties of liquids
Bulk elasticity is the and gases lies in their bulk elasticity. Gases can be compressed much more readily than
same as liquids, and as a consequence any motion involving appreciable variations in pressure
compressibility
will be accompanied by much larger changes in specific volume in the case of a gas than
in the case of a liquid.
2.2 the continuum hypothesis
The molecules of a gas are separated by vacuous regions with linear dimensions much
larger than those of the molecules themselves. Even in a liquid, in which the molecules
are nearly as closely packed as the stong short-range repulsive forces will allow, the
mass of the material is concentrated in the nuclei of the atoms composing a molecule
and is very far from being smeared uniformly over the volume occupied by the liquid.
Other properties of a fluid, such as composition or velocity, likewise have a violently
non-uniform distribution when the fluid is viewwed on such a small scale as to reveal
the individual molecules. However, fluid mechanics is normally concerned with the
2.3 properties of a fluid 43
behaviour of matter in the large, on a macroscopic scale large compared with the distance
between moelcules, and it will not often happen that the molecular structure of a fluid
need be taken into account explicitly.
We shall suppose for this purpose that the macroscopic behaviour of fluids is the same
as if they were perfectly continuous in structure; and physical quantities such as the
mass and momentum associated with the matter contained within a given small volume
will be regarded as being spread uniformly over that volume instead of, as in strict
reality, being concentrated in a small fraction of it.
We are able to regard the fluid as a continuum when the measured fluid property is
constant for sensitive volumes small on the macroscopic sclae but large on the microscopic
scale.
Our hypothesis implies that it is possible to attach a definite meaning to the notion of
value at a point of the various fluid properties such as density, velocity and temperature;
and that in general the values of these quantities are continuous functions of position in
the fluid and of time. On this basis we shall be able to establish equations governing the
motion of the fluid which are independednt, so far as their form is concerned, of the
nature of the particle structure and indeed, independent of wheter any particle structure So that gases and
exists. A similar hypothesis is made in the mechanics of solids, and the two subjects liquids are treated
together
together are often designated as continuum mechanics.
There is ample observational evidence that the common real fluids, both gases and
liquids, move as if they were continuous, under normal conditions and indeed for
considerable departures from nomrmal conditions, but some of the properties of the
equivalent continuous media need to be determined empirically.
2.3 properties of a fluid
2.3.1 Pressure
Consider first a small piece of solid surface of area, dA, in contact with a fluid. Then the
pressure, p, on that element of surface is the component of fluid-induced force acting
normal to the surface divided by the area, dA. A simple way to visualize this is as
the time-averaged effect of the collisions of the molecules of the fluid with that surface
element. Indeed, by Newtons Law, the pressure is the sum of the normal impulses
imposed on the wall by the molecular collisions per unit time and per unit surface area.
Alternatively this can be evaluated as the net flux of momentum through and normal to
the surface per unit area.
The pressure therefore has units of force per unit area, M/L T2 or N/m2 or kg/m s2 . A
pressure of 1 kg/m s2 is also called a pascal, 1 [Pa]. The above defines pressure at a

solid bounding surface. This definition can be extended to any point within the body of
the fluid as follows:
• Consider a small, imaginary piece of surface of area, dA, at any point within the
fluid. Then the pressure on that imaginary surface can be defined as the net flux
of fluid momentum through and normal to the surface per unit area. In a crude
way it can be visualized as the normal force per unit area acting on a solid surface
that might be placed in the body of the fluid (though one would have to assume in
this thought experiment that the imaginary surface did not alter the characteristic
motions of the molecules).
In general as defined above, the pressure at any point in a fluid could differ depending
on the orientation of the surface area, dA, since the molecular motions may be non-
isotropic, that is to say not the same in all directions at the point under consideration.
Under these circumstances we define a normal stress vector that is a function of both
direction and location within the fluid. However, in a simple fluid at rest the random
molecular motions are commonly isotropic and therefore the net flux of momentum at a
point in the fluid is the same in all directions. Under such circumstances the pressure is
independent of the direction at a given location and is then simply a scalar function of
position within the fluid. How the pressure varies from point to point within a fluid is
the subject of fluid statics.
2.3.2 Density
The density is defined at any point in the body of a fluid by choosing a very small
volume around that point and assessing the mass within that volume divided by the
The density of water volume. It is a fundamental thermodynamic quantity characterizing the state of the fluid
is 1000 kg/m3 at that location within the fluid and can be used with one other thermodynamic variable
and the density
of air such as the pressure, temperature, entropy or enthalpy to completely define the state of
is 1 kg/m3
the fluid at that location. The density therefore has units of mass per unit volume ( M/L3 )
or kg/m3 .
In general, the variation of density with a pair of thermodynamic variables is given
by the equation of state for that fluid. Simple equations of state for the commonly
encountered fluids water and air are as follows. Though the compressibility of water is
sometimes important, it is frequently sufficient to assume that water is incompressible,
that is to say that its density is independent of the pressure and a function only of
temperature. Commonly the temperature is the same throughout the flow and then the
density is constant everywhere. It is more commonly the case that the variations in the
density within an air or gas flow need to be taken into account. Usually, it is sufficient
to assume that these changes can be taken into account by assuming that the density, ρ,
temperature, T, and pressure, p, can be related through the perfect gas law
p = ρRT (2.1)
where R is the gas constant for the gas in question.

In most of this thesis we will focus one of two characteristic though simplistic fluids:
• an incompressible fluid (usually representing water) whose density, ρ, is known,

constant and uniform throughout the fluid flow and
• a perfect gas (usually representing air whose pressure, p, temperature, T, and
density, ρ, are related by the perfect gas law.
While many departures from these two idealized fluids will need to be considered we
shall do so as the circumstances demand it.
2.3.3 Viscocity
Up to this point in our description of the characteristics of a fluid we have focussed on

the properties of a fluid at rest. We now move to discuss the characteristics of a fluid in
motion. First we need to describe the property known as viscosity. Though this initial
visit to the subject is necessarily somewhat superficial, we will be returning at many

points to discuss the complex consequences of this property.
Figure 2.1: Sketch of Couette flow
For present purposes it is sufficient to consider the simple flow depicted in Figure 2.1,
namely the flow generated between two infinite, parallel plates, one of which is held
fixed (the lower one in the figure) and the other (the upper one) is moving at a velocity,
U, in its own plane. The gap between the plates, width h, is filled with a viscous fluid.
This is called a Couette Flow or a simple shear flow.
One of the fluid properties that we will discuss in more detail later is the no-slip
condition. This states that liquid in immediate contact with a solid surface must move
with that surface. Physically this comes about because of the intermolecular forces that
the solid molecules impose on the liquid molecules adjacent to the surface, essentially
locking them to the solid surface. As discussed elsewhere, there are circumstances when
this condition no longer holds but it does hold in many flows of practical interest and we
will confine the present discussion to such a flow. It follows that the liquid in immediate
contact with the lower wall has zero velocity while the fluid in contact with the upper
plate is moving with a velocity, U. The question then arises as to the velocity of the fluid
at intermediate positions within the gap.
To answer this question rigorously we must consider the forces acting on and within
this flow. In order to produce the motion U of the upper plate a force must be applied to
it in the direction of motion, parallel with the plate. By symmetry the force that has to be
applied to a unit area of the plate is the same everywhere and consequently we define
the force applied per unit area as the shear stress, τ, required to maintain a steady motion
and flow. This shear stress is then directly applied by the plate to the fluid in contact
with it and that layer of fluid applies the same shear stress to the next layer beneath it
and so on all the way across the gap. Thus, at any location within the fluid in the gap the
shear stress, τ, is being applied from one layer of fluid to the layer beneath. Therefore, in
this particular flow, the shear stress, τ, in the fluid is uniform everywhere.
The effect of a shear stress, τ, at a point within the fluid is to force one layer of fluid
to slide over the layer beneath. More specifically, if we denote the distance across the
gap by the coordinate, y (see Figure 2.1), and the fluid velocity parallel to the plates at
that location by u(y) then the shear stress, τ, induces a certain velocity gradient, du/dy, in
the fluid at that point. The viscosity, µ, is the property of the fluid that is the factor of
proportionality between the shear stress, τ, and the velocity gradient:
du
τ=µ (2.2)
dy
Assuming that µ is the same across the gap we can the integrate this equation and
apply the boundary conditions (the known velocities, u = 0 at y = 0 and u = U at y = h)
to obtain
µU y
τ= and u=U (2.3)
h h
Thus the velocity varies linearly across the gap and the shear stress τ required to
maintain an upper plate velocity of U for a gap of h is given by µ U/h. We note that the
word shear rate is used to refer to a velocity gradient like U/h or du/dy.
Figure 2.2: Typical results from a Couette viscometer
Experiments can then be run at different speeds and the measured shear stress plotted
against the shear rate, U/h, as illustrated in Figure 2.2. Clearly the slope of the lines in
such a graph is the dynamic viscosity, µ. If all the data lie on a straight line like the blue
curve, the viscosity is independent of the shear rate and the fluid is said to be Newtonian.
Many of the fluids one deals with in practice, such as water or air, are Newtonian and
this is fortunate for this significantly simplifies the analysis of their flows. However there
are other fluids which are non-Newtonian and for which the shear stress is not linearly
proportional to the shear rate.
Some, such as blood, are shear-thinning in that, as the shear rate is increased, the shear
stress is less than would be expected if the behavior were linear. In the case of blood,
this has a physiological advantage since under physical stress the power required by
the heart to pump more blood is less than expected. Conversely. other fluids exhibit a
shear-thickening behavior in which the increase in the shear stress with increasing shear
rate exceeds a linear expectation.
In future analyses of some phenomena we will find that features of the flow are
dependent on the ratio of the dynamic viscosity, µ, to the density, ρ. This property is
called the kinematic viscosity and is denoted by ν = µ/ρ. It has units of m2 /s which

are the units of any diffusivity. We shall see that ν is, in fact, the diffusivity that governs
the diffusion of vorticity in a laminar fluid flow.
2.3.4 Surface tension
There are several macroscopic phenomena that occur as a result of the intermolecular
forces near and at a fluid surface that we will describe in this and some linked sections.
First we address the phenomena known as surface tension and describe some of the
observable effects that result from this fluid property. In the bulk of a liquid nominally
at rest the molecules experience forces exerted by the surrounding molecules which,
averaged over time, are isotropic. The state of the liquid at that point is a function of that
isotropic pressure as well as the temperature. However, the liquid molecules on or near
an interface between the liquid and another fluid experience intermolecular forces that
are different on different sides, most particularly that facing the liquid and that facing
the other fluid.
For simplicity we consider just the circumstance in which the other fluid is a gas or a
vaccuum which exerts little or no force on the liquid molecules; we call this a free surface.
Then, as a result of the asymmetry described above, the liquid molecules are bound
more tightly to the molecules on the bulk liquid side and this tighter binding necessarily
implies an additional strain energy in the free surface of the liquid. This additional stored
energy is called surface energy and, for reasons described below, is characterized by a
property called surface tension. More particularly, there is a certain amount of this surface
energy stored in each unit area of the surface and the surface tension is the amount
of energy per unit surface area. The surface energy or surface tension is a function of
the particular liquid and of its temperature since the intermolecular forces depend on
temperature.
Figure 2.3: An L × L element of liquid surface being stretched
The units of surface energy are therefore those of energy per unit area or kg/s2 or N/m
N/m. It is called surface tension because of what happens when one expands a certain
area of surface. Consider a square section of liquid surface with sides, L, as shown in
Figure 2.3 and denote the surface energy per unit surface area by S. If that area is now
stretched
h i sides measure L + dL, the stored surface energy must increase by
so that the
S ( L + dL)2 − L2 ≃ 2 S L dL. It follows that in order to deposit this additional energy
in the surface we must do work on the surface in order to expand it. We can imagine
doing this by pulling on the top and right-hand-side of the area shown in Figure 2.3.
Since the displacement in both cases is dL and the total work that must be done is
2 S L dL, it follows that the force that must be applied to the top and the right-hand-side
is S L.
In other words the force per unit length of the sides is S. Hence the presence of surface
energy can also be visualized as an additional tension present in the surface of a liquid.
This tension can be visualized by considering making a imaginary incision in a liquid
surface. Then the force that one side of this incision applies to the other side is a tension
or force tangential to the surface equal to S times the length of that incision. It is for
this reason that S is called the surface tension though the property is more accurately
envisaged as stored energy per unit area.
One of the common macroscopic effects of surface tension is the pressure difference it
causes between the two sides of a curved surface.
2.4 fluid statics
In order to determine how the pressure varies from point to point within a fluid at
rest, consider a small circular cylinder (cross-sectional area, A, and height, dy) of fluid
(density ρ) with its axis oriented vertically. As usual we employ a coordinate framework
in which y is oriented vertically upward and the coordinates x and z are in a horizontal
plane. The acceleration due to gravity is denoted by g.
We will evaluate all the forces acting vertically on this cylinder. First the weight of the
fluid within the cylinder, ρ g A dy, acts vertically downward. Since the fluid is at rest
there will be no shear forces acting vertically on the sides of the cylinder. If we define the
pressure acting on the bottom surface as p then the force acting vertically upward on that
surface is p A. Since the pressure on the top surface will therefore be p + (∂p/∂y) dy,it
follows that the force acting downward on that upper surface is { p + (∂p/∂y) dy} A.
Since the fluid cylinder is at rest all these vertical forces must sum to zero and hence it
follows that
−ρ g A dy + p A − { p + (∂p/∂y) dy} A = 0
−ρ g A dy + p A − p A − (∂p/∂y) dy A = 0
(2.4)
∂p
= −ρ g
∂y
Thus the vertical gradient of the pressure is equal to the density multiplied by the
acceleration due to gravity. We will pursue the consequences of this result shortly.
Next we will evaluate all the forces acting horizontally on a similar cylinder with its
axis oriented in a horizontal direction. The weight of the fluid within the cylinder does
not contribute to the horizontal forces; moreover, there will be no shear forces acting
on the sides or ends of the cylinder since the fluid is at rest. Consequently, the forces
due to the pressures on the ends of the cylinder must balance and therefore the pressure
gradient in any horizontal direction must be zero or
∂p ∂p
=0 ; =0 (2.5)
∂x ∂z
In other words in a connected body of fluid at rest, the pressure must be the same
everywhere on any horizontal plane. Note that since the pressure only varies with y the
partial derivative, ∂p/∂y in the above equation 2.303 may be replaced by dp/dy.
The above results completely define the distribution of pressure within a connected
body of fluid at rest. However, it remains to integrate the first result in order to explicitly
generate an expression for the pressure, p, in terms of the vertical coordinate or elevation
y
ˆ y
p = p0 ρ g dy (2.6)
y0
where p0 is the reference pressure at some reference location, y = y0 . If the density

were the same everywhere and the acceleration due to gravity were the same everywhere
then ρ g would be a simple constant and the integration leads to
p = p0 − ρ g ( y − y0 ) (2.7)
2.4 fluid statics 49
This simple result is applicable in many situations where ρ g is close to being constant,
for example in almost all terrestrial tanks of liquid, manometers or in lakes and oceans.
It also leads directly to Archimedes principle. Moreover, the further integration of the
pressure over a surface in contact with the fluid leads to important results for the forces
experienced by walls or submerged objects.
In planetary atmospheres on the other hand the density (and to some extent g) varies
significantly with altitude and the integration is more complex. The interiors of planetary
or celestial objects require a different integration since, even when the density is relatively
constant, the acceleration due to gravity, g, will vary significantly with radius.
2.4.1 Fluid static forces
In this section we derive some basic results for the forces on bounding walls or solid
surfaces as a result of the pressure, p, acting on that surface. Consider a small piece of
solid surface of area, δ A, in contact with a fluid and at an arbitrary inclination relative
to a coordinate framework ( x, y, z). If we denote the outward unit vector normal to the
surface by n then clearly the vector force acting on that surface is p dA n. It is usually
more convenient to consider the components of this force in the x, y and z directions
namely p dA n x , p dA ny , p dA nz respectively. However it is convenient to note that the
terms dA n x , dA ny and dA nz in these expressions are simply the projected areas in the
x, y and z directions. Therefore, the component in any arbitrary direction of the fluid
force on a bounding wall or solid surface due to a pressure, p, is simply that pressure
multiplied by the projected area of the surface in that arbitrary direction. This simple
result can be very convenient when dealing with complicated curved surfaces.
Figure 2.4: Sketch of a two-dimensional dam and reservoir, depth h
As a example, consider the horizontal force imposed on the sidewall containing a

large body of still liquid as sketched in Figure 2.4. This could be the side of a swimming
pool, a dam or any containing surface. For simplicity we will assume two-dimensional
geometry and that the breadth perpendicular to the sketch is b. If, for convenience,
we define a vertically downward coordinate, ζ (so that ζ = −y), with its origin at the
surface of the liquid, then the pressure at any depth in the liquid is p a + ρ g ζ where p a
is the atmospheric pressure (ρ g is assumed constant). Consequently the horizontal force
imposed on the dam by the element, dζ, is ( p a + ρ g ζ ) b dζ. Partly balancing this is the
force on the same element imposed by the atmospheric pressure on the dry side of the
dam which is p a b dζ so that the net horizontal force on the dam due to the element, dζ,
is dFx where
dFx = ρ g b ζ dζ (2.8)
To obtain the total horizontal force, Fz , we integrate this expression

ˆ h
Fx = ρ g b ζ dζ
0 (2.9)
1
= ρ g b h2
2
In addition to the magnitude of the force, Fx , it is important to know the line of action
of this force so that the moments on the dam can be calculated. To obtain the location of
the line of action we begin by denoting its vertical location by ζ m . Then the force, dFx , on
the element, dζ, would have an anticlockwise moment, dM, about the location, ζ = ζ m ,
given by
dM = ρ g b ζ (ζ − ζ m ) dζ (2.10)
Integrating this expression from ζ = 0 to ζ = h we obtain the total fluid induced

moment, M, about the chosen location ζ m
ˆ h
M = ρgb ζ (ζ − ζ m ) dζ
0
3 (2.11)
ζ m h2

h
= ρgb −
3 2
Then the line of action of the force Fx is at the location ζ m for which this moment, M,
is zero. Hence
h3 ζ m h2
− =0
3 2 (2.12)
2h
ζm =
3
and, as might have been anticipated, the force acts at 2/3 of the depth.
2.4.2 Archimedes principle
Archimedes Principle is a very important result that follows directly from the way in
which the pressure varies from point to point within a fluid at rest. We first address the
simple case of an arbitrarily-shaped, impermeable body submerged in an incompressible
liquid of density, ρ, which is at rest and in which the gravitational acceleration, g, is
uniform so that p = p0 − ρ g y where p0 , ρ and g are constants. First we consider dividing
the volume of the body into narrow vertical cylinders as shown in Figure 2.5. Consider
one such cylinder with horizontal cross-sectional area, dA, as indicated.
If we denote the pressure at the bottom end of this cylinder by p then the vertically
upward force on that end is simply p dA whatever the shape or orientation of that end.
Moreover, because of the above way in which the pressure in the fluid varies with y the
pressure acting downward at the top end of the cylinder will be ( p − ρ g h) where h is
the height of the cylinder. Consequently the force acting downward on the top will be
2.4 fluid statics 51
Figure 2.5: submerged body divided into small, vertical, cylindrical elements, dA
( p − ρ g h) dA, again independent of the shape or orientation of that top end. Thus the
net vertically-upward fluid force acting on the cylinder, dF, is
dF = p dA − ( p − ρ g h) dA
= ρ g (h dA) (2.13)
= ρ g dV
where h dA = dV is simply the volume of the cylinder. If we then sum up over

all the cylinders that make up the volume of the entire submerged body, the total
vertically-upward fluid force acting on the body, F, is then
ˆ
F = ρ g dA
ˆ
= ρ g dV (2.14)
= ρgV
where V is the volume of the submerged body. Moreover ρ g V is simply the weight of
the displaced mass of fluid and this is Archimedes principle, namely that the vertically-
upward fluid force (known as the bouyancy force) acting on a submerged body is equal in
magnitude to the displaced mass of fluid.
A companion result yields the line of action of this buoyancy force. As one can confirm
by considering the contribution from all the incremental cylindrical volumes to the
moment acting on the body, the buoyancy force acts through the center of volume of
the body (not the center of mass). This has important implications for the stability of
submerged bodies.
To complete the analysis we should also consider the horizontal forces acting on the
submerged body. To do this we consider subdividing the body into horizontal cylinders
as shown in Figure 2.6. Consider any one of these cylinders with a cross-sectional area
dA. If we denote the pressure at one end by p then the horizontal force acting on that
end is p dA. But the other end is on the same horizontal plane and consequently the
pressure there must also be p. Hence the force on the other end is also p dA and the net
horizontal force is zero. Integrating over the entire volume of the submerged body the
net force acting horizontally is zero as it must be.
Consider now the more general result for a fluid at rest in which the density and the
acceleration due to gravity may not be constant. Again denote the pressure at the bottom
Figure 2.6: A submerged body divided into small, horizontal, cylindrical elements, dA
of the cylinder in Figure 2.6 by p; the force acting on the bottom is again p dA. In this
case the pressure acting on the top end of the cylinder is
ˆ h
p− ρ g dy (2.15)
0
where y is a coordinate measured vertically upward from the bottom end of the
cylinder. Therefore it follows that the vertically upward buoyancy force on the cylindrical
element is
ˆ h
dA ρ g dy (2.16)
0
and this is simply the sum of the weights of the displaced fluid at each point, y, along
the axis of the cylinder since the incremental volume at the point y is dA dy, the mass
of the fluid displaced locally at this point is ρ dA dy and the local weight is therefore
g ρ dA dy. Integrating over the length of the cylinder and then over all the cylinders
making up the volume of the submerged body we conclude that the Archimedes principle
holds whatever the spatial variations in ρ and g.
2.4.3 Pressure differences due to surface tension
One common manifestation of surface tension is the difference in pressure it causes

across a curved surface. For simplicity we consider first a liquid surface which is curved
in only one plane but is flat in a direction perpendicular to that plane. A small section of
such a surface is sketched in Figure 2.7.
Figure 2.7: A small section of curved surface (red) with surface tension, S, and pressures, pO and
p I on the outside and inside respectively
The pressures in the different fluids on either side of this interface are denoted by pO
(on the outside of curve) and p I (on the inside of the curve). Now consider all the forces
acting on a small section of the surface of length, ds, and unit dimension normal to the
sketch. The radius of curvature of the surface is denoted by R as indicated so that the
angle subtended at the center of curvature, dθ, is given by ds = R dθ. Now consider
2.5 descriptions of fluid motion 53
all the forces in the direction n normal to the center of the fluid element. By the result
described in static forces the net force in the outward direction due to forces near and at
a fluid surface that we will describe in this and some linked sections. pO and p I will be
2 R ( p I − pO ) sin dθ. Opposing this will be the components of the surface tension forces
S acting on the two ends of the section of surface which yield an inward force (in the
negative n direction) equal to 2 S sin dθ. Thus in equilibrium
2 R ( p I − pO ) sin dθ = 2 S sin dθ (2.17)
or
p I − pO = S/R (2.18)
Thus the surface tension causes a greater pressure inside the surface and the difference
is the surface tension divided by the radius of curvature (so that a flat surface yields
no pressure difference). Note that it makes no difference whether the liquid is on the
outside or on the inside.
Figure 2.8: Half of a spherical drop of radius, R, (red) with surface tension, S, and pressures, pO
and p I on the outside and inside respectively
Now consider a spherical surface, specifically the spherical drop or bubble shown in
Figure 2.8. If the drop is cut in half as shown the force imposed by surface tension on the
remaining half will be 2π R S. Opposing that will be the pressure difference ( p I − pO )
acting on the projected area π R2 and therefore, in equilibrium,
2π R S = π R2 ( p I − pO ) (2.19)
so that, in the case of a spherical surface,
p I − pO = 2 S/R (2.20)
An appropriate way to visualize this is that, on the surface, the curvatures in each of
the two perpendicular directions contribute equally to the pressure difference caused by
the surface tension. Indeed, a general three-dimensional surface will have two principal
radii of curvature, R1 and R2 , and it can be shown that the resulting pressure difference
in this general case is given by

1 1
p I − pO = S + (2.21)
R1 R2
2.5 descriptions of fluid motion
In order to investigate fluid motion, one must first find an appropriate and useful way
of represesenting that motion in mathematical terms. There are two important ways in
which to approach this. One way would be to imagine fixing attention on a particular,
small element of fluid (conceptually labelling a group of molecules) and to describe how
the position, xi , pressure, p, temperature, T, etc., of this element change with time, t
xi ( t ) ; p(t) ; T (t) (2.22)
where the velocity, ui (t), would then follow simply by differentiation of xi (t). Note
the tensor notation for the velocity and the position; i.e. x1 would represent the x-axis,
x2 the y-axis, x3 the z-axis, etc; while u1 would be the velocity u in the x axis, u2 the
velocity v in the y-axis and u3 the velocity w in the z-axis. This is known as the Lagrangian
description of the motion. Of course, the various elements that one might choose could
have quite different xi (t), p(t), T (t), etc. and therefore, in order to depict the whole fluid
motion we would need to factor into this a unique label for each of the fluid elements.
One possible choice for the unique label would be the position of the fluid element at
some initial point in time which we could denote, say, by Xi . Then the whole motion is
described by
x i ( Xi , t ) ; p ( Xi , t ) ; T ( Xi , t ) (2.23)
This Lagrangian view of the motion has some advantages: for example, most of the
fundamental laws of physics that we might wish to apply to the fluid flow, pertain to a
particular mass of fluid, a mass made up of a particular group of molecules. However, it
also has some disadvantages: for example, this is not a particularly convenient way to
identify the flow conditions at a particular point on the surface of an aircraft.
The second important way in which one might describe the motion is to fix attention
on a particular point in space, say a particular point in some Cartesian coordinate system,
xi and to describe how the velocity, ui , pressure p, temperature T, etc. vary with time at
that and all other fixed points, xi . Then the whole motion is described by
ui ( xi , t ) ; p ( xi , t ) ; T ( xi , t ) (2.24)
This is known as the Eulerian description of motion and is more satisfactory from an
engineering or practical point of view because we can, for example, fix our coordinate
system in the aircraft and thus readily identify the velocity of the flow at a particular
point surrounding the aircraft.
Thus we see that there are two different ways to describe fluid motion mathematically,
the Lagrangian and Eulerian descriptions. The Lagrangian description which focuses
on a particular element of fluid is ideally suited for the application of the laws of
Newtonian physics, conservation of mass, Newton’s laws of motion and the basic laws
of thermodynamics. Much of what we do in developing the equations of fluid motion
is precisely this process of applying these laws to a Lagrangian fluid mass. However,
the Lagrangian description is normally not very convenient from an engineering point
of view since we most often wish to focus on a particular point in space rather than a
particular fluid element. Hence the second step in the development of the equations for
fluid motion involves converting the basic laws to an Eulerian frame of reference.
The first tool that we will need for this conversion process is a relation between the
time derivatives within each of these descriptions of motion.
2.5.1 Lagrangian and Eulerian time derivatives
The first tool that we will need for this conversion process is a relation between the time
derivatives within each of these descriptions of motion. The Lagrangian time derivative
(often called the material time derivative) is denoted by the operator D/Dt and, as its
name implies, is defined as the rate of change with time of some property of the fluid
(denoted here by Q which could be the velocity, density, pressure, etc.) within some
particular fluid element; that is to say as we move along with the fluid. On the other
hand the Eulerian time derivative, denoted here by ∂/∂t is the rate of change with time of
some fluid property at a fixed point within the coordinate frame of reference that we
have chosen for that Eulerian view.
By way of an illustrative example, think of the fluid flow as a highway filled with rush-
hour traffic and consider Q in this example to be the velocity of the vehicles. Then the
Lagrangian acceleration, DQ/Dt, is the acceleration of an individual vehicle as it speeds
up or slows down during its journey. On the other hand, the Eulerian acceleration, ∂/∂t,
would be the rate of increase or decrease in velocity of the vehicles at one particular
location on the highway. Clearly these two accelerations are not necessarily the same.
Though they must be related in some way; that relationship is not immediately obvious.
Therefore one of the first tasks we face in developing the basic equations of fluid
motion is to find the fundamental relationship between the Lagrangian and Eulerian
time derivatives. It can be shown that
D ∂ ∂ ∂ ∂
≡ +u +v +w
Dt ∂t ∂x ∂y ∂z
∂ ∂
≡ + uj (2.25)
∂t ∂x j
∂
≡ + (u · ∇)
∂t
To illustrate the above result and its consequences let us consider, as an example,
the accelerations that can occur in a fluid flow. For this purpose, we examine the
time derivatives of the velocity, ui , namely the Lagrangian accelerations, Du/Dt, and the
Eulerian acceleration, ∂ui/∂t. The former is the acceleration of a particle fluid element
within a flow, the rate of increase of the velocity as we travel with the fluid element. On
the other hand, the latter is the rate of increase of the velocity at a particular, fixed point
in the flow. From the above relation
Dui ∂u ∂u
= i + uj i (2.26)
Dt ∂t ∂x j
or in terms of its Cartesian components:

Du ∂u ∂u ∂u ∂u
= +u +v +w
Dv ∂v ∂v ∂v ∂v
= +u +v +w (2.27)
Dw ∂w ∂w ∂w ∂w
= +u +v +w
Note, most obviously that the Lagrangian and Eulerian accelerations are, in general,
different and the differences are related to the spatial velocity gradients in the flow.
At this point, it is appropriate to define what we mean by the phrase steady flow. A
steady flow in a particular coordinate system is one in which the velocities at every point
are not changing with time so that ∂ui/∂t = 0 and then
Dui ∂u
= uj i (2.28)
Dt ∂x j
A very simple example is to consider the steady flow through a duct as shown in
Figure 2.9. By definition, since the flow is steady the velocity at some fixed point in the
nozzle such as A is unchanging with time. But since the velocity is decreasing from left
to right, the Lagrangian acceleration, the acceleration of a fluid element passing through
the point A is not zero. Thus we demonstrate that Newton’s law applied to that fluid
element must utilize the Lagrangian rather than the Eulerian acceleration.
Figure 2.9: Duct flow
2.5.2 Steady flow
Often the term steady flow will be used in characterizing a flow. This refers to a flow
(and the frame of reference of the observer) in which all the Eulerian time derivatives are
zero. Of course this does depend on the frame of reference of the observer. For example,
a passenger in a vehicle may observe that the flow around the vehicle is steady; but the
stationary observer who watches that vehicle (and its flow) go past sees an unsteady
flow. In fluid mechanics it is always advisable to make a Galilean change to the frame of
reference in order to observe or analyze a steady flow when that is possible. Of course,
there are many circumstances when that is not possible, for example the flow around
any vehicle that is accelerating. The value in switching to a frame in which the flow is
steady can be seen not only in the many analyses described in these pages but can also
be recognized by the experimenter who observes the flow in a wind tunnel or water
tunnel. Of course, almost no flow is truly steady for all time. Thus the flight of an
airplane over the long time from take-off to landing is unsteady; but a plane in cruise
can be seen to manifest a steady flow for short times.
Sometimes there are flows which are mostly steady and in which analyses or observa-
tions in a frame of reference in which most of the Eulerian time derivatives are close to
zero is recommended. An example is the flow of a constant uniform stream around a
fixed object. Most of the flow may be steady but the unsteady motions in a turbulent
boundary layer are manifestly unsteady. Sometimes, as will be described in the sections
on boundary layers, the mean or time-averaged effects of the turbulence on the mean
flow are included in the analysis and the unsteadiness is set aside so the result is an
effectively steady flow.
2.5.3 Transport theorem
Here we derive a fundamental relation known as the Transport Theorem that relates the
Lagrangian rate of change of the total amount of some transportable fluid property in a
Lagrangian volume to the rate of change of that same quantity in an Eulerian volume
that coincides with that Lagrangian volume at the time under consideration. Consider
the Eulerian volume, V, of fluid shown by the red dashed line in Figure 2.10. Next we
consider the Lagrangian fluid volume that coincides with V at the initial time t = 0
under consideration. Though these volumes coincide at t = 0, the difference is that,
being an Eulerian volume, V remains in the same location for all time, whereas, because
of the fluid flow, the Lagrangian volume moves on with the flow and occupies a different
location when t ̸= 0. We denote the position of the Lagrangian volume at some small
time later, t = δt, by V ∗ as shown by the blue dashed line in Figure 2.10. For convenience
we also denote the surface of V by S.
Figure 2.10: Arbitrary Eulerian Volume, V, and coincident Lagrangian Volume, V ∗
It follows that a point A on the surface of V where the velocity at t = 0 is denoted by

the vector u will be displaced to the point B at t = δt where the vector AB is equal to
u δt. Therefore, if we define a small area of the surface of V around the point A by dS,
the volume of the parallelopiped swept out by dS between the times t = 0 and t = δt
will be
(u δt dS) · n = (u · n) δt dS (2.29)
where n is the outward unit normal to the surface S at A.

Now consider the Lagrangian and Eulerian time derivatives of some general trans-
portable property per unit volume in the fluid motion that we will denote by Q. The
total amount of Q in the volume V is then given by the integral
ˆ
Q dV (2.30)
V
The Lagrangian rate of change of time of the total amount of Q in the Lagrangian
volume must therefore be given by
ˆ " ´ ∗
´ #
D V ∗ { Q }t=δt dV − V { Q }t=0 dV
Q dV = (2.31)
Dt V δt
δt→0
Notice in the numerator on the right hand side that the first and second terms have
different integrands and different limits of integration. To progress we divide the first
term into an integral over the volume V plus an integral over the small volume in
between the volumes V and V ∗
ˆ " ´ ´ ∗ − V) −
´ #
D V { Q } t = δt dV + ∗
V −V { Q } t = δt d ( V V { Q } t = 0 dV
Q dV =
Dt V δt
δt→0
" ´ #
∗
´
V ({ Q }t=δt dV − { Q }t=0 ) dV V ∗ −V { Q }t=δt d (V − V )
= +
δt δt
δt→0
(2.32)
Finally, using Gauss’ theorem which states that for any vector field q
ˆ ˆ
q · n dS = ∇ · q dV (2.33)
S V
it follows that
ˆ ˆ ˆ
D ∂Q
Q dV = dV + ∇ · ( Q u) dV
Dt V V ∂t V
ˆ (2.34)
∂Q
= + ∇ · ( Q u) dV
V ∂t
This is the transport theorem. It is most valuable in expressing the Lagrangian rate
of change of many different integral, transportable properties in terms of Eulerian
quantities.
2.6 kinematics of deformation and motion
In any homogeneous and isotropic continuum, whether fluid or solid, the deformation
or motion of any incremental Lagrangian element can be decomposed into four funda-
mental deformations or motions. In these sections we will identify these fundamental
deformations or motions and relate them to the displacements or velocities of the contin-
uum. These relations are purely kinematic in that they do not involve any mechanical or
physical properties of the continuum; consequently they are identical for a solid, fluid or
any continuum.
First we address the case of a solid where it is convenient to focus on the vector of
displacements denoted by Xi that must be measured relative to some initial location or
unstressed state of the substance. The four fundamental displacements are translation,
rotation, shear (or extension) and dilation. Translation is obviously described by the
displacement vector, Xi , and needs little further comment. At the next level of detail
we can identify the deformation that is clearly related to the spatial differences in the
displacements and therefore to the spatial gradient of the displacement, ∂Xi/∂x j . This
tensor is called the displacement gradient tensor. Note that without any displacement
gradient there would be no deformation. Moreover, this displacement gradient tensor
combines both the deformation and the rotation of the substance. Thus we separate
these two displacements into the strain tensor, Eij , and the rotation tensor, Ωij , which are
respectively the symmetric and antisymmetric parts of that displacement gradient tensor.
The strain tensor is

1 ∂Xi ∂X j
Eij = + (2.35)
2 ∂x j ∂xi
2.6 kinematics of deformation and motion 59
and the rotation tensor is

1 ∂Xi ∂X j
Ωij = − (2.36)
2 ∂x j ∂xi
Note that
Eij = Eji and Ωij = −Ω ji (2.37)
We refer to the diagonal components in this tensor as the normal strains and the
off-diagonal components as shear strains. Moreover the sum of the normal strains
Φ∗ = E11 + E22 + E33 (2.38)
is known as the dilation and is a measure of the volumetric expansion of the substance.
In a precisely parallel way the motions of a fluid are characterized by the rates of
change with time of the above measures. Thus the velocities, ui ( xi , t), are just the
Lagrangian rates of change with time of the displacements
DXi
ui = (2.39)
Dt
and the rate of deformation of the fluid is clearly related to the spatial gradients of the
velocities, ∂ui/∂x j , a tensor that is called the velocity gradient tensor. Note that without any
velocity gradient there would be no rate of deformation. Moreover, this velocity gradient
tensor combines both the rate of deformation and the rate of rotation of the fluid. Thus
we separate these by defining the strain rate tensor, eij , and the rate of rotation tensor, ωij∗ ,
that are respectively the symmetric and antisymmetric parts of that velocity gradient
tensor, ∂ui/∂x j . Thus the rate of strain tensor is

1 ∂ui ∂u j
eij = + (2.40)
2 ∂x j ∂xi
and the rate of rotation tensor is

∗ 1 ∂ui ∂u j
ωij = − (2.41)
2 ∂x j ∂xi
Note that
eij = e ji and ωij∗ = −ω ∗ji (2.42)
We refer to the diagonal components in this matrix as the normal strain rates and the
off-diagonal components as shear strain rates. Moreover the sum of the normal strain
rates
Φ = e11 + e22 + e33 (2.43)
is known as the dilation rate and is a measure of the rate of volumetric expansion of
the fluid.
Written out in full the rate of strain tensor (or matrix) has Cartesian components
∂u ∂v ∂w
exx = ; eyy = ; ezz = (2.44)
∂x ∂y ∂z

1 ∂u ∂v
exy = eyx = +
2 ∂y ∂x

1 ∂v ∂w
eyz = ezy = + (2.45)
2 ∂z ∂y

1 ∂w ∂u
ezx = exz = +
2 ∂x ∂z
The other key kinematic property is the rate of rotation tensor, ωij∗ given by

1 ∂ui ∂u j
ωij∗ = −ω ∗ji = − (2.46)
2 ∂x j ∂xi
Since ωij∗ = ω ∗ji this also allows us to define a very important kinematic property of a
flow namely the vorticity, denoted by ωk which is just twice the rate of rotation tensor
ωij∗ so that
ωk = −2 ωij∗
∂u j ∂u (2.47)
= − i
∂xi ∂x j
Written out in terms of its Cartesian components, the vorticity components are

∗ ∗ ∂w ∂v
ωx = 2 ωzy = −2 ωyz = −
∂y ∂z

∗ ∗ ∂u ∂w
ωy = 2 ωxz = −2 ωzx = − (2.48)
∂z ∂x

∗ ∗ ∂v ∂u
ωz = 2 ωyx = −2 ωxy =
∂x ∂y
2.6.1 Fluid element deformations
For the purposes of illustrating the decomposition of fluid motion detailed in the section
on the kinematics of motion, it may be helpful to visualize the effects of the velocity
gradients on a very small fluid element with dimensions that are large compared with the
atomic and molecular dimensions but small compared with any macroscopic dimensions
of the flow under consideration. For simplicity we will do this in two dimensions.
In a two-dimensional planar flow in the xy-plane without any velocity in the z direction,
the components of the strain rate tensor become
∂u
exx =
∂x
∂v
eyy = (2.49)
∂y

1 ∂u ∂v
exy = eyx = +
2 ∂y ∂x
2.6 kinematics of deformation and motion 61
and, written out in terms of its Cartesian components, the only vorticity component,
ωz , which, for simplicity, is denoted by the scalar, ω, is

∂v ∂u
ω = ωz = − (2.50)
∂x ∂y
We visualize a Lagrangian element, ABCD, with dimensions dx × dy × dz as shown

in Figure 2.11 and consider the effects of the velocity gradients on the shape distortions
of this element.
Figure 2.11: A small Lagrangian element of fluid
Consider the dashed element AB∗ C ∗ D ∗ which represents the shape of the Lagrangian
fluid element a short time, dt, later; we have removed the translation of the element by a
Galilean transformation that superimposes the point A on its original location. Then,
because of the velocity gradients in the flow the angle
b ∗= ∂u
B AB dt (2.51)
∂y
and the angle
b ∗= ∂v
D AD dt (2.52)
∂x
and therefore the deformation of the element is given by the difference between the
angles
∂u ∂v π
B∗ AD
b ∗ − B AD
b = dt + dt − (2.53)
∂y ∂x 2
so the rate of deformation is

∂u ∂v
+ (2.54)
∂y ∂x
in accord with the expression 2.49.

Moreover the rotation of the element is given by the angle between the diagonals AC
and AC ∗ which, by simple geometry, is

1 ∂v ∂u
− dt (2.55)
2 ∂x ∂y
so that the rate of rotation is

1 ∂v ∂u
− (2.56)
2 ∂x ∂y
2.7 conservation of mass
In Newtonian mechanics, mass is conserved and since we are omitting all relativistic
effects from the present text, the first basic principle that we need to apply in our
construction of the basic equations of fluid mechanics is the principle of conservation of
mass. The resulting equation is known as the Continuity Equation and we will develop
three different forms of this equation. These three forms are essentially identical but
the result is phrased in three different ways, each of which is appropriate to its own
set of applications. In addition the three different forms illustrate the three different
approaches that can be used in applying other, more complex conservation principles.
We called the three approaches, the Macroscopic Approach, the Differential Approach and
the Integral Approach, in order of mathematical complexity.
A critical step that is common to all of the approaches (and to the application of any
conservation principle) is the choice of a Control Volume (CV for short). We devote a
special section to the process of choosing this control volume.
2.7.1 Control volume
An important fundamental step in setting up the equations of fluid flow in any particular
context is the establishment of a well-defined control volume. This is a volume which is
usually fixed in some coordinate system (commonly an Eulerian volume) and whose
entire surface is precisely defined. In places it may coincide with a fluid/solid interface
and in other places it may cut across the flow. In a given problem or context it may
include only fluid; in other problems or contexts it may include both fluid and structure.
Often the most convenient and productive choice of control volume will depend on the
desired outcome of the analysis applied to the control volume. Therefore several choices
of control volume might be explored before the best choice becomes evident. In almost
all cases the purpose is to apply some conservation principle to the control volume and
therefore it will become neccessary to evaluate the flux of that conserved quantity into
or out of the control volume through all of its surfaces. Consequently this evaluation
needs to be borne in mind when defining the control volume.
A few examples could be useful. The infinitesmal control volume dx × dy × dz shown
in Figure 2.12 (left) is commonly used in deriving the differential form of the equation
that results from the application of a conservation principle to some fluid flow. In the
present text this will be used in applying the principle of conservation of mass and
Newton’s law to quote two examples.
Figure 2.12: Infinitesmal (left) and Arbitrary (right) Eulerian control volumes (dashed red lines)
On the other hand useful global results can often emerge when the conservation
principle is applied to a large, macroscopic control volume such as that depicted in
Figure 2.12 (right). The results that emerge from such an analysis usually take the form
of an integral equation rather than a differential equation. We shall also use this in
2.7 conservation of mass 63
developing integral forms for the conservation of mass and Newton’s law as well as the
second law of thermodynamics (to quote three examples).
It is often the case that a judicious choice of a macroscopic control volume in a particu-
lar engineering problem can yield very useful results. If, for example, a conservation
principle is to be applied to the tank with three connecting pipes shown on the left in
Figure 2.13 then one possible choice of the control volume is indicated by the surface
defined by the dashed red lines. Another possible choice would be the dashed blue line
shown on the right in Figure 2.13. We shall explore these kinds of alternatives in other
sections of this book, for example during our discussions of the thrust produced by jet
engines and rocket engines.
Figure 2.13: Two alternative Eulerian control volumes
2.7.2 Macroscopic form of continuity
The simplest context in which to apply the principle of conservation of mass is to an

internal flow confined within some solid vessel or collection of pipes and in which the
flows into and out of that vessel is simply characterized (or approximately characterized).
Consider, for example, the solid-walled tank depicted in Figure 2.14 that has three pipes
connected to it, labelled A, B and C. For the purpose of the present analyses we choose
a control volume that encloses the entire vessel and cuts across the inlet and exit pipes
as shown by the red, dashed line in the figure.
Figure 2.14: Macroscopic Eulerian control volume
We denote the mass of the fluid inside the control volume by M, the velocity, density
and cross-sectional area of pipe A at the point where the CV cuts across that pipe by u A ,
ρ A and A A respectively. We similarly characterize the flow through pipes B and C (at
the points where the CV cuts across those pipes) by the subscripts B and C. These flow
velocities are denoted as positive when directed out of the vessel.
Since according to the principle of conservation of mass, fluid mass cannot be created
or destroyed, it follows that the rate of flow of mass into the control volume must be
equal to the rate of increase of mass within the control volume. The latter quantity
is readily represented by the time derivative, dM/dt (note that this time derivative is
unambiguous since M is a function only of time and not of position). On the other hand
the rate of flow of mass into the control volume requires a little more construction.
The block of fluid that would cross the surface of the control volume at location A in
a time δt has a volume equal to u A A A δt and therefore the rate of flow of volume (the
volume flux) across that part of the control volume surface is u A A A . Hence the the rate
of flow of mass (the mass flux) across that part of the control volume surface is ρ A u A A A .
Recalling the sign convention for the velocities it follows that the rate at which mass is
entering the control volume is given by
all
− ρ A u A A A − ρ B u B A B − ρC uC AC = − ∑ ρa ua Aa (2.57)
a= A
where the extension to the summation over all conduits into and out of the control
volume is obvious. In passing we note that we have implicitly assumed that at the
location A the velocity, u a and density, ρ A are uniform over the area A A (and similarly
for the other locations). If this is not the case it is clear that the simple forms developed
above will have to be replaced by integrals over the areas such as A A . This is essentially
what is done in the integral approach described later.
It follows that, in this example, conservation of mass requires that
all
dM
+ ∑ ρa ua Aa = 0 (2.58)
dt a= A
and this is a form in which conservation of mass is invoked in a wide range of

applications. If the fluid can be considered incompressible then the densities in all of the
terms of the above equation are identical and the continuity equation therefore reduces
to
all
dV
+ ∑ ua Aa = 0 (2.59)
dt a= A
where V is the volume of the fluid inside the control volume. If the vessel and pipes
are rigid this volume V cannot change with time and so the continuity equation is further
reduced to
all
∑ ua Aa = 0 (2.60)
a= A
This would also be the case if the flow were steady (dV/dt = 0) even if the vessel
and pipes were not rigid. This particularly simple form of the continuity equation is,
perhaps, the most commonly used version in practical applications. It simply states that
the volume flow rate in must equal the volume flow rate out. Perhaps the commonest
example is the simple steady duct flow shown in figure 2 for which
ρ A u A A A = ρB uB AB (2.61)
Figure 2.15: A simple steady duct flow
2.7.3 Differential form of continuity
In the second or differential approach to the invocation of the conservation of mass,

we consider a small Eulerian control volume of fluid within the flow that measures
dx × dy × dz in some fixed Cartesian coordinate system. Depicted in Figure 2.16, this
volume must be small compared with the typical spatial distance within the flow over
which substantial changes in the velocities, pressure, etc. vary. However it must
also be large compared with the molecular dimensions and mean free paths of the
molecules of the fluid, so that it becomes sensible to characterize the fluid motion (and
other properties) using continuum quantities. It should be noted that, though such
intermediate scales between the global flow scale and the molecular scale can be found
in many practical problems, there are flows for which it is not possible to identify such
an intermediate scale. In such circumstances one must resort to other methodologies to
apply the conservation laws.
Figure 2.16: Infinitesmal Eulerian control volume
Assuming the continuum approximation is valid, we then consider the flux of mass
into the differential control volume and equate it with the rate of increase of mass inside
the control volume. Consider first the flux of mass through the two sides perpendicular
to the x-axis. We will define u as the velocity in the x-direction at the center of lefthand of
these two sides. Similarly we define the density of the fluid at the center of lefthand side
as ρ. Then the flux of mass into the control volume through that lefthand side is given by
ρ u dy dz. Then it follows that by Taylor’s series (neglecting all terms of order (dx )2 and
higher which can be shown to have no contribution to the result) the flux of mass out
of the control volume through the righthand side is given by [ρ u + {∂(ρ u)/∂x dx }] dy dz.
Combining the fluxes through these two sides perpendicular to the x-direction, it follows
that the net flux of mass out of the control volume through the sides perpendicular to
the x-direction is {∂(ρ u)/∂x} dx dy dz. The fluxes through the other two pairs of sides
follow from a similar construction so that the net flux of mass out of the control volume
through all of its sides becomes
∂(ρ u j )

∂(ρu) ∂(ρv) ∂(ρw)
+ + dx dy dz = dx dy dz (2.62)
∂x ∂y ∂z ∂x j
where we must use the Eulerian time derivative since the control volume is defined as
an Eulerian volume. Re-arranging and cancelling the differential form of the continuity
equation becomes
∂ρ ∂(ρ u) ∂(ρ v) ∂(ρ w)

+ + + =0 (2.63)
∂t ∂x ∂y ∂z
or, in tensor notation

∂ρ ∂(ρ u j )
+ =0 (2.64)
∂t ∂x j
and in vector notation

∂ρ
+ ∇ · (ρ u) = 0 (2.65)
∂t
If the flow is steady these clearly reduce to
∂(ρ u j )
=0 or ∇ · (ρ u) = 0 (2.66)
∂x j
If the fluid is incompressible (whether steady or not) they reduce to

∂u j
=0 or ∇·u = 0 (2.67)
∂x j
These are the differential forms of the continuity equation in a rectangular Cartesian
coordinate system. There are also many problems in which it is much more convenient
to use an alternate coordinate system such as a polar coordinate system, a cylindrical
coordinate system or a spherical coordinate system.
2.7.4 Integral approach to the continuity equation
The third and last approach to the invocation of the conservation of mass utilizes the
general macroscopic, Eulerian control volume depicted in Figure 2.17. This volume is
denoted by V and its surface by S. We consider a small vector segment of that surface dS
where the magnitude of the vector is the scalar area of the segment dS and the direction
of the vector is the outward normal to the surface at that point. Note that if n is the
outward unit normal to the surface at that point then dS = n · dS.
Figure 2.17: Arbitrary Eulerian control volume

First we evaluate the net flux of mass out of the control volume V. If u is the velocity
of the fluid at the surface area dS then the rate of flow of mass leaving V through that
elemental area is
ρ u · dS = (n · u) ρ dS (2.68)
Invoking the conservation of mass, this must be equal to minus the rate of increase of
mass inside V which is
ˆ ˆ
∂ρ
dV + ρ (n · u) dS = 0 (2.69)
V ∂t S
It can be seen that this is just a generalization of the macroscopic version of the
continuity equation developed earlier. This integral version of the continuity equation is
not only useful in the form given above but is also useful when the last term is converted
from a surface integral to a volume integral by using Gauss’ theorem. This states that for
any general vector quantity, q,
ˆ ˆ
q · dS = ∇ · q dV (2.70)
S V
and therefore, since the last term in the integral form of the continuity equation implies
q = ρ u in this instance, that integral continuity equation can be written as
ˆ ˆ
∂ρ
dV + ∇ · (ρ u) dV = 0 (2.71)
V ∂t V
or
ˆ
∂ρ
+ ∇ · (ρ u) dV = 0 (2.72)
V ∂t
Since the value of this volume integral is zero and since the choice of the volume V
was arbitrary it must follow that the integrand must be everywhere zero and therefore
∂ρ
+ ∇ · (ρ u) dV = 0 (2.73)
∂t
and this is just the differential form of the continuity equation developed earlier. Thus
we see that all three approaches lead to the same answer. However, the various forms
of the continuity equation developed along the way are all useful in the variety of
applications explored in this text.
2.7.5 Streamfunction
In some particular flows, for example planar flow or axisymmetric flow, we can introduce
a special scalar quantity called the stream function which, through its introduction, means
that the continuity equation for that flow is automatically satisfied. Normally the stream
function is denoted by ψ and we begin by defining the stream function for planar,
incompressible flow.
In the section on mass conservation we have seen that the continuity equation for
planar, incompressible flow in the xy plane is
∂u ∂v
+ =0 (2.74)
∂x ∂y
and therefore if we define a stream function ψ( x, y) such that

∂ψ ∂ψ
u= ; v=− (2.75)
∂dy ∂x
the continuity equation is automatically satisfied. Note that in mathematical terms

we have replaced two unknown functions, u( x, y) and v( x, y), with a single unknown
function, ψ( x, y), from which the velocities can be derived.
We now explore the physical meaning of this streamfunction by examining how this
scalar quantity changes from place to place within the planar flow field. As a result of
small incremental displacements, dx and dy, the value of the stream function will change
by dψ where by basic calculus
∂ψ ∂ψ
dψ = dx + dy = −v dx + u dy (2.76)
∂x ∂y
from the definition of ψ. Let us examine first a displacement along a streamline so

that dx = u dt and dy = v dt. Substituting into the above relation it follows that
dψ = −v u dt + u v dt = 0 (2.77)
Therefore the streamfunction is constant along a streamline and we could envision

each streamline being labelled with that value. Now consider how ψ changes with
displacements along lines normal to a streamline. To do so we define a coordinate, n,
normal to a streamline and an angle, θ, that the streamline makes with the x axis at that
point. It follows that a small displacement, dn, in the normal direction (the n direction)
will be given by
dx = −dn sin θ ; dy = dn cos θ (2.78)
and if we also denote the magnitude of the velocity at that point by q (q2 = u2 + v2 ) so
that u = q sin θ and v = q cos θ then it follows that the change in the streamfunction, dψ,
over a displacement dn normal to the streamline will be
dψ = −v dx + u dy = q dn (2.79)
In words, the difference between the streamfunction on a streamline and the stream-
function at a point a distance dn from that streamline (in the normal direction) is equal
to the volume flow rate (per unit breadth perpendicular to the plane of flow) passing
between those two points. Moreover by integrating along a line from one streamline to
another we can also say that the difference in the streamfunction values associated with
those two streamlines is equal to the volume flow rate (per unit breadth) passing along
the streamtube between those two streamlines.
2.8 newton’s law of motion 69
2.8 newton’s law of motion
Having established the form of the first basic conservation law (namely the conservation
of mass) in the context of fluid flow we now turn to the second basic conservation
principle, namely Newton’s first law of motion, and explore the form it takes when
applied to a flowing fluid. Newton’s law states that the net vector force, F, on a
specific mass (of fluid, solid or any combination thereof) is equal to the rate of change
of momentum of that mass. It is particular important to note that this applies to a
Lagrangian mass, M, a particular group of particles whose motion is being followed in
the flow. We write this as
D { M u} Du
F= =M (2.80)
Dt Dt
The second form on the right of the above equation follows since the mass, M, in a
Lagrangian volume does not change with time (if we do not consider relativistic effects).
As in the case of the development of the equation for conservation of mass, we will
develop several applications of Newton’s law using both infinitesmal and macroscopic
control volumes. We begin utilizing an infinitesmal control volume to develop differential
equations that embody Newton’s law and we do this at two levels of complexity, one
which neglects viscous forces and leads to Euler’s equations and the second which
includes those viscous forces and leads to the Navier-Stokes equations. Later we utilize
a macroscopic control volume to develop the very useful momentum thereoms of fluid
mechanics.
Figure 2.18: Infinitesmal Eulerian control volume
We begin by applying Newton’s law to the infinitesmal control volume dx × dy × dz

shown in Figure 2.18 which contains a mass of fluid ρ dx dy dz so that Newton’s law can
be written as
Du
F = ρ dx dy dz (2.81)
Dt
It is useful to delineate several forms of this equation. In tensor form it may be written
as

Fi Dui ∂ui ∂u
=ρ =ρ + uj i (2.82)
dx dy dz Dt ∂t ∂x j
where we have used the relation between the Lagrangian and Eulerian time derivatives
to write the sceond version. Note for future reference that Fi/dx dy dz is the net force (per
unit volume) acting on the infinitesmal control volume and it remains to develop that
quantity.
It is also useful to further develop the vector form of the above equations, namely

F ∂u
=ρ + (u · ∇) u (2.83)
dx dy dz ∂t
by utilizing the following vector identity

2
1 |u|
∇ (u · u) = ∇ = u × (∇ × u) + (u · ∇) u (2.84)
2 2
It remains to evaluate the net force F acting on the control volume. This consists of a
number of contributions divided into two categories, body forces and surface forces. The
so-called body forces such as gravity or electromagnetic forces act on the body of fluid
inside the control volume. Ee shall include only gravity in the present development
for now, though. If the body force per unit volume has components, f i , then that
contribution to Fi/dx dy dz is simply f i . If, like gravity, the body force is conservative (the
energy expended in moving a mass from one location to another is all recovered when
the mass is returned to its original location) then we can define a body force potential
such that f = ∇ (U ) and f i = ∂U/∂xi . If, for example, we define a set of axes such that y
is vertically upward (this is a common choice and universally the case in this text) then
it follows that
U = −ρ g y , f y = −ρ g , fx = fz = 0 (2.85)
where g is the acceleration due to gravity. In the sections which follow the only body
force that we will include will be that due to gravity and the above expressions for U
and f i will be deployed.
There are also surface forces which act on the surfaces of the control volume. Principle
among theses are the forces which the surrounding fluid is imposing on the fluid inside
the control volume. Depending on the nature of the fluid these surface forces can be
quite complex. It is convenient to begin by developing the equations in circumstances
in which the surfaces forces are assumed to be simple and specifically consist only of
forces due to the pressure imposed by the surrounding fluid on the faces of the control
volume. Those forces are normal to the faces on which they act. The result will be the set
of equations of motion known as Euler’s equations. Those equations omit the tangential
forces that act on the surfaces of the control volume which are usually shear stresses
caused by the viscosity of the fluid. Later in this text we return to the control volume
analysis to include these viscous forces; the resulting set of equations of motion are
called the Navier-Stokes equations.
2.8.1 Euler’s equations of motion
As previously derived, Newton’s first law of motion applied to the infinitesmal control
volume dx × dy × dz as shown in Figure 2.18 can be written as

Fi Dui ∂ui ∂ui
=ρ =ρ + uj (2.86)
dx dy dz Dt ∂t ∂x j
or in vector form as

F ∂u
=ρ + (u · ∇) u (2.87)
dx dy dz ∂t
where Fi/dx dy dz is the force acting on the control volume divided by its volume. In
the absence of tangential surface forces, Fi/dx dy dz will consist of body forces f i plus
surface forces imposed by the surrounding fluid. Euler’s equations are that version of
the equations of motion which neglect any tangential surface forces and include only the
normal forces, the forces due to the pressure, p. In a Newtonian fluid this implies that
we are assuming an inviscid fluid in which all viscous forces are neglected and, since
tangential forces are proportional to the viscosity, the tangential forces are zero.
Figure 2.19: The forces due to pressure on faces normal to the x direction
Under these circumstances we need only assess the net pressure forces acting on the
control volume in order to complete the derivation of Euler’s equations. It is simplest
to do this one Cartesian component at a time and we choose first the x direction. Since
the only forces due to pressure acting in the x direction are those on the faces of
the control volume perpendicular to the x direction we need only consider the forces
indicated in Figure 2.19. If we define the pressure acting on the left hand side of
the control volume as p then the force per unit area acting on that side is p dy dz in
the positive x direction. It follows that the pressure acting on the right hand side is
[ p + (∂p/∂x)] and therefore the force acting on the right hand side is [ p + (∂p/∂x)] dy dz in
the negative x direction. Consequently the net force on the control volume in the positive
x direction is [− (∂p/∂x) dx dy dz] and the force per unit volume is − (∂p/∂x) and this is the
contribution to Fx/dx dy dz due to the pressure acting on the control volume. Similarly the
contributions to Fy/dx dy dz and Fz/dx dy dz are − (∂p/∂y) and − (∂p/∂z) respectively. Finally
Euler’s equations of motion for an inviscid fluid become

∂ui ∂ui ∂p
ρ + uj =− + fi (2.88)
∂t ∂x j ∂xi
or equivanlently in vector form

∂u
ρ + (u · ∇) u = −∇ p + f (2.89)
∂t
Moreover, in the case of conservative body forces we may replace f by ∇U where U

is the body force potential (U = −ρ g y for the force due to gravity when y is vertically
upward) and then Equation 2.89 can be written as
2
|u|

∂u
ρ +∇ − u × (∇ × u) = −∇ p + f (2.90)
∂t 2
If, in addition, the flow is incompressible (ρ is constant and uniform) then this can be
written as
| u |2

∂u p
− u × (∇ × u) + ∇ −U + =0 (2.91)
∂t ρ 2
and we will utilize this form in the pages which follow.

Note that in a static fluid (u = 0) under the action of gravity (U = − g y) this equation
(Eq. 2.91) reduces to ∂p/∂y = −ρ g which is consistent with the expression derived in the
section on fluid statics.
Extensive use will be made of Euler’s equations during our discussions of fluid flow
phenomena. Under some conditions the equations can be integrated to yield a scalar
relation between the pressure, velocity and elevation, an important equation known as
Bernoulli’s equation. However in order to properly frame that derivation and integration
it is valuable to digress to discuss a quantity called the vorticity that plays a central role
in our understanding of fluid flow.
2.8.2 Vorticity
The vorticity, ω ( x, t) or ωi ( x, t), in a fluid flow is a vector quantity equal to twice the rate
of rotation of an infinitesmal fluid element. It is simply related to the velocity vector by
ω = ∇×u (2.92)
∂w ∂v ∂u ∂w ∂v ∂u
ωx = − ; ωy = − ; ωz = − (2.93)
∂y ∂z ∂z ∂x ∂x ∂y
Note that just as we defined streamlines to be lines that are everywhere tangential
to the velocity vector, u, we also define vortex lines to be lines that are everywhere
tangential to the vorticity vector, ω. Note also that by virtue of the definition of vorticity,
vortex lines are everywhere orthogonal to streamlines. Also note that in planar flow the
vorticity is perpendicular to the plane of the flow; in such flow it is convenient to use ω
to denote the magnitude of the vorticity and therefore
∂v ∂u
ω= − (2.94)
∂x ∂y
Figure 2.20: Boundary layer at large Reynolds numbers
Though it may be premature to do so, it is useful to anticipate some of the properties

of vorticity even though they will not be proven until later. We begin with a uniform
stream for which ω = 0 since all the velocity gradients are zero. It transpires that when
such a flow encounters a solid object such as the airfoil in Figure 2.20, vorticity is created
at the surface of that object since the fluid in contact with the solid surface is at rest
as a result of the no-slip condition. Consequently the fluid is inclined to roll at the
surface rather like a layer of microscopic ball bearings between two sliding surfaces. But
due to the action of the tangential stresses between fluid layers (caused by the fluid
viscosity), this rotation tends to be transmitted further outward into the fluid. Hence
the vorticity is created at the solid surface due to the no-slip condition and is then
diffused outward into the fluid through the action of viscosity. Of course those outer
layers of fluid are also being carried along in the flow and therefore the vorticity tends
also to be convected along in the flow. Hence there are two transport mechanisms for
vorticity, diffusion across fluid layers by viscosity and convection in the direction of flow.
The latter mechanism dominates when a parameter called the Reynolds number, Re, is
greater than unity (Re = U L/ν where U and L are the typical velocity and dimension of
the flow and ν is the kinematic viscosity of the fluid.) Then, as sketched in Figure 2.20,
the vorticity will be confined to a thin layer near the solid surface, a region that is called
the boundary layer. Beyond the trailing edge of the body the boundary layers form
the wake behind the body. In contrast to the flow inside the boundary layer and wake,
the flow outside tends to remain free of vorticity. Such a flow in which the vorticity is
negligibly small is called an irrotational flow and is therefore characterized by
ω = ∇ (u) = 0 (2.95)
There are many circumstances and applications in which it is useful to solve for the
details of an irrotational flow. One such class consists of flows in which solid boundaries
play a minor role, for example waves on a deep ocean. Another class are flows at large
Reynolds numbers in which the boundary layers are very thin compared with the other
dimensions of the flow. Then, as illustrated in Figure 2.21, a first approximation would be
to neglect the boundary layer and to solve the irrotational flow bounded by the geometry
of the body instead of by the geometry of the outer surfaces of the boundary layers.
Figure 2.21: Thin boundary layer at large Reynolds number
Since the irrotational flow has a tangential velocity at the outer surfaces of the boundary
layers, it would be inappropriate to apply the no-slip condition to that irrotational
solution. But it would be appropriate to apply the condition of zero normal velocity.
These and other details of such solutions are treated in more detail later in this text;
the present intent is simply to indicate that irrotational flow solutions do have practical
value in real fluid flows. Here we proceed to investigate some of the characteristics of
irrotational flows and to develop a number of solution methodologies and examples. To
do so we first introduce the velocity potential.
2.8.3 Velocity potential
Irrotational flow is defined as a flow in which the vorticity, ω, is zero and since
ω = ∇×u (2.96)
it follows that the condition, ω = 0, is automatically satisfied by defining a quantity

called the velocity potential, ϕ, such that
u = ∇ϕ (2.97)
since it is always true that ∇ × ∇ϕ = 0. For this reason irrotational flow is often called
potential flow and we will refer to it as such. Other forms of Equation 2.97 are
∂ϕ ∂ϕ ∂ϕ ∂ϕ
ui = ; u= ; v= ; w= (2.98)
∂xi ∂x ∂y ∂z
It is important not to confuse the velocity potential, ϕ, with the streamfunction, ψ. The
former can be defined in any three-dimensional flow whereas a streamfunction can only
be defined in some two-dimensional flows. For example for incompressible planar flow
∂ϕ ∂ϕ ∂ψ ∂ψ
u= ; v= but u= ; v=− (2.99)
∂x ∂y ∂y ∂x
Figure 2.22: The orthogonal flow net of streamlines and equipotentials in an incompressible
planar potential flow
These are called the Cauchy-Riemann equations and they imply that, in incompressible
planar potential flow, the lines of constant velocity potential (called equipotentials) are
everywhere perpendicular to the lines of constant ψ, namely streamlines. Therefore, in
sketches of incompressible planar potential flow these lines, streamlines and equipoten-
tials, form an orthogonal net that can be visualized as covering the flow as shown by
example in Figure 2.22.
2.8.4 Bernoulli’s equation
Another key feature of irrotational flow is the fact that it allows integration of Euler’s
equation to yield Bernoulli’s equation, one of the most widely used relations in practical
fluid flow problems. Euler’s equation for inviscid flow with conservative body forces is
2
|u|

∂u
ρ +∇ − u × (∇ × u) = −∇ p + ∇U (2.100)
∂t 2
and since ∇ × u = ω, this can be written as

2
|u|

∂u
ρ +∇ − u × ω = −∇ p + ∇U (2.101)
∂t 2
We consider a number of different circumstances in which important scalar relations

can be derived from this:
• If the flow is steady (∂u/∂t = 0) and irrotational (ω = 0) as well as inviscid then

Equation 2.101 becomes
2
|u|
ρ∇ + ∇ p − ∇U = 0 (2.102)
2
• If, in addition, the flow is incompressible so that ρ is uniform and constant then
Equation 2.102 can be written as
ρ | u |2

∇ +p−U = 0 (2.103)
2
• Since the gradient of the quantity in parentheses is everywhere zero it follows that
that quantity can only be a function of time and since the flow is assumed to be
steady it can only be a constant
ρ | u |2
+ p − U = constant and uniform (2.104)
2
• If the only body forces are those due to gravity acting in the y direction so that
U = −ρ g y this becomes
ρ | u |2
+ p + ρ g y = constant (2.105)
2
This is Bernoulli’s equation for steady, incompressible, inviscid, irrotational flow with
conservative body forces due to gravity. It is one of the most widely used equations of
fluids engineering and we have much more to say about it elsewhere in this text.
A version of Bernoulli’s equation for unsteady flow can be obtained by noting that
since the flow is irrotational we can write u = ∇ϕ and thus retain the ∂u/∂t term as
∇ (∂ϕ/∂t) so that Bernoulli’s equation for unsteady, incompressible, inviscid, irrotational
flow with conservative body forces due to gravity is
∂ϕ ρ |u|2
+ + p + p g y = uniform in space, but might be function of time (2.106)
∂t 2
What if the flow is not irrotational and we must retain the term u × ω in Eq. 2.101?
Note that this vector quantity u × ω is perpendicular to both the velocity vector u and to
the vorticity vector ω. Consequently the components of u × ω in the direction of both
the velocity vector and the vorticity vector are zero. It follows that in any steady, inviscid,
incompressible, flow (with a body force potential U ) the quantity
ρ | u |2
+p−U (2.107)
2
is constant along any streamline or along any vortex line. It is therefore constant on any
sheet defined by a set of vortex lines and streamlines. It may however change from one
sheet to another if the flow is not irrotational.
Since Bernoulli’s equation is used extensively in engineering, some groups of terms in
it have traditional names that are useful to mention. Equation 2.105 is often written as
| u |2 p
+ + y = constant (2.108)
2g ρg
The group y + p/ρ g is referred to as the static head or piezometric head whereas the term
|u|2/2 g
is called the dynamic head. The sum of the static and dynamic heads is called the
total head (denoted by H) and equivalently the quantity ρ |u|2/2 + p + ρ g y = constant
is known as the total pressure (denoted by p T ). Thus Bernoulli’s equation states that if
the flow is steady, inviscid, incompressible and irrotational the total head and the total
pressure will be the same everywhere in the flow. On the other hand if there are viscous
losses the total head (or total pressure) will (presumably) decrease in the direction of
flow as energy is diverted to overcome the viscous losses.
Bernoulli’s equation is often used in the non-dimensional form
| u |2 p 2gy
+1 + 2 = constant (2.109)
U2 /2 ρ U 2 U
where U is some reference fluid velocity, often that of an upstream uniform flow.
Frequently, it is convenient in this context to define a non-dimensional pressure, or
coefficient of pressure, C p , as
p − p∞
Cp = 1/2 ρ U 2
(2.110)
where p∞ is a reference pressure often that where the fluid velocity, U, pertains and
where the elevation is y∞ . Then Bernoulli’s equation can be written as
| u |2 2 g (y − y∞ )
+ Cp + = constant (2.111)
U2 U2
In many high speed flows the third term on the left-hand side is negligible. Then
provided p∞ and U are chosen to refer to upstream uniform flow conditions then
Bernoulli’s equation implies that C p and the velocity ratio, |u|/U are simply related by
1/2
| u |2

Cp = 1 − 2 (2.112)
U
2.8.5 Linear momentum theorem
In previous sections we have used Newton’s law of motion applied to an infinitesmal

fluid element to develop the differential equations of fluid motion such as Euler’s
equations. But, just as we found in applying the principle of mass conservation to a
flowing fluid, we can also develop very useful tools by considering the application of
Newton’s law to an macroscopic volume of fluid.
To do so consider the arbitrary Lagrangian volume, VL , which, at the moment of time
under consideration, coincides with the Eulerian volume, V, as depicted in Figure 2.23.
Since the Lagrangian volume always contains the same mass of fluid (and/or solid), it
follows from conservation of mass that
ˆ
D
ρ dVL = 0 (2.113)
Dt VL
Figure 2.23: Arbitrary Eulerian control volume, V, and the coincident Lagrangian volume, VL , at
time, t = 0
Newton’s law states that the net vector force, F, acting on this Lagrangian mass must
be equal to the Lagrangian rate of change of the vector momentum contained within the
volume and since the contained mass remains constant this can be written as
ˆ
D
F= ρ u dVL (2.114)
Dt VL
The problem in developing this relation further is that the right hand side requires us
to evaluate the rate of change with time of an integral whose limits (VL ) are changing
with time and whose integrand is changing with time. The key step in addressing these
difficulties is to use the transport theorem to write the right hand side as
ˆ ˆ
∂(ρ u)
F= dV + (ρ u) u · dS
V ∂t
ˆ ˆS (2.115)
∂(ρ u)
= dV + ρ u (u · n) dS
V ∂t S
so that the integrals are all transformed to volume or surface integrals over the fixed
Eulerian volume. This is the linear momentum equation or theorem. In words it states The LHS remains the
that for any arbitrary Eulerian volume the vector force acting on that volume is equal same since the force
acting on the
to the rate of change of the momentum contained within the Eulerian volume (the first
coincident
term on the right hand side) plus the net vector flux of momentum out of that Eulerian Lagrangian volume
volume. For ease of evaluation note that the momentum flux term is the integral over is the same as the
the surface S of the mass flow rate out of the surface (ρ un dS where un is the outward force acting on the
Eulerian volume
velocity component normal to the surface) times the vector velocity, u.
In other words the linear momentum theorem says that the vector force does two
things: it can accelerate the fluid within the Eulerian control volume and it can produce
a net flux of momentum out of the control volume. This is an important and powerful
result which does not involve any assumption about the fluid (whether compressible
or incompressible, viscous or inviscid) or, indeed, whether the content of the control
volume is solid, fluid or both. In steady flows the first component of the right hand side
is zero since the partial time derivatives are zero. Consequently, in steady flow, the net
force is equal to the net flux of momentum out of the control volume.
It is important to stress that the momentum equation (Eq. 2.115) is a vector equation.
However, it is sometimes easier to apply the theorem in its vectorial components
ˆ ˆ
∂ ( ρ ui )
Fi = dV + ui (ρ un dS) (2.116)
V ∂t S
where un is the component of the vector velocity normal to dS and out of the control
volume. In words this states that the force on the control volume in the i direction is
equal to the rate of change of the momentum in the i direction (called the i-momentum
for convenience) contained within the Eulerian volume plus the net flux of i-momentum
out of that Eulerian volume. It is often easiest to separately apply the linear momentum
theorem in each of three Cartesian directions. Note that then the flux of x-momentum
(for example) out of a segment of control volume surface would be given by the mass
flow rate out of through that segment times the velocity in the x-direction.
To demonstrate the relationship between the momentum equation and the differential
versions of Newton’s law, namely the Euler equations, it is necessary to develop and
substitute expressions for the force, F. This is most simply done for an inviscid fluid and
should result in Euler’s equations. In an inviscid fluid the force F will consist of body
force acting on the volume of fluid in the control volume and forces due to the pressure
acting on the surface. The force due to gravity is then simply
ˆ
− ρ g y dV (2.117)
V
where g is the acceleration due to gravity and y is a unit vector in the vertically upward
direction. The force due to the pressure, p, acting on the surface of the control volume is
p dS and acts opposite to the outward normal, n, and so yields the following contribution
to F
ˆ
− p n dS (2.118)
S
When these two forces are substituted into Equation 2.115, the linear momentum
theorem for the inviscid case becomes
ˆ ˆ ˆ ˆ
∂(ρ u)
− ρ g y dV − p n dS = dV + ρ u (u · n) dS (2.119)
V S V ∂t S
To proceed we employ Gauss’ theorem to convert the surface integrals to volume

integrals and thus obtain:
ˆ ˆ ˆ ˆ
∂(ρ u)
− ρ g y dV − ∇ p dV = dV + ρ (u∇ · u + (u · ∇) u) dV (2.120)
V V V ∂t V
where the last term is most readily obtained by considering Gauss’ theorem applied
to each of the components of u (u · n). By continuity, ∇ · u = 0, and after rearranging
Equation 2.120 becomes
ˆ
∂u
ρ + ρ (u · ∇) u + ∇ p + ρ g y dV = 0 (2.121)
V ∂t
and since the volume V is arbitrary the integrand must be zero everywhere so that
∂u ∇p
+ (u · ∇) u + +gy = 0 (2.122)
∂t ρ
which is precisely Euler’s equation for an incompressible, inviscid flow.

2.9 general equations of motion 79
2.9 general equations of motion
2.9.1 The stress tensor
The general state of stress in any homogeneous continuum, whether fluid or solid,
consists of a stress acting perpendicular to any plane and two orthogonal shear stresses
acting tangential to that plane. Thus, as depicted in Figure 2.24, there will be a similar set
of three stresses acting on each of the three perpendicular planes in a three-dimensional
continuum for a total of nine stresses which are most conveniently denoted by the stress
tensor, σij , defined as the stress in the j direction acting on a plane normal to the i
direction. Equivalently we can think of σ as the 3 × 3 matrix
 
σxx σxy σxz
 σyx σyy σyz  (2.123)
σzx σzy σzz
Figure 2.24: Differential element indicating the nine stresses
in which the diagonal terms, σxx , σyy and σzz are the normal stresses which would
be equal to − p in any fluid at rest (note the change in the sign convention in which
tensile normal stresses are positive whereas a positive pressure is compressive). The
off-digonal terms, σij with i ̸= j, are all shear stresses which would, of course, be zero in a
fluid at rest and are proportional to the viscosity in a fluid in motion. In fact, instead of
nine independent stresses there are only six because, as we shall see, the stress tensor is
symmetric, specifically σij = σji . In other words the shear stress acting in the i direction
on a face perpendicular to the j direction is equal to the shear stress acting in the j
direction on a face perpendicular to the i direction. The proof is as follows: consider the
very small rectangular element depicted in Figure 2.25 and evaluate the moment of the
forces about the axis parallel to the z direction. The only stresses contributing forces
that in turn contribute a moment about an axis parallel to the z-axis are those shown
in Figure 2.25. The two forces on planes perpendicular to the y direction have moment
arms of dy/2 while the two forces on plane perpendicular to the x direction have moment
arms equal to dx/2. Therefore, neglecting the higher order contributions that are of order
(dx )4 , the net moment in the clockwise direction is

dy dx
2 σyx dx dz − 2 σxy dy dz (2.124)
2 2
Figure 2.25: Forces contributing to moment about an axis parallel to z-axis
According to Newton’s law of angular motion this moment is equal to the moment of
inertia multiplied by angular acceleration of the element. The moment of inertia will be
of the order of the density multiplied by the dimension of the element to the fifth power
and since the angular acceleration does not contain a length dimension, it follows that
for an element of infinitesmal size the moment given by Equation 2.123 must be zero. It
therefore follows that
σyx = σxy (2.125)
The same procedure for moments about axes parallel to the x and the y axes leads to
σyz = σzy and σxz = σzx (2.126)
hence proving the symmetry of the stress tensor, σij = σji .

Having established the stress tensor and therefore all the forces acting on the surface
of a differential element, dx dy dy, the net force due to the general stress configuration
can be assessed and substituted into Newton’s law. It is most instructive to separately
evaluate the net force in each of the Cartesian directions. All of the forces in the x
direction are shown in Figure 2.25 and the sum of these in the positive x direction is

∂σyx
σyx + dy dx dz − σyx dx dz+
∂y

∂σzx
σzx + dz dx dy − σzx dx dy+ (2.127)
∂z

∂σxx
σxx + dx dy dz − σxx dy dz
∂x
and this becomes

∂σxx ∂σyx ∂σzx
+ + dx dy dz (2.128)
∂x ∂y ∂dz
A similar analysis for the net force in the y direction yields

∂σxy ∂σyy ∂σzy
+ + dx dy dz (2.129)
∂x ∂y ∂dz
as well as for the net force in the z direction

∂σxz ∂σyz ∂σzz
+ + dx dy dz (2.130)
∂x ∂y ∂dz
and therefore the net force vector acting on the surfaces of the control volume can be
written as
∂σij
dx dy dz (2.131)
∂x j
This is to be compared with the net surface force vector obtained during the derivation
of Euler’s equations and in the absence of viscous shear forces, namely
∂p
− dx dy dz (2.132)
∂xi
Then, if the expression for Eq. 2.131 instead of the expression in Eq. 2.132 is used in
the equations of motion, we obtain the general equations of motion for a homogeneous
continuum; namely

∂ui ∂ui ∂σij
ρ + uj = + fi (2.133)
∂t ∂x j ∂x j
This equation is widely applicable. In solid mechanics when the velocities are zero it
is known as the equilibrium equation. In fluid mechanics it will form the basis of the
Navier-Stokes equations once we have established the relationship between the stresses
and the velocities.
Written out for each of three Cartesian directions this equation becomes

∂u ∂u ∂u ∂u ∂σxx ∂σxy ∂σxz
ρ +u +v +w = + + + fx
∂t ∂x ∂y ∂z ∂x ∂y ∂z

∂v ∂v ∂v ∂v ∂σyx ∂σyy ∂σyz
ρ +u +v +w = + + + fy (2.134)
∂t ∂x ∂y ∂z ∂x ∂y ∂z

∂w ∂w ∂w ∂w ∂σzx ∂σzy ∂σzz
ρ +u +v +w = + + + fz
∂t ∂x ∂y ∂z ∂x ∂y ∂z
2.9.2 Constitutive laws
The constitutive laws for a substance, whether solid or fluid, are the relations between the
forces imposed on that substance and the resulting deformation at a macroscopic level.
In essence, they represent the collaborative effects of the intermolecular forces when
the substance is stretched, twisted, sheared or otherwise deformed. A substance can
have many different constitutive laws pertaining to different deformations. For example,
Laplace’s equation for the relation between the surface tension pressure difference and
the curvature of a liquid interface is one such constitutive law as is Fick’s law for
heat conduction. However, the constitutive law which we will introduce here is that
connecting the state of stress within a substance as defined by the stress tensor, σij ,
and the displacement or velocity gradients caused by those stresses as defined by the
strain tensor, Eij , or the rate of strain tensor, eij . That relationship between the state of
stress and the state of strain within a substance is a material property resulting from the
intermolecular forces and bonds.
Though the focus here will be on the constitutive laws for a fluid, it is useful to
put these in context by also describing some of the basic constitutive laws for a solid.
These allow us to again identify the differences between a fluid and a solid and also
describe the laws for substances having both fluid and solid properties. In the semi-rigid
structure of a solid one might surmise that the obvious constitutive law would be a
linear relationship analogous to that between the load and extension of a simple spring;
in other words a linear relation between the stress tensor, σij , and the strain tensor, Eij .
However, such a relation has to be supplemented by the normal stresses, σii , that result
from uniform isotropic displacement of the material as given by the dilation, Φ∗ . Thus
the constitutive law for a homogeneous isotropic solid, known as Hooke’s Law, is
σij = Λ δij Φ∗ + 2 G Eij (2.135)
where δij is the Kronecker delta ( δij = 1 for i = j and δij = 0 for i ̸= j). The factors of
proportionality are the Lame constant, Λ, and the shear modulus, G; which are related
to Young’s modulus, E, and Poisson’s ratio, Σ, by
EΣ E
Λ= and G= (2.136)
(1 + Σ ) (1 − 2 Σ ) 2 (1 + Σ )
While Equation 2.135 is the most commonly used constitutive law for a simple solid
there are many variants representing different solids and their properties.
In contrast, a fluid is defined as a substance in which the stresses do not depend on
the displacement but do depend on the velocity gradients within the fluid. The most
commonly used constitutive law for a simple fluid, known as a Newtonian fluid, is
σij = δij {− p + (Λ − 2/3 µ) Φ} + 2 µ eij (2.137)
While the last term involving the viscosity, µ, is the expected linear relation for the
stress associated with the shear in the fluid (the factor 2 is included simply so that the
stress and shear rate in a simple Couette flow are directly connected by the viscosity),
the first term, which includes the Kronecker delta and therefore only contributes to the
normal stresses, needs some further explanation. The first part of the first term, − p δij
incorporates the normal stresses that would obviously be associated with a uniform
isotropic pressure, p. The second part is another isotropic contribution which vanishes
for an incompressible fluid in which Φ = 0 and involves the parameter, Λ, the so-called
second coefficient of viscosity. For a monatomic gas Λ = −2/3 µ.
For an incompressible Newtonian fluid the constitutive law becomes

∂ui ∂u j
σij = −δij p + 2 µ eij = −δij p + µ + (2.138)
∂x j ∂xi
and comprises the isotropic component, − p, and the deviatoric stress tensor, σijD
σijD = 2 µ eij (2.139)
In terms of its Cartesian components, the constitutive law for an incompressible

Newtonian fluid are
∂u ∂v ∂w
σxx = − p + 2 µ ; σyy = − p + 2 µ ; σzz = − p + 2 µ (2.140)
∂x ∂y ∂z

∂v ∂u ∂w ∂v ∂u ∂w
σxy = µ + ; σyz = µ + ; σzx = µ + (2.141)
∂x ∂y ∂y ∂z ∂z ∂x
2.9.3 The Navier-Stokes Equations
Substituting the expressions for the stresses in terms of the strain rates from the constitu-
tive law for a fluid into the equations of motion we obtain the important Navier-Stokes
equations of motion for a fluid. In passing we should also note that the same process
using the constitutive law for a solid yields the so-called equations of equilibrium for
that solid. Specifically, substituting the constitutive law for a Newtonian fluid, Equation
2.137, into the equation of motion (Eq. 2.133) yields the Navier-Stokes equations for a
Newtonian fluid with dynamic viscosity µ:

∂ui ∂u ∂p ∂ ∂ui ∂u j ∂ ∂u j
ρ + uj i = − + µ + + Λ + f i (2.142)
∂t ∂x j ∂xi ∂x j ∂x j ∂xi ∂xi ∂x j
For an incompressible Newtonian fluid, the Navier-Stokes equations simplify to

∂ui ∂u ∂p ∂ ∂ui ∂u j
ρ + uj i = − + µ + + fi (2.143)
∂t ∂x j ∂xi ∂x j ∂x j ∂xi
For a uniform viscosity and using the equation of continuity for an incompressible
fluid, namely,
∂ui
=0 (2.144)
∂xi
the Navier-Stokes equations (Eq. 2.142) become
∂2 u i

Dui ∂ui ∂ui ∂p
ρ =ρ + uj =− +µ + fi (2.145)
Dt ∂t ∂x j ∂xi ∂x j ∂x j
We shall refer to the terms in this equation as follows: the term (or terms) on the
left-hand side is the inertial term, the first term on the right-hand side is the pressure
term, the second is the viscous term and the last is the body force term. Written out in its
Cartesian components the equations are
2
∂ u ∂2 u ∂2 u

∂u ∂u ∂u ∂u ∂p
ρ +u +v +w = − +µ + + + fx
∂t ∂x ∂y ∂z ∂x ∂x2 ∂y2 ∂z2
2
∂ v ∂2 v ∂2 v

∂v ∂v ∂v ∂v ∂p
ρ +u +v +w = − +µ + 2 + 2 + fy (2.146)
∂t ∂x ∂y ∂z ∂y ∂x2 ∂y ∂z
2
∂ w ∂2 w ∂2 w

∂w ∂w ∂w ∂w ∂p
ρ +u +v +w = − +µ + 2 + 2 + fz
∂t ∂x ∂y ∂z ∂z ∂x2 ∂y ∂z
and in the forms that we will frequently use in the pages ahead, the Navier-Stokes
equations for planar, incompressible, Newtonian flow in the xy plane are:
2
∂ u ∂2 u

∂u ∂u ∂u ∂p
ρ +u +v = − +µ + 2 + fx
∂t ∂x ∂y ∂x ∂x2 ∂y
2 (2.147)
∂ v ∂2 v

∂v ∂v ∂v ∂p
ρ +u +v = − +µ + 2 + fy
∂t ∂x ∂y ∂y ∂x2 ∂y
In these pages, the Navier-Stokes equations will be deployed to explore a wide range
of flows. However, the number of exact solutions to these equations is small and mostly
for geometrically simple flows. These are described and explored in the sections which
follow. The exact solutions are very limited because the non-linear inertial terms present
an intractable mathematical barrier in all but the simplest geometries. Therefore we will
resort to approximate solutions for many different types of flows and in doing so will
have a need to neglect or approximate some of the terms in the equations in order to
make progress. In the process we will estimate the magnitudes of some of the terms
relative to other terms. In particular we will compare the magnitudes of the viscous
and inertial terms and the ratio of these magnitudes leads to an important fluid flow
parameter known as the Reynolds number, Re.
Consider a Newtonian fluid flow with a typical dimension, L, and a typical velocity,
U. Then one might argue that the inertial terms in the equation of motion Eq. 2.145
would have a typical magnitude of ρ U 2/L and that the viscous terms would have a typical
magnitude of µ U/L2 . Therefore the importance of the inertial terms relative to the viscous
terms would be given by the ratio known as the Reynolds number, Re is defined as
ρ U 2/L ρU L UL
Re = = = (2.148)
µ /L
U 2 µ ν
This number is used throughout the subject of fluid mechanics in order to assess the
relative importance of the inertial and viscous terms and therefore to try to justify some
simplification to the Navier-Stokes equations.
As a first example one might argue that in flows for which Re >> 1 the viscous terms
would be unimportant relative to the inertial terms and therefore one could leave out the
viscous terms. The equations of motion would then revert to Euler’s equations and the
methods utilizing those equations can be used. On the other hand one might also argue
that in flows for which Re << 1 the inertial terms would be unimportant relative to the
viscous terms and therefore one might omit the inertial terms in seeking solutions to the
flow. This approach leads to a set of equations known as the equations of creeping flow
or Stokes’ flow.
If these two arguments were foolproof then fluid mechanics would be a much simpler
subject and the various fluid flow phenomena would be less interesting. The problem is
that the arguments are not uniformly valid and, in both cases, fail in some particular
regions of the flow.
First consider the case of Re >> 1. While it is true that the viscous terms are much
smaller than the inertial terms in most regions of the flow and can therefore be neglected
in seeking solutions to the flows in those regions, there are other regions where this is
not true and the viscous terms therefore play a significant role. As we will describe in
the section on boundary conditions, the flow at the surface of a solid object experiences
what is known as the no-slip condition so that the velocity of the fluid in contact with
the solid surface is equal to that of the solid surface and is therefore zero for a solid
object at rest (for simplicity of demonstration we focus on steady flows past a solid object
at rest). Therefore the inertial terms (which involve the velocity) are identically zero at
that solid surface and are very small close to the solid surface. In contrast the viscous
terms are non-zero since they involve the gradients of the velocity but not the velocity
itself. Hence near the solid surface there is a special region in which the viscous terms
are actually large compared with the inertial terms and must therefore be included in
seeking a solution to the flow in that region. These regions of the flow are called viscous
boundary layers and their study is an important, indeed crucial, component in the study
of flows at high Reynolds numbers. They will be described in much more detail in the
sections that follow. Further out in the flow, away from the solid surface, the viscous
terms are indeed negligible in many cases and that region of the flow can be treated
using Euler’s equations of motion and the results, such as Bernoulli’s equation, that
follow from Euler’s equations.
Second, consider the complementary case of Re << 1. While it is true that the inertial
terms are much smaller than the viscous terms in many regions of the flow and can
therefore be neglected in seeking solutions to the flows in those regions, there are other
regions where this is not true and the inertial terms therefore play a significant role.
Specifically consider again the flow of a uniform stream past a solid object. If Re << 1
then anywhere close to the solid object the inertial terms will be small relative to the
viscous terms and can be neglected in seeking a solution to the flow anywhere near to
the solid object; thus the solution in that region would appropriately utilize the equations
of Stokes’ flow as described above. However, when we consider the flow at a large
distance from the object in a region where the flow is close to being that of a uniform
stream, we recognize that the second derivatives that comprise the viscous terms could
and do become very small and therefore the inertial terms become comparable to and
larger than the viscous terms. In that region, far from the object where the inertial terms
are significant we must use a modified version of the Stokes’ equations, one version of
which are the Oseen equations described in a section that follows.
2.9.4 Vorticity transport equation
For an incompressible Newtonian fluid with a uniform viscosity, the Navier-Stokes are
Dui ∂p ∂2 u i
ρ =− +µ + fi (2.149)
t ∂xi ∂x j ∂x j
and for a conservative force field f i can be written as ∂U/∂xi where U is the body force
potential (in the case of the force due to gravity, U = − g y, where y is the vertically
upward coordinate. Alternatively Equation 2.149 may be written in vector form with
f = ∇U as
| u |2

∂u
ρ +∇ −u×ω = −∇ p + ρ ∇U + µ ∇2 u (2.150)
∂t 2
where, as usual, the vorticity, ω = ∇ × u and we note the vector identity
∇2 u = ∇ (∇ · u) − ∇ × ω (2.151)
Taking the curl of Equation 2.150 in order to eliminate the pressure, p, and using two
additional vector identities, namely
∇2 ω = ∇ (∇ · ω ) − ∇ × (∇ × ω ) = −∇ × (∇ × ω ) (2.152)
since, in the present case, ∇ · ω = 0 and
∇ × (u × ω ) = (ω · ∇) u − ω (∇ · u) − (u · ∇) ω + u (∇ · ω ) = (ω · ∇) u − (u · ∇) ω
(2.153)
again since ∇ · u = 0 and ∇ · ω = 0, then the curl of Equation 2.150 can be written as
Dω ∂ω
= + (u · ∇) ω = (ω · ∇) u + ν ∇2 ω (2.154)
Dt ∂t
or
Dωi ∂ωi ∂u ∂u ∂2 ωi
= + uj i = ωj i + ν (2.155)
Dt ∂t ∂x j ∂x j ∂x j ∂x j
This is the vorticity transport equation for an incompressible fluid with a uniform and
constant viscosity. In using this equation we note that the term (ω · ∇) u is the gradient
of the velocity u in the direction of the vorticity vector multiplied by the magnitude of ω.
This term is zero for planar flow since the velocity vector is perpendicular to the vorticity
vector.
• In a planar, inviscid flow both of the terms on the right hand side of the vorticity
transport equation are zero, the first because the flow is planar and the second
because it is inviscid. But in the flow of a uniform stream around a object, the
vorticity far upstream of the object is zero because the gradient of the velocity
there is zero. Hence the vorticity is zero everywhere in the flow and the flow is
irrotational.
• In a three-dimensional, inviscid flow the vorticity transport equation becomes
Dω
= (ω · ∇) u (2.156)
Dt
Therefore, in a flow that originates with a uniform stream in which the vorticity, ω
is zero, the right hand side of Equation 2.156 is zero and so Dω/Dt = 0 and so the
vorticity cannot change from zero. Therefore without the effects of viscosity, the
vorticity everywhere in such a flow will be zero and the flow will be irrotational.
• In a planar viscous flow the vorticity transport equation becomes
Dω
= ν ∇2 ω (2.157)
Dt
which is a convection/diffusion equation that teaches that vorticity is both con-
vected and diffused in such a flow. Therefore, in the flow that originates with
a uniform stream the zero vorticity upstream only changes because vorticity is
diffused into the flow by the action of viscosity from some source of vorticity. In
many flows that source is the friction with a solid wall where the no-slip condition
produces vorticity that diffuses out into the flow.
• In a three-dimensional viscous flow the right hand side of the vorticity transport
Equation 2.153 or 2.154 teaches that the vorticity in a Lagrangian element of fluid
flowing along a streamline will only change for one of two reasons. Either it
changes because of the diffusion of vorticity into that element through the action of
viscosity. Or it changes because of the term (ω · ∇) u, a phenomenon that is known
as vortex stretching. If we consider a coordinate, s, measured along a line tangent
to the vorticity vector (a vortex line) and denote the velocity in that direction by us
then the term (ω · ∇) u is equal to |ω | dus/ds where |ω | denotes the magnitude of
the vorticity. Since dus/ds is the rate at which the fluid element on the vortex line
is being stretched in the direction of the vorticity, it is appropriate to consider the
term (ω · ∇) u as the contribution of vortex stretching to the rate of change of the
vorticity. This stretching is similar to the way the rotation rate of a ballet dancer or
ice-skater increases as they draw in their arms; it represents a consequence of the
conservation of angular momentum.
2.10 turbulence 87
2.10 turbulence
Fluid flows can become unstable in a number of ways. Some of those lead to large scale
oscillations in the geometry of the flow as exemplified by vortex shedding from a cylinder.
However, in this and the following sections, we will address those circumstances in
which some region of a flow becomes unstable and those local disturbances grow in
magnitude and range of frequency, leading to the phenomenon we call turbulence. Often
this occurs first in a highly sheared region of the flow and spreads out to adjacent parts
of the flow as the unsteadiness envelopes nearby fluid. The most common examples and
the examples that will be addressed in detail in the sections which follow are the flows
in a pipe and the flow in an initially laminar boundary layer.
Consequently, we first address the issue of the stability of a laminar flow and the
methods used to evaluate the stability. Particular attention will be paid to the stability of
quite simple flow, namely planar, parallel flow since the analytical challenges presented
by more complicated flow are very considerable. Once we have addressed the problem
of assessing where and when instability will occur, we briefly address the problem of
understanding the growth of the disturbances from their minute origin to observable
turbulence. Once turbulence has been substantially developed there are flows such as
that in a straight pipe in which the magnitude and frequency content of the turbulence
asymptotes toward a steady state. This is termed fully-developed turbulence and the later
sections in this group focus on the properties and analyses of fully-developed turbulence.
Particular attention will be paid to two common geometric configurations namely to
fully-developed turbulent pipe flow (in both smooth-walled and rough-walled pipes)
and to fully-developed turbulent boundary layers.
The flow instabilities that are the focus of this introduction to turbulence cause very
small disturbances or imperfections in the flow to be amplified. Nominally steady flow
in this context is one in which the time-averaged pressure and velocity is not changing
with time t though it may be a function of position, xi , within the flow. In such flows
we will seek to evaluate the rate of growth of the disturbances with position, the rate of
growth as they are convected downstream. However, we shall also identify the rate of
growth of the disturbances with time in non-steady flows.
In most instances it is difficult to identify the precise source of the very small imperfec-
tions or noise that initiates the process. However, there seem to be at least three different
possible sources of the noise:
• Imperfections in the incoming fluid stream known as free stream turbulence. It is

essentially impossible to eliminate the free-stream turbulence in a water tunnel or
wind tunnel in which experiments are conducted.
• Surface roughness of the object or objects that create the flow; again it is essentially
impossible to eliminate surface roughness effects.
• Acoustic noise and structural vibration in the test facility.
All of this noise will consist of a range of frequencies, f , and both the spectra and
amplitude, A, of this noise is important in an attempt to construct how it will be amplified
by the flow.
As we shall see, it transpires that many of the flows with which we will be concerned,
are highly selective amplifiers. That is to say the amplification rate, dA/dx, is a strong
function of the frequency, f (x could be the longitudinal coordinate in a pipe). At least
during the early phase of amplification, only a narrow range of frequency is amplified
as depicted in Figure 2.26.
Figure 2.26: Representation of the selective amplification of noise in a pipe or boundary layer
Later, as the amplitude increases to a level where frequency dispersion becomes

significant, the waves break to generate a spectrum of frequencies. This progress in the
amplification process can be observed in measurements of the instantaneous velocities in
a flow undergoing transition to turbulence. As we shall see later, the rate of amplification
is also a function of the Reynolds number and the shape of the velocity profile.
2.10.1 Linear stability analysis
The stability of steady laminar flows of an incompressible, Newtonian fluid of kinematic

viscosity, ν, is assessed by a linear stability analysis that begins with the relevant
equations of continuity and motion,
∂ui
=0 (2.158)
∂xi
∂ui ∂u 1 ∂p ∂2 u i
+ uj i = − +ν (2.159)
∂t ∂x j ρ ∂xi ∂x j ∂x j
Each of the flow variables, q (where q represents ui or p) is then decomposed into a

steady or time-independent component denoted by an overline, q, and a small, time-
dependent perturbation, qe. The latter is envisaged as a small perturbation such that
qe << q so that terms that are quadratic (or higher order) in qe quantities can be neglected
leaving only terms that are either independent of or linear in qe quantities. Consequently
various perturbations, qe, can be linearly superposed.
We will conduct what is known as a normal mode analysis by considering perturbations
that are oscillatory in time, t, and in one spatial direction, say x, so that qe may be written
in the form
q = q( xi ) + qe( xi , t)
(2.160)
= q( xi ) + Re {qe∗ ( xi ) exp [i (k x − ω t)]}
where i is the square root of −1, Re {} denotes the real part of and qe∗ is the amplitude
of the perturbation. The wavenumber, k, and the radian frequency, ω, may both be
complex so that
k = kR + i k I and ω = ωR + i ω I (2.161)
where the subscripts R and I denote the real and imaginery parts. This allows two
types of solution that are of particular interest namely:
2.10 turbulence 89
• oscillatory perturbations that have an amplitude that is growing in space but not
in time so that ω I = 0 and ω = ω R is the perturbation frequency. Then k R is the
wavenumber of the wave-like perturbation and k I is the growth or attenuation rate.
For convenience we refer to this as the spatial growth case.
• oscillatory perturbations that have an amplitude that is growing in time but not in
space so that k I = 0 and 2π/k = 2π/k R is the wavelength of the perturbation. Then
ω R is the spatial wavenumber of the wave-like perturbation and ω I is the spatial
growth or attenuation rate. For convenience we refer to this as the temporal growth
case.
Here we will focus primarily on the first case and examine the rate of growth, ω I , of
perturbations of frequency, ω R , and wavenumber, k R .
When expansions of the form of Eq. 2.160 are substituted into the governing equation
(Eq. 2.159), the terms which are independent of t are isolated and solved to obtain the
mean motion. The terms that are linear in the perturbations qe∗ are the linear stability
equations that are to be solved to determine the stability of the flow.
In practice, the implementation of this procedure is very difficult unless the basic mean
flow is simple. Here we shall limit the implementation to simple planar, parallel flows in
which the unperturbed flow consists of a velocity, u = U (y), in the x direction, v = 0
and p is uniform and constant. The equations of motion (Eq. 2.159) for this parallel flow
are
2
∂ u ∂2 u

∂u ∂u ∂U 1 ∂p
+U +v =− +ν +
∂t ∂x ∂y ρ ∂x ∂x2 ∂y2
2 (2.162)
∂ v ∂2 v

∂v ∂v 1 ∂p
+U =− +ν + 2
∂t ∂x ρ ∂y ∂x2 ∂y
e and pressure, pe, will be represented by

The perturbations in the streamfunction, ψ,
e = Re { f (y) exp [i (k x − ω t)]}
ψ and pe = Re { g(y) exp [i (k x − ω t)]} (2.163)
so that

df
ue = Re exp [i (k x − ω t)] and ve = Re {−i k f (y) exp [i (k x − ω t)]} (2.164)
dy
and substituting these expressions into the equations of motion, Eq. 2.160 and 2.159,
yields
d3 f

df dU ikg 2 df
(i k U − i ω ) −ik f =− + ν −k + 3
dy dy ρ dy dy
(2.165)
d2 f

2
1 dg 3
k U−kω f = − +ν ik f −ik
ρ dy dy
Eliminating the function g from these two equations results in

2
d2 U
4 2

d f 2 d f 2 d f 4
(ω − k U ) −k f +k f = iν −2k +k f (2.166)
dy2 dy2 df4 dy2
This equation which must be solved for the perturbation f (y) is called the Orr-
Sommerfeld Equation. The version in which the viscous terms are neglected is
2
d2 U

d f 2
(ω − k U ) 2
− k f +k f =0 (2.167)
dy dy2
and is called the Rayleigh Equation. Notice that both versions are homogeneous in f
and represent eigenvalue problems for which we need to identify boundary conditions.
It remains to discuss the boundary conditions under which these equations must be
solved. At any solid boundary parallel with the x direction at, say, y = 0, the zero normal
velocity and no-slip conditions require that

df
( f ) y =0 = 0 and =0 (2.168)
dy y=0
Furthermore, in the case of the boundary layer problem we must require that
( f )y→∞ → 0 (2.169)
in order that the perturbation velocities decay to zero at large y. In the inviscid
case governed by the Rayleigh equation (Eq. 2.167), these three boundary conditions
(Equations 2.168 and 2.169) complete the eigenvalue problem. The calculation for the
spatial problem using a shooting method can proceed as follows. Given U (y) and a
real frequency, ω R , we choose guessed vales for both k R and k I and begin the numerical
integration for f at y = 0 where both f and d f/dy are zero. The Rayleigh equation then
yields d2 f/dy2 and we integrate using, for example, a Runge-Kutta procedure to find f and
d f/dy at the next y mesh point (this is best done along a complex y contour in order to
avoid potential singularities on the real y axis). As the integration approaches large and
real values of y the boundary condition as y → ∞ must be satisfied and this provides a
criterion by which both k R and k I must be adjusted in order to satisfy that condition. An
iterative method can then be used to determine the final values of k R and k I . Note that
the temporal problem for a particular k R can be addressed in a precisely analogous way
except that this involves initial guessed and finally determined values of ω R and ω I .
2.10.2 Amplification of disturbances
Given the information on amplification rates provided by the stability analyses it is

possible to envisage computing the amplitude of the Tollmein-Schlicting waves as the
flow proceeds downstream and hence, perhaps, being able to estimate where that
amplitude becomes comparable to the wavelength so that the waves begin to break up
and generate turbulence. Such calculations have, indeed, been formulated in an effort
to predict transition to turbulence. For example, if in a spatial amplification case, we
denote the disturbance amplitude for a particular disturbance frequency, ω R , by a then
analyses would provide the disturbance amplification rate, −k I in the x direction where,
in the present discussion, k I is negative. It would follow that
1 da
= −ki ( x ) (2.170)
a dx
2.10 turbulence 91
and therefore by integration
ˆ
 x 

a( x ) = a( x0 ) exp −k I ( x ) dx (2.171)
 
x0
Then, if one knew the amplitude of the initial disturbance, a( x0 ), when that frequency
first became unstable (and the location, x0 , where it first became unstable) one could
evaluate the amplitude of that frequency at all locations downstream. However, there are
three serious difficulties with such a calculation. The first is the difficulty of determining
the amplitude where the flow first becomes unstable for usually this is a very small
quantity. The second is the difficulty of knowing what amplitude constitutes the initiation
of turbulence. And the third is the fact that the analyses of stability are linear analyses
and non-linear effects will undoubtedly become important as the end of this growth
process is approached.
Despite these difficulties, methods of the above type have been developed in an effort
to predict transition to turbulence, for example by A.M.O. Smith and his colleagues at
Douglas Aircraft. Smith recommended a rule of thumb that transition would occur when
a( x )/a( x0 ) = en where n is of the order of 8 or 9. This can only be regarded as a very crude
estimate but one that was based on experimental observations.

Though quantifying the progress toward turbulence is difficult, more is known of
the qualitative mechanisms that occur during the process of transition to turbulence.
Though these mechanisms can be quite flow specific, the general pattern is that some
flow instability leads to some fairly large scale disturbance(s) or eddies in the flow field.
As these disturbances gather energy from the mean flow, they begin to spawn smaller
disturbances or eddies which, in turn spawn even smaller eddies. This process ends
because, eventually, the eddies reach a size for which viscous effects become critical and
the very small eddies are damped out by viscosity. Eventually, the spectrum of spatial
or temporal eddy sizes reaches a fully-developed state in which energy is continually
cascading down to smaller-sized eddies and the disturbance energy for any one size of
eddy becomes relatively constant. We discuss more of the details of this process in the
next section.
The large scale disturbances that occur at the beginning of the above process of
transition can take a number of forms. One are the Tollmein-Schlicting waves. Another
common one are the eddies shed due an instability in the flow around a bluff body.
2.10.3 Turbulence spectra and scale
As we described in the section on amplification, transition to turbulence begins when

some flow instability leads to some fairly large scale disturbance(s) or eddies in the
flow field. As these disturbances gather energy from the mean flow, they begin to
spawn smaller disturbances or eddies which, in turn spawn even smaller eddies. This
process ends because, eventually, the eddies reach a size for which viscous effects
become important and the very small eddies are damped out by viscosity. Eventually,
the spectrum of spatial or temporal eddy sizes reaches a fully- developed state in which
energy is fed from the mean flow into large eddies and then continually cascades
down to smaller and then smaller eddies eventually reaching a size at which viscosity
becomes important and damps out those small eddies. In this fully-developed state the
disturbance energy for any one size of eddy becomes relatively constant though it can,
of course, continue to change with the flow conditions. To examine this process further,
the fluid velocities, stresses and pressures are subdivided into mean, time- averaged
quantities denoted by an overbar and unsteady components with zero time averages
denoted by a prime:
ui = ui + ui′ ; p = pi + pi′ (2.172)
and similarly for the individual velocity components, u, v and w, and all the compo-
nents of the stress tensor, σij . To be specific, the mean or overbar steady components are
defined by averaging the quantity over a period of time, T, which is much larger than
any of the periods of the turbulent fluctuations so that, for example
ˆT +t
1
ui = ui dt (2.173)
T
t
so it necesarilly follows that
ˆ
t+ T
1
ui′ dt = 0 (2.174)
T
t
The mean kinetic energy (per unit mass) associated with the turbulent motions, ui′ , is
denoted by E and is given by
1 ′ ′
E= u u (2.175)
2 i i
and we visualize this energy as being distributed either spatially over eddies of many
sizes (or wavenumbers, k) or over eddies of many frequencies, ω, so that we can define a
turbulent energy density, e(k ) or e(ω ), such that
ˆ∞ ˆ∞
E= e(k ) dk or E= e(ω ) dω (2.176)
0 0
Thus by plotting e(k ) against k or e(ω ) against ω we can present a spectrum of

the turbulent fluctuations. It was G.I.Taylor who hypothesized that these spatial and
temporal distributions were equivalent. A example of a temporal spectrum is presented
in Figure 2.27 in which the amplitude of the turbulent energy density is plotted against
the frequency of those fluctuations. The largest eddies (lowest frequencies) on the left
side of the spectrum are generated by the mean flow and the energy cascades down to
smaller and smaller eddies (larger frequencies) until they are damped out by viscosity.
Various scales are used to characterize these spectra and processes. G.I. Taylor was the
first to use statistical methods to analyse turbulence and to suggest a critical intermediate
eddy size below which viscosity would begin to damp out those eddies (called the
Taylor microscale). However, it was A.N. Kolmogorov who substantially advanced the
understanding of turbulence by using a combination of physical insight and dimensional
analysis to relate the dimensional quantities
• The wave number k L−1 ,

• The turbulent knietic energy per unit mass E L2 T−2

• The turbulent kinetic energy density per unit mass e(k ) L3 T−2

2.10 turbulence 93
Figure 2.27: Characteristics of a turbulence spectrum
• The turbulent kinetic energy flux per unit mass ϵ L2 T−3

where the last quantity is defined below.

Kolmogorov argued that there exists an intermediate range of wavenumbers, called
the inertial range, bounded on the high side by the injection wavenumber, k i , separating
the wavenumbers at which the mean flow is creating large eddies from this intermediate,
inertial range and bounded on the low side by the dissipation wavenumber, k d , separating
the wavenumbers of the inertial range from the wavenumbers that are substantially
attenuated by viscosity. He argued that in this inertial range, k i < k < k d , neither the
mean flow creation nor the viscosity are explicitly important, but instead the energy flux
down the cascade of eddy size and the local wavenumber, k, are the only controlling
parameters. Then, in a steady state, the energy flux, ϵ, flowing down the spectrum to
smaller eddies must be relatively constant though it will later decline in the dissipation
range, k > k d . Then dimensional analysis requires that, in the inertial range,
E ≃ C ϵ /3 k− /3
2 5
(2.177)
where C is some assumed universal constant. Many turbulent spectra, including those
shown in Figures 2.25, exhibit such an inertial range.
Kolmogorov also argued that the dissipation wavenumber, k d , would be related to the
kinematic viscosity, ν, as well the energy flux, ϵ, and therefore by dimensional analysis
k d ∝ ν− /4 ϵ /4
3 1
(2.178)
The inverse of this, λ, is known as the Kolmogorov length scale and the corresponding
Kolmogorov time scale is also given by (ν/ϵ) /2 using dimensional analysis.
1
To provide further perspective on these results, consider a mean flow with velocity, U,
and typical dimension, L. The largest eddies would have a velocity and a dimension also
given by U and L so that, by dimensional analysis ϵ ∝ U 3/L and hence the Kolmogorov
1
length scale would be proportional to ν3/4 L /4/U 3/4 so that
λ 3
∝ ( Re)− 4 (2.179)
L
where Re is the Reynolds number of the flow. This result has important consequences
in demonstrating that the higher the Reynolds number of the flow the smaller the
significant eddies are relative to the dimension of the flow. Therefore, in order to resolve
the details of the flow in a computation with a mesh, the mesh size must become smaller
and smaller the larger the Reynolds number.
2.10.4 Reynolds stresses
In this section we will investigate the effect of the turbulence on the mean flow. It will be
assumed that the turbulent spectrum has reached a steady state so that the turbulence
can be regarded as fully developed. For example, fully developed turbulent pipe flow
would have the same statistical properties at any axial location in the pipe.
We wish to identify the mechanism by which the unsteady motions in the turbulence
effect the mean motions. We choose to limit the present investigation to incompressible,
Newtonian fluids of uniform and constant viscosity for which the equations of continuity
and motion may be written as
∂u j
=0 (2.180)
∂x j
and
∂ui ∂ ( ui u j ) 1 ∂σij 1 ∂p ∂2 u i
+ = =− +ν (2.181)
∂t ∂x j ρ ∂x j ρ ∂xi ∂x j ∂x j
where we have omitted the body force term. Substituting the expressions for the mean
velocity and pressure and then integrating each term over the large time interval, T, and
assuming that the order of the differential and integral operators can be interchanged
(an assumption which could introduce some error), we obtain the following expressions
that govern the mean motions
∂u j
=0 (2.182)
∂x j

∂ ui′ u′j

∂ ui u j 1 ∂p ∂2 u i
+ =− +ν (2.183)
∂x j ∂x j ρ ∂xi ∂x j ∂x j
Notice that the integration over T has eliminated all the terms containing a single
fluctuation but that the mean of a quadratic combination of fluctuation terms is not
necessarily zero and
so the convective inertial term containing such a combination of
velocities, namely ∂ ui u′j /∂x j , must be retained. Indeed this is the sole contribution of the
′
turbulent fluctuations to the mean motion and will be seen to distinguish a turbulent
flow from its laminar counterpart. These additional terms in the governing equations for
turbulent flow are called Reynolds stress terms for the following reason.
It is noticeable that the additional term in the governing Equation 2.183 due to the
turbulent motions has precisely the same mathematical form as the stress terms. It is
therefore convenient to combine these and write Equation 2.183 as

∂ σ − ρ u ′ u′
∂ui 1 ij i j
uj = (2.184)
∂x j ρ ∂x j
where we have also used the continuity equation (Eq. 2.182). Consequently the effect
of the turbulent motions is to add additional ’stresses’ to what are otherwise the same
equations that govern laminar flow. These additional ’stresses’ are called Reynolds
stresses though it is important to recognize that they are not stresses but are instead
2.10 turbulence 95
additional momentum fluxes due to the unsteady turbulent motions. Thus the effective
total stress tensor in the turbulent flow denoted by σij∗ is given by
σij∗ = σij − ρ ui′ u′j (2.185)
and written out in its components this represents the following individual stresses
∗ ∗ ∗
σxx = σxx − ρ u′ u′ ; σyy = σyy − ρ v′ v′ ; σzz = σzz − ρ w′ w′ (2.186)
where the second terms are called the Reynolds normal stresses and
∗ ∗ ∗
σxy = σxy − ρ u′ v′ ; σyz = σyz − ρ v′ w′ ; σzx = σzx − ρ w′ u′ (2.187)
where the second terms are called the Reynolds shear stresses.
It is instructive to demonstrate how the turbulent motions give rise to additional
momentum fluxes and can then be interpreted as additional effective stresses. As an
example we choose to focus on just one of the Reynolds stresses, in particular the shear
stress, ρ u′ v′ . Consider the elemental control volume, dx dy dz, sketched in Figure 2.28
and placed in a turbulent flow such that v = w = 0 and u ̸= 0.
Figure 2.28: An elemental control volume placed in a turbulent flow such that v = w = 0 and
u ̸= 0
We anticipate that ρ u′ v′ is associated with the fluxes of x-momentum (momentum

in the x direction) through the sides ABCD and EFGH. The instantaneous flux of
x-momentum into the control volume through ABCD is
ρ dx dz v′ × u + u′

(2.188)
and the instantaneous flux of x-momentum out of the control volume through EFGH
is

′ ′ ∂ ′
v u + u′

ρ dx dz v u + u + dy (2.189)
∂y
and therefore the net flux of x-momentum out of the control volume through the sides
ABCD and EFGH is
∂
v′ u + u′

ρ dx dy dz (2.190)
∂y
and the mean or time-averaged component of this net flux of x-momentum out of the
control volume is
∂
u′ v′

ρ dx dy dz (2.191)
∂y
Now an additional shear stress, ∆σxy , acting on this control volume would pro-
duce a force −dx dz ∆σxy acting on the surface ABCD in the x direction and a force,
dx dz ∆σxy + dy /∂y ∆σxy acting on the surface EFGH and so yield a net force in

∂
the x direction equal to

∂∆σxy
dx dy dz (2.192)
∂y
Therefore, the net flux of x-momentum given by the expression of Eq. 2.191 is
equivalent to the negative of the expression of Eq. 2.192 and it follows that
∆σxy = −ρ u′ v′ (2.193)
and is the Reynolds shear stress. Similar explanations follow for all the other Reynolds
stresses.
To summarize, the equations for the steady component of fully-developed turbulent
flows are
∂u j
=0 (2.194)
∂x j
∂ui 1 ∂p ∂2 u i ∂ ′ ′
uj =− +ν − ui u j (2.195)
∂x j ρ ∂xi ∂x j ∂x j ∂x j
and written out in components for the planar flows that will be the focus of the
following sections these become
∂u ∂v
+ =0 (2.196)
∂x ∂y
∂2 u ∂2 u

∂u ∂u 1 ∂p ∂ ∂
u′ u′ − u′ v′

u +v =− +ν + 2 −
∂x ∂y ρ ∂x ∂x2
∂x ∂y ∂y
2 2
(2.197)
∂v ∂v 1 ∂p ∂ v ∂ v ∂ ∂
v′ v′ − v′ u′

u +v =− +ν 2
+ 2 −
∂x ∂y ρ ∂y ∂x ∂y ∂y ∂x
In order to proceed to solve these equations for a turbulent flow we must find some
way to relate the Reynolds stresses to the mean flow velocities. Such relations are known
as turbulence models and these will be addressed in sections to follow.
2.10.5 Computing turbulent flows
In this section we address methods that might be used to compute turbulent flows. Of
course, the most obvious approach would be to discretize or enmesh the entire region of
the flow and to solve numerical approximations to the Navier-Stokes equations. This is
known as the DNS or Direct Numerical Simulation method. However to be thorough
2.10 turbulence 97
it would rely on the mesh being substantially finer than the smallest eddies in order to
accurately resolve the flow. The problem with such an approach, as discussed at the end
of last section, is that even for quite small Reynolds numbers this requires a very refined
mesh and very large computer storage and run times. Moreover those demands increase
substantially as the Reynolds number of the desired fluid flow increases.
At the other end of the spectrum of methods is the RANS or Reynolds Averaged
Navier Stokes equation approach in which no effort is made to compute the unsteady
flow eddies. Instead the method focusses on models of the Reynolds stress terms in
the averaged Navier Stokes equations. These models attempt to relate the Reynolds
stress terms to the mean flow and spatial gradients of the mean flow. We will focus on
these RANS methods here since they are the commonest methods used in computing
turbulent flows. But it is also worth mentioning that a hybrid approach is also widely
used in which one attempts to model the large eddies but then uses a model to include
the averaged effect of the smaller eddies. One such version of this is called LES or Large
Eddy Simulation.
One of the commonest elements in RANS methods is the original model suggested by
Prandtl and called Prandtl’s mixing length model.
In order to solve the equations for turbulent flow it is necessary to find relations for
the Reynolds stresses in terms of the mean flow properties, specifically the mean flow
velocities. Many such models have been proposed over the years and we will only give a
brief review here. It is appropriate to begin with Prandtl’s original idea, ocusing on a
typical boundary layer near a wall as shown in Figure 2.29 and on a typical turbulent
eddy of size, l, embedded in that layer.
Figure 2.29: Notation for Prandtl’s mixing length model
Prandtl argued that fluid with typical x velocity, u, closer to the wall would be
transported to a position a distance l (known as Prandtl’s mixing length) more distant
from the wall where the local mean velocity is u + l (du/dy). If it carried with it the
original velocity u then a typical local velocity fluctuation at that further position would
be l (du/dy) and so a reasonable estimate for u′ (or v′ ) would be l (du/dy). Consequently
Prandtl argued that the Reynolds shear stress might be given by
2
′ ′ 2 ∂u
u v ≃l (2.198)
∂y
The second step is to estimate l and since the Reynolds shear stress must be zero at the
wall it seemed appropriate to set l = K y where K was some constant (ambitiously termed
the universal constant) for which the experimental data suggest the not unreasonable
value of about 0.4. This yields
2
′ ′ 2 2 ∂u
u v ≃K y (2.199)
∂y
2.10.6 Turbulent Couette flow
Prandtl’s mixing length model applied to turbulent Couette flow produces the following
result with notation as described in Figure 2.30.
Figure 2.30: Notation for turbulent Couette flow
We focus on the lower half of the flow since the mean velocity, u(y), must be symmetric
about y = h/2 with
U
u(h − y) = U − u(y) and u(h/2) = (2.200)
2
and satisfy the no-slip conditions u(h) = U and u(0) = 0. As in any Couette flow the
shear stress, σxy , must be uniform across the flow and therefore
∂ u′ v′

∂2 u du
=ν 2 or u′ v′ = ν +C (2.201)
∂y ∂y dy
where we have assumed there is no pressure gradient in the x direction, C is an

integration constant and we have replaced the partial derivatives with d/dy since all
quantities vary only with y and not with x. Then substituting the Prandtl mixing length
model (Eq. 2.199), so the equation that msut be solved to find u(y) becomes
2
2 2 du du
K y =ν +C (2.202)
dy dy
2.10.7 Law of the Wall
Another valuable tool in understanding and analyzing turbulent flows is obtained by

dimensional analysis of the flow near a solid boundary. It is known as the Law of the
Wall and is derived by assuming that the turbulence near that boundary is a function only
of the flow conditions pertaining at that wall and is independent of the flow conditions
further away. For the non-dimensional analysis we can identify the following limited set
of conditions and fluid properties (Figure 2.31):
2.10 turbulence 99
• The distance y [L] from the wall,

h i
• The mean velocity (or velocity profile) u(y) L T−1 ,
h i
• The shear stress τW M L−1 T−2 at the wall,
• The fluid density ρ M L−3 ,

h i
• The fluid kinematic viscosity, ν L2 T−1
Figure 2.31: The universal velocity profile
where the dimensions (mass [ M ], length [ L], and time [ T ]) of each quantity are
indicated in square brackets and the density and viscosity are assumed uniform and
constant. It is convenient to introduce a quantity called the friction velocity, uτ , defined
by
1/2
τW
uτ = (2.203)
ρ
and dimensional analysis then yields only two dimensionless quantities namely
1/2
y ∗ τW uτ y
Dimensionless length, y = =
ν ρ ν
−1/2 (2.204)
∗ τW u
Dimensionless velocity, u = u =
ρ uτ
Absent any other conditions, properties or influences it must follow that u∗ is a

function only of y∗ . In practice, of course, boundaries or conditions further away must
begin to have an effect on u∗ at large y∗ but leave that for later consideration.
It therefore seems valuable to try to identify the function u∗ (y∗ ), the so-called universal
velocity profile for turbulent flow near a wall. First and most obviously a laminar flow
near the wall will necessarily be a simple shear flow in which
∂u
τw = ρ ν (2.205)
∂y
which with the present notation translates simple to
u∗ = y∗ (2.206)
Moreover, in a turbulent flow since the turbulent fluctuations must go to zero at the
wall it follows that there will always be a very small layer next to the wall in which the
flow is essentially laminar and in which u∗ = y∗ . This is called the laminar sub-layer and
is present whatever the flow outside it may be. The only question is how deeply do
significant turbulent perturbations penetrate this laminar sublayer and thus determine
its thickness.
Further from the wall where the turbulent fluctuations dominate we require some
model of the turbulence to construct the function u∗ (y∗ ). As an instructive example, we
will use Prandtl’s mixing length model to write
2
τW ′ ′ 2 2 ∂u
= −u v ≃ K y (2.207)
ρ ∂y
and, assuming some constant value for K this integrates to

1 2.303
u∗ = ln y∗ + C = log10 y∗ + C (2.208)
K K
where C is an integration constant. It remains therefore to determine the integration

constant, C, and to match the velocity profile in the laminar sublayer, u∗ = y∗ to that in
the turbulent region further away from the wall; this has been done by comparing with
experimental data in order to generate a universal velocity profile of the turbulent flow
near a solid boundary. The following are the typical results for such a universal velocity
profile:
• The laminar sublayer in which u∗ = y∗ extends out to about y∗ = 5 which in terms

of the dimensional quantities means that the laminar sub-layer thickness, δLSL , is
given approximately by
5ν
δLSL = (2.209)
(τW/ρ) /2
1
• There is a transitional buffer zone that extends from y∗ = 5 to about y∗ = 30.

• The fully turbulent flow begins about y∗ = 30 and exhibits a velocity profile given
by
u∗ = C1 log10 (y∗ ) + C2 (2.210)
where the constant C1 lies between about 5.6 and 5.75 and the constant C2 lies
between about 4.9 and 5.5. It transpires that this turbulent profile extends out into
the flow much further than one might expect; it other words the modifications
caused by other boundaries or flow features are smaller than one would expect.
For example, in the analysis of turbulent pipe flow this profile is successfully used
all the way to the center of the pipe.
There are two important footnotes to this summary of the ’universal’ velocity profile.
First note that the above applies to a smooth wall though it will remain to be seen what
constitutes a smooth wall. That question and the universal velocity profile for a rough
wall are both detailed in later sections. Second, although the profile detailed above might
be an accurate fit to the experimental data, its functional form may not be the most
2.11 unsteady flows 101
convenient for use in some applications. Therefore the following simpler but cruder
form of the velocity profile,
u∗ = 8.7 (y∗ ) /7
1
(2.211)
is also widely used. Known as the one-seventh power law profile it was suggested by
Blasius.
2.11 unsteady flows
Unsteady fluid flows have already been addressed in many sections of this book. How-
ever, this brief review of basic fluid dynamics would not be complete without some
discussion of other unsteady flow phenomena and analyses. In the next section we
address the issue of the added mass properties that manifest themselves in unsteady
flows. When a body is accelerated in a fluid, in addition to the force connected with the
acceleration of the body, there is a force necessary to accelerate the mass of fluid that is
accelerated along with the body. That mass is called the added mass and the next section
is devoted to data and analyses of that added mass. Nevertheless, first we shall address
the solution for basic flows given the Navier-Stokes equations.
The following section addresses another important unsteady fluid flow phenomenon,
namely the process of vortex shedding in flows in a particular range of Reynolds numbers.
One of the important consequences is the resulting forces to which the object that is
shedding the vortices is subjected. Some massive structures have been destroyed by
those forces.
2.11.1 Exact solution to the Navier-Stokes Equations
2.11.1.1 Couette and planar Poiseuille flow
Couette and planar Poiseuille flow are both steady flows between two infinitely long,
parallel plates a fixed distance, h, apart as sketched in Figures 2.32a and 2.32b.
(a) Couette flow

(b) Planar Poiseuille flow
Figure 2.32: Sketch of flows
The difference is that in Couette flow one of the plates has a velocity U in its own
plane (the other plate is at rest) as a result of the application of a shear stress, τ, and
there is no pressure gradient in the fluid. In contrast in planar Poiseuille flow both plates
are at rest and the flow is caused by a pressure gradient, dp/dx, in the direction, x, parallel
to the plates. It is however, convenient, to begin the analysis of these flows together. We
will omit any conservative body forces like gravity since their effects are can be simply
added to the final solutions. Then, assuming that the only non-zero component of the
velocity is ux and that the velocity and pressure are independent of time the resulting
planar continuity equation for an incompressible fluid yields
∂u x
=0 (2.212)
∂x
so that u x (y) is a function only of y, the coordinate perpendicular to the plates. Using
this the planar Navier-Stokes equations for an incompressible fluid of constant and
uniform viscosity reduce to
∂p ∂2 u x
=µ
∂x ∂y2 (2.213)
=0
The second of these shows that the pressure, p( x ), is a function only of x and hence
the gradient, dp/dx, is well defined and a parameter of the problem. This allows the first
of these equations to be integrated so that the velocity, u x , can be written as
1 dp y2

ux = + C1 y + C2 (2.214)
µ dx 2
where C1 and C2 are integration constants to be determined by the application of the

boundary conditions at the two plates. Here the solutions for Couette flow and planar
Poiseuille flow diverge.
Addressing first Couette flow for which dp/dx = 0 and applying the no-slip conditions
at the upper and lower plates, namely
( u x )y=h = U and ( u x ) y =0 = 0 (2.215)
yields
U C2
C1 = − and C2 = 0 (2.216)
h h
And so the solution to Couette flow is

Uy
ux = (2.217)
h
and the shear stress at the walls is τ = µ U/h. Indeed a simple application of the
momentum theorem to a rectangular control volume within the device will show that
the shear stress, σxy , anywhere within the fluid is equal to µ U/h.
The solution for planar Poiseuille flow proceeds along similar lines except, of course,
that dp/dx is not zero. Applying the no-slip conditions at the lower and upper walls,
namely
( u x ) y =0 = 0 and ( u x ) y =0 = 0 (2.218)
yields

h dp
C2 = 0 and C1 = − (2.219)
2 dx
and so the solution to planar Poiseuille flow is

1 ∂p y
ux = − (h − y) (2.220)
µ ∂x 2
where the pressures, p1 and p2 , could be measured at two different x locations a

distance l apart in order to determine dp/dx = ( p1 − p2 )/l . The velocity distribution in the
fluid is parabolic with a maximum velocity on the centerline of
h2

dp
(u x )y=h/2 = − (2.221)
8µ dx
and the volume flow rate, Q̇, per unit depth normal to the plane of the flow is
ˆh
h3

dp
Q̇ = u x dy = − (2.222)
12 µ dx
0
so that the average velocity of the flow, u, is
h2

dp
u= − (2.223)
12 u dx
so the average is 2/3 of the maximum. The shear stresses, τ, at the walls are

h dp h dp
τy=0 = − and τy=h = − − (2.224)
2 dx 2 dx
and the shear stress within the fluid varies linearly between the plates according to

dp h
σxy = − −y (2.225)
dx 2
We note, parenthetically, that these expressions Eq. 2.224 and Eq. 2.225 for the shear
stresses can be derived using the momentum theorem applied to a simple rectangular
control volume within the fluid. The results are independent of the viscosity and, in fact,
independent of the constitutive properties of the fluid (or solid) contained between the
two plates. Finally we should note that the above results for planar Poiseuille flow only
have practical application up to Reynolds numbers of about 2000 for above that value
the flow will transition from laminar to turbulent.
2.11.1.2 Poiseuille flow
Poiseuille flow is the steady, axisymmetric flow in an infinitely long, circular pipe of
radius, R, as sketched in Figure 2.33.
Figure 2.33: Poiseuille flow

The flow is caused by a pressure gradient, dp/dx, in the axial direction, x. The resulting
axisymmetric continuity equation for an incompressible fluid yields
∂u x
=0 (2.226)
∂x
so that the axial velocity, u x (r ), is a function only of r, the radial coordinate. Using
this the axisymmetric Navier-Stokes equations for an incompressible fluid of constant
and uniform viscosity reduce to
2
∂p ∂ ux 1 ∂u x
=µ +
∂x ∂r2 r ∂r (2.227)
=0
The second of these shows that the pressure, p( x ), is a function only of x and hence
the gradient, dp/dx, is well defined and a parameter of the problem. This allows the first
of these equations to be integrated so that the velocity, u x , can be written as
r2

dp
u x (r ) = + C1 ln r + C2 (2.228)
4 µ dx
where C1 and C2 are integration constants to be determined by the application of the

boundary conditions. On the axis the velocity cannot be infinite therefore C1 must be
zero. Moreover, the no-slip boundary condition on the pipe wall requires that u x ( R) = 0
and so
R2 dp

C2 = − (2.229)
4 µ dx
and so the solution to Poiseuille flow in a circular pipe is

1 dp
R2 − r 2

u x (r ) = − (2.230)
4µ dx
2.11.1.3 Rayleigh and Ekman flows
Now, some valuable exact solutions to the Navier-Stokes equations involve unsteady
planar flows in a viscous, incompressible fluid (with constant and uniform viscosity)
bounded by a single flat plate that moves in its own plane as depicted in Figure 2.34.
Figure 2.34: Rayleigh and Ekman flows
Two classic solutions will be presented here. In the first the fluid is initially at rest
and the plate begins to move with a constant velocity, U, in its own plane at time, t = 0.
This is known as Rayleigh flow. In the second the plate is oscillating with velocity,
U = U ∗ sin fˆ t, in its own plane with radian frequency, fˆ. This is known as Ekman flow.
Indeed other solutions of this type are viable, for example, U = U ∗ exp α t. We begin by
delineating the equations that apply to this whole class of flows.
Since the flow at every x location is the same, it must be true that uy = uz = 0 and
∂p/∂x= 0 and therefore these uni-directional flows must satisfy the following Navier-
Stokes equation for u x (y, t)
∂u x ∂2 u x
=ν (2.231)
∂t ∂x2
which is a diffusion equation featuring a diffusivity of ν. Since the vorticity, ω, in

these flows is
∂uy ∂u x ∂u x
ω= − =− (2.232)
∂x ∂y ∂y
it follows that the vorticity satisfies a similar diffusion equation
∂u x ∂2 u x
=ν (2.233)
∂t ∂x2
Indeed this illustrates the typical process of diffusion that vorticity satisfies. We now
examine the particular solutions, starting with Ekman flow and proceeding to Rayleigh
flow.
Ekman flow is solved by deploying separation of variables to write the solution to
Equation 2.231 in the form
u x = F (y) G (t) (2.234)
where, after substitution into Equation 2.231, the initially unknown functions, F (y)
and G (t), are found to satisfy
1 dG
= − k2
G dt
(2.235)
ν d2 F
=−
F dy2
where k is an arbitrary constant. Solving these two implied ordinary differential

equations it follows that
u x = {C1 sin k t + C2 cos k t} {C3 exp [(k/ν1/2 ) y] + C4 exp [− (k/ν1/2 ) y]} (2.236)
where C1 , C2 , C3 , and C4 are arbitrary constants to be determined. We now apply

the boundary conditions. First since the velocity must tend to zero as y → ∞ it follows
that C3 = 0 and we can choose C4 = 1 without loss of generality. Second the no-slip
condition at the plate, namely

(u x )y=0 = U ∗ sin fˆ t (2.237)
requires that
C2 = 0 and C1 = U ∗ and k = fˆ (2.238)
so that the solution becomes
u x = U ∗ sin(k t) exp − fˆ/ν1/2 y

(2.239)
and the corresponding vorticity is
fˆ U ∗
sin(k t) exp − fˆ/ν1/2 y

ω=− 1/2
(2.240)
ν
A typical velocity profile for Ekman flow is shown in Figure 2.35.
Figure 2.35: Profiles of velocity and maximum velocity for Ekman flow
The velocity oscillates like a standing wave with an amplitude that declines expo-
nentially with distance from the plate. The vorticity profile is similar in kind. The
decrease in the amplitudes with y allows definition of a boundary layer thickness, δ, as
the distance from the plate at which the velocity magnitude (or the vorticity magnitude)
has decreased to 1 % of the value at the plate surface.
We turn now to Rayleigh flow in which the plate is suddenly set in motion with
velocity U at time t = 0. As in other, equivalent unsteady diffusion problems (for
example, in heat transfer) it proves to be appropriate to seek a similarity solution in
the case of a sudden change in the boundary condition. The appropriate similarity
variable is s where s = y/(4 ν t)1/2 (the 4 and the ν are not necessary but are included for
later convenience) and, with this similarity variable, the partial differential equation (Eq.
2.231) can be reduced to the ordinary differential equation
d2 u x du x
2
+2s =0 (2.241)
s ds
where the advantage of including the 4 and the ν in the definition of the similarity
variable becomes apparent because it follows that this governing equation (Eq. 2.241)
contains no parameters. Moreover the boundary conditions the solution must satisfy are
• u x → 0 as y → ∞ that is as s → ∞.
• u x = 0 for t = 0, y > 0, that is at s = ∞.
• u x = U for y = 0, t > 0, that is at s = 0.
The solution to Equation 2.241 is obtained by setting

du x
q= (2.242)
ds
so that Eq. 2.241 becomes

dq
= −2 s ds and so q = C1 exp −s2 (2.243)
q
which can be then integrated to yield

ˆs
exp −z2 dz + C2

u x = C1 (2.244)
0
where C1 and C2 are integration constants and z is a dummy s variable. Substituting

back from the definition of s this solution can be written as
y/(4 ν t)1/2
ˆ !
C1 (π ) /2
1
2
y
u x = C1 exp −z dz + C2 = erf + C2 (2.245)
(4 ν t) /2
1
2
0
where erf() is the error function. Then applying the three boundary conditions listed
above it transpires that C2 = U and C1 = −2 U/(π )1/2 so that the final solution is
" !#
y
u x = U 1 − erf (2.246)
(4 ν t) /2
1
and
y2

U
ω= exp − (2.247)
(4 ν t) /2
1
4νt
The form of this solution is shown graphically in Figure 2.36.
Figure 2.36: Profiles of velocity (left) and vorticity (right) for Rayleigh flow
The velocity profile expands outward as time progresses. The vorticity begins at t = 0
as an infinitely line line of infinite vorticity. That vorticity then diffuses out into the
fluid as the value at the surface of the plate decreases. Notice also how in the absence
of viscosity (ν = 0) the velocity must be zero throughout the fluid; in other words the
no-slip condition cannot be satisfied
These features of Rayleigh flow are particularly instructive. They reveal how the
no-slip condition produces vorticity which then diffuses out into the fluid through the
action of viscosity. This is a common feature in all flows involving solid surfaces.
2.11.2 Added mass
Whenever acceleration is imposed on a fluid flow either by acceleration of a body in the

fluid or by acceleration externally imposed on the fluid, additional fluid forces will act
on the surfaces in contact with the fluid. These fluid inertial forces can be of considerable
importance in many practical situations. In this and the sections which follow we will
review the state of knowledge of these forces and, in particular, identify the added mass
matrices that can be used to characterize them.
Perhaps the most fundamental view of the phenomenon of added mass is that it
defines the necessary work that is needed to change the kinetic energy associated with a
fluid motion. Any fluid motion such as that which occurs when a body moves through
the fluid implies a certain, positive, non-zero kinetic energy associated with the fluid
motion. We will confine attention to an incompressible fluid of density, ρ, in which case
the total kinetic energy, T, is given by
ˆ
ρ
u21 + u22 + u23 dv

T=
2 V
ˆ (2.248)
ρ
= u j u j dv
2 V
where u j , j = 1, 2, 3 are the Cartesian components of the fluid velocity and V is

entire volume or domain of fluid. If the motion of the body is one of steady rectilinear
translation at a velocity U through a fluid otherwise at rest then clearly the total kinetic
energy is finite and constant; it must in fact be equal to the work that had to be done
on the body to get it up to that velocity after starting form rest with all velocities equal
to zero. Moreover it is likely that the fluid velocities, u j , will, in some manner, be
proportional to U in which case T will be proportional to U2 . In such cases
ˆ
ρ U2 uj uj
T= I where I= dv (2.249)
2 V U U
and the integral I will be some simple, unvarying number. This is indeed the case
with some fluid flows such as potential flow or Stokes’ flow at low Reynolds numbers.
However, it may not be true for the complicated, vortex-shedding flows that occur at
intermediate Reynolds numbers.
Now consider that the body begins to accelerate or decelerate. The kinetic energy in
the fluid will also begin to change as U changes. If the body accelerates the kinetic energy
will increase in all probability. However, this energy must be supplied through work
done by the body on the fluid in order to increase T. Moreover the rate of additional work
required is simply the rate of change of T, dT/dt. This additional work is experienced by
the body as an additional drag, such that the rate of additional work done, − F U = dT/dt,
where the negative sign results from the choice that F is positive in the same direction as
U. Moreover, if the pattern of the flow is constant so that I remains unchanged
1 dT
F=−
U dt
dU
= −ρ I (2.250)
dt
dU
= −M
dt
This force has the same form and sign as that required to accelerate the solid mass, m,
of the body, namely m dU/dt. Consequently it is often convenient to consider the mass,
M = ρ I, as an added mass that is being accelerated along with the body. Of course, there
is no such identifiable fluid mass; rather all of the fluid is accelerating to some degree as
the total kinetic energy is increasing.
Note, parenthetically and obviously, that F is not the only drag force experienced
by the body. During steady translation through a real viscous fluid there is, of course,
a steady drag associated with the necessary work which must be done to balance the
steady rate of dissipation of energy in the viscous fluid. When the body accelerates there
will be a similar though not necessarily equal drag associated with the instantaneous
magnitude of U. Furthermore there may be delayed effects associated with the entire
previous history of translation as exemplified by the Basset force.
We will now consider a finite three-dimensional body moving and accelerating in
a fluid at rest far from that body. If that body experiences a general motion with
translational accelerations, A j , j = 1, 2, 3 in any or all three directions and rotational
accelerations, A j , j = 4, 5, 6 in the same directions then it will, in general, experience
fluid forces, Fi , i = 1, 2, 3, and moments Fi , i = 4, 5, 6, in all the same directions. We
seek to find the relations between those accelerations and forces. However, unless those
relations are linear the construction becomes extremely complicated and not readily
addressed analytically. Fortunately there are some fluid flows in which those relations
are linear, specifically in the case of potential flow or Stokes’s flow at asymptotically
small Reynolds numbers. We confine the following discussion to those circumstances.
The linear relations allow flows to be superposed and the effects of individual motions
and accelerations in one direction to be superimposed on or isolated from those in
another direction. We might however note that even in flows which are not superposable,
the methodology that follows might be a useful first approximation.
Given superposability and linear relations between the accelerations, A j , j = 1 → 6,
and the forces, Fi , i = 1 → 6, they induce, we can define an added mass matrix, Mij , as
Fi = − Mij A j (2.251)
When the flow is superposable it is convenient to define uij as the induced fluid
velocities caused by unit velocity of the body in the j direction (j = 1 → 6). Then, if the
body velocities are denoted by Uj , j = 1 → 6, it follows that the fluid velocities are
ui = uij Uj (2.252)
and the total kinetic energy can then be written as

1
T= A Uj Uk (2.253)
2 jk
where the matrix, A jk , is composed of elements

ˆ
A jk = ρ uij uik dV = M jk (2.254)
V
and it can be shown (Yih 1969, p.102) that the matrix A is, in fact, the added mass
matrix, M. It is certainly clear that the diagonal terms, A11 , A22 , and A33 , are identical
to the added masses in the introduction (to establish this define the direction x of that
introduction as either x1 , x2 , or x3 ; then uij and uik are identical to the velocity ui/U in the
introduction). Moreover, Equation 2.254 also demonstrates that the added mass matrix
must be symmetric when the flow is superposable since exchanging the indices j and k
in that equation does not change the value of the right hand side. Hence superposability
implies symmetry of the added mass matrix. Consequently, in general, the added mass
matrix will contain 21 different coefficients, 6 diagonal values and 15 off-diagonal values
since a force applied externally to the body will, in general, cause accelerations in all six
directions, translational and rotational.
To complete the formulation of the inertial terms in the equation of motion for a body
we would add to the left hand side of Equation the inertial matrix due to the mass and
moments of inertia of the material of the body itself. If the center of mass of the body is
chosen as the origin for the body mass matrix then that matrix will be symmetric and
will contain only 7 different, non-zero values (namely the mass and the six different
components of the moment of inertia matrix). This contrasts with the 21 independent
coefficients in the added mass matrix. This number can however be reduced when one
considers the simplifications caused by geometric symmetries.
The simplifications introduced by geometric symmetries of the body are fairly easily
established. Consider, for example, a body with a single plane of symmetry, for example
an airplane. It is clearly convenient to select axes such that this plane of symmetry
corresponds to the x3 = 0 plane. Then any acceleration confined to this plane, namely
any combination of A1 , A2 and A6 , will produce no added mass force F3 , F4 or F5 . The
only possible non-zero forces will be F1 , F2 and F6 . It follows that for such a body the
following 9 components of the added mass matrix will be zero
Mij = 0 for i = 3, 4, 5 ; j = 1, 2, 6 (2.255)
If, in addition, the flow is assumed to be potential flow such that the matrix is
symmetric then M ji = 0 for the same domains of i and j. The number of independent,
non-zero values required to define the matrix is 12, namely
Mii , i = 1 → 6 and M12 , M34 , M35 , M45 , M16 , M26 (2.256)
Bodies which have two planes of symmetry (for example, a hemisphere) yield a further
reduction in the number of non-zero values. Suppose axes are chosen such that both
x2 = 0 and x3 = 0 are planes of symmetry. Then not only must Equation 2.254 be true
but also
Mij = 0 for i = 2, 4, 6 ; j = 1, 3, 5 (2.257)
One other form of Equation 2.254 is useful in dealing with potential flows. If ϕj
denotes the velocity potential of the steady flow due to motion with unit velocity in the j
direction, then it follows that
∂ϕj
uij = (2.258)
∂xi
Then substitution into Equation 2.254 and application of Green’s theorem yields
ˆ
∂ϕ
M jk = A jk = −ρ ϕj k dS (2.259)
S ∂n
where S is the surface of the body and n is the outward normal to that surface. In
many potential flows it is clearly easier to evaluate the surface integral in equation 2.259
than the volume integral in equation 2.254.
Even in a viscous fluid, the added mass for a body accelerating from rest in a fluid
previously at rest is given by the potential flow value at that first moment when the
velocity is still zero. This is because, when the velocity is zero, the vorticity is zero and
therefore we can define what is known as the acceleration potential, ϕ′ , such that
∂u
= ∇ϕ′ (2.260)
∂t
Then conservation of mass for an incompressible fluid leads to Laplace’s equation and
potential flow for the acceleration. Hence we have potential flow even when the later
flow is dominated by viscous effects. It follows that in this first moment the forces and
accelerations in a viscous flow are related in the same way as in conventional potential
flow.
2.11.3 Fluid acceleration far from the body
In the preceding sections, the introduction and the section discussing added mass, we
have confined the presentation to those cases in which the fluid far from the body was
at rest. Clearly, unless this is the case the total kinetic energy integral, T, defined by
Equation 2.248 becomes infinite and the subsequent analysis becomes meaningless. We
turn attention now to those circumstances in which the fluid far from the body is either
moving with a constant, uniform velocity or accelerating.
Examine first the case where the fluid fluid far from the body is moving with a
constant velocity, and consider the case in which the fluid far from the body has some
uniform and constant velocity, Wi . It is clear that since the inertial forces cannot be
altered by a simple Galilean transformation that the appropriate definition of T under
those circumstance should be
ˆ
ρ
T= (ui − Wi ) (ui − Wi ) dv (2.261)
2 V
The value of this integral is then finite and the problem is resolved. In other words the
appropriate velocity, ui , to be used in equation 2.248 is the velocity of the fluid relative
to the fluid velocity far from the body. This leads to no change in the formulation of the
fluid inertial forces. Thus a more universal expression for those forces is

d Uj − Wj dUj
Fi = − Mij = − Mij (2.262)
dt dt
In other words, since the derivative of Wj is zero we recover the previous expression.
Case in which Wj is a function of time is, however, more complicated yet important
because it is a common occurrence and because some of the experiments carried out to
measure the inertial forces use an accelerated fluid flow rather than an accelerated body.
We begin with a case with a constant, uniform velocity, Wj , j = 1, 2, 3 far from the body
whose center of volume is moving at a velocity, Uj , j = 1 → 6 and which is accelerating
with components, A j . Equation 2.262 is appropriate and superposability is required.
Then we choose to apply an additional, globally uniform acceleration, dWj/dt, j = 1, 2, 3
to both the body and the fluid so that the new acceleration of the body is A j + dWj/dt. It
transpires as long as we begin with a superposable set of equations governing the fluid
flow (for example, the equations of potential flow or Stokes’ flow) and a set of boundary
conditions that are also superposable, then the equations governing this new flow with
the added global acceleration are identical to those without the global acceleration except
that where the pressure, p, occurs it is replaced by p − ρ xj dWj/dt. Consequently the forces
that the fluid exerts on the body are identical except for an additional contribution in the
flow with the additional global acceleration due to the additional pressure −ρ xj dWj/dt.
When this is integrated over the surface of the body the additional, buoyancy-like force
on the body becomes ρ Vb dWj/dt where Vb is the volume of fluid displaced by the body.
Therefore the inertial force becomes
dWi
Fi = − Mij A j + ρ Vb (2.263)
dt
where the acceleration of the body, dUj/dt, is now

dUj dWj
= Aj + (2.264)
dt dt
and dWj/dt is the acceleration of the fluid far from the body. Substituting for A j in
Equation 2.263 produces the final result for the flow with fluid acceleration far from the
body namely
dUj dWi
Fi = − Mij + Mij + ρ Vb δij for j = 1, 2, 3 (2.265)
dt dt
where δij is the Kroneker delta. Therefore the added mass associated with the fluid
acceleration, dWj/dt, is the sum of the true added mass, Mij , and the mass of the displaced
fluid, ρ Vb . This distinction is essential in interpreting and comparing the results of
experiments designed to measure inertial effects. For example, in the experiments of
Keulegan and Carpenter (1958) the body is held at rest while the fluid far from the body
undergoes sinusoidal acceleration so that the fluid inertial effects are proportional to
Mij + ρ Vb . In contrast, in the experiments of Skop, Ramberg and Ferer (1976) the body
is accelerated while the fluid far away is not and therefore their fluid inertial effects are
proportional to Mij . The term added mass should be reserved for Mij in order to avoid
confusion.
2.12 internal flows
We shall analyze the effects of viscosity on an incompressible, internal flow. The viscous
effects in a flow result in the introduction of the Reynolds number as the ratio of inertial
force to the viscous force
Vρl
RE = (2.266)
µ
When this ratio becomes large, it is expected that the inertial forces may dominate
the viscous forces. This is usually true when short, sudden geometric changes occur;
for long reaches of pipe or open channels, this is not the situation. When the surface
areas, such as the wall area of a pipe, are relatively large, viscous effects become quite
important and must be included in our study.
At sufficiently high Reynolds numbers, a turbulent flow occurs. We will consider
laminar flows first and then turbulent flows.
2.13 entrance flow and developed flow
When considering internal flows, we are interested primarily in developed flows in

conduits. A developed laminar flow results when the velocity profile ceases to change in
the flow direction. Let us first focus our attention on laminar flows.
2.14 laminar flow in a pipe 113
Figure 2.37: Entrance region of a laminar flow in a pipe or a wide rectangular channel
In the entrance region of a laminar flow, the velocity profile changes in the flow direction,
as shown in Fig. 2.37. The idealized flow from a reservoir begins at the inlet as a uniform In reality, there is a
flow; the viscous wall layer then then grows over the inviscid core length Li until the thin viscous layer on
the wall
viscous stresses dominate the entire cross section. The profile then continues to change
in the profile development region due to viscous effects until a developed flow is achieved.
For a turbulent flow the situation is slightly different, as shown in Fig. 2.38. A
developed flow results when all characteristics of the flow cease to change in the flow
direction. The inviscid core exists followed by the velocity profile development region,
which terminates at x = Ld . And additional length is needed however, for the detailed
structure of the turbulent flow to develop.
Figure 2.38: Velocity profile development in a turbulent pipe flow
2.14 laminar flow in a pipe
We shall investigate incompressible, steady, developed laminar flow in a pipe as sketched

in Fig. 2.39.
Figure 2.39: Developed flow in a circular pipe

For developed flows in a circular pipe, the streamlines are parallel to the wall with
Note that u = vz no swirl, so that if U = u x + v y + w z then vr = vθ = 0 and u = u(r ) only. Given the
and the z-coordinate momentum equation in cylindrical coordinates, the x-component Navier-Stokes equation
has been replaced by
is
x
2
1 ∂2 u ∂2 u

∂u ∂u vθ ∂u ∂u ∂p ∂ u 1 ∂u
ρ + vr + +u = − + γ sin θ + µ + + 2 2+ 2
∂t ∂r r ∂θ ∂x ∂x ∂r2 r ∂r r ∂θ ∂x
(2.267)
Fear not, there are many terms that we shall not consider:
• Steady flow, so ∂u/∂t = 0.

• The streamlines are parallel to the wall, so vr = 0.
• There is no swirl, so vθ/r = 0.
• Developed flow, so ∂u/∂x = 0.
• Symmetric flow, so ∂2 u/∂θ 2 = 0.
• Developed flow, so ∂2 u/∂x2 = 0.
Note that there is no acceleration (the LHS is zero) fo the fluid particles as they move
in the pipe. So, Eq. 2.267 simplifies to
2
∂p ∂ u 1 ∂u
0=− + γ sin θ + µ +
∂x ∂r2 r ∂r
2

1 ∂p ∂ u 1 ∂u
+ γ sin θ = 2 + (2.268)
µ ∂x ∂r r ∂r

1 ∂ 1 ∂ ∂u
( p + γ h) = r
µ ∂x r ∂r ∂r
where notice that we have used sin θ = −dh/dx. The first two terms in the parenthesis
of the RHS have been combined. The LHS is at most a function of x, and the RHS is at
most a function of r. Since x and r can be varied independently, we must have

1 d du
r =λ (2.269)
r dr dr
If λ = f ( x ), then u where λ is a constant and we have used ordinary derivatives since u depends only on
would depend on x, the variable r. We can multiply both sides by r and integrate to get
which is not
acceptable for this du λ
developed flow. The r = r2 + A (2.270)
dr 2
same for λ = f (r )
Then, we divide both sides by r and integrate again
λ 2
u (r ) = r + A ln r + B (2.271)
4
The velocity must remain finite at r = 0; hence A = 0. Also, at r = r0 , we have u = 0;

thus B can be evaluated and we have
λ 2
u (r ) = (r − r02 ) (2.272)
4
2.15 laminar flow between parallel plates 115
Replacing the value for the constant λ

1 d
u (r ) = ( p + γ h) (r2 − r02 ) (2.273)
4 µ dx
This is the parabolic velocity ditribution for flow in a pipe, often referred to as Poiseuille
flow.
2.14.1 Pipe flow quantities
The average velocity can be found by solving

ˆ r0 ˆ
Q 1 2 r0 1 d ( p + γ h ) 2
V= = u ( r ) 2π r dr = (r − r02 ) r dr (2.274)
A π r02 0 r02 0 4 µ dx
which results in an average velocity V of
r02 d( p + γ h)
V=− (2.275)
8µ dx
Expressing the pressure drop ∆p in terms of the average velocity, we have

8µV L
∆p = (2.276)
r02
where we have used ∆p/L = −dp/dx since dp/dx is a constant for developed flow. NOte
that the pressure drop is a positive quantity, wheres the pressure gradient is negative.
The maximum velocity is found at r = 0, so
r02 d( p + γ h)
umax = − (2.277)
4µ dx
so that
1
V= umax (2.278)
2
The shearing stress is determined to be
du r d( p + γ h)
τ = −µ =− (2.279)
dr 2 dx
Letting τ = τ0 at r = r0 , we can determine the pressure drop ∆p over a length L of a

horizontal section of pipe as
2 τ0 L
∆p = (2.280)
r0
2.15 laminar flow between parallel plates
Consider the incompressible, steady, developed flow of a fluid between parallel plates,
with the upper plate moving with velocity U, as shown in Fig. 2.40. We shall derive the
velocity distribution.
Figure 2.40: Developed flow between parallel plates
For developed flows between parallel plates, the streamlines are parallel to the plates,
so that u = u(y) only and v = w = 0. The Navier-Stokes equation for the x-direction is
2
∂ u ∂2 u ∂2 u

∂u ∂u ∂u ∂u ∂p
ρ +u +v +w = − +µ + 2 + 2 + γ sin θ (2.281)
∂t ∂x ∂y ∂z ∂x ∂x2 ∂y ∂z
Just like the last case, there are several terms that we shall not consider
• Steady flow, so ∂u/∂t = 0.

• Developed flow, so ∂u/∂x = 0.
• Parallel plates, so v = w = 0.
• Developed flow, so ∂2 u/∂x2 = 0.
• Wide channel, so ∂2 u/∂z2 = 0.
If the parallel plates are the top and bottom of a channel, the channel must be wide; the
analysis then applies to the midsection away from the side walls. Using sin θ = −dh/dx,
Eq. 2.281 reduces to
∂2 u 1 d
2
= ( p + γ h) (2.282)
∂y µ dx
The LHS is at most a function of y, and the RHS is at most a function of x. Hence, we
must have
∂2 u
=λ (2.283)
∂y2
where again λ is a constant, since x and y are independent variables. This can be
integrated twice to yield
λ 2
u(y) = y +Ay+B (2.284)
2
such that A and B are arbitraty constants of integration. We demand that u = 0 at

y = 0 and u = U at y = a. This gives
U λa
A= − and B=0 (2.285)
a 2
Using the constants, we can find the velocity profile as

λ 2 U
u(y) = (y − a y) + y (2.286)
2 a
2.16 turbulent flow in a pipe 117
And replacing the value of the constant λ we get

1 d U
u(y) = ( p + γ h ) ( y2 − a y ) + y (2.287)
2 µ dx a
If the flow is due to the motion of the plate only (a linear profile), the flow is called
a Couette flow. If the motion is due to the pressure gradient only; i.e., U = 0, it is a
Poiseuille flow.
2.15.1 Simplified flow situation
Let U = 0, so we can write the expression for the velocity distribution between the fixed
plates as
1 d
u(y) = ( p + γ h ) ( y2 − a y ) (2.288)
2 µ dx
Using this distribution, we can find the flow rate per unit width to be
ˆ ˆ a
1 d( p + γ h) 2 a3 d ( p + γ h )
Q = u dA = (y − a y)dy = − (2.289)
0 2µ dx 12 µ dx
So the average velocity V is found to be
Q a2 d ( p + γ h )
V= =− (2.290)
a×1 12 µ dx
This can be expressed as the pressure drop in terms of the average velocity; for a
horizontal channel we have
12 µ V L
∆p = (2.291)
a2
where we have used ∆p/L = −dp/dx, since dp/dx is constant for developed flows.
2.16 turbulent flow in a pipe
Unlike in laminar flows, in a turbulent flow all three velocity components are non-zero.
If we measure the compoenents as a function of time, graphs similar to those shown in
Fig. 2.41 would result for a flow in a pipe where u, v and w are in the x, r and θ direction
respectively. There is seldom any interest in the details of the randomly fluctuating
velocity components; hence we introduce the notion of a time-average quantity.
So, the velocity components may be written as
u = ū + u′ v = v̄ + v′ w = w̄ + w′ (2.292)
where the bar over the quantity denotes a time average and a prime denotes the
fluctuating part. Using the component u as an example, the time average is defined as
ˆ
1 T
ū = u(t) dt (2.293)
T 0
Figure 2.41: Velocity components in a turbulent pipe flow (a) x-component velocity, (b) r-
component velocity and (c) θ-component velocity
where T is a time increment large enough to eliminate all time dependencies from ū.
In a developed flow, ū would be non-zero and v̄ = w̄ = 0, as is observed in Fig. 2.41.
2.16.1 Differential equation
The differential equation that must be solved to provide us with the time-average velocity
distribution is derived by using a particle approach. Consider the situation for a turbulent
flow in a horizontal pipe, as shown in Fig. 2.42.
Figure 2.42: Turbulent flow in a horizontal pipe
We use the velocity components u and v in the x and y directions respectively. Fluid
particles move randomly throught the flow. At an instant in time, a fluid particle moves
through the incremental area dA due to the velocity fluctuation v′ ; it enters a neighboring
layer of fluid which is moving with a higher x-components velocity, thereby providing a
retarding effect on the neighboring layer. A fluid particle that moves to a neighboring
layer that is traveling with a lower x-component velocity would tend to accelerate the
slower moving fluid. The x-component force that results due to the random motion of a
fluid particle passing through the incremental are dA would be
dF = −ρ v′ dA u′ (2.294)
where u′ is the negative change in x-component velocity due to the momentum

exchange and ρ v′ dA is the mass flux through the area; the negative sign provides a
positive dF. If we divide both sides by the area dA, we obtain a ’stress’ that we call the
turbulent shear stress
dF
τtur = = −ρ u′ v′ (2.295)
dA
where we know that (u′ v′ ) is, on the average, a negative quantity since a positive v′
produces a negative u′ . This ’shear stress’ is acually a momentum exchange, but since it
has the same effect as a stress, we call it shear stress.
2.16 turbulent flow in a pipe 119
The time-average turbulent shear stress, often called apparent shear stress, is
¯ = −ρ u′ v′
τtur (2.296)
Note that u′ w′ would be zero, since a ω ′ -component (in the θ-direction) would not
move a fluid particle into a layer of higher or lower x-component velocity. Also, v′ w′ = 0
using the same logic.
The total shear stress at a particular location would be due to both the viscosity and
the momentum exchange described above; that is
∂u
τ = τ lam + τ tur = µ − ρ u′ v′ (2.297)
∂y
The shear stress can be related to the pressure gradient by summing forces on the
horizontal cylindrical element shown in Fig. 2.41. There results
r dp r ∆p
τ=− = (2.298)
2 dx 2L
which shows that the shear stress distribution is linar for a turbulent flow as well as a
laminar flow. The turbulent shear obviously goes to zero at the wall since the velocity
perturbations are zero at the wall, and the total shear is zero at the centerline where
r = 0 or y = r0 . The viscous shear is nonzero only in a very thin viscous wall layer, of
thickness δv′ near the wall. Note that the turbulent shear reaches a maximum near the
wall in the viscous wall layer.
The differential equation that must be solved if the time-average velocity distribution
is to be determined is found by combining the two preceding equations; i.e.
r dp du
= ρ u′ v′ + µ (2.299)
2 dx dr
where we have used r + y = r0 so that dy = −dr. For developed flows, we know that
dp/dx = const.; hence if we know how u′ v′ varied with r, the differential equation could
be solved. The quantity u′ v′ cannot be determined analytically, however, the solution to
Eq. 2.299 cannot be attempted until an empirical expression is found for u′ v′ . Rather
than finding an empirical expression for u′ v′ and then soling the differential equation
for u, we will simply present the empirical result obtained for the velocity profile u(y).
However, before we do this we shall introduce the eddy viscoity, the mising length and
the correlation coefficient.
Instead of using the quantity u′ v′ as the unknown, we often introduce the eddy viscocity
η, defined by the relationship
du
u′ v′ = η (2.300)
dy
Note that it has the same dimensions as the kinematic viscosity. In terms of the eddy
viscosity, the differential equation becomes
r dp du
= ρ(v + η ) (2.301)
2 dx dr
If we view the turbulent process as the random and chaotic mixing of particles of fluid,
we may choose to introduce the mixing length lm , the distance a particle travels before
interacting with another particle. Based on reasoning related to momentum interchange,

we relate the eddy viscosity to the mixing length with
2 du
η = lm (2.302)
dy
The correlation coefficient Kuc , a normalized turbulent shear stress, often used when
describing turbulent motions, has limits of ±1 and is defined as
u′ v′
Kuv = p p (2.303)
u ′2 v ′2
The quantities η, lm and Kuv are functions of r (or y) that simply replace the variable
u′ v′ . They do not simplify the differential equation 2.299; they allow the equation to
take slightly different forms. Since we cannot derive an expression for η, lm or Kuv , we
cannot find u(r ) using analytical methods.
S O LY D M E C H A N I C S F U N D A M E N TA L S
3
3.1 Coupled two degrees of freedom spring-mass system analysis 121
3.2 Free vibration 122
3.3 Forced vibration 129
3.1 coupled two degrees of freedom spring-mass system analysis
Consider the coupled, undamped spring-mass system shown in Fig. 3.1. The displace-
ment of the top and bottom masses at any time, m1 = m and m2 = m respectively, can
be specified by the vertical motions x1 (t) and x2 (t).
Figure 3.1: Two-degree-of-freedom undamped spring mass system
Each of the masses mi : i ∈ [1, 2] in the system have one possible type of motion, in
the vertical direction. Thus, it is said that the system has two degrees of freedom
Degrees of freedom = Number of masses × Number of possible types of motion of each mass
(3.1)
Therefore, we will encounter equations of motion for each of masses, in the form
of coupled differential equations - that is, each equation involves all the coordinates. If a
harmonic solution is assumed for each coordinate, the equations of motion will lead to
a frequency equation that will give two natural frequencies ωi for the system. If we
would give some suitable initial excitation, the system would vibrate at one of these
natural frequencies, or in a superposition of both.
121
122 solyd mechanics fundamentals
3.2 free vibration
During free vibration at one of the natural frequencies, the amplitudes of the two degrees
of freedom are related in a specific manner, and the configuration is called a mode. Thus,
the two degrees freedom system is said to have two modes. However, given an arbitraty
excitation to the system, the resulting vibration will still be a superposition of the two
modes.
So, we can completely describe the motion the motion of the system in Fig. 3.1 by
the coordinates xi (t), which define the position of the masses mi at any time t from the
equilibrium positions.
We will consider k1 = k3 , and k2 ̸= k1 in the system shown in Fig. 3.1. So given the
free body diagrams shown in Fig. 3.2
Figure 3.2: Free body diagrams of the system
we can apply Newton’s second law of motion to each of the masses, which gives the
equations of motion
m1 x1¨(t) − k2 ( x2 (t) − x1 (t)) + k1 x1 (t) = 0

(3.2)
m2 x2¨(t) + k1 x2 (t) + k2 ( x2 − x1 ) = 0
m1 x1¨(t) + x1 (t) (k1 + k2 ) − k2 x2 (t) = 0

(3.3)
m2 x2¨(t) − k2 x1 + x2 (k1 + k2 ) = 0
If we define k = (k1 + k2 ), and consider that m1 = m2 = m and xi (t) = xi , then
m x¨1 + k x1 − k2 x2 = 0
(3.4)
m x¨2 − k2 x1 + k x2 = 0
Eqs. 3.4 are the equations of motion that represent the dynamics of the system. We
are interested in knowing wether m1 and m2 can oscillate harmonically with the same
frequency and phase angle, but with different amplitudes. If we assume it is indeed
possible to have harmonic motion of the masses at the same frequency ω and phase
angle ϕ, then we can assume solutions of the system as
x1 (t) = X1 cos(ω t + ϕ)
(3.5)
x2 (t) = X2 cos(ω t + ϕ)
The harmonic solutions in Eq. 3.5 verify that
d2 ( X1 cos(ω t + ϕ))
m + k X1 cos(ω t + ϕ) − k2 X2 cos(ω t + ϕ) = 0
dt2 (3.6)
d2 ( X2 cos(ω t + ϕ))
m − k2 X1 cos(ω t + ϕ) + k X2 cos(ω t + ϕ) = 0
dt2
3.2 free vibration 123
−m ω 2 X1 cos(ω t + ϕ) + k X1 cos(ω t + ϕ) − k2 X2 cos(ω t + ϕ) = 0

(3.7)
−m ω 2 X2 cos(ω t + ϕ) − k2 X1 cos(ω t + ϕ) + k X2 cos(ω t + ϕ) = 0
cos (ω t + ϕ) X1 k − m ω 2 − k2 X2 = 0

(3.8)
cos (ω t + ϕ) X2 k − m ω 2 − k2 X1 = 0

Since the solution must be satisfied for all values of the time t, then we take
X1 k − m ω 2 + X2 (−k2 ) = 0

(3.9)
X1 (−k2 ) + X2 k − m ω 2 = 0

It is clear that Eq. 3.9 is satisfied by the trivial solution X1 = X2 = 0, which would
imply that there is no vibration. For a non-trivial solution we have
k − m ω2

(−k2 )
det =0 (3.10)
(−k2 ) k − m ω2
k − m ω2 k − m ω 2 − k22 = 0

(3.11)
k2 − 2 k m ω 2 + m2 ω 4 − k22 = 0
Finally, we end up with
m2 ω 4 − 2 k m ω 2 + k2 − k22 = 0

(3.12)
Eq. 3.12 is called the frequency equation. The solutions of the frequency equation
will yield the natural frequncies of the system.
So let u = ω 2 ∴ u2 = ω 4 ; then
m2 u2 − 2 k m u + (k2 − k22 ) = 0
q q
2 k m ± 4 k2 m2 − 4 m2 k2 + 4 m2 k22 2 k m ± 2 m k2 − k2 + k22
ui = = (3.13)
2 m2 2 m2
k ± k2
ui =
m
From the last equation, the natural frequencies of the system ωi can be defined as
r
k ± k2
ωi = (3.14)
m
√ √
so that ω1 = (k1 +2 k2 )/m and ω2 = k1/m. But, the values of Xi remain to be
determined, which will depend upon the natural frequencies.
If we input the first natura frequency ω1 into the condition for the harmonic solution
of Eq. 3.9, we get
X1 k − m ω12 + X2 (−k2 ) = 0

(3.15)
X1 (−k2 ) + X2 k − m ω12 = 0

(1) k2 (1)
X1 = X (3.16)
k − m ω12 2
(1)
X2 k − m ω12
r1 = (1)
= (3.17)
X1 k2
(1) (1)
X2 = r 1 X1 (3.18)
(1)
where the notation Xi denotes the values of X1 and X2 corresponding to ω1 ; and we
define the amplitude ratio r1 given that the system is homogeneous.
Applying the same logic for the second natural frequency ω2
X1 k − m ω22 + X2 (−k2 ) = 0

(3.19)
X1 (−k2 ) + X2 k − m ω22 = 0

(2)
X2 k − m ω22
r2 = (2)
= (3.20)
X1 k2
(2) (2)
X2 = r 2 X1 (3.21)
Computing the actual values of the amplitude ratios given the natural frequencies
from Eq. 3.14
k − m ((k+k2 )/m) k − k + k2
r1 = = =1
k2 k2
(3.22)
k − m ((k−k2 )/m) k − k − k2
r2 = = = −1
k2 k2
So the amplitudes are related by

(1) (1)
X2 = X1
(3.23)
(2) (2)
X2 = − X1
With this in mind, we can define the modal vectors of the system as
( ) ( ) ( )
(1) (1) (1)
X X X
X (1) = 1
(1) = 1
(1) = 1
(1)
X2 r 1 X1 X1
( ) ( ) ( ) (3.24)
(2) (2) (2)
(2) X1 X1 X1
X = (2) = (2) = (2)
X2 r 2 X1 − X1
So we can express the solutions of the system in terms of the two possible modes
( ) ( )
(1) (1)
(1) X1 X1 cos(ω1 t + ϕ1 )
x (t) = (1) = (1) , first mode
X2 r1 X1 cos(ω2 t + ϕ1 )
( ) ( ) (3.25)
(2) (2)
X X cos ( ω 1 t + ϕ2 )
x (2) ( t ) = 1
(2) = 1
(2) , second mode
X2 r2 X1 cos(ω2 t + ϕ2 )
(
(1) √ )
(1) X1 cos( (k+k2 )/m t + ϕ1 )
x (t) = (1) √ , first mode
X1 cos( (k−k2 )/m t + ϕ1 )
( √ ) (3.26)
(2)
X cos ( (k+k2 )/m t + ϕ2 )
x (2) ( t ) = 1
(2) √ , second mode
− X1 cos( (k−k2 )/m t + ϕ2 )
(1) (2)
where the constants X1 , X1 , ϕ1 and ϕ2 shall be determined by the initial conditions.
It can be seen that when the system vibrates in its firts mode, the amplitudes of the
two masses remain the same. This implies that the length of the middle spring remains
constant. Thus, the motions of the two masses are in phase, as it is shown in Fig. 3.3a.
When the system vibrates in its second mode, the displacements of the two masses
have the same magnitude with opposite signs: the motions of the masses are 180◦ out of
phase. In this case, the mid-point of the middle spring remains stationary, as shown in
Fig. 3.3b; such point is called a node.
Figure 3.3: Modes of vibration
3.2.1 Initial conditions to excite an specific mode
We shall find the initial conditions that need to be applied to the system in order to make
it vibrate in one of its two modes.
In order to do this, we can express the general solution of the system as a superposition
of the two modes
(1)
x 1 ( t ) = X1 cos(ω1 t + ϕ1 ) + X12 cos(ω2 + ϕ2 )
(3.27)
(1)
x 2 ( t ) = X1 cos(ω1 t + ϕ1 ) − X12 cos(ω2 + ϕ2 )
(1) (2)
The unknown constants X1 , X1 , ϕ1 and ϕ2 can be determined from the initial
conditions; which we will denote as
x1 (t = 0) = x10
x2 (t = 0) = x20
(3.28)
x˙1 (t = 0) = x˙10
x˙2 (t = 0) = x˙20
This leads to
(1) (2)
x10 = X1 cos(ϕ1 ) + X1 cos(ϕ2 )
(1) (2)
x20 = X1 cos(ϕ1 ) − X1 cos(ϕ2 )
(3.29)
(1) (2)
x˙10 = −ω1 X1 sin(ϕ1 ) − ω2 X1 sin(ϕ2 )
(1) (2)
x˙20 = −ω1 X1 sin(ϕ1 ) + ω2 X1 sin(ϕ2 )
The amplitude of the firts mass for the natural frequency ω1 can be expressed as
(2)
(1) x10 − X1 cos ϕ2
X1 = (3.30)
cos ϕ1
which can be used to express the ampitude for ω2

(2)
!
x10 − X1 cos ϕ2 (2)
x20 = cos ϕ1 − X1 cos ϕ2
cos ϕ1
(2) (3.31)
x20 = x10 − 2 X1 cos ϕ2
(2) x10 − x20
X1 =
2 cos ϕ2
In turn, this results for ω1

− x20
x10 − x210cos cos ϕ2
(1) ϕ2
X1 =
cos ϕ1
x − 1/2 x10 − 1/2 x20 (3.32)
= 10
cos ϕ1
x10 − x20
=
cos ϕ1
Now we shall find the expressions for the phase angles ϕi . We will use the velocity
(1) (2)
initial conditions for this, given the results we found for the amplitudes X1 and X1
x10 − x20 x10 − x20

x˙10 = −ω1 sin ϕ1 − ω2 sin ϕ2
2 cos ϕ1 2 cos ϕ2
= −ω1 1/2 ( x10 − x20 ) tan ϕ1 − ω2 1/2 ( x10 − x20 ) tan ϕ2 (3.33)
− x˙10 − 1/2 ( x10 − x20 ) tan ϕ2 − x˙10 + 1/2 ω2 ∆x0 tan ϕ2
tan ϕ1 = =
1/2 ω1 ( x10 − x20 ) −1/2 ω1 ∆x0
where we defined ∆x0 = x20 − x10 and ∆ x˙0 = x˙20 − x˙10 . With this expression for tan ϕ1 ,
we can find the phase angle ϕ2 as a function of the initial conditions
x˙20 = −1/2 ω1 ( x10 − x20 ) tan ϕ1 + 1/2 ω2 ( x10 − x20 ) tan ϕ2

= x˙10 + 1/2 ω2 ( x10 − x20 ) tan ϕ2 + 1/2 ω2 ( x10 − x20 ) tan ϕ2
= x˙10 + ω2 ( x10 − x20 ) tan ϕ2 (3.34)
x˙20 − x˙10 ∆ x˙0
tan ϕ2 = =−
ω2 ( x10 − x20 ) ω2 ∆x0
Finally
∆ x˙0

−1
ϕ2 = tan (3.35)
ω2 ∆x0
We can use this expression to find the phase angle ϕ1

x˙10 ω2 ∆ x˙0
tan ϕ1 = 1/2 ω1 ∆x0
−
ω1 ω2 ∆x0
2 x˙10 − ∆ x˙0
= (3.36)
ω1 ∆x0
2 x˙10 − ∆ x˙0

ϕ1 = tan−1
ω1 ∆x0
Given the final expressions for the phase angles ϕi , we can redefine the amplitudes.
Starting with the amplitude for the frequency ω1
(1) −∆x0 −∆x0

X1 = = (3.37)
2 cos ϕ1 ˙ −∆ x˙0
2 cos tan−1 2 xω10
1 ∆x0
We can use the trigonometric identity

p
1 + φ2
cos(tan−1 φ) = (3.38)
1 + φ2
(1)
So the amplitude X1 results
2
˙ −∆ x˙0
2 x10
−∆x0 1 + ω2 ∆x2
(1) 1 0
X1 = v
2
u
2 x ˙ −∆ x˙
1+ 102 2 0
u
t ω ∆x0
2 1+ 1
ω12 ∆x02

−∆x0 ω1 ∆x0 ω12 ∆x02 + (2 ẋ10 − ∆ x˙0 )2
= q
2 ω12 ∆x02 ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2
h i2
2 − ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2 (3.39)
(1)
X1 = q 2
2
2 ω1 ω1 ∆x0 + (2 ẋ10 − ∆ ẋ0 )
2 2
h i2
ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2
=
4 ω12 ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2
ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2

=
4 ω12
So finally, we can express the amplitude for the frequency ω1 as a function of the
initial conditions
1
q
(1)
X1 = ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2 (3.40)
2 ω1
Now, we have to do the same for the the amplitude corresponding to the frequency
ω2
(2) x10 − x20 −∆x0
X1 = =
2 cos ϕ2 2 cos ω∆2 ∆x ẋ0
0
∆ ẋ02

−∆x0 1 + ω2 ∆x2
1 0
= r
∆ ẋ0
2
2 1 + ω2 ∆x 2
1 0
−∆x0 ω22 ∆x02 + ∆ ẋ02

= r
ω22 ∆x02 +∆ ẋ02
2 ω2 ∆x0
2 2
ω22 ∆x02
− ω22 ∆x02 + ∆ ẋ02

(3.41)
=
2 ω22 ∆x0
q
ω2 ∆x0 ω22 ∆x02 + ∆ ẋ02
2
− ω22 ∆x02 + ∆ ẋ02

(2) 2

X1 = h q i2
2 ω2 ω22 ∆x02 + ∆ ẋ02
2
ω22 ∆x02 + ∆ ẋ02
=
4 ω22 ω22 ∆x02 + ∆ ẋ02

ω22 ∆x02 + ∆ ẋ02

=
4 ω22
(1) (2)
We finally have expressions for the four constants X1 , X1 , ϕ1 and ϕ2 as functions of
the initial conditions xi0 and ẋio
1
q
(1)
X1 = ω12 ∆x02 + (2 ẋ10 − ∆ ẋ0 )2
2 ω1
1
q
(2)
X1 = ω22 ∆x02 + ∆ ẋ02
2 ω2
(3.42)
−1 2 x˙10 − ∆ x˙0

ϕ1 = tan
ω1 ∆x0
∆

−1 x˙0
ϕ2 = tan
ω2 ∆x0
For the first mode we had identical motion of the masses

( )
(1)
X cos ( ω 1 t + ϕ1 )
x (1) ( t ) = 1
(1) (3.43)
X1 cos (ω1 t + ϕ1 )
So if we look at the general solution

(1)
x 1 ( t ) = X1 cos(ω1 t + ϕ1 ) + X12 cos(ω2 + ϕ2 ) (3.44)
(2)
we can see that the identical motion will only be given if and only if X1 = 0; which
will be only satisfied if both ∆x0 = ∆ ẋ0 = 0. This implies that
x10 = x20 , for the first mode

(3.45)
ẋ10 = ẋ20
3.3 forced vibration 129
Now, for the second mode we had

( )
(2)
X cos ( ω 2 t + ϕ2 )
x (2) ( t ) = 1
(2) (3.46)
X1 cos (ω2 t + ϕ2 )
So once again looking at the general solution

(1)
x 2 ( t ) = X1 cos(ω1 t + ϕ1 ) − X12 cos(ω2 + ϕ2 ) (3.47)
(1)
we can conclude that X1 = 0 for morion in the second mode. This implies
∆x0 = 0 → x10 = x20

(3.48)
∆ ẋ0 = 0 → 2 ẋ10 − ∆ ẋ0 = 0 → 3ẋ10 = ẋ20
3.3 forced vibration
If the system vibrates under the action of an external harmonic force, the resulting forced
harmonic vibration will take place at the frequency of the applied force. That is, if we
consider the external forces to be harmonic
Fj = Fj0 exp(i ω t) j ∈ [1, 2] (3.49)
we call ω the forcing frequency. The forcing frequency is such that the amplitudes
of the two degrees of freedom will be maximum when it is equal to one of the natural
frequencies of the system; i.e. ω = ωi . This condition is called resonance.
In order to find the solutions of the system, we can rewrite the equations of motion
from Eq. 3.4 to account for the external forces
m x¨1 + k x1 − k2 x2 = F1
(3.50)
m x¨2 − k2 x1 + k x2 = F2
Therefore, we can assume the steady-state solutions to be harmonic as well
x j (t) = X j exp (i ω t) (3.51)
The equations of motion then result
−m ω 2 X1 exp (i ω t) + k X1 exp (i ω t) − k2 X2 exp (i ω t) = F10 exp (i ω t)

(3.52)
−m ω 2 X2 exp (i ω t) − k2 X1 exp (i ω t) + k X2 exp (i ω t) = F20 exp (i ω t)
−m ω 2 X1 + k X1 − k2 X2 = F10
(3.53)
−m ω 2 X2 − k2 X1 + k X2 = F10
−m ω 2 + k

−k2 X1 F
= 10 (3.54)
−k2 −m ω 2 + k X2 F20
Where we will define the matrix

−m ω 2 + k

−k2
Z (ω ) = (3.55)
−k2 −m ω 2 + k
as the mechanical impedance of the system. Thus, the equations of motion can be
solved to obtain
Z X = F 0 → X = Z −1 F 0 (3.56)
The inverse of the mechanical impedance matrix is easy to compute for this system
−m ω 2 + k

1 k2
Z −1 = (3.57)
(−m ω 2 + k2 )2 − k22 k2 −m ω 2 + k
Hence, we can express the solution as
−m ω 2 + k F10 + k2 F20

X1 =
(−m ω 2 + k)2 − k22
(3.58)
k2 F10 + −m ω 2 + k F20

X2 =
(−m ω 2 + k)2 − k22
Then, the natural frequencies correspond to the frequencies at which the system
oscillates without any external forcing; namely, the frequencies ωi that we found in the
previous section from Eq. 3.14.
F L U I D - S O L I D I N T E R A C T I O N F U N D A M E N TA L S
4
4.1 Dimensionless fundamental equations 132
4.2 Solid with a still fluid 141
4.3 Coupling with a fast fluid 157
4.4 Coupling with a slow flow 169
4.5 Coupling with any type of flow 174
What do we mean by fluid-solid interaction (FSI)? Consider the following case schemat-
ically
Figure 4.1: Cylinder deformed by an external flow
The cyllinder is deformed by the flow, which itself is modified by the deformation of
the cylinder. Therefore, the fluid and solid mechanics are coupled, and cannot be solved
indepedently.
There are namely many phenomenons that can fall into the domain of fluid-structure
interactions (FSI)
• Flutter, as when the the tail of an airplane starts to vibrate in reaction to the air
around it.
• Dolphin skin effect, as the analysis of whether the flexibility of the dolphin skin is
beneficial in reducing friction - does a dolphin go faster because it has soft skin?
• Wind on crop effect, as in the waves due to an FS interaction.
• Hard disk drive head, the reading head in a conventional hard disk drive floats
above the surface of the disk, interacting with the air entrained by the disk. So,
predicting the dynamics of the head requires that the interaction with the flow is
properly taken into account.
• Inflatable dam, which is a rather flexible structure in contact with the water that
it’s holding. So an FSI analysis would yield results when there’s an earthquake
and the dam starts to oscillate.
It is evident that there are many cases that can be analyzed with FSI tools, and they
are all quite different from each other. If we ant to develop a general approach to model
131
132 fluid-solid interaction fundamentals
them, we first must find a way to classify all these couplings - the variety is so large that
it is not feasible to find a single model applicable to all the cases.
Once we have classified the cases, then we shall try and build relevant models for
them.
4.1 dimensionless fundamental equations
4.1.1 Dimensional analysis
Let us consider as an example the flow around a sphere - pure fluid mechanics.
Figure 4.2: External flow around a sphere
We may define the configuration by
• The upstream velocity U.

• The density of the fluid ρ.
• The fluid viscocity µ.
• The diameter of the sphere L.
Although the configuration is defined, all of the above are dimensional quantities:
their magnitude refers to a sclae of units. For instance, the velocity must be given in
length [ L] by time [ T −1 ] units.
However, out of these four dimensional quantities we may define a dimensionless
number
ρU L [ M ] [ L −3 ] × [ L ] [ T −1 ] × [ L ]
RE = = (4.1)
µ [ M ] [ L −1 ] [ T −1 ]
This is the so called Reynolds number R E , and it is dimensionless in the sense that we
do not need any sclae of units to express it. It is useful for classifying types of flows.
Dimensional analysis always refers to physical laws; i.e., quantities used to represent
the system that is being considered. In a general form, such a physical law would read
f ( x1 , x2 , . . . , x n ) = 0 (4.2)
We take as a principle that

A physical law must relate only dimensionless parameters.
This way, if the units are changed the magnitude will remain the same and the law
will still be satisfied. This can be easily seen with an example. Consider that we want
find the physical law that related the force exerted by the fluid, namely the drag D, to
the parameters that define the configuration
4.1 dimensionless fundamental equations 133
Figure 4.3: Drag on a sphere
The law would read
f ( D, U, ρ, µ, L) = 0 (4.3)
Where all the parameters are dimensional. So, by combining them we can find two
dimensionless parameters, which are

D ρU
F 2 2
, = F Cdrag , R E = 0 (4.4)
ρU L µ
The second parameter is clearly the Reynolds number as defined earlier, and the first
one is often referred as the drag coefficient Cdrag . Note that we have considerably reduced
the dimension of the parameter space to explore; it had a dimension of 5 and now it has
a dimension of 2.
So generally, how do we go from dimensional quantities x1 , . . . , xn to dimensionless
quantities X1 , . . . , X p ? There is a rather general theorem that states
The number of dimensionless quantities P is equal to the dimensional quantities N
minus the rank R of the dimension exponents matrix.
P= N−R (4.5)
For the previous example, we knew that we could find two dimensionless quantities
since N = 5, and the rank of the exponents matrix is
D U µ L
 
ρ
L 1 1 −3 −1 1 
R = rank 
M 1
=3 (4.6)
0 1 1 0
T −2 −1 0 −1 0
and then P = 5 − 3 = 2 dimensionaless quantities are guaranteed. Note that the π

theorem does not tell you what the dimensionless variables are, just how many of them
there are. Therefore, there is no good choice of dimensionless parameters, there are just
efficient choices that will allow to express a physical law for a problem.
4.1.2 Independent dimensional analysis in the fluid alone
We need to specify what quantities we want to use to define our problems, and what we
are looking for.
For the case of the fluid, let us say that we are looking for the local velocity U in
relation to the coordinate of the point we consider x and the time t. So we can list the
parameters to analyze
• Fluid
• Coordinates x
• Time t
• Velocity field U
• Viscosity µ
• Size L
• Gravity g
• Density ρ
• Boundary conditions U0
This is, we know the velocity will depend upon the fluid viscocity, the density and
the gravity. The velocity also will change if the alter the domain size L. And lastly, it is
going to depend on some boundary condition; e.g. U0 .
So, we assume that there exists a physical law that relates the fluid velocity with all
other parameters in the fluid alone
f ( x, t, U, µ, L, g, ρ, U0 ) = 0 (4.7)
This means that the flow is not going to depend upon the deformation of the solid, for
instance, because the stifness E is not included in the law above.
Using the π Buckingham theorem, we can rest assured that there are at least P =
N − R = 8 − 3 = 5 independent dimensionless parameters; being R = 3 the rank of the
dimension exponents matrix. This are
!
U x U0 t ρ U0 L U0
F , , , ,p =0 (4.8)
U0 L L µ gL
where we can spot the Reynolds number and a new quatity defined as the Froude
√
number FR = U0/ g L relating the gravity g with the dimensions of the domain L and
the flow U0 . Note that the five members are independent from one another; you cannot
get one by a combination of the others.
Let’s go back to the ratio between U0 T over L (third member). We can rewrite it as
follows
U0 t t t
= L = (4.9)
L /U0 TFluid
Here, TFluid represents the time taken by a particle of velocity U0 to travel across the
distance L. Therefore, TFluid is a time scale associated with convection in the fluid.
4.1.3 Independent dimensional analysis in the solid alone
Now, for the solid, we might want to know the displacement ξ at position x at time t.
Aditionally, it may depend upon the stifness of the solid E, it’s density ρs and on gravity
g. It will also depend upon the size L and there is somewhere the magnitude of the
displacement that is set, i.e. ξ 0 .
• Solid
• Coordinates x
• Time t
• Displacement field ξ
• Stiffness E
• Size L
• Gravity g
• Density ρs
• Boundary conditions ξ 0
So we look for a relation that satisfies

f x, t, ξ, E, L, g, ρs , ξ 0 = 0 (4.10)
where the displacement ξ is our unknown. Using again the π Buckingham theorem,
we can find the minimum amount of dimensionless parameters, which is 5. This are
√
ξ x t E/ρs ξ 0 ρs g L
F , , , , =0 (4.11)
L L L L E
We can identify some interesting quantities. The first, we define as the Displacement
Number D = ξ 0/L; when it is large, the displacementes are large with regards to the size
of the domain. The second one, we shall define as the Elastogravity Number G = ρs g L/E;
when it is large, it means that the deformations induced by gravity in the solid are large.
For example in a jelly cake, the shape is really affected by gravity.
As we did in the fluid, there is one last quatity relating the time scale. If we define
√
c = E/ρs , we can rewrite it as follows
tc t t
= L = (4.12)
L /c TSolid
So TSolid represents the scale of elastic wave velocities inside the solid. This is, TSolid is
the time that an elastic wave takes to go across the solid.
4.1.4 Dimensional analysis of the interactions
We are going to use the smae method that we used for the fluid and solid sone but now
considering the fluid and solid parameters simultaneously.
The variables of interest are the fluid velocity U and the solid displacement ξ fields.
We must look for a a relation between one of the variables of interest and all the other
parameters. If we start with the fluid velocity
f (U, x, t, µ, ρ, U0 , L, g, E, ρs , ξ 0 ) = 0 (4.13)
Where you can note that there are some quantities that are common to both domains,
such as the gravity g and the scale of lengths L, and others that are only defined for the
fluid, like the viscocity µ, and the stifness E in the solid. But here, we relate them all to
find the law for the fluid velocity U for the positions x and time t being considered.
Applying the π Buckingham theorem, we now that we can find P = 11 − 3 = 8

dimensionless independent quantities. We already now five of these quantities from the
analysis done for the fluid alone in Eq. 4.8. And from the ones we know from the solid
side, i.e. the Displacement Number D and the Elastogravity Number G we have seven of
them
!
U x U0 t ρ U0 L U0 ξ 0 ρs g L
F , , , ,p , , ,A =0 (4.14)
U0 L L µ gL L E
were we have called A to our missing last quantity. We know that A cannot be in
combination of the other ones because we are looking for a set of independent quantities.
Also, it must mix parameters from the fluid and the solid side, otherwise we would have
found it before when doing the uncoupled cases.
So we can define different dimensionless quantities combining fluid and solid dimen-
sional parameters. For example, we have the Mass Number
ρ
M= (4.15)
ρs
which tells you that is diffeent for a solid to interact with fluids of differnts densities;
e.g., with air or water.
We can also find a Reduced Velocity
U0
UR = E/ρs
(4.16)
that contains information of the way the two dynamics are related.
Lastly, we can define the Cauchy Number combining stresses or stifness
ρ U02
CY = (4.17)
E
which tells you that the higher CY , the more the solid is elastically deformed by the
flow.
The Mass Number M, the Reduced Velocity UR and the Cauchy Number CY are the
most common quantities that relate parameters in both the fluid and the solid. Therefore,
our last quantity A could be any of these three depending on the problem to analyze.
4.1.5 Coupled equations for fluids and solids
So far, we have defined the dimensionless parameters that need to be used for fluid-solid
interactions, and we have done this without using any of the equations of fluid mechanics
or solid mechanics. We just worked on the parameters that would be needed to describe
the system. Now, we must use the models for some of the laws that satisfy the varibles
in each domain, and write a condition at the interface.
In the fluid side, we know that a system can be considered as a given collection of
fluid particles, so the mass must remain fixed. The net flux of mass entering the element
must be equal to the rate of change of the mass of the element; that is
∂
ṁin − ṁout = melement (4.18)
∂t
So, for the case of an incompressible flow in which the density of the fluid particles do
not change as it travels along, the continuity equation states
∇·U = 0 (4.19)
But that is just one equation that has three unknowns - the three compoenents of the
velocity. Now, the differential momentum equation is a vector equation that results from
applying Newton’s Second Law to a fluid particle, and it will provide us with three
scalar equations.
Many fluids exhibit a linear relationship between the stress components and the
velocity gradients. Such fluids are called Newtonian fluids and include common fluids
such as water, oil, and air. If in addition to linearity, we require that the fluid be isotropic,
it is possible to relate the stress components and the velocity gradients using only two
fluid properties, the viscosity µ and the second coefficient of viscosity λ. The momentum
equations that relate said quantities are the Navier-Stokes equations
dU
ρ = −∇ p + ρ g + µ ∇2 U (4.20)
dt
The Navier-Stokes equation describes the balance between the acceleration of the
fluid, the pressure forces, the gravitational forces, and the viscous forces. It captures the
dynamics and interactions of these physical phenomena in fluid flow, allowing us to
understand and predict the behavior of fluids under different conditions.
The LHS represents the rate of change of momentum; it describes the acceleration of
the fluid particles in the fluid.
On the RHS the term −∇ p represents the pressure gradient in the fluid, where the
negative sighn indicates that pressure acts in the directions opposite to the gradient.
It represents the force exerted by the pressure field on the fluid. Then, the term ρ g
represents the gravitational force action upon the fluid, accounting for the vertical
movement and distribution of the fluid in presence of gravity. Lastly, µ ∇2 U shows
the viscous effects in the fluid; it describes the diffusion of momentum due to the
internal friction or viscosity of teh fluid. It accounts for the shearing and dissipation of
momentum within the fluid.
Now, for the solid side, there are several models that we could use, such as Continuum
mechanics. But we shall use as much as possible a modal approximation, or even in
many cases a single mode approximation. This means that we are going to assume that
the displacement ξ in the solid is a product of a function of time q(t) and a function of
space φ( x )
ξ ( x, t) = q(t) φ( x ) (4.21)
where q(t) will be called the modal displacement and φ( x ) the modal shape. We
assume that we know the modal shape φ( x ), which may result from experiments or
computation, so the only unknown that remains is the modal displacement q(t). This is
a really simple framework to describe the kinetics of the motion of a solid, but it will be
sufficient for most study cases.
We shall assume that the modal displacement in the solid satisfies an oscillator equation
d2 q
m +kq = f (4.22)
dt2
Figure 4.4: Modal shape described as as function of coordinates in a moving beam
where m is the modal mass, k the modal stifness and f is the modal load, all knowns.
In the case of the beam shown in Fig. 4.4, the load is just the sum over the solid of the
local loads tiem the modal shape
ˆ
f = F · φ dx (4.23)
We say that the forces are projected on the mode.

So, we have the continuity and momentum equations on the fluid side, and the modal
equation in the solid side. We shall now connect them by the continuity equations at the
interface. This is, at the interface between the solid and the fluid, we have two kinds of
conditions
1. kinematic condition, that connects directly the velocities. We will mostly use a
condition that the two velocities are equal, meaning that there is no mixing between
the two domains, and there is no sliding either.
∂ξ
U=
∂t (4.24)
dq
U ( x, t) = (t) φ( x )
dt
2. dynamic condition, stating an equilibrium of forces at the interface. On the fluid

side, the acting local force is the result of pressure and viscous forces; and on the
solid side we only consider the quantity called the modal force. This means that
we have to equal the sum of all the fluid forces times the modal shape with the
modal force f .
ˆ
− p I + µ ∇U + ∇t U · n · n · φ dS = f

(4.25)
Interface
The dynamic condition establishes an equilibrium of forces at the interface between

the solid and fluid domains. It considers the forces acting on both sides of the interface
and requires their balance. The term ∇U + ∇t U represents the deformation or strain
rate of the fluid. It is the sum of the gradients of the velocity vector U and its transpose.
It accounts for the rate at which the fluid is changing shape or deforming. The integral
over the interface ensures that the sum of the forces on the fluid side, represented by the
terms within the curly brackets, weighted by the modal shape function φ, is equal to the
modal force f on the solid side. This dynamic condition ensures a balance of forces at
the interface between the fluid and solid domains.
By combining the kinematic condition (velocity equality) and the dynamic condition
(force equilibrium), we can establish a connection between the fluid and solid domains,
allowing for the interaction and exchange of forces and velocities at their interface.
4.1.6 Dimensional analysis of the coupled equations
We have the equations that govern the dynamics of our coupled system. But they relate
dimensional quantities. We now want to move to equations relating dimensionless
quantities.
First of all, we have to define the dimensionless variables we want to use out of the
dimensional ones. In the fluid side, we have the coordinate x, the velocity U and the
pressure p. We can define the following dimensionless quantities
• Dimensionless coordinate xe = x/L

e = U/U0
• Dimensionless velocity U
• Dimensionless pressure pe = p/(ρ U02 )
On the solid side, we can do something very similar but for the coordinate x, the
modal displacement q and teh modal force f
• Dimensionless coordinate x = x/L

• Dimensionless modal displacement q = q/ξ 0
• Dimensionless force f = f/(k ξ 0 )
But we have not said anything about the dimensionless time. In Section 4.1.1 we have
defined both a dimensionless time relevant for the fluid elements and the and the solid
elements; referred to the convection time TFluid and the time of elastic waves to propagate
TSolid respectively.
So, these two times have different references. We must choose which one to use in the
next steps, since the sime clock needs to be the same for both fluid and solids if we want
to compare them. We will choose arbitrarily the reference time of the solid
• Dimensionless time t = t/TSolid
So now we have to take the equations in dimensional form from Sec. 4.1.5 and
substitute these variables to obtain equations between our dimensionless quantities.
We start in the fluid-side. The mass balance is simple, it takes the same form
∇·U = 0
(4.26)
∇·U
e =0
For the momentum equation, we end up with

dU
ρ = −∇ p + ρ g + µ ∇2 U
dt
(4.27)
c dU
e gL µ
= −∇ pe − 2 + ∇2 U
e
U0 dt U0 ρ U0 L
We have identified most of the quantities as dimensionless numbers, so we can write a

dimensionless momentum balance as
1 dU
e 1 1
= −∇ pe − 2 + ∇2 U
e (4.28)
UR dt FR RE
where we note we have the 1/UR coefficient in front of the acceleration because we are
using a dimensionless time based upon the solid dynamics.
Now, in the solid-side, things are quite simple. Since we have chosen the reference
time as the time of the solid, we end up with a oscillator equation with frequency 1
d2 q
m +kq = f
dt2
r 2
m d2 q
m ξ0 2 + k ξ0 q = k ξ0 f (4.29)
k dt
d2 q
2
+q = f
dt
Lastly, we shall apply the same analysis for the interface. We start with the kinematic
condition
∂ξ
U=
∂t
dq
U ( x, t) = (t) φ( x )
dt (4.30)
U0 TSolid e ξ 0 dq
U= φ( x )
L L dt
UR U e = D dq φ( x )
dt
which contains the reduced velocity UR and the displacement number D that we have
already defined.
Finally, we have to apply the same operation on the dynamic condition, noting that
the sum of the whole interface involves also the change of space coordinates
ˆ
− p I + µ ∇U + ∇t U · n · n · φ dS = f

Interface
ˆ
ρ U02 L

µ
t e · n · n · φ dS = ξ 0 f
− pe I + ∇U
e +∇ U (4.31)
Interface k ρ U0 L L
ˆ
1 e te
CY − pe I + ∇U + ∇ U · n · n · φ dS = D f
Interface RE
We have now equations where the influence of our dimensions numbers is evident.
We expect that for a particular case, when a given demonstrated number is large or small,
the placement in terms of the equations will be more or less influencing the result. This
way, we can find simpler models by simplyfing terms.
4.2 solid with a still fluid 141
4.2 solid with a still fluid
We have seen that the Reduced Velocity UR = TSolid/TFluid = U0/c is a very useful parameter
to help in classifying problems, since it compares the timescales of the solid dynamics
and the fluid dynamics. We shall explore the effects of this Reduced Velocity.
4.2.1 Small reduced velocities
We will start by analyzing the case when the fluid is apparently not moving; that is,
when its motion seems very slow in comparison with that of the deformation of the solid
(small UR ). Imagine the oscillations of the flexible dam in contact with a very slowly
moving water, or a small boat that oscillates on a river.
Now, let us imagine we follow in time the evolution of a quantity pertaining to the
solid - the displacement somewhere. This quantity evolves with a timescale we defined
as TSolid . Conversely, the quantity pertaining to the fluid dynamics such as the position
of a fluid particle would evolve on a much longer timescale, TFluid , as shown in Fig. 4.5.
Figure 4.5: Evolution of displacement in the solid in comparison to position of a fluid particle
So we can say that the dynamics of the solid, when the timescales are so apart, occur
with a still fluid (blue line in Fig. 4.5). This is a much simpler framework to model FSI.
We have now to see how this materializes in equations. We have to take into account
that there is a significant difference between the dynamics of the fluid and that of the
fluid. We have shown that the fluid domains see two types of boundary conditions: those
at the interface with the solid (kinematic and dynamic conditions) and others anywhere
else (no slip, etc). So, we must analyze how these boundary conditions contribute
respectively to the dynamis of the fluid: we will do this by comparing the order of
magnitude they impose on the velocity.
At the interface, the condition on the velocity is that it is equal to that of the solid
∂ξ
U= (4.32)
∂t
Schemtically, we may have something as shown in Fig. 4.6 in terms of the evolution of
the solid displacement somewhere.
Figure 4.6: Velocity at the interface
Since ξ is a displacement in the solid, we have an order of magnitude of this that we

called ξ 0 . Then, ξ evolves with the characteristic timescale that we called TSolid , so the
quantity ∂ξ/∂t has an order of magnitude

∂ξ ξ0
=U=O (4.33)
∂t TSolid
which gives us the order of the fluid velocity U.

The other boundary conditions on the fluid of course also affect the dynamics. But,
we have defined a quantity U0 which sets the magnitude of our fluid velocity - this may
represent the magnitude of the upstream velocity, for instance. So, we can state that the
fluid boundary conditions set a velocity of order
U = O(U0 ) (4.34)
We are stating that the dynamics of the fluid are governed by a condition of order O(U0 )
and in the other hand byt a condition of order O(ξ/TSolid ). Certainly, if U0 << ξ/TSolid , the
first condition can be set to
U=0 (4.35)
without changing much the results.

So, in terms of the dimensionless numbers, we can rewrite this condition as
ξ0
U0 <<
TSolid
U0 TSolid ξ0 (4.36)
<<
L L
UR << D
Therefore, under the condition UR << D we say that to compute the dynamics of the
solid in contact with a fluid, we can totally disregard the proper dynamics of the fluid.
So we have two dynamics, one slow which is that of the fluid, and one fast which is that
of the coupled still fluid and solid.
Figure 4.7: Scheme of the two fluid dynamics
4.2.2 Adapted dimensionless numbers
So, we have defined a condition of small reduced velocity where the problem of the
coupled dynamics between the fluid and the solid could be considered as independent of
that of the proper dynamics of the fluid. But we have yet to see how this would simplify
the equations.
First of all, we have to reconsider our choice of dimensionless numbers, since we have
been using some quantities that contain the upstream velocity U0 , which is just not
relevant anymore for the dynamics of the coupled problem.
The upstream velocity U0 appears in the
ρ U0 L
Reynolds number, R E =
µ
U0
Froude number, FR = p (4.37)
gL
ρ U02
Cauchy number, CY =
E
which represent the scale of flow velocity U0 with a speed of viscocity diffusion (R E ),
with the velocity of gravity waves (FR ) and with the deformation due to the flow (CY ).
In place of U0 it would be much more relevant to our problem to use in the dimen-
sionless numbers the velocity related to the solid: the velocity of elastic waves c = L/TSolid .
We can find the following dimensionless quantities
RE ρcL
Stokes number, ST = =
UR µ
FR c
Dynamic Froude number, FD = =p (4.38)
UR gL
CY ρ c2 ρ
Mass number, M = 2
= =
UR E ρs
which scale what happens in the solid with the speed of diffusion (ST ), the velocity of
gravity FD and the ratio of masses M. Remember that UR = U0/c.
We have assumed that we could set the fluid boundary conditions at U = 0 because
it has a negligible effect on the dynamics of the fluid. But, we have to be consistent in
terms of the variables that we use in the equations
• Dimensionless fluid coordinate xe = x/L

e = U/U0
• Dimensionless fluid velocity U
• Dimensionless fluid pressure pe = p/(ρ U02 )
• Dimensionless solid coordinate x = x/L

• Dimensionless solid modal displacement q = q/ξ 0
• Dimensionless solid force f = f/(k ξ 0 )
• Dimensionless time (same for both) t = t/TSolid
This is, we are using U0 to scale the velocity and the pressure, but at the same time we
assume that U0 is not relevant to the problem. So we have to make another choice of
variables. We can replace U0 by a velocity referring to the solid; the elastic waves velocity
c again, for instance. So we can rewrite all our fluid variables as
• Dimensionless fluid coordinate x = x/L

• Dimensionless fluid velocity U = U/c
• Dimensionless fluid pressure p = p/(ρ c2 )
whilst the solid’s remain the same. Now we shall rewrite our fluid and interface
equations in terms of these new variables.
In the fluid-side, we first have the mass balance. Replacing U
e with U we end up with
a very similar equation
Old, ∇ · U
e =0
(4.39)
New, ∇ · U = 0
Then, for the momentum balance we will find some new coefficients, which happen to
be the new dimensionless number ST , FD and M that we defined above
1 dU
e 1 1
Old, = −∇ pe − 2 + ∇2 U
e
UR dt FR R E
(4.40)
dU 1 1
New, = −∇ p − 2 + ∇2 U
dt FD ST
which is no surprise, since we have obtained these new dimensionless numbers also
by replacing U0 by c.
Now, at the interface we can do the same kind of substitution to obtain the kinematic
and dynamic conditions. We will start with the kinematic condition
e = D dq φ( x )
Old, UR U
dt (4.41)
dq
New, U = D φ( x )
dt
And the same for the dynamic condition

ˆ
1 e
Old, CY − pe I + ∇U + ∇ t Ue · n · n · φ dS = D f
Interface RE
ˆ (4.42)
1 t

New, M −p I + ∇U + ∇ U · n · n · φ dS = D f
Interface ST
4.2.3 Small solid motion
We have simplified the general problem of fluid-solid interactions by considering the

limit case where the velocity of the fluid is so small that we can neglect its influence.
When considering the motion of a solid in a still fluid, there is also a very interesting
simple case which is that of small amplitudes.
Figure 4.8: Deformation of the solid for different orders of magnitude of D
We have a dimensionless parameter that quantifies the magnitude of displacement

of the solid: the displacement number D = ξ 0/L. Fig. 4.8 shows the impact of D in the
deformation of the solid. When D << 1 there is little change to the shape of the solid,
since the displacement is much smaller than the characteristic length, resulting in only a
slight change to the shape of the solid. So, we will explore how this might simplify our
equations.
We can write expansions of the quantities as Taylor Series. A Taylor series expansion
allows us to approximate a function by expressing it as an infinite sum of terms that
involve the function’s derivatives evaluated at a particular point. So, if we consider D as
a small parameter, this would imply that higher-order terms in the expansion will be
successively smaller compared to the lower-order terms. By neglecting these higher-order
terms, we can focus on the leading-order terms, which capture the dominant behavior of
the system.
To make equations easier to read from now on, we shall omit the overline superscript
over the variables, but remember that they are still dimensionless.
For example, in the fluid, we can write the pressure expansion P as
P = P0 + D p + D2 . . . (4.43)
where P0 is the zeroth order term, p is the first order term and so on. The velocity
might also be developed
U = 0 + D u + D2 . . . (4.44)
We saw that when we consider small amplitudes of solid motion, it implies that the
solid’s displacement and resulting fluid velocities are relatively small compared to the
characteristic scales of the problem. In this case, the zeroth-order term, which represents
the motionless fluid velocity (U = 0), is neglected because we are interested in the
influence of the solid motion on the fluid. The first-order term, U = D u, captures the
dominant contribution to the fluid velocity resulting from the solid motion. The small
u in this term represents the first-order approximation of the fluid velocity field. It
accounts for the influence of the solid’s displacement on the fluid flow.
Now in the solid, the dimension of displacement ξ has alrealdy been writen as
ξ = D q φ when we used a single mode approximation. We can rewrite this again taking
into account the expansion for small D
ξ = 0 + D q φ + D2 . . . (4.45)
4.2.4 Zeroth order approximation
When we refer to a zeroth-order approximation, it means that we neglect terms of the

first order and higher in the expansion. In the context of the velocity equation, the
zeroth-order approximation corresponds to setting the first-order term, D u = 0 .
By considering only the zeroth-order term, we effectively assume that the fluid velocity
(U) is negligible or approximately zero. This approximation is valid when the fluid
motion is significantly slower or has a smaller magnitude compared to the solid motion.
In other words, it assumes that the effect of the fluid flow on the solid is negligible.
The zeroth-order approximation simplifies the problem by disregarding the fluid
dynamics and focusing solely on the solid motion. It allows us to analyze the behavior
of the solid independently of the fluid flow. This approximation is particularly useful
when studying solid structures in still or slow-moving fluids, where the fluid’s influence
can be considered negligible compared to the solid’s motion.
So, for all the equations we have to evaluate what happens at D = 0. In other terms,
once we have inserted the expansions into our equations in the fluid
∇·U = 0
dU 1 1 (4.46)
= −∇ P − 2 + ∇2 U
dt FD S T
We just substitute U = 0 and P = P0 since we are considering the zeroth order

approximation and we end up with much simpler forms
∇·0 = 0
1 (4.47)
0 = −∇ P − +0
FD2
The mass balance is satisfied, sinde the divergence of zero of course is zero. The
momentum equation just relates now the pressure field P0 with the gravity. This shows
that although there is no velocity in the fluid at that order, there is a pressure which
depends only on the gravity force. This is the well known Hydrostatic Pressure.
4.2.5 First order approximation
The first-order approximation takes into account the next level of terms in the expansion,
which in the case of the velocity corresponds to the term D u and for the pressure
corresponds to D p.
Including the first-order term allows us to consider the influence of fluid motion on
the solid. In this approximation, we assume that the fluid velocity (U) has a non-zero
value, but it is small compared to the solid motion. By including the first-order term, we
account for the effects of fluid flow on the solid, although still considering them to be
relatively small.
The first-order approximation provides a more accurate description of the fluid-solid
interaction compared to the zeroth-order approximation. It allows us to capture some of
the coupling between the fluid and solid behaviors, taking into account the influence of
the fluid flow on the solid’s motion. However, it is important to note that the first-order
approximation is still limited to small fluid velocities or small fluid-solid interaction
effects.
In terms of equations, we start from the same ones
∇·U = 0
dU 1 1 (4.48)
= −∇ P − 2 + ∇2 U
dt FD ST
but we make now the substitutios gathering the terms that are proportional to D; that
is u for the velocity and p for the pressure
∇·u = 0
∂u 1 (4.49)
= −∇ p + ∇2 u
∂u ST
and obtain the equations in terms of u and p which represent the variations of the
pressure and velocity in the fluid when the solid moves.
The mass balance is still satisfied. Then, the momentum balance is much simpler,
since the convective term in the acceleration dissapars because it would give a D2 term.
Secondly, the gravity is not here anymore, because it is a constant and therefore a zeroth
order term.
Then, at the interface, the kinematic condition is quite simple. Since the fluid velocity
should be equal to the solid velocity
dq
U=D φ (4.50)
dt
and because U = D u for the first order approximation, we have the displacement
number D on both sides of the equation and it simply results as
dq
u= φ (4.51)
dt
For the dynamic condition we have to be more careful. The dynamic interface condition
that we have used allows to derive the fluid loading projected to the mode of interest.
This was done by summing over the interface the product of the local fluid loadings
time the modal shape φ. But this integration is done on the instantaneous position of
the interface. So, since the interface is deformed, a point x has moved to the position
x + D q φ as depicted in Fig. 4.9.
We should expand the geometry at the first order in D. But we have to be careful
considering this change of geometry. Let us imagine that the displacement of the solid is
a pure traslation, as shown in Fig. 4.10.
Figure 4.9: Deformation of the interface
Figure 4.10: Pure traslation fo the displacement of the solid
This means that φ is a vector independent of the interface. In that case, we can take
out φ from the dynamic condition integral and focus on all other terms, noting that the
normals n are unchanged
ˆ
1 t

Mφ· −P I + ∇U + ∇ U · n dS = D f (4.52)
Interface ST
Now, we can take advantage of the smallest placement of the interface to use quantities
that are defined on the original undeformed position of the interface. For instance,
because D is small, the pressure at the new position may be developed as
P ( X + D q φ) = P0 ( X + D q φ) + D p ( X + D q φ) + . . .
(4.53)
= P0 ( X ) + D q φ · ∇ P0 + D p ( X ) + . . .
where we say that the pressure at the new position ( X + D q φ) might be estimated
by the pressure at the original position X plus the gradient times the displacement
D q φ · ∇ P0 .
With regards to the velocity, the first terms remain the same as before
U ( X + D q φ) = 0 + D u ( X ) + . . . (4.54)
Therefore, now we have all the quantities involved defined at the same reference X
of the undeformed interface. Now, we just have to put this P and U in the integral that
defines the fluid loading
ˆ ˆ
1
∇u + ∇t u

Mφ· −p I + · n dS − M q φ · (∇ P0 · φ) · n dS = f
Interface ST Interface
(4.55)
We can note two very different terms in Eq. 4.55. The first one represents a projected
force on the solid, resulting from the pressure and viscous stress in the fluid. This is,
since this pressure p and velocity u are those directly caused by the motion of the solid,
this represents just a feedback from the fluid to the solid motion.
The second term is quite different. It does not depend neither on p nor u, but only
on the zeroth order pressure P0 . It means that it is not related to the motion of the fluid
induced by that of the solid, but rather to the motion of the solid in a pre-stressed fluid.
It is important to note that all the first order equations are now linear. We linearized
them when taking advantage of the small motion, so they can be solved now.
4.2.6 Added stiffness
The key result we obtained was that the force f exerted by the fluid on the solid was
made of two terms: one related to the fluid motion caused by the solid and other that
does not depend on the motion of the fluid. This means that you can compute this last
term without doing any fluid mechanics - it is an added stiffness term.
If we look at this part of the force acting on the solid and disregard the other term
ˆ
−M q φ · (∇ P0 · φ) · n dS = f (4.56)
Interface
we basically have a force f that depends upon the modal displacement q. Then, the
∇ P0 is given by the momentum balance at the zeroth order from Eq. 4.50
1
0 = −∇ P − (4.57)
FD2
It is related to the loading by gravity, which is in the dynamic Froude number FD . So,
if we substitute the gradient of the pressure P0 and take FD out of the integral, the force
f takes the following form
ˆ
M
f stifness = q 2 ( φ · ez ) ( φ · n) dS (4.58)
FD Interface
Since this force f stifness is proportional to the displacement q we define it as a stifness

force. A stiffness force refers to the force exerted on a body due to its deformation or
displacement from its equilibrium position. It arises from the resistance of the body to
deformation and is proportional to the stiffness or rigidity of the material. In practical
terms, the stiffness force can be understood as the force required to stretch, compress,
bend, or twist a material.
When a solid body is subjected to external loads or displacements, it tends to resist
the deformation and return to its original shape. This resistance is characterized by the
material’s stiffness, which is a measure of its ability to withstand deformation under the
influence of applied forces.
So we can simplify this stifness force as
f stifness = −q k F (4.59)
where the coefficient k F is a constant defined by

ˆ
M
kF = − 2 ( φ · ez ) ( φ · n) dS (4.60)
FD Interface
From the point of view of the solid, moving in a fluid with a pressure gradient is
strictly equivalent to being connected to an elastic spring, as depicted in Fig. 4.11.
Figure 4.11: Fluid-induced stifness
This means that we can compute the stifness of this spring once and for all for a given
problem. We do not have to recompute the force at each position of the solid.
4.2.7 Added mass
We shall now explore the first term of the fluid force, disregarding the second one
ˆ
1 t

f = Mφ· −p I + ∇u + ∇ u · n dS (4.61)
Interface ST
It is a sum over the interface of the force resulting from the pressure p and the velocity
u in the fluid. It corresponds to the to the effect on the solid of the motion induced in
the fluid, by the motion of the solid itself. It is a feedback term which depends on how
the fluid reacts to the motion of the interface.
Although it was not the case for the stifness terms we analyzed in Sec. 4.2.6, now we
will have to compute the motion of the fluid in order to compute the force.
In the fluid domain, the equations that we must solve are
∇·u = 0
∂u 1 (4.62)
= −∇ p + ∇2 u
∂u ST
where we can notice, that along with the interface dynamic condition from Eq. 4.61,
there is only one dimensionless parameter: the Stokes number ST . The Stokes number is
based upon the Reynolds number R E , but it is not based on the mean velocity bur rater
on the velocity of the solid. So it appears on the equations to bring the effect of viscocity
in the momentum equation and in the fluid force.
As an example, for a common practice case where L = 1 [m], µ/ρ = 10−6 [m2 s−1 ],
TSolid = 1 [s], the Stokes number can be computed as
ρcL ρ L2
ST = = = 106 (4.63)
µ µ TSolid
So the term 1/ST is quite small, and can be disregarded in many practical cases. So we
can add another approximation to our model: high Stokes number. This simplifies the
equations in several ways.
In the fluid domain, the momentum balance simplifies to
∇·u = 0
∂u (4.64)
= −∇ p
∂u
which means that the viscous term can be neglected. Then, in the conditions at the
interface
u · n = q̇ φ · n
ˆ
(4.65)
f =M (− p n) · φ dS
Interface
where we notice that in the dynamic condition the viscous shear can be neglected, so
we only have the sum over the interface of the pressure loading times the solid shape.
But the kinematic condition has also changed, because in the absence of viscous effects,
it is just impossible to enforce a condition on the tangential compoenents of the velocity.
The only condition that remains is that the normal velocities must be equal, meaning
that there is no mass transfer or separation at the interface.
The fluid velocity at the kinematic condition takes the form of a function of time q̇(t)
times a function of space φ( x ). This comes from the single mode description of the solid.
We may then look for a velocity that would have the smae form, not only on the interface,
but rather everywhere in the fluid
u( x, t) = q̇(t) φu ( x ) (4.66)
We can apply the same idea but for the pressure. Because it is related to accelerations,
it would depend upon q̈(t) rhater than q̇
p( x, t) = q̈(t) φ p ( x ) (4.67)
With this in mind, the mass balance now implies that
∇·u = 0
∇ · (q̇(t) φ( x )) = 0
(4.68)
q̇(t) ∇ · φu ( x ) = 0
∇ · φu = 0
Then, in the momentum balance the two q̈(t) cancel out, so we end up with
∂u
= −∇ p
∂u
dq̇(t)
φu ( x ) = −∇ q̈(t) φ p (t) (4.69)
dt
φu ( x ) q̈(t) = −q̈( T )∇ φ p (t)
φu = −∇ φ p
Now, the kinematic condition takes the form
u( x, t) · n = q̇(t) φ( x ) · n
(q̇(t) φu ( x )) · n = q̇(t) φ( x ) · n
(4.70)
q̇(t) ( φu ( x ) · n) = q̇(t) φ( x ) · n
φu · n = φ · n
And finally, in the dynamic condition we can take the ddotq(t) out of the integral in
the space and end up with
ˆ
f =M (− p n) · φ dS
ˆ Interface

f =M −q̈(t) φ p ( x ) n · φ dS (4.71)
Interface
ˆ
f = −q̈ M φ p ( x ) n · φ dS
Interface
This is a major result of fluid-structure interaction theory. The term in the brackets is
actually a constant, since we have the mass number times a sum over the interface of
quantities that depend only on space. Therefore, in this approximation, the fluid force
resulting from the motion of the solid is an inertia force, and the coefficient of inertia
is called the added mass
f inertia = −m A q̈ (4.72)
Also, it is worth noting that the response of the fluid to the motion of the solid is
instantaneous; the force at a fiven time only depends on the acceleration at that same
moment.
Figure 4.12: Concept of added mass
Schematically, if we look at Fig. 4.12, we can say that the problem on the left, where we
have a solid moving in interaction with a fluid, is reduced to the very simple problem on
the right, where we can analyze the same solid with just an augmented mass. So we can
can compute this single quantity m A once and for all, just like the stiffness coefficient k F .
So, to compute the added mass m A we need to compute φ p first. We know that it is
related to the velocity shape φu through Eq. 4.69, and we know that φu was the solution
of Eq. 4.67 and Eq. 4.81 at the interface. So if we take the divergence of the momentum
equation
φu = −∇ φ p
∇ · φu = −∇ · ∇ φ p
(4.73)
0 = −∇ · ∇ φ p
0 = ∇2 φ p
we end up with a Laplace equation for the field φ p at the interface. Once we solve this
Laplace equation we will obtain the added mass by a simple sum over the interface. But,
the most important advantage of this method, is that we only have to solve this equation
once - we do not have to iterate at all.
4.2.8 Added damping
We have considered up until now that the Stokes number was going to be high. But
that is not the case for all fluid-structure interaction problems. Consider for example the
vibrations inside a bearing, as shown in Fig. 4.13, where the size of the fluid domain is
small and the viscocity of the oil is thousand times or more the than the water.
Figure 4.13: Bearing where ST << 1
So, such an small Stokes number means that the viscous term in the momentum
equation dominates over the acceleration term. This means that we can neglect the term
∂u/∂t
∂u 1
= −∇ p + ∇2 u
∂t ST
∂u (4.74)
ST = −∇ p ST + ∇2 u
∂t
0 = −∇ p + ∇2 u
where we have assumed in the last equation that the pressure is already rescaled by
the Stokes number; i.e. p ← p ST .
The mass balance and the kinematic condition remain the same. But, we can write the
dynamic, disregarding the stiffness terms, as
ˆ
M
φ · − p I + ∇u + ∇t u · n dS

f = (4.75)
ST Interface
In order to solve this, we can use the same technique that we used for the low Stokes
number and added mass. This is, since the kinematic condition at the interface is
u = q̇ φ (4.76)
We can assume a general velocity in the fluid domain as
u( x, t) = q̇(t) φu ( x ) (4.77)
and a general pressure field, but one that now depends upon the velocity and not the
acceleration
p( x, t) = q̇(t) φ p ( x ) (4.78)
Notice that the pressure fields φu ( x ) and φ p ( x ) are of course different to those of the
added mass.
The mass balance then implies that
∇·u = 0
∇ · (q̇(t) φ( x )) = 0
(4.79)
q̇(t) ∇ · φu ( x ) = 0
∇ · φu = 0
Then, in the momentum balance we get
0 = −∇ p + ∇2 u
0 = −∇ q̇(t) φ p ( x ) + ∇2 (q̇(t) φu ( x ))

(4.80)
0 = q̇(t) −∇ φ p + ∇2 φu

0 = −∇ φ p + ∇2 φu
The kinematic condition takes the form

u( x, t) · n = q̇(t) φ( x ) · n
(q̇(t) φu ( x )) · n = q̇(t) φ( x ) · n
(4.81)
q̇(t) ( φu ( x ) · n) = q̇(t) φ( x ) · n
φu · n = φ · n
And, in the dynamic condition, if we disregard the stifness terms and use our new
single mode description of the pressure and velocity, we can find
ˆ
M
φ · − p I + ∇u + ∇t u · n dS

f =
ST Interface
ˆ
M
φ · −q̇ φ p I + ∇(q̇ φu ) + ∇t (q̇ φu ) · n dS

f = (4.82)
ST Interface
ˆ
M t

f =− − φ · − φ p I + ∇ φu + ∇ φu · n dS q̇
ST Interface
All the terms inside the curly brackets are time independent, so the quantity is just a
constant. This means that the force induced by the fluid on the solid is a damping force,
proportional to the velocity of the solid q̇. We can write this as
f damping = c A q̇ (4.83)
where the coefficient c A defined as

ˆ
M
φ · − φ p I + ∇ φu + ∇t φu · n dS

cA = − (4.84)
ST Interface
represents an added damping. And again, as it happened with the added mass, the
fluid loading on the solid at time t only depends on the velocity at the same time q̇(t)
and therefore this is an instantaneous feedback.
So, we found that for large Stokes number, the effect of viscocity in the dynamics
of the fluid could be neglected - the motion of the fluid is not influenced by viscous
diffusion, which is really slow. This resulted in an instantaneous inertial response of
the fluid, and had the consequence of an instantaneous added mass effect (Fig. 4.14b).
Conversely, for small Stokes number, the effect of viscocity is dominant. The motion
of the fluid now governed by viscous diffusion which is very fast. This results in an
instantaneous viscous response of the fluid, and as a consequence, an instantaneous
added damping effect (Fig. 4.14a).
(a) Damped oscillator

(b) Free oscillator
Figure 4.14: Effects of added damping and added mass
4.2.9 Memory effect
We showed that the force acting on the solid was made of an added stiffness part and,
depeding on the value of the Stokes number, an added mass or added damping part.
These extreme cases where obtained when the effects of viscocity were negligible (added
mass) or conversely, dominant (added damping). The question now arises: what happens
for intermediate Stokes number?
We will have to change our approach, since we must now keep the two terms involving
the velocity in the momentum balance equation - the inertial and the viscous terms.
None of them can be neglected in this case. So, any form of the velocity in the form of a
function of space and time is going to mix in that equation the function of time and its
derivative.
This means that there is a time scale inside the equations for the fluid domain. For
example, when you have an equation such as
dy 1
+ y=0 (4.85)
dt τ
the solution is y = exp (−t/τ ). This is, the dynamics of y includes the time scale τ. So,
if there is a time scale in the fluid dynamics there is no way that the fluid response can
be instantaneous. We did had instantaneous response for low and high Stokes numbers,
because we have instantaneous viscous diffusion and instantaneous intertial response
respectively. In between, there is competition between these two effects.
To try and solve this, let us consider an idealized problem of what is called the
impulsive response. Imagine that the motion of the solid is an impulsive start as shown
in Fig. 4.15. This is going to cause some fluid motion of velocity u I ( x, t) and pressure
p I ( x, t), where the superscript I denotes impulse. By integrating at the interface, these
velocities and pressure will give us the force acting on the solid
ˆ
I I 1 t

f (t) = M φ · −p I + ∇u + ∇ u · n dS (4.86)
Interface ST
Figure 4.15: Solid motion as an Heaviside step function H (t)
As this is a response of the system to an impulse displacement of the solid we shall

call it the impulse force f I (t). And we don’t know exactly how this force evolves with
time at this moment, but we can be certain that it depends upon time.
If we imagine a more general case with an arbirtrary motion of the solid defined by
any q̇(t), we can approximate it as the sum of succesive impulses as shown in Fig. 4.16.
Figure 4.16: Arbitrary solid motion as a sum of Heaviside step functions
So we can write the solid motion as the sim of succesive impulses of magnitude q̈(t)
times the Heaviside step function
ˆ t
q̇(t) = q̈(t) H (t − τ ) dτ (4.87)
0
Because all the equations governing the dynamics of the fluid are linear, the response
of the fluid, and as a consequence the force, is going to be the sum of successive impulse
responses to the change of velocities (the acceleration). So this will take the form of a
convolution product between the acceleration of the solid and the impulse force
ˆ t
f (t) = q̈(t) f I (t − τ ) dτ (4.88)
0
Therefore, at time t, the force exerted by the fluid on the solid, f (t), results from all
fluid motion started earlier as a consequence of previous changes in velocity of the solid.
The force f (t) keeps the memory of the past. This is because the impulsive response
was not instantaneous, due to the fact that the fluid now has its own time scale. So, any
motion of the solid is going to start a progressive diffusion of momentum in the fluid,
and progressively the feedback force will take this into account. This is sketched in Fig.
4.17.
Imagine a motion defined by the velocity q̇(t) as sketched in Fig. 4.18. At low Stokes
number, the force will have the same shape, being proportional to q̇(t). When the motion
stops, the force stops. On the other hand, at very high Stokes numbers the force is
proportional to the acceleration q̈(t), but it will also stop when the motion stops. In
4.3 coupling with a fast fluid 157
Figure 4.17: Memory effect on the force
between, because the delay in the response, the force will continue to exist even after the
solid has stopped its motion.
Figure 4.18: Effect of the Stokes number
This is a very important new result: between these cases of added damping and added
mass, the fluid feedback force will depend on the history of loading, not an added
damping nor an added mass, but something like an internal clock.
So, we can be certain now that the effect of viscocity is all in the Stokes number. If the
Stokes number is large, we can neglect viscocity and consider only an added mass. If it
is very small, we shall neglect the intertial and consider an added damping. In between,
the copetition between inertia and viscous effects will result in a memory effect that will
give you a force depending on the history of motion of the solid.
In practice, if we consider vibrations of a solid in a still fluid, at large Stokes numbers
the fluid force will be in phase with the acceleration, at small Stokes numbers it will
be in phase with the velocity, and in between the phase will be in between both the
acceleration and velocity. So, by measuring the phase between displacements and the
force, we can estimate the Stokes number and thereby the fluid viscocity.
4.3 coupling with a fast fluid
4.3.1 Large reduced velocities and aeroelasticity approximation
Up until now, we have explores the case of small reduced velocities UR . Let us now
analyze the effect of flows with high reduced velocities; i.e.
TSolid
UR = >> 1 (4.89)
TFluid
A very large reduced velocity means that the time scale of the solid dynamics TSolid is
much onger than that of the fluid dynamics TFluid . In other terms, the time of oscillation
or of wave propagation in the solid is now much longer than the time of convection of a
free particle across the length scale L.
Let us imagine now that we follow in time the evolution of a quantity pertaining to
the solid; for instance, the displacement of a material point. This quantity evolves with
a long time scale that we called TSolid , as depicted in Fig. 4.19. Conversely, a quantity
pertaining to the fluid dynamics, such as the position of a fluid particle, evolves on a
much shorter time scale, TFluid , also as shown in the figure.
Figure 4.19: Comparison on the evolution of solid a fluid quantities over time
Intuitively, we may imagine that there is a limit when these time scales are very
different, such that the dynamics of the fluid occurs in interaction with a fixed solid (red
line in Fig. 4.19). This would certainly be a much more simple framework to model
fluid-solid interaction if the solid does no seem to move. Note though, that by stating
that the solid is fixed, it does not mean that it actually does not move. It just means that
its own motion occurs so slowly that we can neglect it when computing the motion of
the fluid - but we will have to compute the motion of the solid at some point nonetheless.
In contrast with the small reduced velocities case, the dimmensionless number we
chose to derive our fundamental equation in Sec. 4.1.5
ρ U0 L
Reynolds number, R E =
µ
U0
Froude number, FR = p (4.90)
gL
ρ U02
Cauchy number, CY =
E
will be useful in this case, since they all refer to a quantity that is relevant to our
problem: U0 .
But now, it would be more appropiate to use a dimensionless time based upon the
dynamics of the fluid instead of the solid
t t
et = = (4.91)
TFluid L/U0
And we shall use dimensionless velocities and pressures that are based upon the
upstream velocity U0 ; i.e.
e= U
U
U0
p (4.92)
pe =
ρ U02
We have to rewrite now the dimensionless equations in terms of these new variables.
To do this, we can replace the dimensional variables by the dimensionless ones in the
original dimensional equations.
In the fluid part, we have the mass balance and momentum equation as
∇·U
e =0
dU
e 1 1 (4.93)
= −∇ pe − 2 ez + ∇2 U
e
dt
e FR
R E
On the solid side, we can state that the displacement can be represented using a single
mode approximation. For this mode, we can use the dimensionless oscillator equation
ξ ( x, t) = D q(t) φ( x )
∂2 q (4.94)
UR2 + q = f FS
∂et
And lastly, at the interface, we can write kinematic and dynamic conditions in dimen-
sionless forms. They sate that the velocities are continuous at the interface (kinematic)
and that the force that applies on the mode f FS is the projection of the local fluid loadings
on the modal shape φ
Ue = ∂ξ
∂et ˆ (4.95)
1 e
te
D f FS = CY − pe I + ∇U + ∇ U · n · φ dS
∂Ω FS RE
The reduced velocity is going to appear in the scaling of the boundary conditions on
the fluid domain, as depicted in Fig. 4.20.
Figure 4.20: Scaling of the fluid boundary conditions
The scalar U0 corresponds to the reference velocity of the fluid, which we can symboli-
cally write as a boundary condition on the fluid. The velocity there scales as U0 so that
e = O (U0/U0 ) = O(1).
in the dimensionless value it is of the order of U
Then, the kinematic condition states that at the interface the fluid velocity must be
equal to the solid velocity. In order to compute the solid velocity, we know that the
solid displacement scales as ξ 0 and evolves in a time scale TSolid . This means that the
dimensionless displacements scale as ξ 0/L, which is exactly the displacement number
D. This displacement evolves in a dimensionless timescale of TSolid/TFluid , which we know

represents the reduced velocity UR . This is depicted in Fig. 4.21.
Figure 4.21: Boundary conditions
So, ξ = O( D ) and it varies over a time scale of UR . Then, we can say that the
= O( D/UR ) and then U
dξ/dt e = O( D/UR ).
Therefore, the dynamics of the fluid are governed by a condition of order U e = O (1)
and in the other hand a condition of order U = O( /UR ) at the interface. Then, if
e D
UR >> D, the second condition can be set to zero without changing much the results.
This corresponds exactly to what we meant by neglecting the solid dynamics. The solid
moves so slowly that its velocity is not expected to play any role in the fluid dynamics.
In the general case, we have an interface between the fluid and the solid, which has a
deformation and a velocity. In the approximation that we just built, it has deformation
but no velocity, as depicted in Fig. 4.22.
Figure 4.22: Quasi-static aeroelasticity
The motion of the solid is, from the point of view of the fluid, quasi-static. We shall
say that the position of the interface is frozen in time.
With this in mind, in the general case the fluid dynamics and the solid dynamics
are coupled by the kinematic and dynamic conditions at the interface, and they evolve
simultaneously. But in our approximation we have two dynamics:
• one slow, which is that of the solid

• one fast, which is that of the fluid
The solid dynamics gives the position of the interface through a kinematic condition
that is considered as time independent for the fluid dynamics. This means that we are
back to the classical problem of fluid dynamics with a rigid boundary. Then, the solution
of the fluid dynamics gives a load at the interface. The solid dynamics gives the position
of the interface through a kinematic condition that is considered as time independent
for the fluid dynamics. This means that we are back to the classical problem of fluid
dynamics with a rigid boundary. Then, the solution of the fluid dynamics gives a load at
the interface. And with that we can compute the solid dynamics, and so on - the two
dynamics are coupled, but because we have a separation of timescales, we do not have
to solve them simultaneously. This is depicted schematically in Fig. 4.23.
Figure 4.23: Schematic coupling of the two dynamics
4.3.2 Flow-induced static instability
The question arises: if we use the simplest framework for the dynamics of the solid, that
of a single mode approximation, what will be the form of the fluid loading and what
would be the consequences on the dynamics of the solid?
In the single mode approximation, the position of the interface depends on the single
parameter q
ξ ( x, t) = D q(t) φ(t) (4.96)
Considering that the interface is fixed at the position defined by q, we have in the fluid
domain a steady-state problem with a boundary condition dependent on q as well
pe( D q)
(4.97)
e ( D q)
U
Therefore, the pressure and velocity in the fluid will depend on that value q, and so
will the fluid loading projected on the mode projected on the mode FFS . We may state
that the fluid loading at a given time will depend on the position of the interface at that
same time
ˆ
1 e te
FFS ( R E , D q, . . .) = CY − pe I + ∇U + ∇ U · n · φ dS (4.98)
∂Ω FS RE
This means that the projected fluid loading is actually proportional to the Cauchy
number CY . So, if we write the force as CY times a function of other parameters such as
the Reynolds number and the interface displacement q
FFS = CY F ( R E , D q, . . .) (4.99)
Then, for small motion we have D << 1, so we can expand the fluid loading as a
function of the small parameter D

0 ∂F
FFS = CY F + D CY q+... (4.100)
∂( D q)
The first term represents the permanent loading corresponding to the reference position
of q = 0. The next term is D times the fluctuation of the fluid loading force - it’s the
flow induced force resulting from the motion of the interface, so it is proportional to the
modal displacement on q. This is a flow induced stifness force as a spring force

∂F
f FS = D CY q (4.101)
∂( D q)
Now, from the point of view of the dynamics of the solid, the coupling to the flow is
identical to a coupling with a spring, as shown in Fig. 4.24.
Figure 4.24: Coupling of the solid with the fluid
We can represent the stifness k F of the spring as
f FS = −k F q
0 (4.102)
∂F
k F = −CY
∂q
It can be postive as well as negative; the sign just depends on the way the flow reacts
to the displacement of the boundary. Moreover, the magnitude of the stiffness depends
upon the flow velocity being proportional to the Cauchy number; and the Cauchy
number varies as U 2 . Therefore, we have a flow induced stiffness with a sign and a
magnitude that depends only on the fluid dynamics. So, it is a flow-induced stifness on
the dynamics of the solid.
The dynamics of the mode are governed by an oscillator equation that we obtained in
dimensionless form as
∂2 q
UR2 + q = f FS (4.103)
∂et2
But, if we are interested in the dynamics of the solid, we should now change the
referece time and use the dimensionless time t = t/TSolid , and we end up with a simple
oscillator equation for the quantity q(t)
q̈ + q = f FS (4.104)
Then, in the presence of flow, the oscillator equation is modified by the RHS force
f FS , which we just derived in Eq. 4.101. As this force is proportional to the modal
displacement q, we can incorporate it in the oscillator equation
q̈ + q − f FS = 0
0
∂F
q̈ + q − CY q=0
∂q (4.105)
" 0 #
∂F
q̈ + 1 − CY q=0
∂q
This means that in the presence of flow, we have a new oscillator with a total stiffness
that varies with flow velocity. Depending on the sign of the fluid induced stiffness, the
total stfness may decrease as the flow velocity is increased.
You can also notice that when
0
∂F
CY =1 (4.106)
∂q
all stiffness will be lost. At 0 stiffness, instability will occur because the frequency
becomes imaginary. This instability is often referred to as static instability in the sense
that only involves displacements terms, not inertia or velocity.
If we consider the response to a pertubation, at 0 flow velocity the solid oscillates.
Then, for velocities below the instability threshold, the pertubation also results in an
oscillation, but of smaller frequency. Lastly, above the threshold any pertubation will be
exponentially aplified in time. This is depicted in Fig. 4.25.
Figure 4.25: Response to a pertubation below, at and above the instability threshold
4.3.3 Flow-induced dynamic instability
There are some flow induced instabilities that behave as an oscillation of growing
amplitude, not just as an increase in displacement - that is, if the motions are in more
than one mode.
We will base our new model to analyze this problem on a richer approximation of the
kinematics of the solid, using a two modes approximation for the motion
ξ ( x, t) = D q1 (t) φ1 ( x ) + D q2 (t) φ2 ( x ) (4.107)
This two modes kinematics may be of several kinds; for instance, it could be a
combination of a rigid body motion in two different directions - one mode along the
x axis and another along the y axis, and the combination of the two modes gives any
position in space as shown in Fig. 4.26.
To compute the dynamics of the solid, we have to solve the modal equations of motion
with the fluid loading on each mode
(1)
m1 q¨1 + k1 q1 = f FS
(4.108)
(2)
m2 q¨2 + k2 q2 = f FS
Figure 4.26: Two modes approximation
This is identical to what we did in Sec. 4.3.2 with one mode, except that now we have
two modes and different modal masses and stiffness, with the same fluid loading to
be projected on the two modes. We have enriched the model of the kinematics of the
solid, but in terms of timescales we keep the approximation that these motions are much
slower than that of the fluid. We are still in quasistatic aeroelasticity.
Now, if we have two modes, the steady state fluid mechanics problem will depend
upon the instantaneous value of both q1 and q2 , as shown in Eq. 4.107. So, the resulting
fluid loading needs to be projected on the two modes
pe( D q1 , D q2 )
(4.109)
e ( D q1 , D q2 )
U
Therefore, the two modal forces will depend on q1 and q2 as

(1)
FFS = CY F1 ( R E , D q1 , D q2 , . . .) , projection on φ1
(4.110)
(2)
FFS = CY F2 ( R E , D q1 , D q2 , . . .) , projection on φ2
(i )
Let us now expand each of the forces FFS in terms of the small quantity D, the
amplitude of displacements, since we assumed it to be negligible
(0) (0)
(i ) (0) ∂Fi ∂Fi
FFS = CY Fi + D CY q1 + D CY q2 + . . . (4.111)
∂q1 ∂q2
(0)
The first term CY Fi represents the permanent loading. But, as before we are only
interested in the load that depends on the motion of the solid, and more particularly the
first coefficients of the expansion.
We define
∂Fi (0)

Kij = (4.112)
∂q j
and the modal equations now take the simple form
m1 q̈1 + k1 q1 = CY K11 q1 + CY K12 q2

(4.113)
m2 q̈2 + k2 q2 = CY K21 q1 + CY K22 q2
Now, we have a set of two equations that are coupled by a flow induced stiffness
term. These terms vary with the fluid velocity through the Cauchy number. Note
that the diagonal terms of the matrix corresond to the previous analysis where fluid
forces resulted in a modified stiffness of a mode, but what is new here is that we have
off-diagonal terms, and these induce a coupling between the two modes.
Consider first the effect of the diagonal terms CY K11 q1 and CY K22 q2 . Each of the two
modes will see its frequency change when the Cauchy number is varied. Depending on
the values of the flow induced stifness K11 and K22 , these two frequencies may part from
each other or may come closer; as depicted in Fig. 4.27.
Figure 4.27: Two modes coupled through flow-induced stiffness forces
In this case, we may have something called a coincidence of frequencies of the two modes.
There is a simple way to predict what happens to this dynamical system at coincidence.
We shall move to the LHS of Eq. 4.113 the diagonal terms
m1 q¨1 + (k1 − CY K11 ) q1 = CY K12 q2

(4.114)
m2 q¨2 + (k2 − CY K22 ) q2 = CY K21 q1
Since the frequencies are now equal, it will have thw two modal equations identical
and scaled at a frequency of (k1 − CY K11 ) = (k2 − CY K22 ) = 1. Then, let us assume that
the coupling stiffness is very small in comparison to the stifness of the mode; i.e. ε << 1.
So we can now consider two cases
• A symmetrical case, when both stiffness are equal
q̈1 + q1 = ε q2
(4.115)
q̈2 + q2 = ε q1
• An antisymmetrical case, when both stiffness are the opposite
q̈1 + q1 = ε q2
(4.116)
q̈2 + q2 = −ε q1
4.3.4 Symmetrical coupling
We look for the modes of the coupled system from Eq. 4.115 in the form of the real part
of
!
q1 q1
= Re exp(i ω t) (4.117)
q2 q2 0
By inserting this in the modal equations, the frequencies then need to satisfy the
following condition
−ω 2 q10 exp(i ω t) + q10 exp(i ω t) = εq20 exp(i ω t)

(4.118)
(−ω 2 + 1) q10 exp(i ω t) = ε q20

(4.119)
(−ω 2 + 1) q20 exp(i ω t) = ε q10
Which in matrix form results as

1 − ω2

−ε q1 0
2 exp(i ω t) = (4.120)
−ε 1−ω q2 0 0
which implies that the determinant is equal to 0
1 − ω2 −ε
=0 (4.121)
−ε 1 − ω2
And because ε << 1, we have two simple solutions to this

ε
ω ≃ 1± (4.122)
2
We shall call the two modes of the system A and B respectively, and we can express
them by
ε
ωA = 1 +
2

q1

1 (4.123)
=
q2 A −1
ε
ωB = 1 −
2

q1

1 (4.124)
=
q2 B 1
Note that both modes A and B have real frequencies, and the eigenvectors are combi-
nations of the original modes (1) and (2), and they are also real. This represents a weak
coupling that makes the system slightly different, with a slight difference of frequencies
between motions along the two eigenvectors.
4.3.5 Antisymmetrical coupling
Let us do the same derivation but now, we look for the modes of the coupled system
from Eq. 4.116. The modes we look for shall have the same real part
!
q1 q1
= Re exp(i ω t) (4.125)
q2 q2 0
By inserting this in the modal equations Eq. 4.116, the frequencies then need to satisfy
the following condition

(4.126)
−ω 2 q20 exp(i ω t) + q20 exp(i ω t) = −εq10 exp(i ω t)
(−ω 2 + 1) q10 exp(i ω t) = ε q20

(4.127)
(−ω 2 + 1) q20 exp(i ω t) = −ε q10
Which in matrix form results as

1 − ω2

−ε q1 0
2 exp(i ω t) = (4.128)
+ε 1−ω q2 0 0
So now, the determinant must satisfy that
1 − ω2 −ε
=0 (4.129)
+ε 1 − ω2
And because again ε << 1, we have now the following solutions

ε
ω ≃ 1±i (4.130)
2
Unlike the symetrical case, the frequencies now have an immaginary part. So, the two
modes of this system are the following, with complex frequencies
ε
ωA = 1 + i
2

q1

1 (4.131)
=
q2 A −i
ε
ωB = 1 − i
2

q1

1 (4.132)
=
q2 B i
Even the eigenvectors have an imaginary part. But, what does this represent?
Let us analyze the two modes separatedly, starting with the first one: mode A. If we
insert the solutions q1 and q2 into the condition of Eq. 4.133 we get
! !
q1 q1 1
= Re exp(i ω A t) = Re exp [i (1 + i ε/2) t] (4.133)
q2 q2 A −i A
Because both the eigenvectors and the frequency are complex quantities, we obtain a
q1 and q2 that oscillate as a cosine and sine that decrease exponentially in time

q1 cos t
= exp (−ε t/2) (4.134)
q2 sin t
This is a very different case from the previous one. If we start with an arbitrary
condition along the mode, the amplitude of q1 and q2 decrease with time. In the q1 − q2
plane, we have a decreased spiral, as sketched in Fig. 4.28. So we call this mode the
damped mode.
Figure 4.28: Damped mode in asymmetrical coupling
Now, let us analyze the second mode: mode B. We start with the same criterion
! !
q1 q1 1
= Re exp(i ω B t) = Re exp [i (1 − i ε/2) t] (4.135)
q2 q2 B i B
We obtain now a q1 and q2 that also oscillate as a cosine and sine, but they increase
exponentially in time

q1 cos t
= exp (ε t/2) (4.136)
q2 sin t
Therefore, this is an unstable mode. If we start with an initial condition on the q1 − q2

plane, we shall have an spiral growing in time as depicted in Fig. 4.29.
Figure 4.29: Unstable mode in asymmetrical coupling
So, these mode corresponds to what we shall call a dynamic instability. Dynamic as
opposed to static since there is here an exponential growth and oscillation. In the static
instability case, although we did have an exponential growth, there was no oscillation.
4.3.6 Conservative and non-conservative forcing
In the case of symmetric coupling, there is a potential
Φ = ε q1 q2 (4.137)
such that the RHS of the modal equations reads

∂Φ
q̈1 + q1 = ε q2 =
∂q1
(4.138)
∂Φ
q̈2 + q2 = ε q1 =
∂q2
4.4 coupling with a slow flow 169
This is what we call conservative forcing. Any cycle in the q1 − q2 plane is going to
produce a net zero energy transfer in the modes.
Conversely, for antisymmetric coupling, there is no potential. This is a non-conservative
forcing, with energy input or output at each cycle.
So what about our original case? We had expressed it as follows
m1 q¨1 + (k1 − CY K11 ) q1 = CY K12 q2

(4.139)
m2 q¨2 + (k2 − CY K22 ) q2 = CY K21 q1
The question arises: is it symmetrical or antisymmetrical? In general, it is neither of

the two, because the coefficients are CY K12 and CY K21 , and Kij tells the dependence
of the projected fluid force on the mode i to the displacement of the mode j. There is
absolutely no reason for them to be equal or opposite, but if the coupling is not exactly
symmetric, it contains a non-zero antisymmetric part and is therefore non-conservative.
So, as soon as they are not equal the system is non-conservative, and there will be an
unstable mode.
4.4 coupling with a slow flow
Up until now, we have obtained general results on the form of the force applied on
the solid as a consequence of its interactions with the fluid. For fast flows, or more
precisely for high reduced velocities UR , the force acting on the solid is a stiffness force
which stiffness depends on the flow velocity. We have seen that this may cause static
instabilities and even dynamic instabilities. But, to obtain such results, we had to make
a very strong assumption on the flow velocity: it had to be so high that the velocity of
the fluid-solid interface could be comparatively neglected. This was the assumption of
quasi-static aeroelasticity.
But, there are cases where the velocity does not satisfy such a condition.
4.4.1 Pseudo-static aeroelasticity
In terms of the reduced velocity UR = TSolid/TFluid , we want to explore cases where UR is

neither small nor large. This means that the two timescales for the solid and fluid should
be of the same order of magnitude.
Using the same ideas than in the last section, if we compare the evolution in time of
two quantities, one pertaining to the dynamics of the solid and one to the dynamics of
the fluid; if the TSolid is very large in comparison with TFluid , we in the case of a fast flow
or quasi-static velocity. Then, if TSolid is larger than TFluid , but not as large, we may be in
the domain we want to explore now. This is illustrated in Fig. 4.30.
Let us now have a closer look at the boundary conditions at the interface. We have
already estimated in the previous section the order of magnitude of the two types of
boundary conditions on the fluid. At the interface, the dimensionless velocity is of the
order of

Ue =O D (4.140)
UR
Figure 4.30: Intermediate reduced velocities in quantities pertaining to the solid and fluid over
time
On the other boundaries, it is of the order of

e = O (1)
U (4.141)
But, contrary to the analysis for a fast flow, now neither of them can be neglected. We
shall analyze the magnitude of the variation of the interface conditions, the acceleration,
which is the variation in time of the velocity
∂U
e ∂2 ξ
e=
γ = 2 (4.142)
∂et ∂t
We know that ξ = O( D ), and it varies over a timescale of UR . Then, we can say that
the order of
∂2 ξ

D
2
= O (4.143)
∂t UR2

D
So the acceleration is of the order of γ = O UR2
. Then, the dynamics of the fluid are
governed on one hand by a condition of order U e = O(1) and in the other by a condition
of order Ue = O ( D/UR ) at the interface that varies in time. This is, during a unit time
interval ∆et = 1, its variation is of the order of ∆U
e=γ e · ∆et = O ( D/UR2 ).
Now certainly, if UR2 >> D, the variation can be set to zero ∆U e = 0 without changing
much the results. This corresponds exactly to what we meant by neglecting the variation
of the solid dynamics. The solid moves so slowsly that its velocity can be considered
constant in time in terms of the fluid dynamics. This is a bit more complicated than the
very strong approximation we made in quasi-static aeroelasticity. Here, we do not ask
the solid to be fixed, but rather we ask for the solid to have a fixed velocity.
Figure 4.31: Comparison between the cases

So, in the general case, we have an interface within the fluid and the solid which has
a deformation and a velocity. In the quasi-static aeroelasticity approximation, it has
a deformation but no velocity - we say then that the deformation is frozen in time in
comparison with the fluid dynamics. Now, we have defined what is called the pseudo-
static aeroelasticity approximation where the interface has a deformation and a velocity,
but the velocity is frozen in time. This is shown schematically in Fig. 4.31.
In pseudo-static aeroelasticity, we still have two dynamics, one slow and one fast. The
solid dynamics give the position and the velocity of the interface through a kinematic
condition that is considered as time independent for the free dynamics. This means that
in the fluid, we are back again to the classical problem of fluid mechanics with time
independent boundary conditions, as depicted in Fig. 4.32. Then, the solution of the
fluid dynamics gives the load at the interface. And with that, we can compute the solid
dynamics, and so on - the dynamics are coupled, but because we have a separation of
timescales, we do not have to solve them simulataneously.
Figure 4.32: Pseudo-static aeroelasticity approximation
4.4.2 Flow-induced dynamic instability
We are now trying to develop models in an intermediate range of reduced velocities,

where the dynamics of the solid is coupled to a flow that is not that fast.
We will start by considering the most simple framework for the dynamics of the solid:
that of a single mode approximation, where the position of the interfaces depends upon
a single parameter, q
ξ ( x, t) = D q(t) φ( x ) (4.144)
Considering that the interface is fixed by q, and with a velocity q̇, we have in the fluid
domain a steady-state problem for the boundary conditions that is dependent on both q
and q̇ as well - the pressure and velocity in the fluid will depend upon this values
pe ( D q, D q̇)
(4.145)
U
e ( D q, D q̇)
and so will the fluid loading projected on the mode FFS
FFS ( R E , D q, D q̇) (4.146)
By this, we may state that the fluid loading on the solid at a given time, will depen on
the position and velocity of the interface at the same time.
The dimensionless form of the force exerted by the fluid on the solid projected on the
mode could be written as
ˆ
1 e
FFS = CY − pe I + ∇U + ∇ t U
e · n · φ dS (4.147)
∂Ω FS RE
It is clear that the force is actually proportional to the Cauchy number CY , so we can
write it as the Cauchy number times a function of the other parameters
FFS = CY F ( R E , D q, D q̇, . . .) (4.148)
And since we are considering small motion, D << 1, we may expand the fluid loading
as a function of the small parameter D
(0)
FFS = FFS + D f FS + . . . (4.149)
where
(0)
FFS permanent loading corresponding to the reference position q = 0
f FS is the
flow-induced force D f FS fluctiation of the fluid loading force
resulting from the This flow induced force f FS can be written as
motion of the
interface
∂F ∂F
f FS = CY q + CY q̇ (4.150)
∂q (0) ∂q̇ (0)
The first term proportional to the modal displacement q is the flow-induced stiffness
force that we already obtained, but it also contains a new term highlighted in Eq. 4.150
that is proportional to q̇. This is a new flow-induced damping force.
From the point of view of the dynamics of the solid, the coupling to the flow is
equivalent to a coupling with a spring plus a damper, as depicted in Fig. 4.33.
Figure 4.33: Flow-induced stiffness and damping from the point of view of the solid
And the characteristics of this spring and damper depend on the fluid mechanics in
two aspects
1. They increase in magnitude through the Cauchy number, which is proportional to

the square of the fluid velocity.
These are the ∂F/∂q 2. The spring and damper coefficients depend on the sensitivity of the fluid loading to
and ∂F/∂q̇ terms the deformation and velocity of the interface.
Such a fluid induced damping has consequences on the dynamics of the solid. To
analyze that, we need to begin from the dynamics of our mode, which is governed by
the dimensionless oscillator equation
∂2 q
UR2 + q = f FS (4.151)
∂et2
But now, we have a model for the fluid force f FS that applies on it, given by Eq. 4.150.
Let us simply incorporate this force in the dimensionless forme of the oscillator equation
∂2 q

∂F ∂F
UR2 + q = CY q + CY q̇ (4.152)
∂et2 ∂q (0) ∂q̇ (0)
If we are interested in the dynamics of the solid, we should now change the reference
time and use the dimensionless time t based on TSolid

∂F CY ∂F
q̈ + q = CY q+ q̇ (4.153)
∂q (0) UR ∂q̇ (0)
The equation keeps the same form, but the original oscillator part changes since the Remember that TSolid
time differntiation has changed. depends upon UR
Now, since the fluid force contains a stiffness term and a damping term, we may just
incorporate them in the classical oscillator equation
q̈ + c F q̇ + (1 + k F ) q = 0 (4.154)
where

CY ∂F
cF = −
UR ∂q̇ (0)
(4.155)
∂F
k F = − CY
∂q (0)
This is, in the presence of flow, we have a new oscillator with damping c F and stiffness
k F that vary with the flow velocity.
4.4.3 The damping coefficient
First of all, there is no reason for the damping coefficient c F to be possitive or negative.
• If it is positive, then the flow is going to damp the free motion of the solid.
• Conversely, if it is negative, any perturbation will be exponentially amplified in
time.
The above is shown in Fig. 4.34.
Figure 4.34: Damped and unstable conditions due to the damping coefficient
So, it is clear that when c F > 0, we have a dynamic instability. We already found
dynamics instabilities in Chap. 4.3, but there is a major difference of mechanism here:
• In the instability by coincidence and non-symmetric coupling that we analyzed in

4.3, the first effect of the flow was to bring the frequencies of two modes together,
and then a combined mode with negative damping appeared.
In practice, this • But, in the dynamic instability we just obtained, the mode is unstable itself as soon as
instability will set on the fluid force acts.
only then the total
damping of the mode So, if there is an initial mechanincal damping c0 , then the total damping will decrease
reaches zero and
under the effect of flow and reach zero at the critical value of the reduced velocity URCRIT ,
becomes negative
as shown in Fig. 4.35.
Figure 4.35: Total damping as a function of the reduced velocity UR
This critical reduced velocity value may be predicted as a function of the initial
damping c0 , of the mass number M anf of the sensibility of the fluid loading to a velocity
at the interface
c
URCRIT = 0 (4.156)
M ∂F∂q̇ (0)
We found out that if the velocity of the solid at the interface cannot be neglected,
there is a possibilty that the vibration of the solid negatively damped and the negative
dampign induced by the flow wouyld then increase with the flow velocity. The dynamic
instability caused by this mechanism seems much more dangerous than the one we
found about in Chap. 4.3 because
• it does not require a coincidence of two modes and,

Any simple single • does not even require that the solid has a kinematic condition that is based on a
mode motion may be two mode approximation
unstable
4.5 coupling with any type of flow
Using our dimensionless parameters, particularly the reduced velocity, we have seen that
we could build simple models for the interactions between a fluid and a solid. These
models were aplicable for specif range of these dimensionless parameters - they are
patches of models.
Now, we are going to analyze the case for systems that see a progressive increase of
the fluid velocity from zero to very large. In other terms, we will study how to jump
from one model to another.
4.5 coupling with any type of flow 175
4.5.1 The garden hose instability
It represents the problem of a flexible pipe that flutters when some flow velocity is put
in it, as depicted in Fig. 4.36. We cannot address this problem in the general case - this is
why we had to make models in parts.
Figure 4.36: The fluid-conveying pipe instability
Consider a straight pipe with flow inside. Depending on the velocity,
• At zero flow velocity, the perturbation, just pulling and releasing, results in a
damped motion of the pipe, more or less damped depending on the pipe material.
• If the flow velocity is now increased, the same perturbation results in an oscillatory
motion of similar form, but more damped.
• Then, as the flow velocity is further increased, this trend is reversed and eventually,
a situation of zero damping arises.
• Above this critical flow velocity, the pipe starts to oscillate by itself. A true dynamic
instability in the sense we defined in the previous Chapters.
Figure 4.37: Garden hose damping depending on the reduced velocity
The above is shown schematically in Fig. 4.37. If we want to see it as damping in

function of the reduced velocity UR , as the velocity is increased in the pipe, we have
an increase of damping, then a decrease and eventually an instability. Then, above the
critical velocity, the amplitude grows until a regular self-sustained motion sets on.
In order to relate all these motions to the models we have built in previous Chapters,
we have to start by defining our dimensionless parameters.
Clearly, this mass We can define the mass number M as the ratio of fluid mass over solid mass by unit
number will be length of the pipe
different for a gas or
liquid conveying ρS
pipe M= (4.157)
m
Then, we have to define the reduced velocity. Quite naturally, it may be defined in
terms of the ratio of times
TSolid
UR = (4.158)
TFluid
The time characteristics of the solid dynamics are, for instance, the period of oscillation
of the first mode without fluid. Then, TSolid = 1/ f . The time characterizing the fluid
dynamics can be defined, for instance, as the time a fluid particle takes to go from the
entrance to the exit of the pipe. Then, TFluid = L/U , where L is the length of the pipe and
U is a mean flow velocity in the pipe section. So, with this in mind, the reduced velocity
can be written as
U
UR = (4.159)
fL
Another key aspect of the models we developed so far was the level of approximation
of the dynamics of the solid. We used a single mode approximation and in some cases two
modes approximations. Or even, a combination of the two modes. For the bending pipe,
this is shown schematically in Fig. 4.38.
Figure 4.38: Modes of the bending pipe
So, if we would now like to use our models of fluid-solid interactions to predict what
happens on a fluid-conveying pipe for a large range of reduced velocities, we can start
by the simplest model - the fluid-conveying pendulum.
4.5.2 The fluid-conveying pendulum
A single mode approximation of the dynamics of the pipe would follow the first mode
of bending as shown in the right-most figure of Fig. 4.38. We can model this by lumping
all the flexibility at on ened with a rotation spring of stiffness C. And then, we can put
all the mass on the other end as a point-mass m. This is shown in Fig. 4.39.
This way, we have a simple mass-spring system. A pendulum, but not stiffened by
gravity, just by the rotational spring. The position of the pendulum is given by the
inclination angle Θ.
Figure 4.39: Modeling of the first mode bending as a fluid-conveying pendulum
4.5.3 Analysis of the solid alone
Without fluid, the equation of motion of this pendulum would be just
m L2 Θ̈ + C Θ = 0 (4.160)
We must go dimensionless now. We can define a dimensionless time based on the

frequency of the pendulum
t
t= √ (4.161)
L m/C
Then, we can use an angle parameter of order one that equals
Θ(t)
θ (t) = (4.162)
Θ0
where Θ0 is the magnitude of angular motion. This is just like the q(t) that we have
been using so many times. So, the dimensionless equation becomes
θ̈ + θ = 0 (4.163)
4.5.4 Analysis of the fluid alone
Let us consider that there is flow of uniform velocity U and density ρ that enters the
pendulum at the fixed end and exits at the free end, as depicted in Fig. 4.40.
Figure 4.40: Fluid flow in the fluid conveying pendulum

Let S be the pipe section where the fluid flows. We can define a mass number M as
the ratio between the fluid mass and the pendulum mass
ρSL
M= (4.164)
m
Then, we can define the reduced velocity UR as teh ratio between the period of
√
oscillation, L/ C/m and the time of convection along the pipe L/U
TSolid L 1 U
UR = =√ = √ (4.165)
TFluid C/m L/U C/m
4.5.5 Small reduced velocity
We have a fluid-solid system whereby a straight pipe that conveys the fluid at a velocity
UR oscillates about its upper support. So, we have to analyze what is the effect of the flow
velocity. We shall start by asking ourselves what happens at small reduced velocities.
So, we neglect the fluid-flowing velocity. The fluid is just entrained b the motion of
the pipe. As we have seen in previous Chapters, we will have an added mass effect. But,
because we are considering rotation around the axis of the pendulum, this added mass
effect will actually be an added moment of inertia.
In the dimensionless form, this added moment of inertia reads
ρ S L3
I= (4.166)
3
Therefore, the dimensionless added mass is M/3, and the equation of the fluid pendu-
lum for very small reduced velocities reads

M
1+ θ̈ + θ = 0 (4.167)
3
The fluid just brings
an added inertia
Because we have neglected the effect of the fluid velocity, it is not surprising that we
do not see any added damping, a phenomenon that clearly depends on the fluid velocity.
For this, we have to go to the other limit, that of a high reduced velocity.
4.5.6 Large reduced velocity
In the limit which we called quasi-static aeroelasticity, we consider the deformed solid
interface, but not the velocity at the interface. So, we have a deformed pipe, but the
angle of deformation θ is frozen in time. The fluid just goes through the pipe.
Of course, this does not bring any torque - it just goes straight through. In that case,
there is no effect of the fluid and the system equation is still
θ̈ + θ = 0 (4.168)
We do not see any effect of the flow velocity either.

4.5.7 Intermediate reduced velocity
For intermediate reduced velocity, we developed an approximation, that of pseudo-static

aeroelasticity. In that case, the solid velocity was taken into account, but as a constant in
time. This means that we have a pipe rotating at a constant rate, θ̇, and a fluid going
through it. In other words, θ̇ is frozen in time. Then, what is the effect of a flowing fluid
on a rotating pipe, as depicted in Fig. 4.41a?
(b) Mass balance of the

fluid
(a) Representation of the fluid
Figure 4.41: Fluid through a rotating pipe
The fluid must satisfy the mass balance, so that its value all along the axis of the pipe
is UR . So, the fluid velocity has a velocity component along the axis of the pipe t that We have assumed
reads UR t. Its value on the direction perpendicular to the pipe must be that of the pipe that the flow is
uniform in the
walls. This is, the velocity of the pipe walls depend on the distance to the axis of rotation,
cross-section for
say x. Then, it reads x θ̇ n. This is shown schematically in Fig. 4.41b. simplicity
So we can write the fluid velocity as
U = UR t + x θ̇ n (4.169)
Now, we need the torque T exerted by the fluid on the solid, becuase we are interested
in its rotation. We can use the balance of angular momentum to get the torque. So, the
anguar momentum of a fluid particle will be the vector product of the distance to the
rotation axis x, and fluid momentum M U
r = x×mU (4.170)
The fluid velocity is UR along the axis of the pipe, but this does not count in the vector
product, since it is perpendicular to x. Across the pipe, the velocity across the pipe is x θ̇;
so, the angular momentum is
r = x × m x θ̇ n → r = x2 M θ̇ (4.171)
Using the angular momentum theorem, we can tell that the torque TSF exerted on the
fluid domain by the solid equals the variation in time of the total momentum plus the
flux of momentum over the boundary Do not mistake the
ˆ ˆ nomenclature T used
∂ for the torque and T
x2 M θ̇ dV + x2 M θ̇ (U · n) dS

TSF = (4.172)
∂t Fluid domain Fluid boundaries used for period
We can see that the first term of the RHS of Eq. 4.172, the variation in time of the
total momentum, will be zero. This is because θ̇ is a constant in our model, frozen in
time. Then, with regards to the second term, we consider the fluxes. At the entrance,
the angular momentum of the fluid is zero - so the flux is 0; i.e., the fluxes on the sides
cancel out. At the exit, the angular momentum is M θ̇ UR , so the torque can be simply
written as
TSF = M θ̇ UR (4.173)
This is the torque exerted by the solid on the fluid. So, the torque exerted by the fluid
on the solid is just the opposite
TFS = − TSF = − M θ̇ UR (4.174)
As a consequence, we can write the motion of the fluid conveying pendulum as
θ̈ + M UR θ̇ + θ = 0 (4.175)
So we have an oscillating pipe that is dampened by the flow. And the damping
increases with the velocity, as we have shown in Fig. 4.37. This is due to the highlighted
term in Eq. 4.175; and it is called the Coriolis damping, because it originates in the angular
momentum balance. It is the reorientation of the fluid momentum that requires a torque
proportional to the fluid velocity. That damps the solid motion.
4.5.8 A model for all fluid velocities
In Eq. 4.172, we used the Angular Momentum Theorem to derive the torque. We said
that the first term was the time dependence of the total fluid angular momentum, and
because we were under the approximation of pseudo-static aeroelasticity, the angular
velocity θ̇ was frozen in time and so the term was equal to zero.
If now the velocity is time dependent, which is the general case, then this term would
read
ˆ
˙ M
For any velocity θ (t) → x2 M θ̇ dV = θ̈ (4.176)
3
We now have the most general form of the torque exerted by the solid on the fluid
M
TSF = θ̈ + M UR θ̇ (4.177)
3
If we include this in the equation of the fluid pendulum, we recover added mass and
added damping at the same time

M
1+ θ̈ + M UR θ̇ + θ = 0 (4.178)
3
And this is valid for all reduced velocities UR . This means that added mass not only
exists at very small reduced velocities, but at all reduced velocities. The same goes for
added damping, it exists at all reduced velocities, not only at intermediate ones. For our
fluid conveying pendulum, the simple models of each range of velocity were right, but
they could not give us more than the dominant effect, let it be added mass or added
Figure 4.42: Models for fluid-solid interactions
damping. The full model combines all of this: effects add up. This is shown schematically
in Fig. 4.42.
The fluid-solid interactions effects do not replace one by the other. At low reduced
velocities, all effects were independent of the reduced velocity. In presudo-static aeroelas-
ticity, the fluid induced damping effects that gave us galloping were proportional to the
reduced velocity UR . And, in quasi-static aeroelasticity, the fluid induced stiffness effect
that gave us divergence or flutter, were proportional to the Cauchy number that varies
like the square of the reduced velocity UR2 . So, depending on the range of reduced veloc-
ity, each type of effect domainates. At low reduced velocities, added mass dominates; at
intermediate velociteis, added damping and at higher reduced velocities flow-induced
stiffness.
4.5.9 The fluid-conveying bi-pendulum
We have found that in the fluid-conveyin pendulum the sinble mode approximation was
enough to give us the added mass and the fluid induced Coriolis damping, but that does
not explain the famous garden-hose instability. In the fluid conveying pendulum, the
higher the velocity, the higher the damping . So, we have to enrich the model - and the The effect is only
only direction of improvement is to have a better approximation of the dynamics of the stabilizing
pipe as a combination of two modes. We are going to work with a bi-pendulum model,
as depicted in Fig. 4.43.
Figure 4.43: Fluid-conveying bi-pendulum model
Without the fluid dynamics, the bi-pendulum deformation will be defined by two
angles: θ and φ. Using the same dimensionless as we used for the standard pendulum,
the dynamics of the solid will be governed by two equations that couple the evolutions
in time of θ and φ
2 θ̈ + φ̈ + 2 θ − φ = 0
(4.179)
θ̈ + φ̈ − θ + φ = 0
So we have two degrees of freedom which evolutions are coupled. Solving the solid
dynamics, we can find out that
√
• The first mode has a frequency ω1 = 2 − 1, and the motion is quite pendular with
the two masses moving in the same direction.
√
• The second mode has a higher frequency ω2 = 2 + 1, and the shape shows the two
masses moving in opposite direction.
Figure 4.44: Modes of motion of the bi-pendulum
This is, the first mode has the two masses in phase, and the second mode out of
phase. This is shown in Fig. 4.44. We can write the state of the bi-pendulum using
a combination of the two modes with the modal variables q1 and q2 as we have done
in previous Chapters so that q1 and q2 satisfy the simple uncoupled modal oscillator
equations

θ (t) 1
√ 1
√
= q1 ( t ) + q2 ( t )
φ(t) 2 − 2
2 (4.180)
q̈1 + ω1 q1 = 0
q̈2 + ω22 q2 = 0
Now that we have the model for the dynamics of the solid, let us include the interaction
with the fluid. We can analyze the fluid-conveying bi-pendulum in a deformation state
frozen in time, as depicted in Fig. 4.45, making use of the pseudo-static aeroelasticity
model for large reduced velocities. We are using this pseudo-static aeroelasticity case
becuase we already had models for the motions of two-mode solid, which means that
Figure 4.45: Fluid-conveying bi-pendulum at very large reduced velocity


θ
, are frozen in time (4.181)
φ
We can analyze the fluid loading on the solid in this geometry with a reduced velocity
UR a the classical problem of flow in a pipe with an elbow. Using the fluid momentum
balance, the magnitude of the force exerted by the flow on an elbow is
F = ρ S U 2 (θ − φ) , in dimensional form (4.182)
In order to make the force equation Eq. 4.182 dimensionless, we can define the Cauchy
number based on the stiffness of the springs
ρ S U2 L
CY = (4.183)
C
And the dimensionaless force reads simply
F = CY (θ − φ) (4.184)
Then, what is the effect of this force on the dynamics of the pipe? To answer this, we
have to project the same force on the two modes. Because F depends on both θ and φ, it
shall also depend on q1 and q2 including new terms, depending on the Cauchy number
√ √
2 2−1 2+1
q̈1 + ω1 q1 = −CY √ q1 + CY √ q2
4+2 2 4+2 2
√ √ (4.185)
2 2−1 2+1
q̈2 + ω2 q2 = −CY √ q1 + CY √ q2
4+2 2 4+2 2
On the LHS of Eq. 4.185 we have the modal oscillator terms for q1 and q2 . Then, on the
RHS, all the terms are proportional to the Cauchy number as we expected. Because there
are terms in q1 and q2 in both equations, we can put on the LHS the stiffness terms of
each equations, and we get something very similar to the cases that we have seen before
√ ! √
2−1 2−1
q̈1 + ω1 + CY √ q1 = CY √ q2
4+2 2 4+2 2
√ ! √ (4.186)
2−1 2−1
q̈2 + ω2 + CY √ q2 = −CY √ q1
4+2 2 4+2 2
On the first mode, we have a stiffness that is going to increase with the Cauchy number.
Conversely, on the second mode we have a stiffness that it is going to decrease, and we
have a coupling between the two modes that is not symmetric at all (notice the negative
sign in the RHS of the equation). This is going to result in an instability with a merging
of the two frequencies and non-symmetric coupling.
We can compute the modes of the system as a function of the Cauchy number, as
depicted in Fig. 4.46. At zero flow velocity, we have two modes with frequencies ω1
and ω2 as before. Then, as the flow velocity is increased, the frequencie come closer and
closer. When they meet, the couplings give two modes, one stable and one unstable.
Figure 4.46: Modes in terms of the Cauchy number
Figure 4.47: Motion of the unstable mode
The unstable mode has an out of phase motion, as depicted in Fig. 4.47. And it is
unstable because over one cycle, the force exerted by the flow on the elbow brings a
positive work.
4.5.10 Vortex induced vibrations
The case of vortex induced vibrations (VIV) is very different from the phenomenons we
have been analyzing up until now. Consider a flexible system that oscillates under flow,
as shown in Fig. 4.48. The vibration is simultaneous with the shedding of vortices from
the solid boundary into the flow.
This phenomenon only occurs in a limited range of reduced velocities, say from 5 to
10, for example; and then disappears at higher flow velocities. This does not look like the
instabilities we have found. For example, when an airfoil or a garden hose are unstable,
increasing the flow velocity made things worse.
So it is clear that the phenomenon of VIV is different. It seems related to some internal
dynamics of the fluid - here, the shedding of vortices. But, so far we have considered
only one time scale in the fluid domain - that is, TFluid = L/U , the time corresponding to
the convection of a particle of velocity U across the length L. In VIV, we have another
time scale; let us call it TVortex , because the oscillations correspond to the shedding of
vortices behind the cyllinder.
If there is another time scale for the fluid, then all our models which are based upon
the comparison between UR = TSolid/TFluid are just not relevant.
Figure 4.48: VIV representation
Figure 4.49: Time scale of vortex shedding
We know that the time scale of vortex shedding, TVortex , can be estimated from the
oscillation of a quantity such as pressure in the wake. On an instantaneous image of the
flow, we can also estimate it from the wavelengths of the pattern of vortices. Both of this
cases are depicted in Fig. 4.49.
This new timescale obeys what is called the Strouhal Law
1
TVortex = TFluid , Strouhal Law (4.187)
S
This is, TVortex is proportional to TFluid by the inverse of the Strouhal number S. Now,
remember that the reduced velocity was built to compare the dynamics of the fluid
and of the solid, UR = TSolid/TFluid . When UR = 1, the time of convection and the time of
oscillation of a solid were the same TSolid = TFluid . Now, using the Strouhal law, we can
make a comparison between TSolid and TVortex
TSolid T T
= 1 Solid = S Solid = S UR (4.188)
TVortex /S TFluid TFluid
This means that when S UR = 1, the shedding of vortices and the motion of the solid
happen at the same time scale. So, something is hapening here between the vortices
and the motion of the solid - and we have oscillations in the flow at a period TVortex .
Certainly, this shall cause oscillation of the lift on the cylinder at the same period. If that
period of forcing is equal to the period of free oscillation of the cylinder, we might have
a resonance.
Let FVortex be that oscillating lift caused by vortex shedding on the cylinder per unit
time; we can write it as

1 2 t
FVortex = ρ U L Cl sin 2π (4.189)
2 TVortex
This force oscillates at the frequency of vortex shedding. Its magnitude is defined by
Cl , called the fluctuating lift coefficient. Now, if the cylinder is a mass-spring system
that is allowed to move in the lift direction, as depicted in Fig. 4.50, we have a forced
oscillator equation.
Figure 4.50: Cylinder as a mass-spring system allowed to move in the lift direction
We can write this forced oscillator equation as

1 2 t
m ÿ + k y = ρ U L Cl sin 2π (4.190)
2 TVortex
where the LHS represents the dynamics of teh cylinder and the RHS the forcing by
the wake. But, we know how an oscillator responds to a sinusoidal forcing
M Cl UR2
Y= (4.191)
2 π 3 1 − S2 UR2

such that Y = y/L and M = (ρ π L2 )/(4 m). Fig. 4.51 shows the response curve. We
If we have damping, can see that, at the reduced velocity value of 1/S we have a resonance. But there is also
the resonance is something special about this resonance curve: it does not decrease after the resonance, it
limited in amplitude
stays on a plateau. This is because if you increase the velocity, you increase the frequency
of the forcing, but also the magnitude of the forcing.
Figure 4.51: Response curve in dimensionless of the forced oscillator

4.5.11 Lock-in phenomenon
We saw that the frequency of vortex shedding f Vortex is proportional to the reduced
velocity UR by the Strouhal law
TSolid
f Vortex = = S UR (4.192)
TVortex
When the cylinder is let free to move under the force caused by the wake, the
frequency of vortex shedding deviates from this, and very significantly. Fig. 4.52 shows
this schematically.
Figure 4.52: Lock-in of the frequency of vortex shedding
If we increase our flow velocity above the point wher there is resonance, the frequency
of shedding is said to lock-in the frequency of oscillation of the cylinder. It very strongly
deviates from the Strouhal law, until eventually at higher reduced velocities, it jumps
back to the Strouhal law.
This lock-in phenomenon results in something like an extended resonance. The wake
continues to excite the cylinder at its own frequency, even at higher reduced velocities.
Instead of having VIV only when the frequencies match, we have them over a wide
range. The amplitude of motion remains high even after the pure resonance condition is
satisfied. Fig. 4.53 represents this.
Figure 4.53: Lock-in of the frequency of vortex shedding
The basic model of VIV is based upon a lift force that oscillates at the frequency of the
Strouhal law
1
FVortex = ρ U 2 L Cl sin (S UR t) (4.193)
2
Then, this force is applied on the oscillating cylinder, and if resonance occurs, the
amplitude is of the motion is large. But here, by construction, the Strouhal law is
respected and cannot be lock-in. So we have to improve our model.
We have to keep in mind that the cause of oscillation of the lift is an oscillatory
dynamics that sets on in the flow. It is a result of what is called hydrodynamic instability.
The uniform flow solution is not stable anymore and an oscillating solution emerges.
So it seems reasonable to take into account somehow that the lift force is actually the
consequence of something that oscillates - something that satisfies an oscillator equation.
What we are looking for is a way to take into account the feedback from the solid to the
wake. This means that we want to take into account the displacement Y in the equations
that govern the lift F. To do this, we have to analyze how the wake responds to a motion
of the cylinder that causes the wake itself. People have done plenty of experiments on
wakes behind a moving cylinder, and what they have found out is that the wakes are
sensitive to the acceleration of the cylinder. The forcing term is actually proportional to Ÿ
F̈ + (S UR )2 F = A Ÿ
1 (4.194)
Ÿ + Y = M UR2 Cl F
2
This set of equations is non-linear, so there is no way that it is going to give us

an amplitude of motion or an amplitude of the force. But, we can use it to try and
understand lock-in, which is just a question of frequencies.
To know the frequencies, we just have to compute the modes. Let us look for solutions
in the forms of

Y Y
= 0 exp(i ω t) (4.195)
F F0
Because we have two degrees of freedom, we have two modes, and therefore two
frequencies. And for each value of UR , we will have a different set of frequencies. Fig.
4.54 shows schematially the result of this computations.
Figure 4.54: Coupled model of vortex-induced vibration
At S UR = 1, the frequency of shedding is expected to be equal to that of the cylinder.

But, in this coupled model, at zero reduced velocity we have two frequencies - one is
that of the solid and other zero because the frequency of shedding is zero with no flow.
With a non-zero flow velocity, the frequency of wake is close to the Strouhal law. But, as
the reduced velocity is increased further, we can see two frequencies comign closer and
closer until they merge and there is only one frequency, and then they split again. The
wake frequency goes up and back to the Strouhal law.
What happens then in the yellow zone? Fig. 4.55 represents that
Figure 4.55: Growth rate of the modes
We can see that the growth rate of the modes is opposite to the imaginary part of
the frequencies. It tells us if a mode is damped or negative, neutral or unstable when
positive. For low and high reduced velocity, they are equal to zero. But, in the range
we have identified before, one of the modes has a positive growth rate, and one has a
negative one. This tells us we defenitively have an unstable coupled mode.
So we have coupling between a solid mode and a wake mode. And these coupling
bring some instability in a limited range of reduced velocity. But, what is very special is
that instead of havin gtwo solid modes being coupled by the flow, we have one solid
mode coupled to one fluid mode - the unstable hydrodynamic mode.
MULCAHY’S LEAKEAGE FLOW MODEL FOR A SINGLE PIPE WITH
5
H O R I Z O N TA L WA L L S
5.1 Mathematical model 191
Given Mulcahy’s study on the geometry depicted in Figure 5.1, we shall analyze the
leakage flow-instability that occurs in this configuration. The geometry consists of a
single pipe of length l1 and variable height s(t), through which fluid passes. Aditionally,
two restrictions with heights s I and sO are located at the inlet and outlet, respectively.
sO
sI s(t)
L
x
Figure 5.1: Scheme of Mulcahy’s geometry for a single pipe with horizontal walls
At this point, no rotation of the walls is permited - this problem shall be analyzed in
the next chapter. Thus, no abrupt changes in the geometry are considered.
5.1 mathematical model
Our objective will be to deduce three partial differential equations for the system
variables, i.e. the fluid flow Q, the channel size s and the pressure p. This equations
shall arise from
• The mass balance

• The momentum balance
• The solid equation of motion
5.1.1 Mass balance
The differential form of the mass balance is given by Equation 2.66 for an incompressible
fluid
∂u j
ρ =0 (5.1)
∂x j
191
192 mulcahy’s leakeage flow model for a single pipe with horizontal walls
However, it is worth noting that the equation above is derived considering a differential
control volume V where the change in size of V is not taken into consideration. Therefore,
we must recall the integral form of the mass balance in a control volume, given by
Equation 2.69
ˆ ˆ
∂
ρ dV + ρ (n · u) dS = 0 (5.2)
V ∂t S
where S is a surface associated to the control volume V , and n is a normal vector to S .

We can apply Gauss’ integral theorem to rewrite Equation 5.2 in the form of a volume
integral as
ˆ ˆ
∂
dV + ∇ · u dV = 0 (5.3)
V ∂t V
where we have lost the term ρ since it is constant and it can be taken out of the
integrals; then, dividing all terms by ρ we reach the equation above.
Then, we can solve the first integral as
ˆ
∂V
+ ∇ · u dV = 0 (5.4)
∂t V
or in index notation
ˆ
∂V ∂ui
+ dV = 0 (5.5)
∂t V ∂x j
5.1.2 Momentum balance
For an incompressible fluid with uniform viscocity, and using the mass balance equation,
the momentum balance can be written in the form of Equation 2.145; i.e.
Dui ∂p ∂τij
ρ =− + + fi (5.6)
Dt ∂xi ∂x j
where D/Dt is the Lagrangian derivative given by Equation 2.25, τ is the stress tensor
and f i represent the body forces. The only body force that we shall take into account in
this analysis is the gravitational force, so we may rewrite Equation 5.6 in the form
Dui ∂p ∂τij
ρ =− + + ρ gi (5.7)
Dt ∂xi ∂x j
so that we consider the gravity gi in the downwards vertical direction; i.e. g1 = 0 and
g2 = − g, where g is the gravitational acceleration.
5.1.3 Kinematic assumptions
The channel walls are allowed to oscillate in the vertical direction with the following law
s = s ( xi ) + ε e
s(t) (5.8)
5.1 mathematical model 193
where s denotes the mean value of the wall displacement, and es is the wall fluctuation.
Then, we introduce a small dimensionless parameter ε << 1 to restrain the fluctuations
of the channel in a different order of magnitude than its mean value.
We may write the rest of the kinematics in this form
ui = ui ( xi ) + ε uei ( xi , t)
(5.9)
p = p( xi ) + ε pe( xi , t)
where we shall consider a bidimensional flow, so that xi = ( x1 , x2 ).

In order to reach conclusions regarding the fundamental equations, we shall make use
of the tool of dimensional analysis. We may scale the independent variables as
xi t
Xi = and T= (5.10)
li tF
where li = (l1 , l2 ) so that l1 denotes the channel length and l2 = s a representative

channel height. Then, t F represents a fluid time scale, such as the time a fluid particle
takes to travel through the entire channel length. Thus, this time scale can be expressed
in the form
l1
tF = (5.11)
u1
We can also scale the velocity, in order to make it nondimensional

ui
Ui = (5.12)
ui
Replacing the new dimensionless variables into the mass balance, Equation 5.5, we get
ˆ
∂ (l1 l2 ) ∂T ∂ (Ui ui ) ∂Xi
− = dV
∂T ∂t V ∂Xi ∂xi
ˆ
∂ (l l2 ) ∂T ∂Ui ∂Xi
− 1 = ui dV (5.13)
∂T ∂ ( T t F ) V ∂Xi ∂ ( Xi li )
ˆ
∂ ( l2 u 1 ) ui ∂Ui
− = dV
∂T li V ∂Xi
where, since the fluid is bidimensional, the volume V is in fact represented by a surface
of order l1 l2 .
Then, comparing the orders on the RHS, we have
u1 u2
=
l1 l2
(5.14)
l2
u2 = u1
l1
Thus, since l2 << l1 , then it must be true that u2 << u1 ; and therefore, the vertical
component of the velocity can be neglected. If we introduce this result in the dimensional
equation, we obtain
ˆ
∂V ∂u1
− = dV
∂t ∂x1 V
∂ ( l l2 ) ∂u
− 1 = 1 ( l1 l2 ) (5.15)
∂t ∂x1
1 ∂s ∂u
− = 1
s ∂t ∂x1
Now that we have found a simplified form of the mass balance given the kinematic
assumptions, we shall move on to the momentum balance. We will assume a shear stress
that opposes the flow velocity, with the same magnitude as in a steady-state turbulent
flow; i.e.
1
γ= c ρ ( u1 )2 (5.16)
2
where c is a constant determined from the Darcy-Weisbach empirical relation, for

losses assumed proportional to the kinetic energy of the fluid flow. Then, using this
model, the divergence of the tensor field might be written as
∂τij γ c ρ ( u1 )2
→ 2 x̂1 = x̂1 (5.17)
∂x j 2 s
With all these assumptions, and neglecting the gravitational forces, we obtain the
following simplified form of the momentum balance

∂u1 ∂u1 ∂p γ
ρ + u1 = − −2 (5.18)
∂t ∂x1 ∂x s
where the Langrangian derivative has been expanded to make explicit the fact that we
are only considering the x1 -direction.
Summary:
• Mass balance
1 ∂s ∂u
− = 1 (5.19)
s ∂t ∂x1
• Momentum balance

∂u1 ∂u1 ∂p γ
ρ + u1 = − −2 (5.20)
∂t ∂x1 ∂x s
5.1.4 Linearized governing equations
Substituing Equations 5.8 and 5.9 into the mass balance, Equation 5.19, we obtain
1 ∂ ∂
− (s + ε e
s) = (u1 + ε ue1 )
(s + ε e
s) ∂t ∂x1

∂es ∂u1 ∂ue
−ε = + ε 1 (s + ε e s)
∂t ∂x1 ∂x1
(5.21)
∂es ∂u1 ∂ue1 ∂ue1 ∂ue1
−ε = s+ε es+ε s + ( ε )2 s
e
∂t ∂x ∂x ∂x1 ∂x1
1 1
∂es ∂u1 ∂u1 ∂ue1
−ε = s + s+
e s ε
∂t ∂x1 ∂x1 ∂x1
where we have ignored the higher order terms, O (ε)2 , and we note that ∂s/∂t = 0

since s = s( x1 ).
Therefore, the terms that are not multiplied by ε we shall define as zeroth order terms,
and they are
∂u1
s=0 (5.22)
∂x1
or
∂u1
=0 (5.23)
∂x1
Then, the terms that are multiplied by ε we define as first order terms, and they
constitute
s
∂e ∂u ∂ue1
− = 1e s+ s (5.24)
∂t ∂x1 ∂x1
But, given Equation 5.23, the equation above reduces to

s
∂e ∂ue
− = 1s (5.25)
∂t ∂x1
We can do the same procedure for the momentum balance. Replacing Equations 5.8
and 5.9 into Equation 5.20, we obtain

∂ ∂u1 ∂ue1 ∂
ρ (u1 + ε ue1 ) + (u1 + ε ue1 ) +ε =− ( p + ε pe) −
∂t ∂x1 ∂x1 ∂x1
(5.26)
e1 )2
1/2 c ρ ( u1 + 2 ε u
−2
(s + ε es)
However, given Equation 5.23 and disregarding higher order terms, O((ε)2 ), we get
c ρ (u1 )2 + 2 ε ue1 u1

∂ue1 ∂ue1 ∂p ∂ pe
ρ ε + ε u1 =− −ε − (5.27)
∂t ∂x1 ∂x1 ∂x1 (s + ε e
s)
If we multiply each member by a quantity (s + ε e s), it yields

∂ue1 ∂ue1 ∂p ∂ pe
ρ ε + ε u1 s) = −
(s + ε e s) − ε
(s + ε e (s + ε e
s)
∂t ∂x1 ∂x1 ∂x1 (5.28)
− c ρ (u1 )2 + 2 ε ue1 u1

Then, we can group in terms of ε, disregarding once again the terms of order O((ε)2 )

∂ue1 ∂ue1 ∂p 2 ∂p p
∂e
ρs + ρ u1 s ε = −s − c ρ (u1 ) + −e s −s − 2 c ρ ue1 u1 ε (5.29)
∂t ∂x1 ∂x1 ∂x1 ∂x1
It is clear now that the zeroth order terms are

∂p c ρ ( u1 )2
=− (5.30)
∂x1 s
and the first order terms are

∂ue1 ∂ue ∂p ∂ pe
ρs + ρ u1 s 1 = −e
s −s − 2 c ρ ue1 u1 (5.31)
∂t ∂x1 ∂x1 ∂x1
If we recall Equation 5.25, we can replace the second term in the LHS by
∂ue1 s
∂e ∂p ∂ pe
ρs − ρ u1 s = −e
s −s − 2 c ρ ue1 u1 (5.32)
∂t ∂t ∂x1 ∂x1
Summary
• Zeroth order steady state equations

• Mass balance
∂u1
=0 (5.33)
∂x1
∂p c ρ ( u1 )2
=− (5.34)
∂x1 s
• First order steady state equations

• Mass balance
s
∂e ∂ue
− = 1s (5.35)
∂t ∂x1
∂ue1 s
∂e ∂p ∂ pe
ρs − ρ u1 s = −e
s −s − 2 c ρ ue1 u1 (5.36)
∂t ∂t ∂x1 ∂x1
5.1.5 Solution to the linearized equations
We shall seek a solution for u, ue, p and et in the mass balance and momentum balance
equations, Eqs. 5.33 to 5.36.
From the zeroth order mass balance, Equation 5.33, we can integrate to yield
u( x1 ) = γ1 (5.37)
where γ1 is a constant of integration referred to the mass balance that must be defined
from the boundary conditions. Note that from now on, we shall denote u1 = u, since we
will only treat velocities in the x1 -direction. Also, since u is a mean value, then it has no
dependence on the time, thus u = u( x1 ).
Then, we can also integrate the zeroth order momentum balance, Equation 5.34, and
we get
c ρ f ( u )2
p ( x1 ) = − x1 + γ2 (5.38)
s
where γ2 is a constant of integration referred to the momentum balance.

Now, we can integrate the first order mass balance, Equation 5.35, and we get
x1 ∂e
s
ue( x1 , t) = − + γ3 (t) (5.39)
s ∂t
so that γ3 represents an integration constant reffered to the mass balance, that in the
most general case depends on time.
Lastly, the first order momentum balance, Equation 5.36, gives
ρ ∂2 e

s c γ1 ρ ∂e
s 2 s c ρ γ1
ρ γ1 ∂e ∂γ3 cρ
pe( x1 , t) = 2
+ 2
( x1 ) + + 2
s−ρ
e + γ3 x1
2 s ∂t (s) ∂t s ∂t (s) ∂t s
+ γ4 (t)
(5.40)
such that γ4 (t) is an integration constant referred to the momentum balance that at
most depends upon time.
MULCAHY’S LEAKAGE FLOW MODEL FOR A SINGLE PIPE WITH
R O TAT I N G WA L L S
6
6.1 Fluid fundamental equations 200
6.2 Approximation based upon dimensional analysis 209
6.3 Lubrication approximation 216
6.4 Solid fundamental equations 220
6.5 Linear solution 222
6.6 Nonlinear solution 224
Leakage flow, a fundamental concept in fluid dynamics, refers to the unintended

and often undesirable passage of fluid through small gaps or clearances in a confined
system. This phenomenon is encountered in a diverse range of phenomena and can have
significant implications for the performance, efficiency, and safety of many a system.
The behavior of leakage flow is influenced by the complex interplay of fluid properties,
geometrical configurations, and boundary conditions. Fluids attempting to flow through
narrow gaps experience intricate interactions with the surrounding surfaces, resulting in
complex flow patterns and pressure distributions. Viscous effects become prominent in
such scenarios, leading to phenomena like boundary layer separation, vortex shedding,
and pressure gradients.
Efforts to understand and manage leakage flow involve both theoretical and exper-
imental approaches. Theoretical models, often based on the Navier-Stokes equations
governing fluid motion, provide insights into the fundamental mechanisms underlying
leakage flows. Computational methods offer a means to simulate and visualize complex
flow behaviors in intricate geometries.
Developing effective strategies to control leakage flows in order to harvest energy from
such a instability is a significant challenge. Design considerations, material selection,
seal mechanisms, and surface treatments play crucial roles in optimizing performance.
We shall then develop a mathematical model for the FSI interaction of a single pipe
configuration with rotating walls. To fully attend this problem, we propose two different
approaches:
• A linear formulation, obtained by the linearization of the partial differential equa-

tions that describe the dynamics of the configuration. This assumes a perturbed
solution that shall be capable of only predicting the bifurcation point of the system.
• A non-linear formulation, where the dynamics of the system are resolved using
numerical methods. The software used to solve such a configuration is pyFSI,
which will be explained in details in the next chapters of this text.
The scheme of the system is depicted in Figure 6.1, where the horizontal rotation of
the wall through the angle coordinate θ is considered for the analysis.
199
200 mulcahy’s leakage flow model for a single pipe with rotating walls
θ(t)
sI
s(x,t) sO
L
x
Figure 6.1: Model of a single pipe with rotating walls
6.1 fluid fundamental equations
The following hypothesis have been assumed for the modeling:
• Incompressible fluid, i.e. flow with low Mach number M < 0.3 (low speed).
• Newtonian fluid, so that the stresses are proportional to the rate of change of the
fluid’s velocity vector. This accounts for well known fluid flows such as air or
water.
• Narrow channel, where the length of the channel, L, is much bigger than a reference
channel height, s0 ; i.e. L >> s0 .
• Closure relation, where non-dimensional parameters are consisten with lubrication
theory.
• Turbulent flow, so that the velocity profile function may change with time; i.e.
u( x, y, t).
• Small angle rotation, for small θ.
The model will be defined in three unkown variables, namely
Q1 , the fluid flow in the x-direction

p, the fluid pressure (6.1)
s, the channel size
Thus, three equations must be stated in order to describe the dynamics of the system:
the mass balance, the momentum balance and the solid equation.
6.1.1 Conservation of mass
We shall derive the conservation of mass for the system; first in its integral form (see Sec.
2.7.4) and then in its differential form (see Sec. 2.7.3). Although we will only use the
differential expression in the formulation, the integral approach is still valid, and some
of the expressions derived in it will be useful in sections to come.
6.1 fluid fundamental equations 201
6.1.1.1 Integral form
Unlike Sec. 2.7.4, we need to express the continuity equation in terms of the variables of
our system; i.e. in terms of the fluid flow Q instead of the fluid velocity ui .
So, let us assume a control volume V (t) = b L s( x, t), where b represents the channel
width, L the channel length and s( x, t) the displacement of the walls. Then, we evaluate
the velocity of the fluid at a surface area S(t), so that, given conservation of mass, the
rate of flow of mass leaving the control volume V (t) should be equal to the negative rate
of increase of mass inside V
ˆ ˆ

i
∂ρ f
ρ f ui n dS = − dV (6.2)
S V ∂t
where we assumed a unit vector normal to the control surface S(t), where ni are the
contravariant components of the velocity ui ; and ρ f represents the density of the fluid.
Since we assumed an incompressible fluid, the density ρ f may be considered as
constant through the control volume. Then, given that the flow is two-dimmensional
ui = 0 ∀ i ̸= 1, 2; i.e. u1 = u x , u2 = uy and u3 is inherently zero in two-dimensional flows.
So Eq. 6.2 might be rewritten as
ˆL ˆs/2 ˆL ˆs/2
∂
ρf u1 n1 + u2 n2 dx1 dx2 = −ρ f b dx1 dx2 (6.3)
∂t
0 −s/2 0 −s/2
If we assume that n has components only in the x-direction; i.e. n I = −ê1 = −ê x and
nO = ê1 = ê x for the inlet and outlet respectively, then u2 n2 = 0 and we can solve the
LHS of Eq. 6.3 as
ˆL ˆs/2 ˆs/2
ρf u1 dx1 dx2 = b ρ f (u1 (l1 , y, t) − u1 (0, y, t)) dx2 (6.4)
0 −s/2 −s/2
´L
where the term b = 0 dx1 accounts for the channel width, and ensures that the
integration of the velocity component u1 is calculated over the entire cross-sectional area
of the channel. Then, for the RHS of Eq. 6.3, we can consider that
ˆs/2
dx2 = s( x, t) (6.5)
−s/2
and so
ˆL ˆs/2 ˆL
∂ ∂s
−ρ f b dx1 dx2 = −b ρ f dx2 (6.6)
∂t ∂t
0 −s/2 0
Therefore, we can write Eq. 6.3 in terms of Equations 6.4 and 6.6
ˆL ˆs/2
∂s
0 = b ρf dx1 + b ρ f (u1 (l1 , y, t) − u1 (0, y, t)) dx2 (6.7)
∂t
0 −s/2
Now, we shall define the flow rates into the configuration (through the inlet; i.e.
at coordinate x = 0) as Qin and out of the configuration (through the outlet; i.e. at
coordinate x = L) as Qout . This can be expressed as
ˆs/2 ˆs/2
Qin = u1 (0, y, t) dx2 and Qout = u1 ( L, y, t) dx2 (6.8)
−s/2 −s/2
So the final form of the integral form of the continuity equation might be written as
ˆL
∂s
0= dx1 + Qin − Qout (6.9)
∂t
0
where all terms have been divided by a factor ρ f b. As expected,it can be noted that
each term represents a function of time only - there is independence with all of the
spatial coordinates.
6.1.1.2 Differential form
We consider now a small Eulerian control volume of fluid within the configuration of
Fig. 6.1, which is depicted in Fig. 6.2. As in Section 2.7.3, we consider that the flux of
mass entering the infinitesimal CV must be equal to the rate of increase of mass inside
such a volume.
n̂ 3 A3
dz
A1 n̂ 2
y s(x,t)
v(x1) v(x2)
x
z
n̂ 1
A2
dx
x1
x2
n̂ 4
A4
Figure 6.2: Infinitesimal control volume for the single pipe with rotating walls model
So we begin with the original continuity equation in terms of the velocity
∂ρ f ∂(ρ f u j )
+ =0 (6.10)
∂t ∂x j
where once again, given a planar flow, j = 1, 2. Now, if we want to rewrite this in terms
of the fluid flow rate Q instead of the velocity u j , we may integrate the conservation of
mass equation across the width of the channel
ˆs/2
∂(ρ f u j )

∂ρ f
+ dx2 = 0
∂t ∂x j
−s/2
ˆ
s/2
ˆs/2
∂ρ f ∂(ρ f u j )
dx2 + dx2 = 0 (6.11)
∂t ∂x j
−s/2 −s/2
ˆs/2 ˆs/2
∂ ∂
ρf dx2 + ρ f u j dx2 = 0
∂t ∂x j
−s/2 −s/2
Then, given the results in Eq. 6.5, we can proceed with Equation 6.11 by rewritting it
as
ˆs/2
∂s ∂
ρf + ρf u j dx2 = 0 (6.12)
∂t ∂x j
−s/2
If we evaluate the second term in Equation 6.12, given its components we get
ˆs/2 ˆs/2 ˆs/2

∂ ∂u1 ∂u2
ρf u j dx2 = ρ f dx2 + ρ f dx2 (6.13)
∂x j ∂x1 ∂x2
−s/2 −s/2 −s/2
The first term of Equation 6.13 is non-intengrable. However, the second term is
integrable, and it results in
ˆs/2 s/2
∂u2
ρf dx2 = ρ f u2
∂x2
−s/2
(6.14)
−s/2
=0
where, given the no-slip condition, the velocity ui evaluated in the walls, (s/2, −s/2),
is null. Therefore, the second term in Equation 6.14 is null, and we may then rewrite
Equation 6.12 in terms of the x-velocity only
ˆs/2
∂s ∂
ρf + ρf u1 dx2 = 0 (6.15)
∂t ∂x1
−s/2
This is an import result, since it tells us that we shall only consider the velocity in the
x-direction in the mass balance. It is important to note though that this does not mean that
we have assumed axial flow. The y-velocity is simply null due to boundary conditions, and
thus the y-direction will be evaluated in the momentum balance as shown in the next
section.
Let us now define the mass flow rate Q1 across a section of the channel in the x-
direction by
ˆs/2
Q1 = u1 dx2 (6.16)
−s/2
where u1 = u1 ( x, y, t) and s = s( x, t). So, if we differentiate Eq. 6.16 with respect to

the x, we get
ˆ
 s/2 
d d 
( Q1 ) = u1 dx2  (6.17)
dx dx1
−s/2
Now, according to the Leibinz rule for differentiation under the integral sign, we may
rewrite the above expression as
 
ˆ
s( x )/2
d  d s( x ) d −s( x )
u1 ( x, y, t) dx2  = u1 ( x, s(x)/2, t) ( /2) − u1 ( x, −s(x)/2, t) ( /2)

dx1 dx1 dx1

−s(x)/2
ˆ
s( x )/2
∂u1 ( x, y, t)
+ dx2
∂x1
−s(x)/2
(6.18)
If the channel size s varies symmetrically with x, we may argue that

d s( x ) d −s( x )
u1 ( x, s(x)/2, t) ( /2) = u1 ( x, −s(x)/2, t) ( /2) (6.19)
dx1 dx1
And then, Equation 6.18 simply reduces to

 
ˆ
s( x )/2
ˆ
s( x )/2
d  ∂
u ( x, y, t ) dx = u1 ( x, y, t) dx2 (6.20)

1 2
dx1

∂x1
−s(x)/2 −s(x)/2
All in all, in terms of the fluid flow rate in the x direction, this means that according
to Eq. 6.17
ˆ
s( x )/2
∂Q1 ∂
= u1 ( x, y, t) dx2 (6.21)
∂x1 ∂x1
−s(x)/2
where the derivation d/dx1 was changed to ∂/∂x1 because we note that Q1 = Q1 ( x, t).
Finally, if we divide all the terms in Equation 6.13 by the fluid density ρ f , we may reduce
it to
∂s ∂Q1
+ =0 (6.22)
∂t ∂x1
And Eq. 6.22 is finally the form of mass balance in terms of the problem variables (the
fluid flow rate Q1 and channel size s) that we were looking for.
6.1.2 Conservation of momentum
We recall the expression for the momentum equation derived in Eq. 2.116, where the
force on the control volume in the x-direction is equal to the rate of change of momentum
in the x-direction contained within a Eulerian volume as shown in Fig. 6.2, plus the net
flux of momentum in the x-direction out of the CV
ˆ ˆ
∂ ( ρ f ui )
dV + ui ρ f un dS = Fi
V ∂t
ˆ ˆ S (6.23)
∂
ρf ui dV + ρ f ui (un dS) = Fi
∂t V S
where we consider un as the component of the velocity vector normal to dS and out
the CV, and we have taken the density ρ f out of the integral because incompressibility
has been assumed.
We can write the RHS of Eq. 6.23 in terms of the forces acting upon the fluid
˛ ˆ
j
Fi = σij n dS + ρ f gi dV (6.24)
V
S
where σij denotes the stress tensor (Sec. 2.9.1), gi is a body force acting upon the fluid;
e.g. gravity g2 = − g, and n j are the contravariant normal vector components to the
surfaces as depicted in Fig. 6.2.
We may then rewrite Equation 6.23 in terms of the bodies forces expressed in Equation
6.24. This results in
ˆ ˆ ˛ ˆ
∂ j
ρf ui dV + ρ f ui (un dS) = σij n dS + ρ f gi dV (6.25)
∂t V S V
S
Each one of the terms in Equation 6.25 shall be explored; starting with the LHS. We
may rewrite the first term given the volume to analyze; i.e.
ˆ ˆ
 s/2 
∂ ∂ 
ρf ui dV = ρ f dx1 dx3 ui dx2  (6.26)
∂t V ∂t
−s/2
where given the definition for the fluid flow rate
ˆs/2
Qi = ui dx2 (6.27)
−s/2
we may write the first term of Eq. 6.25 as

ˆ
∂ ∂Qi
ρf ui dV = ρ f dx1 dx3 (6.28)
∂t V ∂t
Now, in order to give a detailed description of the second LHS term or Equation
6.25, we need to introduce the the normal vectors to each surface in Fig. 6.2. We define
n1 = −êx , n2 = êx , n3 = − sin θ êx + cos θ êy and n4 = − sin θ êx − cos θ êy , where θ is
the angle between the walls and horizontal lines, as depicted in Fig. 6.1. Therefore,
considering the flow through the inlet and outlet, since, because of the no-slip condition
the velocity is null at surfaces S3,4 , we shall then write the second term of Eq. 6.25 in
terms of n1 and n2 , so that un = ui n1i = u j n j and then
ˆ ˆ
ρf ui (un dS) = ρ f ui u j n j dS (6.29)
S Si
where n j represent the contravariant components of the normal vector with respect to
the basis vectors ei . Let us now expand the tensor components from Equation 6.29
ˆ ˆ
j
ρf ui u j n dS = ρ f u1 u j n j + u2 u j n j dS
Si S
ˆ i ˆ (6.30)
= ρf u1 u j n j dS + ρ f u2 u j n j dS
S1 S2
We shall explore each of the integrals over the surfaces S1,2 separatedly. Starting with
the integral over the inlet surface S1 , if once again we expand the tensor components we
get
ˆ ˆ ˆ
1
ρf u1 u j n j dS = ρ f u1 u1 n dS + ρ f u1 u2 n2 dS
S1
ˆ S1 S1
(6.31)
1
= ρf u1 u1 n dS
S1
where u2 n2 = 0 since these components are perpendicular to each other. Considering

the surface area to evaluate as in Fig. 6.2, then S1 = dx2 dx3 in the channel from −s/2
to s/2 and given the normal vector n1 = −ê x and n2 = ê x then n1 = −1 and n2 = 1
respectively. Thus, Eq. 6.31 takes the form
ˆ ˆs/2
j
ρf u1 u j n dS = −ρ f dx3 u21 dx2 (6.32)
S1
−s/2
This represents the contribution of the u21 term to the rate of change of momentum in
the x-direction due to the x-component velocity squared across the channel height s and
through the depth dx3 . Given the definition for the fluid flow rate, Equation 6.27, one
might be tempted to try and find a relation between the expressions above and Qi , but
they cannot be related directly.
This same procedure is valid for the integral over the outlet surface S2 in Equation
6.30. Expanding the tensor components, we get
ˆ ˆ ˆ
j 1
ρf u2 u j n dS = ρ f u2 u1 n dS + ρ f u2 u2 n2 dS
S2
ˆ S2 S2
(6.33)
= ρf u1 u2 dS
S2
Therefore, we have shown that the expression for the second term of Equation 6.25
can be written as
ˆ ˆ ˆ
2
ρf ui (un dS) = ρ f u1 dS + ρ f u1 u2 dS (6.34)
S S1 S2
Now, applying Gauss theorem, we may rewrite this term the form
ˆ ˆ
∂
ρf ui (un dS) = ρ f ui u j n j dV
S V ∂x j
ˆs/2 (6.35)
∂
= ρ f dx1 dx3 ui u j n j dx2
∂x j
−s/2
Thus, for i = 1 and j = 1 we have
ˆs/2
∂
ρ f dx1 dx3 (u1 u1 ) dx2 (6.36)
∂x1
−s/2
then, for i = 2 and j = 1 we get
ˆs/2 s/2
∂
ρ f dx1 dx3 (u2 u1 ) dx2 = ρ f dx1 dx3 (u2 u1 )
∂x2 −s/2 (6.37)
−s/2
=0
where the equation above is null considering that the velocities are evaluated at the
walls, (−s/2, s/2), are also null due to the no-slip condition. Therefore, we may write the
second term of Eq. 6.25 neglecting the y-direction and resulting only in Equation 6.36.
Let us now move on to the RHS of Equation 6.25. The body force term, considering it
as a gravity term, results in
ˆ ˆs/2 ˆs/2
ρf gi dV = ρ f dx1 dx3 g1 dx2 + ρ f dx1 dx3 g2 dx2 (6.38)
V
−s/2 −s/2
where we consider the gravity in the downwards vertical direction; i.e. g1 = gx = 0

and g2 = − g. Thus, Eq. 6.38 gives
ˆ ˆs/2
ρf gi dV = −ρ f dx1 dx3 g dx2
V (6.39)
−s/2
= −ρ f dx1 dx3 g s
Now, the last term of the RHS of Equation 6.25 involving the stress tensor is not as
straigthforward
˛
σij n j dS (6.40)
S
As it has been introduced in Sec. 2.9.2, we may write the stress tensor as

∂ui ∂u j
σij = − p δij + µ f + (6.41)
∂x j ∂xi
where δij is the Kronecker delta and µ f is the fluid dynamic viscosity. Introducing this
definition for the stress tensor into Equation 6.40, we get
˛ ˛ ˛
j j ∂ui ∂u j
σij n dS = − p δij n dS + µ f + n j dS (6.42)
∂x j ∂xi
S S S
We can note clearly the pressure terms and viscous terms in Equation 6.42. Let us
explore the pressure term first. Applying Gauss’ theorem to the surface in the control
volume V, we get
˛ ˆ
j ∂
− p δij n dS = − ( p n j ) dV
V ∂x j
S
ˆs/2 (6.43)
∂p
= −dx1 dx3 dx2
∂x j
−s/2
Given that the term ∂p/∂x j does not depend on the y-direction, it may be taken out of
the integral. Thus, the pressure terms results
ˆs/2
∂p ∂p
−dx1 dx3 dx2 = −dx1 dx3 s (6.44)
∂x j ∂x j
−s/2
Let us now move on to the viscous term in Eq. 6.42. Once again, by applying Gauss’
Divergence theorem, we get
˛ ˆ
∂ui ∂u j j ∂ ∂ui ∂u j
µf + n dS = µ f + dV (6.45)
∂x j ∂xi V ∂x j ∂x j ∂xi
S
It is of interest to develop the integrand in the last equation. It gives
∂2 u i ∂2 u j

∂ ∂ui ∂u j
+ = + (6.46)
∂x j ∂x j ∂xi ∂x j ∂x j ∂x j ∂xi
where it is worth noting that the second term might be rewritten as
∂2 u j

∂ ∂u j
= (6.47)
∂x j ∂xi ∂x j ∂x j
thus, given the incompressibility condition

∂u j
=0 (6.48)
∂x j
this term results null. Therefore, we can write the viscous term, Eq. 6.45, as
˛ ˆ
∂2 u i

∂ui ∂u j j
µf + n dS = µ f dV (6.49)
∂x j ∂xi V ∂x j ∂x j
S
6.2 approximation based upon dimensional analysis 209
It might be illustrative to type out the components of the velocity derivatives

ˆ ˆ 2
∂2 u i ∂2 u2 ∂2 u1 ∂2 u2

∂ u1
µf dV = µ f + + + dV (6.50)
V ∂x j ∂x j V ∂x1 ∂x1 ∂x1 ∂x1 ∂x2 ∂x2 ∂x2 ∂x2
Now that each of the terms of Equation 6.25 have been developed, we can write the
general equation of motion. This equation will be given from the results in Equations
6.26, 6.34, 6.39, 6.43 and 6.45. Thus,
ˆs/2 ˆs/2 ˆ
     s/2 
∂ ∂
 ρ f dx1 dx3 ui dx2  +  ρ f dx1 dx3 ui u j n j dx2  =  ρ f dx1 dx3 gi dx2 
∂t ∂x j
−s/2 −s/2 −s/2
ˆs/2
 
∂p
− dx1 dx3 dx2 
∂x j
−s/2
ˆs/2
 
∂2 u i
− µ f dx1 dx3 dx2 
∂x j ∂x j
−s/2
(6.51)
Since the integrals are all the same, then the integrands must be equal. This way,
dividing all terms by the fluid density ρ f and the surface area dx1 dx3 , we get
∂ ∂ 1 ∂p µ f ∂2 u i
ui + u i u j n j = gi − − (6.52)
∂t ∂x1 ρ f ∂x j ρ f ∂x j ∂x j
We may define the viscous forces as
∂2 u i
f vis,i = µ f (6.53)
∂x j ∂x j
we end up with the general momentum balance for the system

∂ ∂ 1 ∂p
ui + u i u j n j = gi − + f vis,i (6.54)
∂t ∂x1 ρf ∂x j
6.2 approximation based upon dimensional analysis
Although Equation 6.54 is valid, it does not describe the model in terms of the three
unknown variables we have defined; i.e. the fluid flow rate Q, the pressure p and the
channel height s. Thus, we shall make use of tools such as dimensional analysis and
lubrication theory in order to find these relationships.
In order to find the dimensionless relationships, we will chose the independent
variables as follows
x2 x2 t
X1 = ; X2 = ; T= (6.55)
l1 l2 TF
where we denote the dimmensionless variables in capital letters; then TF = L/u1 is the
time scale of the fluid, l2 is a reference size for the channel height and l1 is a reference
length taken as the channel length.
The dependendt variables we chose to scale as

ui p s
Ui = ; P= ; S= (6.56)
ui p l2
where ui and p are reference velocities and pressure respectively.
6.2.1 Dimensionless mass balance
Let us start with the mass balance derived in Equation 6.22. The idea is to neglect terms
from the equation in order to draw conclusions that are valid for our model. We will
begin by trying to find the dimensionless form of the first LHS term in Eq. 6.22. Given
the definitions in Equation 6.56, we note that we can write s = S l2 ; and then
∂s ∂(S l2 ) ∂T
=
∂t ∂T ∂t
∂S ∂ (t/T f )
= l2
∂T ∂t
l2 ∂S ∂t (6.57)
=
T f ∂T ∂t
l2 u1 ∂S
=
l1 ∂T
where we recall that the dimensionless fluid referece time was defined as T f = l1/u1
while T = t/T f .
Now, for the second term in Equation 6.22, let us evaluate it in its integral form; i.e.
ˆs/2
∂Q1 ∂u1
= dx2 (6.58)
∂x1 ∂x1
−s/2
And from Equation 6.58, let us begin by evaluating the derivative ∂u1/∂x1 alone. This
gives
∂u1 ∂(u1 U1 ) ∂X1

=
∂x1 ∂X1 ∂x1
∂U1 ∂ ( x1/l1 )
= u1 (6.59)
∂X1 ∂x1
u1 ∂U1
=
l1 ∂X1
thus, we can introduce this result in Equation 6.58 and then this gives
ˆs/2 ˆ
S l2/2
∂u1 u1 ∂U1
dx2 = d ( l 2 X2 ) (6.60)
∂x1 l1 ∂X1
−s/2 −S l2/2
We define then the dimensionless fluid flow in the x-direction as

ˆ
S l2/2
∂U1
Q1∗ = dX2 (6.61)
∂X1
−S l2/2
with this result and Equation 6.57, we can write the dimensionless form of the mass
balance as
u1 l2 ∂S u l2 ∂Q1∗
+ 1 =0 (6.62)
l1 ∂T l1 ∂X1
As it can be seen in Equation 6.62, both terms are of the same order of magnitude,
O = u1 l2/l1 . Therefore, neither of them may be neglected, and we are able to write the
dimensionless mass balance as
∂S ∂Q1∗
+ =0 (6.63)
∂T ∂X1
where the only conclusion that we arrive from the dimensional analysis applied to the
mass balance is that both terms are of the same order of magnitude.
6.2.2 Dimensionaless incompressibility condition
We have used the incompressibility condition all throghout thsi analysis. This is, for a
Newtonian fluid, the incompressibility condition states that
∂ui
=0 (6.64)
∂xi
Given the variables scaling, ui = Ui ui and xi = X L, so we can non-dimensionalize

the derivative of the velocity with respect to the cartesian coordinates
∂ui ∂(Ui ui ) ∂Xi

=
∂xi ∂Xi ∂xi
∂Ui ∂( i/li )
x
= ui (6.65)
∂Xi ∂xi
u ∂U
= i i
li ∂Xi
Thus, the dimensionless incompressiblity condition may be expressed as

ui ∂Ui
=0 (6.66)
li ∂Xi
Then, if we assume that both terms in Equation 6.66 have the same order, then it must
be true that

u 1 l2
O ( u2 ) = O (6.67)
l1
This is an important result, since it establishes a relationship between the order of

magnitude of the velocity in the y-direction and the velocity in the x-direction. We shall
refer to this result when applying dimensional analysis to the momentum balance in the
next section.
6.2.3 Dimensionless momentum balance
In order to write the dimensionless momentum balance, let us begin by exploring the
viscous force term in Equation 6.54, f vis,i , as defined in Equation 6.53. We shall explore
the dimmension of each term, with components as defined in Equation 6.50
∂2 u1 µ f u1 ∂2 u1 µ f u1

O µf 2
= 2
; O µf 2
=
(∂x1 ) ( l1 ) (∂x2 ) ( l2 ) 2
(6.68)
∂2 u2 µ f u2 ∂2 u2 µ f u2

O µf = ; O µf =
(∂x1 )2 ( l1 ) 2 (∂x2 )2 ( l2 ) 2
Due to the geometry of the channel, it is clear that s0 << L; i.e. l2 << l1 . Thus, under
this assumption, we may disregard several terms in the equation above, since they will
have little to no effect in the final net viscous force. In other words, the terms divided
by a factor l2 will have much more significance; then we may as well discard the other
terms.
So, given Equation 6.68, the viscous
force in the x-direction might be approximated by
2
the term of order O µ f ∂ u1/∂x2 ∂x2 = µ f u1/(l2 )2
∂2 u1
f vis,1 ≃ µ f (6.69)
∂x2 ∂x2
and in the y-direction, the approximation to the viscous force is given by the term of
order O µ f ∂2 u2/(∂x2 )2 = µ f u2/(l2 )2
∂2 u2
f vis,2 ≃ µ f (6.70)
∂x2 ∂x2
Notice how the equations above are consistent with the fact that viscous effects are
stronger near walls.
Now, we may pursue a dimensionless momentum balance. Let us start by applying
dimensional analysis to the first term of the LHS of Eq. 6.54
∂ui ∂(Ui ui ) ∂T
=
∂t ∂T ∂t
∂Ui ∂(t/TF )
= ui
∂T ∂t
(6.71)
ui ∂Ui ∂t
=
TF ∂T ∂t
u u ∂Ui
= i 1
l1 ∂T
where we recall the definition of TF = l1/u1 . It might be instructive to develop

the components of the above equations for both the x-direction and the y-direction
respectively
( u1 )2 ∂U1
, for the x-direction
L ∂T (6.72)
u1 u2 ∂U2
, for the y-direction
L ∂T
since l1 = L. Then, we approach the second term of the LHS of Eq. 6.54
∂ ( ui u j n j ) ∂(Ui ui Uj u j n j ) ∂X1
=
∂x1 ∂X1 ∂x1
∂Ui Uj j ∂ x1/l1
= ui u j n (6.73)
∂X1 ∂x1
ui u j ∂Ui Uj j
= n
l1 ∂X1
Once again, we develop the components in the x and y-directions. Taking into account
the results obtained in Equation 6.34 and the fact that for j = 2 the normal component of
the velocity u j=2 n j=2 = 0, we get
(u1 )2 ∂(U1 )2
, for the x-direction
L ∂X1
(6.74)
u1 u2 ∂(U1 U2 )
, for the y-direction
L ∂X1
Let us now move on to the RHS of Equation 6.54. The body force term, gi , it is already
given in a dimensionless form, since it does not depend upon any of the dimensionless
variables defined in Equations 6.55 and 6.56. We continue our analysis with the pressure
term in the general momentum equation
1 ∂p 1 ∂( P p) ∂Xi
=
ρ f ∂xi ρ f ∂Xi ∂xi
1 ∂P ∂( xi/li )
= p (6.75)
ρf ∂Xi ∂xi
p ∂P
=
ρ f li ∂Xi
Now, when treating the viscous force term, it may be wiser to non-dimensionalize the
approximations to the componensts as derived in Equations 6.69 and 6.70. Starting with
the viscous force in the x-direction
µ f ∂2 u1

µf ∂ ∂u1
=
ρ f ∂x2 ∂x2 ρ f ∂x2 ∂x2

µ f ∂ ∂X2 ∂(U1 u1 ) ∂X2
=
ρ f ∂X2 ∂x2 ∂X2 ∂x2

µ f ∂( x2/l2 ) ∂ ∂U1 ∂( x2/l2 )
= u1 (6.76)
ρ f ∂x2 ∂X2 ∂X2 ∂x2
µ f u1 ∂2 U1
=
ρ f l2 l2 ∂X2 ∂X2
µ f u1
= Fvis,1
ρ f l2 l2
Now, for the approximation to the viscous force in the y-direction, we get
µ f ∂2 u2

µf ∂ ∂u2
=
ρ f ∂x2 ∂x2 ρ f ∂x2 ∂x2

µ f ∂ ∂X2 ∂(U2 u2 ) ∂X2
=
ρ f ∂X2 ∂x2 ∂X2 ∂x2

µf ∂( 2/l2 ) ∂
x ∂U2 ∂( x2/l2 )
= u2 (6.77)
ρf ∂x2 ∂X2 ∂X2 ∂x2
µ f u2 ∂2 U2
=
ρ f l2 l2 ∂X2
µ f u2
= Fvis,2
ρ f l2 l2
Thus, it is clear from Equations 6.77 and 6.76 that we may define a general dimension-
less expression for the viscous force as
µ f ui
Fvis,i (6.78)
ρ f l2 l2
Finally, from Equations 6.71, 6.73, 6.75 and 6.78, we may write a dimensionless
expression for the momentum balance in terms of
ui u1 ∂Ui ui u j ∂Ui Uj j p ∂P µ f ui
+ n = gi − + Fvis,i (6.79)
l1 ∂T l1 ∂X1 ρ f li ∂Xi ρ f l2 l2
We note that, if we introduce the contravariant components n j , such that u j=2 n j=2 = 0,
to the second term of the LHS, we may rewrite it as follows
ui u1 ∂Ui Uj j µ f ui

ui u1 ∂Ui p ∂P
+ n = gi − + Fvis,i (6.80)
l1 ∂T l1 ∂X1 ρ f li ∂Xi ρ f l2 l2
and we obtain the same result. This way, we are able to divide each term by a factor
ui u1/l1 .
This results in
∂Ui ∂Ui Uj j l1 gi p l1 ∂P µ f l1
+ n = − + Fvis,i (6.81)
∂T ∂X1 u i u1 ρ f li ui u1 ∂Xi ρ f l2 l2 u 1
We may define the Reynolds number as

ρ f u 1 l2
Re = (6.82)
µf
and thus Eq. 6.81 may be rewritten as
∂Ui ∂Ui Uj j l1 gi p l1 ∂P 1 l1
+ n = − + Fvis,i (6.83)
∂T ∂X1 u i u1 ρ f li ui u1 ∂Xi Re l2
which represents the general form of the dimensionless momentum balance.

Now that we have a dimensionless expression for the momentum balance, we may
analyze the order of each one of the terms. This way, we may determine how each
term contributes to the general momentum of system, and we could disregard the terms
whose contribution is negligible.
Let us start by analyzing the pressure term. For a non-negligible pressure variation in
the x-direction, the following condition must be satisfied

p
O =1
ρ f u1 u1 (6.84)

O ( p ) = O ρ f u1 u1
If we want to analyze next the order of the pressure variation in the y-direction, we
must recall the relationship between the order of velocities in the x and y-direction that
we found back in Equation 6.67. Thus, also taking into account the condition for the
order of the pressure in Eq. 6.84, we get

p l1 p l1 l1
O =O
ρ f l2 u 2 u 1 ρ f l2 l2 u 1 u 1
ρ f u 1 u 1 l1 l1

=O (6.85)
ρ f l2 l2 u 1 u 1

l1 l1
=O
l2 l2
Since we have the geometrical condition l2 << l1 , then it follows that O ((l1 l1 )/(l2 l2 )) >>
1. Thus, this is the most relevant term in the LHS of Equation 6.83. Physically, this means
that the transient terms can be neglected in the momentum balance in the y-direction.
We shall analyze the order of magnitude of the gravity terms. Recall that it has been
defined only in the downwards y-direction as g1 = 0 and g2 = − g, then

l1 g g l1 l1
O =O (6.86)
u2 u1 u 1 u 1 l2
Equation 6.86 indicates that the gravity effects are going to be significant only for low
fluid velocities; i.e.
g l1 l1
>> 1
u 1 u 1 l2
r (6.87)
g
u1 << l1
l2
This way, considering the dimensional analysis as done above, if we recover our
dimensional equations, we notice that we end up with a simple hydrostatic equation in
the y-direction
1 ∂p
0=− −g
ρ f ∂x2
(6.88)
∂p
= −g ρ f
∂x2
This is an important conclusion, since, this way, we may only focus on the momentum
balance in the x-direction from now on. Then, given in dimensional form, the momentum
equation can be written purely in terms pertaining to the x-direction
∂2 u1

∂ ∂ 1 ∂p
u1 + ( u1 u1 ) = − + µf (6.89)
∂t ∂x1 ρf ∂x1 ∂x2 ∂x2
We may now integrate this equation over y to get
ˆs/2 ˆs/2 ˆs/2 ˆs/2

∂ ∂ 1 ∂p µf ∂2 u1
u1 dx2 + (u1 u1 ) dx2 = − dx2 + dx2 (6.90)
∂t ∂x1 ρf ∂x1 ρf ∂x2 ∂x2
−s/2 −s/2 −s/2 −s/2
If we define
ˆs/2
N1 = (u1 u1 ) dx2 (6.91)
−s/2
then, recalling Eq. 6.27, we may rewrite Equation 6.90 as
ˆs/2 y=s/2
∂Q1 ∂N1 1 ∂p µ f ∂u1
+ =− dx2 + (6.92)
∂t ∂x1 ρf ∂x1 ρ f ∂x2 y=−s/2
−s/2
summary
Up until this point, we have defined the mass balance as

∂s ∂Q1
+ =0 (6.93)
∂t ∂x1
and the momentum balance as

ˆs/2 y=s/2
∂Q1 ∂N1 1 ∂p µ f ∂u1
+ =− dx2 + (6.94)
∂t ∂x1 ρf ∂x1 ρ f ∂x2 y=−s/2
−s/2
with a pressure variation in the vertical y-direction as given by

∂p
= −g ρ f (6.95)
∂x2
6.3 lubrication approximation
We shall use the concepts of lubrication theory in order to try and find an approximation
to the velocity profile based upon experimental data.
6.3.1 Laminar flow
We begin by analizing the most simple case, that of a laminar flow. Thus, for a fully
developed laminar flow, we may neglect the transient terms (LHS of Eq. 6.94) and the
momentum balance yields
ˆs/2 ˆs/2
1 ∂p µf ∂2 u1
dx2 = dx2 (6.96)
ρf ∂x1 ρf ∂x2 ∂x2
−s/2 −s/2
6.3 lubrication approximation 217
Multiplying both terms by the the fluid density ρ f , and then noting that as the integrals
are the same, the integrands must be equal, we arrive to the following equation
∂p ∂2 u1
= µf (6.97)
∂x1 ∂x2 ∂x2
We may integrate the equation above with respect to the y-direction

ˆ ˆ
∂p ∂2 u1
dx2 = µ f x2
∂x1 ∂x2 ∂x2
(6.98)
∂p ∂u1
x2 = µ f + f 1 ( x1 , t )
∂x1 ∂x1
where f 1 ( x1 ) is an arbitrary function of x1 that arises due to integration. Now, we have

an equation involving the velocity u1 in the x-direction, and its partial derivative with
respect to x2 . Thus, we may integrate the equation above again in order to solve for u1
ˆ ˆ
∂p ∂u1
x2 dx2 = µf + f 1 ( x1 , t) dx2
∂x1 ∂x1
(6.99)
x2 x2 ∂p
= µ f u1 + f 1 ( x1 , t ) x2 + f 2 ( x1 , t )
2 ∂x1
and once again, f 2 ( x1 , t) represents another integration constant. This way, we can
find an expression for the x-velocity for a fully developed laminar flow

1 x2 x2 ∂p
u1 = − f 1 ( x1 , t ) x2 − f 2 ( x1 , t ) (6.100)
µf 2 ∂x1
We can work on the integration constants given the no-slip boundary conditions
u1 ( x1 , −s/2, t) = 0
(6.101)
u1 ( x1 , s/2, t) = 0
At the bottom wall, x2 = −s/2, we have

!
1 (−s/2)2 ∂p s
0= − f 1 ( x1 , t ) − − f 2 ( x1 , t )
µf 2 ∂x1 2
!
2 (s)2 ∂p (6.102)
f 1 ( x1 , t ) = − − f 2 ( x1 , t )
s 8 ∂x1
2 s ∂p
f 1 ( x1 , t ) = f 2 ( x1 , t ) −
s 4 ∂x1
Then, at the top wall, x2 = s/2, the velocity equation yields

!
1 (s/2)2 ∂p s
0= − f 1 ( x1 , t ) − f 2 ( x1 , t )
µf 2 ∂x1 2
(s)2 ∂p

2 s ∂p s
0= − f 2 ( x1 , t ) − − f 2 ( x1 , t )
8 ∂x1 s 4 ∂x1 2 (6.103)
(s)2 ∂p (s)2 ∂p
0= − 2 f 2 ( x1 , t ) +
8 ∂x1 8 ∂x1
2
(s) ∂p
f 2 ( x1 , t ) =
8 ∂x1
Thus, we conclude that the first integration constant must be equal to

2
2 (s) ∂p s ∂p
f 1 ( x1 , t ) = −
s 8 ∂x1 4 ∂x1
s ∂p s ∂p (6.104)
= −
4 ∂x1 4 ∂x1
=0
Inserting the results from Equations 6.104 and 6.103 back in the velocity equation; i.e.
Eq. 6.100, we obtain
(s)2 ∂p

1 x2 x2 ∂p
u1 = −
µf 2 ∂x1 8 ∂x1
2 2
(6.105)
4 ( x2 ) − (s) ∂p
=
8 µf ∂x1
Now that we have an expression for the x-velocity for a developed laminar flow, we
can use it back in the complete momentum balance, Eq. 6.94. Although it is true that this
simplified expression does not capture all the details of the transient terms, which are
related to the unsteady behavior of the flow, in many cases this simplified solution could
be used if the transient behavior is of secondary importance or if the flow has reached a
quasi-steady state. Thus, by substituing back this expression into the original equations,
we can see how well it holds up under the assumption that the flow is dominated by the
viscous forces, leading to a rapid attainment of a quasi-steady state. Therefore, given Eq.
6.21, we may integrate the equation above in order to find an expression for the fluid
flow rate
ˆs/2
Q1 = u1 dx2
−s/2
ˆs/2
4 ( x2 )2 − (s)2 ∂p

= dx2
8 µf ∂x1
−s/2
ˆs/2 ˆs/2
 
∂p 1 
= 4 ( x2 )2 dx2 − (s)2 dx2 
∂x1 8 µ f (6.106)
−s/2 −s/2
s/2 s/2
 
∂p 1  4 ( s ) 2
= ( x2 )3 − ( x2 )2 
∂x1 8 µ f 3 2
−s/2 −s/2
( "
2
3 2 3 # " 2 2 2 #)
∂p 1 4 (s) (s) ( s )2 ( s )2 (s)
= + − +
∂x1 8 µ f 3 2 2 2 2 2
(s)3 ∂p
=
12 µ f ∂x1
6.3 lubrication approximation 219
Then, we can apply the exact same logic to find an expression for the term N1
ˆs/2
N1 = (u1 )2 dx2
−s/2
2
( s )5 (6.107)

∂p
=
120 (µ f )2 ∂x1
6 ( Q1 )2
=
5s
Then, the pressure term may be integrated to yield
ˆs/2 ˆs/2
1 ∂p 1 ∂p
− dx2 = − dx2
ρf ∂x1 ρ f ∂x1
−s/2 −s/2 (6.108)
s ∂p
=−
ρ f ∂x1
Lastly, given the expression for the viscous term in Equation 6.94, we may write them
in terms of the fluid flow rate as
s/2
µ f ∂u1 12 µ f
=− Q1 (6.109)
ρ f ∂x2 ρ f s2
−s/2
Therefore, we can write the momentum for a quasi-steady state laminar flow in terms
of our problem’s unknowns
( Q1 )2 12 µ f

∂Q1 ∂ s ∂p
+ ξ1 =− − Q1 (6.110)
∂t ∂x1 s ρ f ∂x1 ρ f s2
where we have assumed a profile factor ξ 1 = 6/5, resulting from Equation 6.107.
6.3.2 Fanning friction factor model
This model is based upon a redefinition of the viscous effects as
f 1 ( Q1 ) ( Q1 )2
f vis,1 = − (6.111)
4 ( s )2
where f 1 ( Q1 ) is the Fanning friction factor, an empirical quantity that models both
laminar and turbulent stresses. It is given by the followin relationship
(
48/Re if Re < 1000
f1 = (6.112)
48/Re0.24 if Re ≥ 1000
Moreover, the profile factor ξ 1 in this model is also modified depending upon the
Reynolds number
(
6/5 if Re < 1000
ξ1 = (6.113)
1 if Re ≥ 1000
Thus, the momentum equation depending upon the Fanning factor becomes
( Q1 )2 f ( Q1 )2

∂Q1 ∂ s ∂p
+ ξ1 =− − 1 (6.114)
∂t ∂x1 s ρ f ∂x1 4 ( s )2
which is valid for both laminar and turbulent profiles.
6.3.3 Eddy viscocity model for turbulent profiles
It is well known that at high Reynolds numbers, the flow becomes turbulent. Thus, the
velocity profile function found in Eq. 6.105 will not be satisfied under this assumption.
There are several models to predict the velocity functions for turbulent profiles - one
well-accepted empirical model is known as the Power Law. For a pipe in cylindrical
coordinates, it takes the form
ur r n
= (6.115)
Umax R
where n is a constant which value depends upon the Reynolds number, and R denotes
the radius of the pipe.
The model includes turbulent stresses in the RANS equations defined as
∂p ∂
( x1 , t ) = (τ + τtur ) (6.116)
∂x1 ∂x2
where
∂u1 ∂u1
τ = µf ( x1 , x2 , t ) and τtur = µtur ( x1 , x2 , t ) (6.117)
∂x2 ∂x2
are the laminar and turbulent shear stresses respectively, µtur is an empirical coefficient
known as the Eddy or turbulent viscocity and p, u1 denote the time-averaged pressure and
x-velocity respectively.
TODO - develop velocity profile function for this model
6.4 solid fundamental equations
The scheme with the boundary restrictions for the wall, to solve for the solid dynamics,
is depicted in Figure 6.3. TODO - add figure desciption
The vertical displacement, as mesured from point O, may be expressed as a linear
combination of two rigid body motions - a vertical displacement h(t) and a rotation
angle θ (t) along the z-axis
s( x, t) = h(t) + x θ (t) = g T ( x ) a(t) (6.118)
where a(t) is a generalized coordinate vector, and g( x ) is the mode shape function

a= h θ and g= 1 x (6.119)
6.4 solid fundamental equations 221
kw cw
aG aR
aO
mw
kt O G θ(t)
R
p(x,t)
x=0
Figure 6.3: Solid scheme
Our objective is to obtain the equation of motion of the solid. In order to do so, we
may apply the Langrange equation, defined as

∂ ∂L ∂L
− = Qk (6.120)
∂t ∂q̇k ∂qk
where L = T − U represents the Langrangian functional, qk is the k-th generalized

coordinate of the system and Qk is the k-th generalized force of the system. The
expressions above can be depicted in a more compact way, considering linear vibrations

∂ ∂T ∂T ∂D ∂V
− + + = fr (6.121)
∂t ∂q̇k ∂qk ∂q̇k ∂qk
where f r represents the external force, T and V the kinetic and potential energies
respectively, and D is the dissipation or Rayleigh function.
All these scalar quantities are defined as
n n
1 T 1
T=
2
ȧ M ȧ =
2 ∑ ∑ ȧ mij ȧ j
i =1 j =1
n n
1 T 1
V=
2
ȧ K a =
2 ∑ ∑ a kij a j (6.122)
i =1 j =1
n n
1 T 1
D=
2
ȧ C ȧ =
2 ∑ ∑ ȧ cij ȧ j
i =1 j =1
where M, K and C are the mass, stiffness and damping matrices respectively.
Given the definition for a in Equation 6.119, we have
1 2 1
T= mw ḣ + aG θ̇ + Iw θ̇ 2
3 2
1 1
V = k w h2 + k t θ 2 (6.123)
2 2
1 1
D = cw ḣ + ct θ̇ 2
2
2 2
Substituing the expressions above back into Equation 6.121, we can obtain the differen-
tial equation of motion
mw ḧ + mw aG θ̈ + cw ḣ + k w h = R cos θ
(6.124)
a2G

Iw + mw θ̈ + mw aG ḧ + ct θ̇ + k t θ = R a R cos θ
where R is the resulting pressure force obtained by integrating p( x1 , t)
ˆL
R= p( x1 , t) dx (6.125)
0
and a R is the pressure center
ˆL
aR = TODO (6.126)
0
These equations of motion can be expressed in matrix form as
Ms ä + Cs ȧ + Ks a = F (6.127)
so that

mw a G mw
Ms =
aG mw Iw + a2G mw

cw 0
Cs =
0 ct
(6.128)
kw 0
Ks =
0 kt

R cos θ
F =
R a R cos θ
TODO - tbc
6.5 linear solution
Our objective is to linearize the equations of motion, in order to study the stability of the
system. To do this, we shall introduce a small parameter ε to the variables related to the
fluid flow in the x-direction; namely the fluid flow, the pressure and the channel size
Q1 ( x1 , t ) ≃ Q1 ( x1 ) + ε Q
f1 ( x1 , t)
p( x1 , t) ≃ p( x1 ) + ε pe( x1 , t) (6.129)
s ( x1 , t ) ≃ s ( x1 ) + ε e
s ( x1 , t )
where the variables with overlines denote mean values which are independent of time,
and the variables with tildes represent fluctuations measured from the mean value. A
detailed description can be found on Sec. 2.16.
6.5 linear solution 223
6.5.1 Mass balance
Substituing the expanded variables into the mass balance, Equation 6.93, we get
∂ ∂
(s + ε e
s) + Q1 + ε Q
f1 = 0
∂t ∂x1
∂s ∂(ε e s) ∂Q1 ∂( Q
f1 ε)
+ + + =0 (6.130)
∂t ∂t ∂x1 ∂x1
!
∂Q1 s
∂e ∂Q
e
+ε + =0
∂x1 ∂t ∂x1
where ∂s/∂t = 0 since s does not depend upon time.

Now, we shall separate terms into ε orders, so that we may categorize them based
upon their dependence on the small parameter ε. That is, we have assumed that ε is
a small deviation from the equilibrium; therefore, by categorizing terms into different
orders of ε, we are organizing them based on how significant they are in terms of that
deviation.
6.5.1.1 Zeroth order terms
For the zeroth order, ε = O(1), and this represents the terms that are not multiplied by ε
at all. Thus, Equation 6.130 becomes
∂Q1
=0 (6.131)
∂x1
As it is expected, we have equilibrium terms that do not change with the perturbation.
Equation 6.130 represents the behavior of the system at its equilibrium state.
It is trivial to integrate Equation 6.131 to find its solutions
ˆ
∂Q1
dx1 = 0
∂x1 (6.132)
Q1 = c
where c an integration constant that must be determined from the boundary conditions.
6.5.1.2 First order terms
This consists of terms that multiplied by a single ε in Equation 6.130, and represent the
effect of the perturbation from the equilibrium state. They quantify how the system
responds to small changes from the equilibrium state.
s
∂e ∂Qe
+ =0
∂t ∂x1
∂Qe s
∂e (6.133)
=−
∂x1 ∂t
ˆ
f1 = s
∂e
Q dx1
∂t
Then, since
s ( x1 , t ) = e
e h(t) + x1 θe(t) (6.134)
Equation 6.133 yields
f1 = − x1 e ( x1 )2 e
Q h(t) − θ (t) + ce(t) (6.135)
2
where ce is another integration constant that can be found from the boundary conditions,
and in the most general case shall depend upon time t only.
6.5.2 Momentum balance
Now, we shall substitute the expanded variables from Equation 6.129 into the momentum
equation given by the Fanning friction factor model, i.e. Eq. 6.114. This gives
 2 
∂ Q 1 + ε Q
f1
s+εes ∂ f1 )2
f 1 ( Q1 + ε Q
f1 + ξ 1 ∂ 

Q1 + ε Q =− ( p + ε pe) −

s+εe s 4 (s + ε e s )2

∂t ∂x1 ρ f ∂x1
(6.136)
We shall apply the same method that we used for the mass balance in order to
categorize the terms given their dependence on the small parameter ε
6.5.2.1 Zeroth order
The terms that are not multiplied by ε are given by
( Q)2 ds s dp f ( Q )2
−ξ 1 + + =0 (6.137)
(s)2 dx1 ρ f dx1 4 (s)2
6.5.2.2 First order
Then, the first order terms imply that

!
e − ξ1 2 Q Qe ds Q d Q
e 3 ( s )2 e
s dp s 1
Q 2
−2 + + + f QQ
e=0 (6.138)
(s) dx1 s dx1 ρf x1 ρf 2
6.6 nonlinear solution
6.6.1 Mass balance integration
We shall integrate Eq. 6.93 in order to solve the mass balance

ˆx1
∂s(η, t) ∂Q1 (η, t)
+ dη = 0 (6.139)
∂t ∂η
0
6.6 nonlinear solution 225
where we use a dummy variable η in order to establish a cummulative sum over the
range (0, x1 ). In other words, since we are using the coordinate x1 in the limits of
integration, we integrate using a dummy variable. Further developing the expression
above, it gives
ˆx1
∂s(η, t)
dη + Q1 ( x1 , t) − Q1 (0, t) = 0 (6.140)
∂t
0
or
Q1 ( x1 , t) = Q1 (0, t) − ė
s(η, t) (6.141)
where we have defined

ˆx1
∂s(η, t)
dη = ė
s ( x1 , t ) (6.142)
∂t
0
which represents the differential general expression for the fluid flow rate when the
wall allows a displacement on x1 .
6.6.2 Momentum balance integration
From the expression for the momentum balance in Equation 6.114, we can obtain
( Q1 )2 f ( Q1 )2

∂p 1 ∂Q1 ∂ξ 1 ∂
= −ρ f + + (6.143)
∂x1 s ∂t ∂s ∂x s 4 ( s )2
Thus, we can integrate on the coordinate x1 using the dummy variable η

ˆx1 ˆx1 ˆx1
( Q1 (η, t))2

∂p(η, t) ρ f ∂Q1 (η, t) ρ f ξ1 ∂
dη = − dη − dη −
∂η s ∂t s ∂η s(η, t)
0 0 0
(6.144)
ˆx1
f ( Q1 (η, t))2
− dη
4 (s(η, t))2
0
where we shall refer to the first, second and third terms on the RHS as transient,
convective and viscous terms respectively. Then, we can solve the LHS for
ˆx1
∂p(η, t)
dη = p( x1 , t) − p(0, t) (6.145)
∂η
0
so that Equation 6.144 might be written as

ˆx1 ˆx1
( Q1 (η, t))2

ρ f ∂Q1 (η, t) ρ f ξ1 ∂
p( x1 , t) = p(0, t) − dη − dη −
s ∂t s ∂η s(η, t)
0 0
(6.146)
ˆx1
f ( Q1 (η, t))2
− dη
4 (s(η, t))2
0
Let us now explore each one of the terms in the RHS.
6.6.2.1 Transient term
The transient term will be noted as

ˆx1
Q̇1 (η, t)
T = −ρ f dη (6.147)
s(η, t)
0
where, recalling the expression for the mass balance in Equation 6.141, we might
rewrite the term Q1 as
ˆx1
Q̇1 (0, t) − ë
s(η, t)
T = −ρ f dη
s(η, t)
0
(6.148)
ˆx1 ˆx1
1 s(η, t)
= −ρ f Q̇1 (0, t)
ë
dη + ρ f dη
s(η, t) s(η, t)
0 0
This way, we may define a transient operator as

ˆx1
T ( f ( x1 )) = ρ f s(η, t)−1 f (η ) dη (6.149)
0
Thus, the transient term might be rewritten depending upon its operator
T ( x1 , t) = −T( Q̇1 ) = −T(1) Q̇1 (0, t) + T(ë

s) (6.150)
where, when not explicetely remarked, it is assumed that Q1 = Q1 ( x1 , t), s = s( x1 , t)

and the same is true for all the other variables.
6.6.2.2 Convective term
The convective term is denoted by

ˆx1 !
1 ∂ ( Q1 (η, t))2
C = −ρ f ξ 1 dη (6.151)
s(η, t) ∂η s(η, t)
0
where again, if we take into account the result from the mass balance, Equation 6.141,
we get
ˆx1 2 !
1 ∂ Q1 (0, t) − ė
s(η, t)
C = −ρ f ξ 1 dη
s(η, t) ∂x1 s(η, t)
0
ˆx1 ˆx1
2 ∂ 1 s(η, t))2
∂ (ė
= −ρ f ξ 1 ( Q1 (0, t)) dη − ρ f ξ 1 dη + (6.152)
∂η (s(η, t))2 ∂η (s(η, t))2
0 0
ˆx1
s(η, t)
∂ ė
+ 2 ρ f ξ 1 Q1 (0, t) dη
∂η (s(η, t))2
0
so that we can define a convective operator

ˆx1
∂ f (η )
C( f ( x1 )) = −ρ f ξ 1 dη (6.153)
∂η (s(η, t))2
0
and therefore, the convective term might be written as
C ( x1 , t) = −C(( Q1 )2 ) = −C(1) ( Q1 (0, t))2 − C((ė

s)2 ) + 2 Q1 (0, t) C(ė
s) (6.154)
6.6.2.3 Viscous term
Lastly, given the expression for Q1 in the mass balance (Eq. 6.141), the viscous term
reads
ˆx1 ˆx1 2
f ( Q1 (η, t))2 f ( Q1 (0, t))2 + ė
s(η, t) − 2 Q1 (0, t) ė
s(η, t)
V=− 2
dη = − 2
dη (6.155)
4 (s(η, t)) 4 (s(η, t))
0 0
In the same fashion as before, we define the viscous operator

ˆx1
f
V ( f ( x1 )) = (s(η, t))−2 f (η ) dη (6.156)
4
0
this way, the viscous term might be written as

2
s)2 + 2 Q1 (0, t) V(ė
V ( x1 , t) = −V ( Q1 )2 = −V(1) ( Q1 (0, t)) − V (ė

s) (6.157)
6.6.2.4 Pressure equation
We shall define a new operator
W ( f ( x1 )) = C ( f ( x1 )) + V ( f ( x1 )) (6.158)
Then, the pressure equation might be written in compact for as
p( x1 , t) = p(0, t) − T( Q̇1 ) − W ( Q1 )2

(6.159)
or explicitely as
s − W(1) ( Q1 (0, t))2 −

p( x1 , t) = p(0, t) − T(1) Q̇1 (0, t) + T ë

(6.160)
− W (ės)2 + 2 W ė

s Q1 (0, t)
6.6.2.5 Double reservoir boundary condition
The value of p(0, t) in Equation 6.160 will be given by the boundary conditions. Note that
we need two boundary conditions due to the extra order of ∂/∂x in the mass balance. Thus,
we define reservoir boundary conditions both for the inlet and for the outlet. Considering
a loosy inlet and outlet, the simplest way is to assume an empirical treatment in which a
portion of the pressure is lost to the kinetic energy, such as
ζ
∆p = ρ (U1 )2 (6.161)
2 f
where ζ is the loss coefficient, which value depends on the Reynolds number and some
geometrical parameters; then U1 is the average velocity along the duct in the primary
direction, obtained by integration.
If we consider the Bernoulli equation at the inlet, we obtain
! 2
ζ i Q1 (0, t) − ė

s(0, t) ζ i Q1 (0, t)
p(0, t) = pi − ρ f = pi − ρ f (6.162)
2 s(0, t) 2 s(0, t)
where pi and ζ i are the reservoir pressure and the inlet loss coefficient respectively,
s(0, t) = 0. Also, we note that the velocity has been replaced by the
and we recall that ė
analogous quantity Q1/s.
Then, if the outlet pressure pe , is known, then we can write Bernoulli’s expression as
!2
ζ o Q1 (0, t) − ė s ( l1 , t )
p ( l1 , t ) = p e + ρ f (6.163)
2 s ( l1 , t )
From the mass balance, Equation 6.141, we had and expression for Q( x1 , t). Therefore,
the outlet boundary condition at x1 = l1 might be rewritten as
2
ζ o Q 1 ( l1 , t )
p ( l1 , t ) = p e + ρ f (6.164)
2 s ( l1 , t )
Substituing the inlet boundary condition, Equation 6.162, into the pressure equation,
Eq. 6.159, we get
ζ i Q1 (0, t) 2

p ( x1 , t ) = p i − ρ f − T( Q̇1 ) − W(( Q1 )2 ) (6.165)
2 s(0, t)
Then, we can evaluate the pressure equation at the coordinate x1 = l1 to yield
ζ i Q1 (0, t) 2

− Tl1 Q̇1 − Wl1 ( Q1 )2

p ( l1 , t ) = p i − ρ f (6.166)
2 s(0, t)
where the quantities Tl1 and Wl1 represent T( f ( x1 )) and W( f ( x1 )) evaluated at

x1 = l1 ; i.e
ˆl1
l1
T ( f ( x1 )) = ρ f s(η, t)−1 f (η ) dη (6.167)
0
and the same is true for all the other operators.

Then, Equation 6.166 must be equal to the outlet boundary condition, Equation 6.164,
so the following relationship must be satisfied
2 2
ζ o Q 1 ( l1 , t ) ζ i Q1 (0, t)
− Tl1 Q̇1 − Wl1 ( Q1 )2 (6.168)

pe + ρ f = pi − ρ f
2 s ( l1 , t ) 2 s(0, t)
6.6.2.6 Flow rate equation
Equation 6.168 might be written explicitely as

2 2
ζ o Q 1 ( l1 , t ) ζ i Q1 (0, t)
pe + ρ f = pi − ρ f −
2 s ( l1 , t ) 2 s(0, t) (6.169)

− T (1) Q̇1 (0, t) + T
l1 l1
s − W l1 ( Q 1 ) 2

ë
Thus, we can now write a differential equation in terms of Q̇1 (0, t) as

" !
1 1 ( Q1 (l1 , t))2 ( Q1 (0, t))2
Q̇1 (0, t) = l −∆p − ρ f ζ o + ζi −
T 1 (1) 2 (s(l1 , t))2 (s(0, t))2
# (6.170)

2
− Wl1 ( Q1 ) + Tl1 ë

s
where ∆p = pe − pi . The equation above might be written explicity to yield

"
1
Q̇1 (t) = − l ∆p+
T 1 (1)
!
ζo ζi
+ ρf 2
+ ρf 2
+ Wl1 (1) ( Q1 (0, t))2 +
2 (s(l1 , t)) 2 (s(0, t))
! (6.171)
s ( l1 , t )
ζ o ė
− Wl1 2 ė

+ −ρ f s Q1 (0, t)+
(s(l1 , t))2
2 !#
ζ o ės ( l1 , t ) 2
− T ë
l1
s +W l1

+ ρf s
2 (s(l1 , t))2
ė
6.6.2.7 Initial equilibrium
We assume that Q̇1 (0, 0) = 0 in order to evaluate the equilibrium. Thus, ë

s = ė
s = 0 and
Equation 6.171 yields
" ! #
1 ζo ζi
Q̇1 (0, 0) = − ∆p + ρf + ρf + W(1) ( Q1 (0, t))2
T(1) 2 (s(l1 , t)) 2
2 (s(0, t)) 2
(6.172)
which results in
v
u −∆p
Q1 (0, 0) = t (6.173)
u
ρf ζo
+ ρf ζi
+ W l1 ( 1 )
2(s(l1 ,t))2 2 (s(0,t)) 2

Main

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Main

Uploaded by

Copyright:

Available Formats

F L U I D - S T R U C T U R E I N T E R A C T I O N F U N D A M E N TA L S

universidad tecnológica nacional - facultad regional de bahía blanca

1 Nonlinear dynamics fundamentals 5

2 Fluid mechanics fundamentals 41

3 Solyd mechanics fundamentals 121

4 Fluid-solid interaction fundamentals 131

is an ordinary differential equation, because it involves only ordinary derivatives dx/dt

is a partial differential equation - it has both time, t, and space, x, as independent

Figure 1.1: Pendulum trajectory

1.1 one dimmensional flows on the line

which denotes a one dimmensional system. Here, x (t) is a real-valued function of

1.1.1 A geometric way of thinking

To evaluate the constant C, suppose that x = x0 at t = 0. Then, C = ln | csc x0 + cot x0 |,

This result is exact, but a headache to interpret. In contrast, a graphical analysis of

Figure 1.2: Graphical representation of Equation 1.9

Figure 1.3: Fixed points and stability

1.1.2 Linear stability analysis

since x ∗ is constant. Thus, η̇ = ẋ = f ( x ) = f ( x ∗ + η ). Now using Taylor’s expansion,

f ( x ∗ + η ) = f ( x ∗ ) + η f ′ ( x ∗ ) + O(η 2 ) + . . . + O(η n ) (1.14)

Figure 1.4: Impossibility of oscillations

Thus, if a fixed point is regarded as an equilibrium solution, the approach to equi-

For example, suppose a mass m is attached to a nonlinear spring whose restoring

Figure 1.5: Mechanical analog: overdamped systems

1.1.3 Euler’s method

x (t0 + ∆t) ≃ x1 = x0 + f ( x0 ) ∆t (1.18)

This is the simplest possible numerical integration scheme. It is known as Euler’s

Figure 1.6: Visualization of Euler’s method

Figure 1.7: Buckle of a beam

1.2.1 Saddle-node bifurcation

where r is a parameter, which may be positive, negative, or zero. When r is negative,

Figure 1.8: Prototypical example of a saddle-node bifurcation

Figure 1.9: Bifurcation diagram of a saddle-node bifurcation

1.2.2 Transcritical bifurcation

Figure 1.10: Prototypical example of a transcritical bifurcation

Figure 1.11: Bifurcation diagram of a transcritical bifurcation

1.2.3 Pitchfork bifurcation

1.2.3.1 Supercritical pitchfork bifurcation

The normal form of the supercritical pitchfork bifurcation is

Figure 1.13: Bifurcation diagram of a supercritical pitchfork bifurcation

1.2.3.2 Subcritical pitchfork bifurcation

then we would have a subcritical pitchfork bifurcation.

Figure 1.14: Bifurcation diagram of a subcritical pitchfork bifurcation

There is no loss in generality in assuming that the coefficients of x3 and x5 are 1.

Figure 1.15: Bifurcation diagram for Eq. 1.24

Figure 1.16: Jumps and hysteresis in the bifurcation diagram

1.2.4 Overdamped bead on a rotating hoop

(a) Bead on a rotating hoop

Figure 1.17: Depiction of the system

m r ϕ̈ = −b ϕ̇ − m g sin ϕ + m r ω 2 sin ϕ cos ϕ (1.25)

Figure 1.18: Forces diagram

b ϕ̇ = −m g sin ϕ + m r ω 2 sin ϕ cos ϕ

Figure 1.19: Plot of cos ϕ vs ϕ

Figure 1.21: Fixed points as a function of γ

where T is a characteristic time scale to be chosen later. When T is chosen correctly,

Hence, Equation 1.25 becomes

Now since this equation is a balance of forces, we nondimensionalize it by dividing by

Each of the terms in parentheses is a dimensionless group. We recognize the group

Then the condition r/g T2 << 1 becomes

Then Eq. 1.33 becomes

Unfortunately, there is something fundamentally wrong with our idea of replacing

1.2.4.1 Phase plane analysis

Figure 1.22: Representation of the phase plane

Along with the definition ϕ′ = Ω, this yields the vector field

Figure 1.23: Pattern of trajectories