Malham-Fluid Mechanics Notes

Lecture notes
Introductory uid mechanics

Simon J.A. Malham
Simon J.A. Malham (17th March 2014)

Maxwell Institute for Mathematical Sciences
and School of Mathematical and Computer Sciences
Heriot-Watt University, Edinburgh EH14 4AS, UK
Tel.: +44-131-4513200
Fax: +44-131-4513249
E-mail: S.J.Malham@ma.hw.ac.uk
Simon J.A. Malham
1 Introduction
The derivation of the equations of motion for an ideal uid by Euler in 1755, and
then for a viscous uid by Navier (1822) and Stokes (1845) were a tour-de-force of
18th and 19th century mathematics. These equations have been used to describe and
explain so many physical phenomena around us in nature, that currently billions of
dollars of research grants in mathematics, science and engineering now revolve around
them. They can be used to model the coupled atmospheric and ocean ow used by
the meteorological oce for weather prediction down to any application in chemical
engineering you can think of, say to development of the thrusters on NASAs Apollo
programme rockets. The incompressible NavierStokes equations are given by
u
+u
t
u=
1
p + f,
u = 0,
where u = u(x, t) is a three dimensional uid velocity, p = p(x, t) is the pressure and
f is an external force eld. The constants and are the mass density and kinematic
viscosity, respectively. The frictional force due to stickiness of a uid is represented by
the term 2 u. An ideal uid corresponds to the case = 0, when the equations above
are known as the Euler equations for a homogeneous incompressible ideal uid. We will
derive the NavierStokes equations and in the process learn about the subtleties of uid
mechanics and along the way see lots of interesting applications.
2 Fluid ow, the Continuum Hypothesis and conservation principles

A material exhibits ow if shear forces, however small, lead to a deformation which is
unboundedwe could use this as denition of a uid. A solid has a xed shape, or at
least a strong limitation on its deformation when force is applied to it. With the category of uids, we include liquids and gases. The main distinguishing feature between
these two uids is the notion of compressibility. Gases are usually compressibleas we
know from everyday aerosols and air canisters. Liquids are generally incompressiblea
feature essential to all modern car braking mechanisms.
Fluids can be further subcatergorized. There are ideal or inviscid uids. In such
uids, the only internal force present is pressure which acts so that uid ows from a
region of high pressure to one of low pressure. The equations for an ideal uid have been
applied to wing and aircraft design (as a limit of high Reynolds number ow). However
uids can exhibit internal frictional forces which model a stickiness property of the
uid which involves energy lossthese are known as viscous uids. Some uids/material known as non-Newtonian or complex uids exhibit even stranger behaviour,
their reaction to deformation may depend on: (i) past history (earlier deformations),
for example some paints; (ii) temperature, for example some polymers or glass; (iii)
the size of the deformation, for example some plastics or silly putty.
For any real uid there are three natural length scales:
1. Lmolecular , the molecular scale characterized by the mean free path distance of
molecules between collisions;
2. Luid , the medium scale of a uid parcel, the uid droplet in the pipe or ocean
ow;
3. Lmacro , the macro-scale which is the scale of the uid geometry, the scale of the
container the uid is in, whether a beaker or an ocean.
And, of course we have the asymptotic inequalities:
Lmolecular
Luid
Lmacro .
Continuum Hypothesis We will assume that the properties of an elementary volume/parcel of uid, however small, are the same as for the uid as a wholei.e. we
suppose that the properties of the uid at scale Luid propagate all the way down
and through the molecular scale Lmolecular . This is the continuum assumption. For
everyday uid mechanics engineering, this assumption is extremely accurate (Chorin
and Marsden [3, p. 2]).
Our derivation of the basic equations underlying the dynamics of uids is based on
three basic conservation principles:
1. Conservation of mass, mass is neither created or destroyed;
2. Newtons 2nd law/balance of momentum, for a parcel of uid the rate of change of
momentum equals the force applied to it;
3. Conservation of energy, energy is neither created nor destroyed.
In turn these principles generate the:
1. Continuity equation which governs how the density of the uid evolves locally and
thus indicates compressibility properties of the uid;
2. NavierStokes equations of motion for a uid which indicates how the uid moves
around from regions of high pressure to those of low pressure and under the eects
of viscosity;
3. Equation of state which indicates the mechanism of energy exchange within the
uid.
3 Trajectories and streamlines

Suppose that our uid is contained with a region/domain D Rd where d = 2 or
3, and x = (x, y, z)T D is a position/point in D. Imagine a small uid particle or
a speck of dust moving in a uid ow eld prescribed by the velocity eld u(x, t) =
(u, v, w)T . Suppose the position of the particle at time t is recorded by the variables
T
T
x(t), y(t), z(t) . The velocity of the particle at time t at position x(t), y(t), z(t) is
d
x(t) = u x(t), y(t), z(t), t ,
dt
d
y(t) = v x(t), y(t), z(t), t ,
dt
d
z(t) = w x(t), y(t), z(t), t .
dt
Denition 1 (Particle path or trajectory) The particle path or trajectory of a uid
particle is the curve traced out by the particle as time progresses. If the particle starts
at position (x0 , y0 , z0 )T then its particle path is the solution to the system of dierential
equations (the same as those above but here in shorter vector notation)
d
x(t) = u(x(t), t),
dt
with initial conditions x(0) = x0 , y(0) = y0 and z(0) = z0 .
Simon J.A. Malham
Denition 2 (Streamline) Suppose for a given uid ow u(x, t) we x time t. A streamline is an integral curve of u(x, t) for t xed, i.e. it is a curve x = x(s) parameterized
by the variable s, that satises the system of dierential equations
d
x(s) = u(x(s), t),
ds
with t held constant.
Remark 1 If the velocity eld u is time-independent, i.e. u = u(x) only, or equivalently
u/t = 0, then trajectories and streamlines coincide. Flows for which u/t = 0 are
said to be stationary.
Example. Suppose a velocity eld u(x, t) = (u, v, w)T is given by
u
y
v = x
w
0
for some constant > 0. Then the particle path for a particle that starts at (x0 , y0 , z0 )T
is the integral curve of the system of dierential equations
dx
= y,
dt
dy
= x
dt
and
dz
= 0.
dt
This is a coupled pair of dierential equations as the solution to the last equation is
z(t) = z0 for all t 0. There are several methods for solving the pair of equations, one
method is as follows. Dierentiating the rst equation with respect to t we nd
d2 x
dy
=
dt
dt2
d2 x
= 2 x.
dt2
In other words we are required to solve the linear second order dierential equation for
x = x(t) shown. The general solution is
x(t) = A cos t + B sin t,
where A and B are arbitrary constants. We can now nd y = y(t) by substituting this
solution for x = x(t) into the rst of the pair of dierential equations as follows:
1 dx
dt
1
= (A sin t + B cos t)
= A sin t B cos t.
y(t) =
Using that x(0) = x0 and y(0) = y0 we nd that A = x0 and B = y0 so the particle

path of the particle that is initially at (x0 , y0 , z0 )T is given by
x(t) = x0 cos t y0 sin t,
y(t) = x0 sin t + y0 cos t
and
z(t) = z0 .
This particle thus traces out a horizontal circular particle path at height z = z0 of
2
radius x2 + y0 . Since this ow is stationary, streamlines coincide with particle paths
0
for this ow.
Example. Consider the two-dimensional ow
u
v
u0
,
v0 cos(kx t)
where u0 , v0 , k and are constants. Let us nd the particle path and streamline for
the particle at (x0 , y0 )T = (0, 0)T at t = 0. Starting with the particle path, we are
required to solve the coupled pair of ordinary dierential equations
dx
= u0
dt
dy
= v0 cos(kx t).
dt
and
We can solve the rst dierential equation which tells us

x(t) = u0 t,
where we used that x(0) = 0. We now substitute this expression for x = x(t) into the
second dierential equation and integrate with respect to time (using y(0) = 0) so
dy
= v0 cos (ku0 )t
dt
t
v0 cos (ku0 ) d
y(t) = 0 +
y(t) =
0
v0
ku0
sin (ku0 )t .
If we eliminate time t between the formulae for x = x(t) and y = y(t) we nd that the
trajectory through (0, 0)T is
y=
v0
sin
ku0
x .
u0
To nd the streamline through (0, 0)T , we x t, and solve the pair of dierential equations
dx
dy
= u0
and
= v0 cos(kx t).
ds
ds
As above we can solve the rst equation so that x(s) = u0 s using that x(0) = 0. We can
substitute this into the second equation and integrate with respect to sremembering
that t is constantto get
dy
= v0 cos(ku0 s t)
ds
s
v0 cos(ku0 r t) dr
y(s) = 0 +
0
y(s) =
v0
sin(ku0 s t) sin(t) .
ku0
If we eliminate the parameter s between x = x(s) and y = y(s) above, we nd the

equation for the streamline is
y=
v0
sin(kx t) + sin(t) .
ku0
Simon J.A. Malham
The equation of the streamline through (0, 0)T at time t = 0 is thus given by
y=
v0
sin(kx).
ku0
As the underlying ow is not stationary, as expected, the particle path and streamline
through (0, 0)T at time t = 0 are distinguished. Finally let us examine two special limits
for this ow. As 0 the ow becomes stationary and correspondingly the particle
path and streamline coincide. As k 0 the ow is not stationary. In this limit the
particle path through (0, 0)T is y = (v0 /) sin(x/u0 ), i.e. it is sinusoidal, whereas the
streamline is given by x = u0 s and y = 0, which is a horizontal straight line through
the origin.
Remark 2 A streakline is the locus of all the uid elements which at some time have
past through a particular point, say (x0 , y0 , z0 )T . We can obtain the equation for a
streakline through (x0 , y0 , z0 )T by solving the equations (d/dt)x(t) = u(x(t), t) asT
suming that at t = t0 we have x(t0 ), y(t0 ), z(t0 )
= (x0 , y0 , z0 )T . Eliminating t0
between the equations generates the streakline corresponding to (x0 , y0 , z0 )T . For example, ink dye injected at the point (x0 , y0 , z0 )T in the ow will trace out a streakline.
4 Continuity equation
Recall, we suppose our uid is contained with a region/domain D Rd (here we will
assume d = 3, but everything we say is true for the collapsed two dimensional case
d = 2). Hence x = (x, y, z)T D is a position/point in D. At each time t we will
suppose that the uid has a well dened mass density (x, t) at the point x. Further,
each uid particle traces out a well dened path in the uid, and its motion along
that path is governed by the velocity eld u(x, t) at position x at time t. Consider an
arbitrary subregion D. The total mass of uid contained inside the region at
time t is
(x, t) dV.
where dV is the volume element in Rd . Let us now consider the rate of change of mass
inside . By the principle of conservation of mass, the rate of increase of the mass in
is given by the mass of uid entering/leaving the boundary of per unit time.
To compute the total mass of uid entering/leaving the boundary per unit time,
we consider a small area patch dS on the boundary of , which has unit outward
normal n. The total mass of uid owing out of through the area patch dS per unit
time is
mass density uid volume leaving per unit time = (x, t) u(x, t) n(x) dS,
where x is at the centre of the area patch dS on . Note that to estimate the uid
volume leaving per unit time we have decomposed the uid velocity at x , time t,
into velocity components normal (u n) and tangent to the surface at that point.
The velocity component tangent to the surface pushes uid across the surfaceno uid
enters or leaves via this component. Hence we only retain the normal component
see Fig. 2.
Fig. 1 The uid of mass density (x, t) swirls around inside the container D, while is an
imaginary subregion.
u.n
dS
Fig. 2 The total mass of uid moving through the patch dS on the surface per unit time,
is given by the mass density (x, t) times the volume of the cylinder shown which is u n dS.
Returning to the principle of conservation of mass, this is now equivalent to the

integral form of the law of conservation of mass:
d
dt
(x, t) dV =
u n dS.
The divergence theorem and that the rate of change of the total mass inside equals
the total rate of change of mass density inside imply, respectively,
(u) dV =
(u) n dS
and
d
dt
dV =
dV.
t
Using these two relations, the law of conservation of mass is equivalent to
+
t
(u) dV = 0.
Now we use that is arbitrary to deduce the dierential form of the law of conservation
of mass or continuity equation that applies pointwise:
+
t
(u) = 0.
This is the rst of our three conservation laws.
Simon J.A. Malham
5 Transport theorem
Recall our image of a small uid particle moving in a uid ow eld prescribed by the
velocity eld u(x, t). The velocity of the particle at time t at position x(t) is
d
x(t) = u(x(t), t).
dt
As the particle moves in the velocity eld u(x, t), say from position x(t) to a nearby
position an instant in time later, two dynamical contributions change: (i) a small instant
in time has elapsed and the velocity eld u(x, t), which depends on time, will have
changed a little; (ii) the position of the particle has changed in that short time as it
moved slightly, and the velocity eld u(x, t), which depends on position, will be slightly
dierent at the new position.
Let us compute the acceleration of the particle to explicitly observe these two
contributions. By using the chain rule we see that
d2
d
x(t) = u x(t), t
dt
dt2
u dx
u dy
u dz
u
=
+
+
+
x dt
y dt
z dt
t
dx
dy
dz
u
+
+
u+
dt x
dt y
dt z
t
u
.
=u u+
t
=
Indeed for any function F (x, y, z, t), scalar or vector valued, the chain rule implies
d
F
F x(t), y(t), z(t), t =
+u
dt
t
F.
Denition 3 (Material derivative) If the velocity eld components are

u = (u, v, w)T
and
+v
+w ,
x
y
z
then we dene the material derivative following the uid to be

D
:=
+u
Dt
t
Suppose that the region within which the uid is moving is D. Suppose is a subregion
of D identied at time t = 0. As the uid ow evolves the uid particles that originally
made up will subsequently ll out a volume t at time t. We think of t as the
volume moving with the uid.
Theorem 1 (Transport theorem) For any function F and density function satisfying
the continuity equation, we have
d
dt
F dV =
t
DF
dV.
Dt
We will use the transport theorem to deduce both the Euler and Cauchy equations of
motion from the primitive integral form of the balance of momentum; see Sections 10
and 15. We now carefully elucidate the steps required for the proof of the Transport
theoremsee Chorin and Marsden [3, pp. 611]. Importantly the concepts of the ow
map in Step 1 and the evolution of its Jacobian in Step 3 will have important ramications in coming sections. The four steps are as follows.
Step 1: Fluid ow map. For a xed position x D we denote by (x, t) = (, , )T
the position of the particle at time t, which at time t = 0 was at x. We use t to denote
the map x (x, t), i.e. t is the map that advances each particle at position x at
time t = 0 to its position at time t later; it is the uid ow-map. Hence, for example
t () = t . We assume t is suciently smooth and invertible for all our subsequent
manipulations.
Step 2: Change of variables. For any two functions and F we can perform the
change of variables from (, t) to (x, t)with J(x, t) the Jacobian for this transformation given by denition as J(x, t) := det (x, t) . Here the gradient operator is with
respect to the x coordinates, i.e.
= x . Note for t we integrate over volume elements dV = dV (), i.e. with respect to the coordinates, whereas for we integrate
over volume elements dV = dV (x), i.e. with respect to the xed coordinates x. Hence
by direct computation
d
dt
F dV =
t
d
dt
d
dt
( F )(, t) dV ()
( F )((x, t), t) J(x, t) dV (x)
d
( F )((x, t), t) J(x, t) dV
dt
d
d
( F )((x, t), t) J(x, t) + ( F )((x, t), t) J(x, t) dV
dt
dt
=
=
=
d
D
( F ) ((x, t), t) J(x, t) + ( F )((x, t), t) J(x, t) dV.
Dt
dt
Step 3: Evolution of the Jacobian. We establish the following result for the Jacobian:
d
J(x, t) =
dt
u((x, t), t) J(x, t).
We know that a particle at position (x, t) = (x, t), (x, t), (x, t)
at x at time t = 0, evolves according to
, which started
d
(x, t) = u (x, t), t .
dt
Taking the gradient with respect to x of this relation, and swapping over the gradient
and d/dt operations on the left, we see that
d
(x, t) =
dt
u (x, t), t .
Using the chain rule we have

xu
(x, t), t =
u (x, t), t
x (x, t)
10
Simon J.A. Malham
Combining the last two relations we see that

d
=(
dt
Abels Theorem then tells us that J = det

d
det
dt
u)
evolves according to
= Tr(
u) det
where Tr denotes the trace operator on matricesthe trace of a matrix is the sum of
its diagonal elements. Since Tr( u) u we have established the required result.
Step 4: Conservation of mass. We see that we thus have
d
dt
D
( F ) + ( F )
Dt
F dV =
D
( F ) +
Dt
=
t
u
u F
((x, t), t) J(x, t) dV

dV
DF
dV,
Dt
where in the last step we have used the conservation of mass equation.
We have thus completed the proof of the Transport Theorem. A straightforward
corollary proved in a manner analogous to that of the Transport Theorem is as follows.
Corollary 1 For any function F = F ((t), t) we have
d
dt
F dV =
t
F
+
t
(F u) dV.
6 Incompressible ow
We now characterize a subclass of ows which are incompressible. The classic examples
are water, and the brake uid in your car whose incompressibility properties are vital
to the eective transmission of pedal pressure to brakepad pressure. Herein we closely
follow the presentation given in Chorin and Marsden [3, pp. 1011].
Denition 4 (Incompressible ow) A ow is said to be incompressible if for any subregion D, the volume of t is constant in time.
Corollary 2 (Equivalent incompressibility statements) The following statements are
equivalent:
1. Fluid is incompressible;
2. Jacobian J 1;
3. The velocity eld u = u(x, t) is divergence free, i.e.
u = 0.
11
Proof Using the result for the Jacobian of the ow map in Step 3 of the proof of the
Transport Theorem, for any subregion of the uid, we see
d
d
vol(t ) =
dt
dt
=
dV ()
t
d
dt
J(x, t) dV (x)
u((x, t), t) J(x, t) dV (x)
u(, t) dV ().
=
t
Further, noting that by denition J(x, 0) = 1, establishes the result.

( u) =
The continuity equation and the identity,
+u
t
u+
u, imply
u = 0.
Hence since > 0, a ow is incompressible if and only if
+u
t
= 0,
i.e. the uid density is constant following the uid.

Denition 5 (Homogeneous uid) A uid is said to be homogeneous if its mass density
is constant in space.
If we set F 1 in the Transport Theorem we get
d
dt
dV = 0.
t
This is equivalent to the statement

(x, t) dV =
t
(x, 0) dV
(x, t), t J(x, t) dV =
(x, 0) dV
where we made a change of variables. Since is arbitrary, we deduce

(x, t), t J(x, t) = (x, 0).
From this relation we see that if the ow is incompressible so J(x, t) 1 then
(x, t), t = (x, 0). Thus if an incompressible uid is homogeneous at time t = 0
then it remains so. If we combine this with the result that mass density is constant
following the uid, then we conclude that is constant in time.
12
Simon J.A. Malham
7 Stream functions
A stream function exists for a given ow u = (u, v, w)T if the velocity eld u is
solenoidal, i.e.
u = 0, and we have an additional symmetry that allows us to
eliminate one coordinate. For example, a two dimensional incompressible uid ow
u = u(x, y, t) is solenoidal since
u = 0, and has the symmetry that it is uniform
with respect to z. For such a ow we see that
u=0
u
v
+
= 0.
x
y
This equation is satised if and only if there exists a function (x, y, t) such that
= u(x, y, t)
y
and
= v(x, y, t).
x
The function is called Lagranges stream function. A stream function is always only
dened up to any arbitrary additive constant. Further note that for t xed, streamlines
are given by constant contour lines of (note u = 0 everywhere).
Note that if we use plane polar coordinates so u = u(r, , t) and the velocity
components are u = (ur , u ) then
u=0
1
1 u
(r ur ) +
= 0.
r r
r
This is satised if and only if there exists a function (r, , t) such that
1
= ur (r, , t)
r
and
= u (r, , t).
r
Example Suppose that in Cartesian coordinates we have the two dimensional ow

u = (u, v)T given by
kx
u
,
=
k y
v
for some constant k. Note that
= kx
y
u = 0 so there exists a stream function satisfying

and
= k y.
x
Consider the rst partial dierential equation. Integrating with respect to y we get
= k xy + C(x)
where C(x) is an arbitrary function of x. However we know that must simultaneously
satisfy the second partial dierential equation above. Hence we substitute this last
relation into the second partial dierential equation above to get
= k y
x
k y + C (x) = k y.
We deduce C (x) = 0 and therefore C is an arbitrary constant. Since a stream function

is only dened up to an arbitrary constant we take C = 0 for simplicity and the stream
function is given by
= k xy.
13
Now suppose we used plane polar coordinates instead. The corresponding ow

u = (ur , u )T is given by
ur
u
k r cos 2
.
k r sin 2
First note that

u = 0 using the polar coordinate form for
Hence there exists a stream function = (r, ) satisfying
1
= k r cos 2
r
and
u indicated above.
= k r sin 2.
r
As above, consider the rst partial dierential equation shown, and integrate with
respect to to get
=
1
2
k r2 sin 2 + C(r).
Substituting this into the second equation above reveals that C (r) = 0 so that C is a
constant. We can for convenience set C = 0 so that
=
1
2
k r2 sin 2.
Comparing this form with its Cartesian equivalent above, reveals they are the same.
8 Rate of strain tensor

Consider a uid ow in a region D R3 . Suppose x and x + h are two nearby points
in the interior of D. How is the ow, or more precisely the velocity eld, at x related
to that at x + h? From a mathematical perspective, by Taylor expansion we have
u(x + h) = u(x) +
u(x) h + O(h2 ),
where ( u) h is simply matrix multiplication of the 3 3 matrix

vector h. Recall that u is given by
u by the column
u/x u/y u/z

u = v/x v/y v/z .
w/x w/y w/z
In the context of uid ow it is known as the rate of strain tensor. This is because,
locally, it measures that rate at which neighbouring uid particles are being pulled
apart (it helps to recall that the velocity eld u records the rate of change of particle
position with respect to time).
Again from a mathematical perspective, we can decompose u as follows. We can
always write
u=
1
2
( u) + ( u)T +
1
2
( u) ( u)T .
We set
D :=
R :=
1
2
1
2
( u) + ( u)T ,
( u) ( u)T .
14
Simon J.A. Malham
Note that D = D(x) is a 3 3 symmetric matrix, while R = R(x) is the 3 3

skew-symmetric matrix given by
0
u/y v/x
R = v/x u/y
0
w/x u/z w/y v/z
u/z w/x
v/z w/y .
0
Note that if we set

1 =
w
v
,
y
z
2 =
u
w
z
x
and
3 =
v
u
,
x
y
then R is more simply expressed as
R=
1
2
0 3 2
3 0 1 .
2 1
0
Further by direct computation we see that

1
R h = 2 h,
where = (x) is the vector with three components 1 , 2 and 3 . At this point, we
have thus established the following.
Theorem 2 If x and x + h are two nearby points in the interior of D, then
1
u(x + h) = u(x) + D(x) h + 2 (x) h + O(h2 ).
The symmetric matrix D is the deformation tensor. Since it is symmetric, there is

an orthonormal basis e1 , e2 , e3 in which D is diagonal, i.e. if X = [e1 , e2 , e3 ] then
d1 0 0
X 1 DX = 0 d2 0 .
0 0 d3
By direct computation, the vector eld above is equivalently given by =
u.
Denition 6 (Vorticity eld) For any velocity vector eld u the associated vector eld
given by
= u,
is known as the vorticity eld of the ow. It encodes the magnitude of, and direction
of the axis about which, the uid rotates, locally.
Now consider the motion of a uid particle labelled by x + h where x is xed and h is
small (for example suppose that only a short time has elapsed). Then the position of
the particle is given by
d
(x + h) = u(x + h)
dt
dh
= u(x + h)
dt
dh
1
u(x) + D(x) h + 2 (x) h.
dt
Let us consider in turn each of the eects on the right shown:
15
1. The term u(x) is simply uniform translational velocity (the particle being pushed
by the ambient ow surrounding it).
2. Now consider the second term D(x) h. If we ignore the other terms then, approximately, we have
dh
= D(x) h.
dt
Making a local change of coordinates so that h = X h we get
h1
h1
d 0 0
d 1
h2 = 0 d2 0 h2 .
dt
0 0 d3
h3
h3
We see that we have pure expansion or contraction (depending on whether di
is positive or negative, respectively) in each of the characteristic directions hi ,

1 h2 h3 satises
i = 1, 2, 3. Indeed the small linearized volume element h
d

(h1 h2 h3 ) = (d1 + d2 + d3 )(h1 h2 h3 ).
dt
Note that d1 + d2 + d3 = Tr(D) = u.
3. Let us now examine the eect of the third term 1 h. Ignoring the other two
2
terms we have
dh
1
= 2 (x) h.
dt
Direct computation shows that
h(t) = (t, (x))h(0),
where (t, (x)) is the matrix that represents the rotation through an angle t
about the axis (x). Note also that (x) h = 0.
9 Internal uid forces

Let us consider the forces that act on a small parcel of uid in a uid ow. There are
two types:
1. external or body forces, these may be due to gravity or external electromagnetic
elds. They exert a force per unit volume on the continuum.
2. surface or stress forces, these are forces, molecular in origin, that are applied by
the neighbouring uid across the surface of the uid parcel.
The surface or stress forces are normal stresses due to pressure dierentials, and shear
stresses which are the result of molecular diusion. We explain shear stresses as follows.
Imagine two neighbouring parcels of uid P and P as shown in Fig. 3, with a mutual
contact surface is S as shown. Suppose both parcels of uid are moving parallel to S
and to each other, but the speed of P , say u, is much faster than that of P , say u .
In the kinetic theory of matter molecules jiggle about and take random walks; they
diuse into their surrounding locale and impart their kinetic energy to molecules they
pass by. Hence the faster molecules in P will diuse across S and impart momentum
to the molecules in P . Similarly, slower molecules from P will diuse across S to slow
the uid in P down. In regions of the ow where the velocity eld changes rapidly over
small length scales, this eect is importantsee Chorin and Marsden [3, p. 31].
16
Simon J.A. Malham

P
u
S
P
Fig. 3 Two neighbouring parcels of uid P and P . Suppose S is the surface of mutual contact
between them. Their respective velocities are u and u and in the same direction and parallel
to S, but with |u|
|u |. The faster molecules in P will diuse across the surface S and
impart momentum to P .
dF
n
(1)
x
(2)
dS
Fig. 4 The force dF on side (2) by side (1) of dS is given by (n) dS.
We now proceed more formally. The force per unit area exerted across a surface
(imaginary in the uid) is called the stress. Let dS be a small imaginary surface in the
uid centred on the point xsee Fig. 4. The force dF on side (2) by side (1) of dS in
the uid/material is given by
dF = (n) dS.
Here is the stress at the point x. It is a function of the normal direction n to the
surface dS, in fact it is given by:
(n) = (x) n.
Note = [ij ] is a 3 3 matrix known as the stress tensor. The diagonal components
of ij , with i = j, generate normal stresses, while the o-diagonal components, with
i = j, generate tangential or shear stresses. Indeed let us decompose the stress tensor
= (x) as follows (here I is the 3 3 identity matrix):
= p I + .
Here the scalar quantity p = p(x, t) is dened to be

1
p := 3 (11 + 22 + 33 )
and represents the uid pressure. The remaining part of the stress tensor = (x) is
known as the deviatoric stress tensor. In this decomposition, the term p I generates
the normal stresses, since if this were the only term present,
= p I
(n) = p n.
The deviatoric stress tensor on the other hand, generates the shear stresses. We will
discuss them in some detail in Section 14 prior to our derivation of the NavierStokes
equations in Section 15.
17
10 Euler equations of uid motion

Consider an arbitrary imaginary subregion D identied at time t = 0, as in Fig. 1.
As the uid ow evolves to some time t > 0, let t denote the volume of the uid
occupied by the particles that originally made up . The total force exerted on the
uid inside t through the normal stresses exerted across its boundary t is given
by
(pI) n dS
( p) dV.
t
If f is a body force (external force) per unit mass, which can depend on position and
time, then the body force on the uid inside t is
f dV.
t
Thus on any parcel of uid t , the total force acting on it is

p + f dV.
t
Hence using Newtons 2nd law (force = mass acceleration) we have

d
dt
p + f dV.
u dV =
t
Now we use the transport theorem with F u and that and thus t are arbitrary.
We see that for at each x D and t
0, the following relation must holdthe
dierential form of the balance of momentum in this case:
Du
= p+
Dt
+ f.
Thus for an ideal uid for which we only include normal stresses and completely ignore
any shear stresses, the uid ow is governed by the Euler equations of motion (derived
by Euler in 1755) given by:
u
+u
t
u=
1
p + f,
Now that we have partial dierential equations that determine how uid ows
evolve, we complement them with boundary and initial conditions. The initial condition
is the velocity prole u = u(x, 0) at time t = 0. It is the state in which the ow
starts. To have a well-posed evolutionary partial dierential system for the evolution
of the uid ow, we also need to specify how the ow behaves near boundaries. Here
a boundary could be a rigid boundary, for example the walls of the container the uid
is conned to or the surface of an obstacle in the uid ow. Another example of a
boundary is the free surface between two immiscible uidssuch as between seawater
and air on the ocean surface. Here we will focus on rigid boundaries.
For ideal uid ow, i.e. one evolving according to the Euler equations, we simply
need to specify that there is no net ow normal to the boundarythe uid does not
cross the boundary but can move tangentially to it. Mathematically this is means that
we specify that
un=0
18
Simon J.A. Malham

p
r
a
Fig. 5 Water draining from a bath.
everywhere on the rigid boundary. For many examples of simple Euler ows we refer
the reader ahead to Section 17all the examples there are Euler ows except for the
last one. Extending those examples further, we now examine an every day ow.
Example (sink or bath drain) As we have all observed when water runs out of a bath
or sink, the free surface of the water directly over the drain hole has a depression in
itsee Fig. 5. The question is, what is the form/shape of this free surface depression?
The essential idea is we know that the pressure at the free surface is uniform,
it is atmospheric pressure, say P0 . We need the Euler equations for a homogeneous
incompressible uid in cylindrical coordinates (r, , z) with the velocity eld u =
(ur , u , uz )T . These are
u2
ur
1 p
+ (u )ur =
+ fr ,
t
r
r
u
ur u
1 p
+ (u )u +
=
+ f ,
t
r
r
uz
1 p
+ (u )uz =
+ fz ,
t
z
where p = p(r, , z, t) is the pressure, is the uniform constant density and f =
(fr , f , fz )T is the body force per unit mass. Here we also have
u
= ur
u
+
+ uz .
r
r
z
Further the incompressibility condition
u = 0 is given in cylindrical coordinates by
1 (rur )
1 u
uz
+
+
= 0.
r r
r
z
We need to make some sensible simplifying assumptions to reduce this system of
equations to a set of partial dierential equations we might be able to solve analytically.
We will assume the uid has uniform density , that the ow is steady, and ur = uz = 0,
i.e. only the azimuthal velocity is non-zero so that the water particles move in horizontal
circlessee Fig. 5. We further assume fr = f = 0. The force due to gravity implies
fz = g. The whole problem is also symmetric with respect to , so we will also
19
assume all partial derivatives with respect to are zero. Combining all these facts
reduces Eulers equations above to
u2
1 p
=
,
r
r
1 p
0=
r
1 p
0=
g.
z
The incompressibility condition is satised trivially. The second equation above tells
us the pressure p is independent of , as we might have already suspected. Hence we
assume p = p(r, z) and focus on the rst and third equation above.
The question now is can we nd functions u = u (r, z) and p = p(r, z) that satisfy
the rst and third partial dierential equations above? To help us in this direction
we will make some further assumptions. We will suppose that as the water ows out
through the hole at the bottom of a bath the residual rotation is conned to a core of
radius a, so that the water particles may be taken to move on horizontal circles with
u =
r,
a2
r ,
r > a.
a,
The azimuthal ow we assume for r

a represents solid body rotation in the core
region. The ow we assume for r > a represents two-dimensional irrotational ow
generated by a point source at the origin. With u = u (r) assumed to have this form,
the question now is, can we nd a corresponding pressure eld p = p(r, z) so that the
rst and third equations above are satised?
Assume r a. Using that u = r in the rst equation we see that
p
= 2 r
r
p(r, z) = 1 2 r2 + C(z),
2
where C(z) is an arbitrary function of z. If we then substitute this into the third
equation above we see that
1 p
= g
z
C (z) = g,
and hence C(z) = gz + C0 where C0 is an arbitrary constant. Thus we now deduce

that the pressure function is given by
p(r, z) = 1 2 r2 gz + C0 .
2
At the free surface of the water, the pressure is constant atmospheric pressure P0 and
so if we substitute this into this expression for the pressure we see that
P0 = 1 2 r2 gz + C0
2
z = ( 2 /2g) r2 (C0 P0 )/g.
Hence the depression in the free surface for r a is a parabolic surface of revolution.
Note that pressure is only ever globally dened up to an additive constant so we are
at liberty to take C0 = 0 or C0 = P0 if we like.
20
Simon J.A. Malham
For r > a a completely analogous argument using u = a2 /r shows that

p(r, z) =
2 a4
gz + K0 ,
2 r2
where K0 is an arbitrary constant. Since the pressure must be continuous at r = a, we

substitute r = a into the expression for the pressure here for r > a and the expression
for the pressure for r a, and equate the two. This gives
1 2 a2 gz + K0 = 1 2 a2 gz
2
2
K0 = 2 a2 .
Hence the pressure for r > a is given by

p(r, z) =
2 a4
gz + 2 a2 .
2 r2
Using that the pressure at the free surface is p(r, z) = P0 , we see that for r > a the
free surface is given by
2 a4
2 a2
z=
.
+
2
g
gr
Remark 3 The fact that there are no tangential forces in an ideal uid has some important consequences, quoting from Chorin and Marsden [3, p. 5]:
...there is no way for rotation to start in a uid, nor, if there is any at the
beginning, to stop... ... even here we can detect trouble for ideal uids because
of the abundance of rotation in real uids (near the oars of a rowboat, in
tornadoes, etc. ).
In particular see DAlemberts paradox in Section 12. We discuss some further consequences of Euler ow in Appendix D.
11 Bernoullis Theorem
Theorem 3 (Bernoullis Theorem) Suppose we have an ideal homogeneous incompressible stationary ow with a conservative body force f = , where is the potential
function. Then the quantity
p
H := 1 |u|2 + +
2
is constant along streamlines.

Proof We need the following identity that can be found in Appendix A:
1
2
|u|2 = u
u+u(
u).
Since the ow is stationary, Eulers equation of motion for an ideal uid imply
u
u=
21
Using the identity above we see that

|u|2 u (
1
2
u) =
p
+ =u(
u)
H =u(
2
1
2 |u|
u),
using the denition for H given in the theorem. Now let x(s) be a streamline that
satises x (s) = u x(s) . By the fundamental theorem of calculus, for any s1 and s2 ,
s2
H x(s2 ) H x(s1 ) =
dH x(s)
s1
s2
H x (s) ds
=
s1
s2
u(
u) u x(s) ds
s1
= 0,
where we used that (u a) u 0 for any vector a (since u a is orthogonal to u).
Since s1 and s2 are arbitrary we deduce that H does not change along streamlines.
Remark 4 Note that H has the units of an energy density. Since is constant here,
we can interpret Bernoullis Theorem as saying that energy density is constant along
streamlines.
Example (Torricelli 1643). Consider the problem of an oil drum full of water that
has a small hole punctured into it near the bottom. The problem is to determine the
velocity of the uid jetting out of the hole at the bottom and how that varies with the
amount of water left in the tankthe setup is shown in Fig 6. We shall assume the hole
has a small cross-sectional area . Suppose that the cross-sectional area of the drum,
and therefore of the free surface (water surface) at z = 0, is A. We naturally assume
A
. Since the rate at which the amount of water is dropping inside the drum must
equal the rate at which water is leaving the drum through the punctured hole, we have
We observe that A
dh
dt
A=U
, i.e. /A
dh
dt
U.
A
1, and hence we can deduce

dh
dt
1
U2
1.
Since the ow is quasi-stationary, incompressible as its water, and there is conservative body force due to gravity, we apply Bernoullis Theorem for one of the typical
streamlines shown in Fig. 6. This implies that the quantity H is the same at the free
surface and at the puncture hole outlet, hence
1
2
dh
dt
P0
P
= 1 U 2 + 0 gh.
2
22
Simon J.A. Malham

P = air pressure
0
z=0
z=h
P
0
Typical streamline
Fig. 6 Torricelli problem: the pressure at the top surface and outside the puncture hole is
atmospheric pressure P0 . Suppose the height of water above the puncture is h. The goal is to
determine how the velocity of water U out of the puncture hole varies with h.
Thus cancelling the P0 / terms then we can deduce that

1
gh = 2 U 2
1
2
dh
dt
= 1U2 1
2
1
U2
= 1U2 1
2
dh
dt
1
2U2
for /A
1
1 with an error of order (/A)2 . Thus in the asymptotic limit gh = 2 U 2 so
U=
2gh.
Remark 5 Note the pressure inside the container at the puncture hole level is P0 +gh.
The dierence between this and the atmospheric pressure P0 outside, accelerates the
water through the puncture hole.
Example (Channel ow: Froude number). Consider the problem of a steady ow
of water in a channel over a gently undulating bedsee Fig 7. We assume that the
ow is shallow and uniform in cross-section. Upstream the ow is characterized by ow
velocity U and depth H. The ow then impinges on a gently undulating bed of height
y = y(x) as shown in Fig 7, where x measures distance downstream. The depth of the
ow is given by h = h(x) whilst the uid velocity at that point is u = u(x), which is
uniform over the depth throughout. Re-iterating slightly, our assumptions are thus,
dy
dx
(bed gently undulating)
23
P0
h(x)
H
y(x)
x
Fig. 7 Channel ow problem: a steady ow of water, uniform in cross-section, ows over a

gently undulating bed of height y = y(x) as shown. The depth of the ow is given by h = h(x).
Upstream the ow is characterized by ow velocity U and depth H.
and
dh
dx
(small variation in depth).
The continuity equation (incompressibility here) implies that for all x,

uh = U H.
Then Eulers equations for a steady ow imply Bernoullis theorem which we apply
to the surface streamline, for which the pressure is constant and equal to atmospheric
pressure P0 , hence we have for all x:
1 2
2U
+ gH = 1 u2 + g(y + h).
2
Substituting for u = u(x) from the incompressibility condition above, and rearranging,
Bernoullis theorem implies that for all x we have the constraint
y=
(U H)2
U2
+H h
.
2g
2gh2
We can think of this as a parametric equation relating the uid depth h = h(x) to the
undulation height h = h(x) where the parameter x runs from x = far upstream
to x = + far downstream. We plot this relation, y as a function of h, in Fig 8. Note
that y has a unique global maximum y0 coinciding with the local maximum and given
by
dy
=0
dh
h = h0 =
(U H)2/3
.
g 1/3
Note that if we set

F := U/
gH
2/3
then h0 = HF , where F is known as the Froude number. It is a dimensionless

function of the upstream conditions and represents the ratio of the oncoming uid
speed to the wave (signal) speed in uid depth H.
Note that when y = y(x) attains its maximum value at h0 , then y = y0 where
y0 := H 1 + 1 F2 3 F2/3 .
2
2
24
Simon J.A. Malham
y
y0
Fig. 8 Channel ow problem: The ow depth h = h(x) and undulation height y = y(x) are
related as shown, from Bernoullis theorem. Note that y has a maximum value y0 at height
h0 = HF2/3 where F = U/ gH is the Froude number.
y0 / H
F<1
F=1
F>1
Fig. 9 Channel ow problem: Two dierent values of the Froude number F give the same
maximum permissible undulation height y0 . Note we actually plot the normalized maximum
possible height y0 /H on the ordinate axis.
This puts a bound on the height of the bed undulation that is compatible with the
upstream conditions. In Fig 9 we plot the maximum permissible height y0 the undulation is allowed to attain as a function of the Froude number F. Note that two dierent
values of the Froude number F give the same maximum permissible undulation height
y0 , one of which is slower and one of which is faster (compared with gH).
Let us now consider and actual given undulation y = y(x). Suppose that it attains
an actual maximum value ymax . There are three cases to consider, in turn we shall
consider ymax < y0 , the more interesting case, and then ymax > y0 . The third case
ymax = y0 is an exercise (see the Exercises section at the end of these notes).
In the rst case, ymax < y0 , as x varies from x = to x = +, the undulation
height y = y(x) varies but is such that y(x)
ymax . Refer to Fig. 8, which plots
the constraint relationship between y and h resulting from Bernoullis theorem. Since
y(x)
ymax as x varies from to +, the values of (h, y) are restricted to part
of the branches of the graph either side of the global maximum (h0 , y0 ). In the gure
these parts of the branches are the locale of the shaded sections shown. Note that the
derivative dy/dh = 1/(dh/dy) has the same xed (and opposite) sign in each of the
branches. In the branch for which h is small, dy/dh > 0, while the branch for which
h is larger, dy/dh < 0. Indeed note the by dierentiating the constraint condition, we
have
(U H)2
dy
.
= 1
dh
gh3
25
F<1
F>1
Fig. 10 Channel ow problem: for the case ymax < y0 , when F < 1, as the bed height y
increases, the uid depth h decreases and vice-versa. Hence we see a depression in the uid
surface above a bump in the bed. On the other hand, when F > 1, as the bed height y increases,
the uid depth h increases and vice-versa. Hence we see an elevation in the uid surface above
a bump in the bed.
Using the incompressibility condition to substitute for U H we see that this is equivalent
to
dy
u2
= 1
.
dh
gh
We can think of u/ gh as a local Froude number if we like. In any case, note that since
we are in one branch or the other, and in either case the sign of dy/dh is xed, this
means that using the expression for dy/dh we just derived, for any ow realization the
sign of 1 u2 /gh is also xed. When x = this quantity has the value 1 U 2 /gH.
Hence the sign of 1 U 2 /gH determines the sign of 1 u2 /gh. Hence if F < 1 then
U 2 /gH = F2 < 1 and therefore for all x we must have u2 /gh < 1. And we also deduce
in this case that we must be on the branch for which h is relatively large as dy/dh is
negative. The ow is said to be subcritical throughout and indeed we see that
dh
=
dy
dy
dh
= 1
u2
gh
< 1
d
(h + y) < 0.
dy
Hence in this case, as the bed height y increases, the uid depth h decreases and viceversa. On the other hand if F > 1 then U 2 /gH > 1 and thus u2 /gh > 1. We must be
on the branch for which h is relatively small as dy/dh is positive. The ow is said to
be supercritical throughout and we have
u2
dh
= 1
dy
gh
>0
d
(h + y) > 1.
dy
Hence in this case, as the bed height y increases, the uid depth h increases and viceversa. Both cases, F < 1 and F > 1, are illustrated by a typical scenario in Fig. 10.
In the second case, ymax > y0 , the undulation height is larger than the maximum
permissible height y0 compatible with the upstream conditions. Under the conditions
we assumed, there is no ow realized here. In a real situation we may imagine a ow
impinging on a large barrier with height ymax > y0 , and the result would be some
sort of reection of the ow occurs to change the upstream conditions in an attempt
to make them compatible with the obstacle. (Our steady ow assumption obviously
breaks down here.)
26
Simon J.A. Malham
12 Irrotational/potential ow
Many ows have extensive regions where the vorticity is zero; some have zero vorticity
everywhere. We would call these, respectively, irrotational regions of the ow and
irrotational ows. In such regions
=
u = 0.
Hence the eld u is conservative and there exists a scalar function such that
u=
The function is known as the ow potential. Note that u is conservative in a region

if and only if the circulation
u dx = 0
C
for all simple closed curves C in the region.

If the uid is also incompressible, then is harmonic since
u = 0 implies
= 0.
Hence for such situations, we in essence need to solve Laplaces equation = 0 subject
to certain boundary conditions. For example for an ideal ow, u n = n = /n
is given on the boundary, and this would constitute a Neumann problem for Laplaces
equation.
Example (linear two-dimensional ow) Consider the ow eld u = (kx, ky)T
1
where k is a constant. It is irrotational. Hence there exists a ow potential = 2 k(x2
2
y ). Since u = 0 as well, we have = 0. Further, since this ow is two-dimensional,
there also exists a stream function = kxy.
Example (line vortex) Consider the ow eld (ur , u , uz )T = (0, k/r, 0)T where
k > 0 is a constant. This is the idealization of a thin vortex tube. Direct computation
shows that
u = 0 everywhere except at r = 0, where
u is innite. For
r > 0, there exists a ow potential = k. For any closed circuit C in this region the
circulation is
u dx = 2k N
C
where N is the number of times the closed curve C winds round the origin r = 0.
The circulation will be zero for all circuits reducible continuously to a point without
breaking the vortex.
Example (DAlemberts paradox) Consider a uniform ow into which we place an
obstacle. We would naturally expect that the obstacle represents an obstruction to the
uid ow and that the ow would exert a force on the obstacle, which if strong enough,
might dislodge it and subsequently carry it downstream. However for an ideal ow, as
we are just about to prove, this is not the case. There is no net force exerted on an
obstacle placed in the midst of a uniform ow.
We thus consider a uniform ideal ow into which is placed a sphere, radius a.
The set up is shown in Fig. 11. We assume that the ow around the sphere is steady,
incompressible and irrotational. Suppose further that the ow is axisymmetric. By this
we mean the following. Use spherical polar coordinates to represent the ow with the
27
U
r
U
U
Fig. 11 Consider an ideal steady, incompressible, irrotational and axisymmetric ow past a

sphere as shown. The net force exerted on the sphere (obstacle) in the ow is zero. This is
DAlemberts paradox.
south-north pole axis passing through the centre of the sphere and aligned with the
uniform ow U at innity; see Fig. 11. Then the ow is axisymmetric if it is independent
of the azimuthal angle of the spherical coordinates (r, , ). Further we also assume
no swirl so that u = 0.
Since the ow is incompressible and irrotational, it is a potential ow. Hence we
seek a potential function such that = 0. In spherical polar coordinates this is
equivalent to
1
r2
+
sin
= 0.
r
sin
r2 r
The general solution to Laplaces equation is well known, and in the case of axisymmetry the general solution is given by
An rn +
(r, ) =
n=0
Bn
rn+1
Pn (cos )
where Pn are the Legendre polynomials; with P1 (x) = x. The coecients An and Bn
are constants, most of which, as we shall see presently, are zero. For our problem we
have two sets of boundary data. First, that as r in any direction, the ow eld
is uniform and given by u = (0, 0, U )T (expressed in Cartesian coordinates with the
z-axis aligned along the south-north pole) so that as r
U r cos .
Second, on the sphere r = a itself we have a no normal ow condition
= 0.
r
Using the rst boundary condition for r we see that all the An must be zero
except A1 = U . Using the second boundary condition on r = a we see that all the
1
Bn must be zero except for B1 = 2 U a3 . Hence the potential for this ow around the
sphere is
= U (r + a3 /2r2 ) cos .
28
Simon J.A. Malham
In spherical polar coordinates, the velocity eld u =
is given by
u = (ur , u ) = U (1 a3 /r3 ) cos , U (1 + a3 /2r3 ) sin .

Since the ow is ideal and steady as well, Bernoullis theorem applies and so along a
1
typical streamline 2 |u|2 + P/ is constant. Indeed since the conditions at innity are
uniform so that the pressure P and velocity eld U are the same everywhere there,
this means that for any streamline and in fact everywhere for r a we have
1 2
1
|u| + P/ = U 2 + P /.
2
2
Rearranging this equation and using our expression for the velocity eld above we have
P P
= 1 U 2 1 (1 a3 /r3 )2 cos2 (1 + a3 /2r3 )2 sin2 .
2
On the sphere r = a we see that

P P
= 1U2 1
2
9
4
sin2 .
Note that on the sphere, the pressure is symmetric about = 0, /2, , 3/2. Hence
the uid exerts no net force on the sphere! (There is no drag or lift.) This result, in
principle, applies to any shape of obstacle in such a ow. In reality of course this cannot
be the case, the presence of viscosity remedies this paradox (and crucially generates
vorticity).
13 Kelvins circulation theorem, vortex lines and tubes

We turn our attention to important concepts centred on vorticity in a ow.
Denition 7 (Circulation) Let C be a simple closed contour in the uid at time t = 0.
Suppose that C is carried along by the ow to the closed contour Ct at time t, i.e.
Ct = t (C). The circulation around Ct is dened to be the line integral
u dx.
Ct
Using Stokes Theorem an equivalent denition for the circulation is

u dx =
Ct
u) n dS =
n dS
S
where S is any surface with perimeter Ct ; see Fig. 13. In other words the circulation is
equivalent to the ux of vorticity through the surface with perimeter Ct .
Theorem 4 (Kelvins circulation theorem (1869)) For ideal, incompressible ow without external forces, the circulation for any closed contour Ct is constant in time.
Proof Using a variant of the Transport Theorem and the Euler equations, we see
d
dt
u dx =
Ct
Ct
for closed loops of uid particles Ct .
Du
dx =
Dt
p dx = 0,
Ct
29
S1
S2
Ct
S0
Fig. 12 Stokes theorem tells us that the circulation around the closed contour C equals the
ux of vorticity through any surface whose perimeter is C. For example here the ux of vorticity
through S0 , S1 and S2 is the same.
C
S
Fig. 13 The strength of the vortex tube is given by the circulation around any curve C that
encircles the tube once.
Corollary 3 The ux of vorticity across a surface moving with the uid is constant in
time.
Denition 8 (Vortex lines) These are the lines that are everywhere parallel to the
local vorticity , i.e. with t xed they solve (d/ds)x(s) = (x(s), t). These are the
trajectories for the eld for t xed.
Denition 9 (Vortex tube) This is the surface formed by the vortex lines through the
points of a simple closed curve C; see Fig. 13. We can dene the strength of the vortex
tube to be
n dS
S
u dx.
Ct
Remark 6 This is a good denition because it is independent of the precise crosssectional area S, and the precise circuit C around the vortex tube taken (because
0); see Fig. 13. Vorticity is larger where the cross-sectional area is smaller and
vice-versa. Further, for an ideal uid, vortex tubes move with the uid and the strength
of the vortex tube is constant in time as it does so (Helmholtzs theorem; 1858); see
Chorin and Marsden [3, p. 26].
14 Shear stresses
Recall our discussion on internal uid forces in Section 9. We now consider the explicit
form of the shear stresses and in particular the deviatoric stress tensor. This is necessary
if we want to consider/model any real uid (i.e. non-ideal uid). We assume that the
30
Simon J.A. Malham
deviatoric stress tensor is a function of the rate of strain tensor u. We shall make
three assumptions about the deviatoric stress tensor and its dependence on the
velocity gradients u. These are that it is:

1. Linear: each component of is linearly related to the rate of strain tensor
2. Isotropic: if U is an orthogonal matrix, then
u.
u U 1 U ( u) U 1 .
Equivalently we might say that it is invariant under rigid body rotations.

3. Symmetric; i.e. ij = ji . This can be deduced as a result of balance of angular
momentum.
Hence each component of the deviatoric stress tensor is a linear function of each
of the components of the velocity gradients u. This means that there is a total of
81 constants of proportionality. We will use the assumptions above to systematically
reduce this to 2 constants.
When the uid performs rigid body rotation, there should be no diusion of momentum (the whole mass of uid is behaving like a solid body). Recall our decomposition
of the rate of strain tensor, u = D + R, where D is the deformation tensor and R
generates rotation. Thus only depends on the symmetric part of u, i.e. it is a linear
function of the deformation tensor D. Further, since is symmetric, we can restrict our
attention to linear functions from symmetric matrices to symmetric matrices. We now

lean heavily on the isotropy assumption 2; see Gurtin [7, Section 37] for more details.
First, we have the transfer theorem.
Theorem 5 (Transfer theorem) Let be an endomorphism on the set of 3 3
symmetric matrices. Then if is isotropic, the symmetric matrices D and (D) are
simultaneously diagonalizable.
Proof Let e be an eigenvector of D and let U be the orthogonal matrix denoting
reection in the plane perpendicular to e, so that U e = e, while any vector perpendicular to e is invariant under U . The eigenstructure of D is invariant to such
a transformation so that U DU 1 = D. Thus, since = (D) is isotropic, we have
U U 1 = (U DU 1 ) = (D) and thus U = U . Any such commuting matrices
share eigenvectors since U e = U e = e. Thus e is also an eigenvector of the
reection transformation U corresponding to the same eigenvalue 1. Thus e is pro

portional to e and so e is an eigenvector of . Since e was any eigenvector of D, the
statement of the theorem follows.

Second, for any 3 3 matrix A with eigenvalues 1 , 2 , 3 , the three scalar functions
I1 (A) := Tr A,
I2 (A) :=
1
2
(TrA)2 Tr(A2 )
and
I2 (A) := det A,
are isotropic. This can be checked by direct computation. Indeed these three functions
are the elementary symmetric functions of the eigenvalues of A:
I1 (A) = 1 + 2 + 3 ,
I2 (A) = 1 2 + 2 3 + 2 3
and
I2 (A) = 1 2 3 .
We have the following representation theorem for isotropic functions.
31
Theorem 6 (Representation theorem) An endomorphism on the set of 3 3
symmetric matrices is isotropic if and only if it has the form

(D) = 0 I + 1 D + 2 D2 ,
for every symmetric matrix D, where 0 , 1 and 2 are scalar functions that depend
only on the isotropic invariants I1 (D), I2 (D) and I3 (D).
Proof Scalar functions = (D) are isotropic if and only if they are functions of the
isotropic invariants of D only. The if part of this statement follows trivially as the
isotropic invariants are isotropic. The only if statement is established if, assuming
is isotropic, we are able to show that
Ii (D) = Ii (D ) for i = 1, 2, 3
(D) = (D ).
Since the map between the eigenvalues of D and its isotropic invariants is bijective,
if Ii (D) = Ii (D ) for i = 1, 2, 3, then D and D have the same eigenvalues. Since
the isospectral action U DU 1 of orthogonal matrices U on symmetric matrices D is
transitive, there exists an orthogonal matrix U such that D = U DU 1 . Since is
isotropic, (U DU 1 ) = (D), i.e. (D ) = (D).
Now let us consider the symmetric matrix valued function . The if statement of
the theorem follows by direct computation and the result we just established for scalar
isotropic functions. The only if statement is proved as follows. Assume has three
distinct eigenvalues (we leave the other possibilities as an exercise). Using the transfer
theorem and the Spectral Theorem (see for example Meyer [15, p. 517]) we have
3
i Ei
(D) =
i=1
where 1 , 2 and 3 are the eigenvalues of and the projection matrices E1 , E2 and

E3 have the properties Ei Ej = O when i = j and E1 + E2 + E3 = I. Since we have

span{I, D, D2 } = span{E1 , E2 , E3 },
there exist scalars 0 , 1 and 2 depending on D such that
(D) = 0 I + 1 D + 2 D2 .
We now have to show that 0 , 1 and 2 are isotropic. This follows by direct computation, combining this last representation with the property that is isotropic.
Remark 7 Note that neither the transfer theorem nor the representation theorem require that the endomorphism is linear.
Third, now suppose that is a linear function of D. Thus for any symmetric matrix
D it must have the form

(D) = I + 2 D,
where the scalars and depend on the isotropic invariants of D. By the Spectral
Theorem we have
3
D=
di Ei ,
i=1
32
Simon J.A. Malham
where d1 , d2 and d3 are the eigenvalues of D and E1 , E2 and E3 are the corresponding projection matricesin particular each Ei is symmetric with an eigenvalue 1 and
double eigenvalue 0. Since is linear we have
(D) =
di (Ei )
i=1
3
di I + 2 Ei .
i=1
where for each i = 1, 2, 3 the only non-zero isotropic invariant is I1 (Ei ) = 1 so that
and are simply constant scalars. Using that E1 + E2 + E3 = I we have
= (d1 + d2 + d3 )I + 2D.
Recall that d1 + d2 + d3 =
u. Thus we have
= (
If we set = +
2
3
u)I + 2D.
this last relation becomes

1
= 2 D 3 (
u)I + (
u)I,
where and are the rst and second coecients of viscosity, respectively.
Remark 8 Note that if u = 0, then the linear relation between and D is homoge
neous, and we have the key property of what is known as a Newtonian uid : the stress
is proportional to the rate of strain.
15 NavierStokes equations
Consider again an arbitrary imaginary subregion of D identied at time t = 0, as
in Fig. 1. As in our derivation of the Euler equations, let t denote the volume of the
uid occupied by the particles at t > 0 that originally made up . The total force
exerted on the uid inside t through the stresses exerted across its boundary t is
given by
(pI + ) n dS
( p +
) dV,
where (for convenience here set (x1 , x2 , x3 )T (x, y, z)T and (u1 , u2 , u3 )T (u, v, w)T )
3
]i =
j=1
ij
xj
3
= [ (
u)]i + 2
j=1
3
= [ (
u)]i +
j=1
3
= [ (
u)]i +
j=1
= ( + )[ (
Dij
xj
xj
uj
ui
+
xj
xi
2 uj
2 ui
+
2
xi xj
xj
u)]i +
ui .
33
If f is a body force (external force) per unit mass, which can depend on position and
time, then on any parcel of uid t , the total force acting on it is
p+
+ f dV.
Hence using Newtons 2nd law we have

d
dt
p+
u dV =
t
+ f dV.
Using the transport theorem with F u and that and thus t are arbitrary, we
see for each x D and t 0, we can deduce the following relation known as Cauchys
equation of motion:
Du
= p + + f.
Dt
Combining this with the form for we deduced above, we arrive at
Du
= p + ( + ) (
Dt
u) + u + f ,
where = 2 is the Laplacian operator. These are the NavierStokes equations. If we

assume we are in three dimensional space so d = 3, then together with the continuity
equation we have four equations, but ve unknownsnamely u, p and . Thus for a
compressible uid ow, we cannot specify the uid motion completely without specifying one more condition/relation. (We could use the principle of conservation of energy
to establish as additional relation known as the equation of state; in simple scenarios
this takes the form of relationship between the pressure p and density of the uid.)
For an incompressible homogeneous ow for which the density is constant, we get
a complete set of equations known as the NavierStokes equations for an incompressible
ow :
u
+u
t
u = u
1
p + f,
u = 0,
where = / is the coecient of kinematic viscosity. Note that we have a closed
system of equations: we have four equations in four unknowns, u and p.
Remark 9 Often the factor 1/ is scaled into the pressure and thus explicitly omitted:
since is constant ( p)/ (p/), and we re-label the term p/ to be p.
As for the Euler equations of motion for an ideal uid, we need to specify initial
and boundary conditions. For viscous ow we specify an additional boundary condition
to that we specied for the Euler equations. This is due to the inclusion of the extra
term u which increases the number of spatial derivatives in the governing evolution
equations from one to two. Mathematically, we specify that
u=0
everywhere on the rigid boundary, i.e. in addition to the condition that there must
be no net normal ow at the boundary, we also specify there is no tangential ow
there. The uid velocity is simply zero at a rigid boundary; it is sometimes called
no-slip boundary conditions. Experimentally this is observed as well, to a very high
34
Simon J.A. Malham
degree of precision; see Chorin and Marsden [3, p. 34]. (Dye can be introduced into a
ow near a boundary and how the ow behaves near it observed and measured very
accurately.) Further, recall that in a viscous uid ow we are incorporating the eect
of molecular diusion between neighbouring uid parcelssee Fig. 3. The rigid nonmoving boundary should impart a zero tangential ow condition to the uid particles
right up against it. The no-slip boundary condition crucially represents the mechanism
for vorticity production in nature that can be observed everywhere. Just look at the
ow of a river close to the river bank.
Remark 10 At a material boundary (or free surface) between two immiscible uids,
we would specify that there is no jump in the velocity across the surface boundary.
This is true if there is no surface tension or at least if it is negligiblefor example at
the seawater-air boundary of the ocean. However at the surface of melting wax at the
top of a candle, there is surface tension, and there is a jump in the stress n at the
boundary surface. Surface tension is also responsible for the phenomenon of being able
to oat a needle on the surface of a bowl of water as well as many other interesting
eects such as the shape of water drops.
16 Evolution of vorticity
Recall from our discussion in Section 8, that the vorticity eld of a ow with velocity
eld u is dened as
:= u.
It encodes the magnitude of, and direction of the axis about which, the uid rotates,
locally. Note that u can be computed as follows
i
j
k
w/y v/z
u = det /x /y /z = u/z w/x .
u
v
w
v/x u/y
Using the NavierStokes equations for a homogeneous incompressible uid, we can in
fact derive a closed system of equations governing the evolution of vorticity = u
1
as follows. Using the identity u u = 2 |u|2 u ( u) we see that we can
equivalently represent the NavierStokes equations in the form
u
+
t
1
2
|u|2 u = u
p
+ f.
If we take the curl of this equation and use the identity

(u ) = u (
noting that
u = 0 and
) (
+u
t
u) + (
)u (u
),
u) 0, we nd that we get
= +
u+
f.
Note that we can recover the velocity eld u from the vorticity by using the identity
( u) = ( u) u. This implies
u =
35
and closes the system of partial dierential equations for and u. However, we can
also simply observe that
u =
If the body force is conservative so that f =
).
for some potential , then
f 0.
Remark 11 We can replace the vortex stretching term u in the evolution equation
for the vorticity by D, where D is the 3 3 deformation matrix, since
u = ( u) = D + R = D,
as direct computation reveals that R 0.
17 Simple example ows

We roughly follow an illustrative sequence of examples given in Majda and Bertozzi [13,
pp. 815]. The rst few are example ows of a class of exact solutions to both the Euler
and NavierStokes equations.
Lemma 1 (Majda and Bertozzi, p. 8) Let D = D(t) be a real symmetric 3 3 matrix
such that Tr(D) = 0 (representing the deformation matrix). Suppose that the vorticity
= (t) solves the ordinary dierential system
d
= D(t)
dt
for some initial data (0) = 0 . If the three components of vorticity are thus =
(1 , 2 , 3 )T , set
R :=
1
2
0 3 2
3 0 1 .
2 1
0
Then we have that

1
u(x, t) = 2 (t) x + D(t) x,
p(x, t) = 1
2
dD
+ D2 (t) + R2 (t) x x,
dt
are exact solutions to the incompressible Euler and NavierStokes equations.

Remark 12 Since the pressure is a quadratic function of the spatial coordinates x,
these solutions only have meaningful interpretations locally. Note the pressure eld
here has been rescaled by the constant mass density see Remark 9. Further note
that the velocity solution eld u only depends linearly on the spatial coordinates x;
this explains why once we established these are exact solutions of the Euler equations,
they are also solutions of the NavierStokes equations.
36
Simon J.A. Malham
Proof Recall that u is the rate of strain tensor. It can be decomposed into a direct
sum of its symmetric and skew-symmetric parts which are the 3 3 matrices
D :=
R :=
1
2
1
2
( u) + ( u)T ,
( u) ( u)T .
We can determine how u evolves by taking the gradient of the homogeneous (no
body force) NavierStokes equations so that
( u) + u
t
( u) + ( u)2 = ( u)
p.
Note here ( u)2 = ( u)( u) is simply matrix multiplication. By direct computation

( u)2 = (D + R)2 = (D2 + R2 ) + (DR + RD),
where the rst term on the right is symmetric and the second is skew-symmetric. Hence
we can decompose the evolution of u into the coupled evolution of its symmetric and
skew-symmetric parts
D
+ u D + D2 + R2 = D
t
R
+ u R + DR + RD = R.
t
p,
Directly computing the evolution for the three components of = (1 , 2 , 3 )T from

the second system of equations we would arrive at the following equation for vorticity,
+u
t
= + D,
which we derived more directly in Section 16.

Thus far we have not utilized the ansatz (form) for the velocity or pressure we
1
assume in the statement of the theorem. Assuming u(x, t) = 2 (t) x + D(t) x,
for a given deformation matrix D = D(t), then
u = (t), independent of x,
and substituting this into the evolution equation for above we obtain the following
system of ordinary dierential equations governing the evolution of = (t):
d
= D(t).
dt
Now the symmetric part governing the evolution of D = D(t), which is independent of
x, reduces to the system of dierential equations
dD
+ D 2 + R2 =
dt
p.
Note that R = R(t) only as well, since = (t), and thus

p must be a function of
t only. Hence p = p(x, t) can only quadratically depend on x. Indeed after integrating
we must have p(x, t) = 1 (dD/dt + D2 + R2 ) x x.
2
37
Fig. 14 Strain ow example.
Example (jet ow) Suppose the initial vorticity 0 = 0 and D = diag{d1 , d2 , d3 }

is a constant diagonal matrix where d1 + d2 + d3 = 0 so that Tr(D) = 0. Then from
Lemma 1, we see that the ow is irrotational, i.e. (t) = 0 for all t
0. Hence the
velocity eld u is given by
d1 x
u(x, t) = D(t)x = d2 y .
d3 z
The particle path for a particle at (x0 , y0 , z0 )T at time t = 0 is given by: x(t) = ed1 t x0 ,
y(t) = ed2 t y0 and z(t) = ed3 t z0 . If d1 < 0 and d2 < 0, then d3 > 0 and we see the ow
resembles two jets streaming in opposite directions away from the z = 0 plane.
Example (strain ow) Suppose the initial vorticity 0 = 0 and D = diag{d1 , d2 , 0}
is a constant diagonal matrix such that d1 + d2 = 0. Then as in the last example, the
ow is irrotational with (t) = 0 for all t 0 and
d1 x
u(x, t) = d2 y .
0
The particle path for a particle at (x0 , y0 , z0 )T at time t = 0 is given by: x(t) = ed1 t x0 ,
y(t) = ed2 t y0 and z(t) = z0 . Since d2 = d1 , the ow forms a strain ow as shown in
Fig. 14neighbouring particles are pushed together in one direction while being pulled
apart in the other orthogonal direction.
Example (vortex) Suppose the initial vorticity 0 = (0, 0, 0 )T and D = O. Then
from Lemma 1 the velocity eld u is given by
1 0 y
2
1
u(x, t) = 2 x = 1 0 x .
2
0
The particle path for a particle at (x0 , y0 , z0 )T at time t = 0 is given by: x(t) =
1
1
cos( 1 0 t)x0 sin( 1 0 t)y0 , y(t) = sin( 2 0 t)x0 + cos( 2 0 t)y0 and z(t) = z0 . These are
2
2
circular trajectories, and indeed the ow resembles a solid body rotation; see Fig. 15.
Example (jet ow with swirl) Now suppose the initial vorticity 0 = (0, 0, 0 )T and
D = diag{d1 , d2 , d3 } is a constant diagonal matrix where d1 + d2 + d3 = 0. Then from
38
Simon J.A. Malham
Fig. 15 When a uid ow is a rigid body rotation, the uid particles ow on circular streamlines. The uid particles on paths further from the origin or axis of rotation, circulate faster
at just the right speed that they remain alongside their neighbours on the paths just inside
them.
z
x
y
Fig. 16 Jet ow with swirl example. Fluid particles rotate around and move closer to the
z-axis whilst moving further from the z = 0 plane.
Lemma 1, we see that the only non-zero component of vorticity is the third component,
say = (t), where
(t) = 0 ed3 t .
The velocity eld u is given by
d1 x 1 (t)y
2
u(x, t) = d2 y + 1 (t)x .
2
d3 z
The particle path for a particle at (x0 , y0 , z0 )T at t = 0 can be described as follows.

We see that z(t) = z0 ed3 t while x = x(t) and y = y(t) satisfy the coupled system of
ordinary dierential equations
d
dt
x
y
d1
1
(t)
2
1 (t)
2
d2
x
.
y
If we assume d1 < 0 and d2 < 0 then the particles spiral around the z-axis with
1
decreasing radius and increasing angular velocity 2 (t). The ow thus resembles a
rotating jet ow; see Fig. 16.
We now derive a simple class of solutions that retain the three underlying mechanisms of NavierStokes ows: convection, vortex stretching and diusion.
Example (shear-layer ows) Recall the vorticity evolves according to the partial
dierential system
+ u = + D,
t
39
with u = . The material derivative term /t + u convects vorticity

along particle paths, while the term is responsible for the diusion of vorticity
and Du represents vortex stretchingthe vorticity increases/decreases when aligns
along eigenvectors of D corresponding to positive/negative eigenvalues of D.
We seek an exact solution to the incompressible NavierStokes equations of the
following form (the rst two velocity components represent a strain ow)
x
u(x, t) = y
w(x, t)
1
where is a constant, with p(x, t) = 2 2 x2 + y 2 . This represents a solution to
the NavierStokes equations if we can determine the solution w = w(x, t) to the linear
diusion equation
w
2w
w
x
= 2,
t
x
x
with w(x, 0) = w0 (x). Computing the vorticity directly we get
0
(x, t) = w/x (x, t) .
0
If we dierentiate the equation above for the velocity eld component w with respect
to x, then if := w/x, we get
2
x
= + 2 ,
t
x
x
with (x, 0) = 0 (x) = (w0 /x)(x). For this simpler ow we can see simpler signatures of the three eects we want to isolate: there is the convecting velocity x;
vortex stretching from the term and diusion in the term 2 /x2 . Note that as
in the general case, the velocity eld w can be recovered from the vorticity eld by
x
w(x, t) =
(, t) d.
Let us consider a special case: the viscous shear-layer solution where = 0. In this
case we see that the partial dierential equation above for reduces to the simple heat
equation with solution
G(x , t) 0 () d,
(x, t) =
R
where G is the Gaussian heat kernel

2
1
G(, t) :=
e /4t .
4t
Indeed the velocity eld w is given by

G(x , t) w0 () d,
w(x, t) =
R
so that both the vorticity and velocity w elds diuse as time evolves; see Fig. 17.
It is possible to write down the exact solution for the general case in terms of
the Gaussian heat kernel, indeed, a very nice exposition can be found in Majda and
Bertozzi [13, p. 18].
40
Simon J.A. Malham

w(x,0)
w(x,t)
Fig. 17 Viscous shear ow example. The eect of diusion on the velocity eld w = w(x, t) is
to smooth out variations in the eld as time progresses.
18 Dynamical similarity and Reynolds number

Our goal in this section is to demonstrate an important scaling property of the Navier
Stokes equations for a homogeneous incompressible uid without body force:
u
+u
t
u = u
1
p,
u = 0.
Note that two physical properties inherent to the uid modelled are immediately apparent, the mass density , which is constant throughout the ow, and the kinematic
viscosity . Suppose we consider such a ow which is characterized by a typical length
scale L and velocity U . For example we might imagine a ow past an obstacle such a
sphere whose diameter is characterized by L and the impinging/undisturbed far-eld
ow is uniform and given by U . These two scales naturally determine a typically time
scale T = L/U . Using these scales we can introduce the dimensionless variables
x =
x
,
L
u =
u
U
and
t =
t
.
T
Directly substituting for u = U u and using the chain rule to replace t by t and x by
x in the NavierStokes equations, we obtain:
U u
U2
+
u
T t
L
u =
The incompressibility condition becomes

through by U 2 /L we get
u
+u
t
u =
U
1
u
L
L2 x
1
u
UL x
U 2
u=
p.
u = 0. Using that T = L/U and dividing
If we set p = p/U 2 and then drop the primes, we get

u
+u
t
1
u
Re
p,
p.
41
which is the representation for the NavierStokes equations in dimensionless variables.

The dimensionless number
UL
Re :=
is the Reynolds number. Its practical signicance is as follows. Suppose we want to

design a jet plane (or perhaps just a wing). It might have a characteristic scale L1 and
typically cruise at speeds U1 with surrounding air having viscosity 1 . Rather than
build the plane to test its airow properties it would be cheaper to build a scale model
of the aircraftwith exactly the same shape/geometry but smaller, with characteristic
scale L2 . Then we could test the airow properties in a wind tunnel for example, by
using a driving impinging wind of characteristic velocity U2 and air of viscosity 2 so
that
U1 L1
U L
= 2 2.
1
2
The Reynolds number in the two scenarios are the same and the dimensionless Navier
Stokes equations for the two ows identical. Hence the shape of the ows in the two
scenarios will be the same. We could also for example, replace the wind tunnel by a
water tunnel: the viscosity of air is 1 = 0.15 cm2 /s and of water 2 = 0.0114 cm2 /s,
i.e. 1 /2 13. Hence for the same geometry and characteristic scale L1 = L2 , if we
choose U1 = 13 U2 , the Reynolds numbers for the two ows will be the same. Such
ows, with the same geometry and the same Reynolds number are said to be similar.
Remark 13 Some typical Reynolds are as follows: aircraft: 108 to 109 ; cricket ball: 105 ;
blue whale: 108 ; cruise ship: 109 ; canine artery: 103 ; nematode: 0.6; capillaries: 103 .
19 Stokes ow
Consider the individual terms in the incompressible three-dimensional NavierStokes
equations with no body force (in dimensionless form):
u
+
t
Re1 u
inertia or
convective terms
p,
diusion or
dissipation term
where Re is the Reynolds number. We wish to consider the small Reynolds number limit
Re 0. Naively this means that the diusion/dissipation term will be the dominant
term in the equations above. However we also want to maintain incompressibility, i.e.
u = 0. Since the pressure eld is the Lagrange multiplier term that maintains the
incompressibility constraint, we should attempt to maintain it in the limit Re 0.
Hence we suppose
p = Re1 q,
for a scaled pressure q. Further the ow may evolve on a slow timescale so that
t = Re .
42
Simon J.A. Malham
Making these changes of variables in the NavierStokes equations and taking the limit
Re 0, we obtain
u
= u q,
u
= Re1 u
t
p.
In dimensional variables we have thus derived the equations for Stokes ow :
u
= u
t
p.
More commonly, the stationary version of these equations are denoted Stokes ow.
Some immediate consequences are useful. If we respectively take the curl and divergence
of the Stokes equations we get
= ,
t
p = 0.
Suppose we know a stream function exists for the ow under considerationso the
ow is incompressible and an additional symmetry allows us to eliminate one spatial
coordinate and one velocity component. For example suppose we have a stationary
two-dimensional ow u = (u, v)T in cartesian coordinates x = (x, y)T . Then there
exists a stream function = (x, y) given by
u=
and
v=
.
x
Hence the vorticity of such a ow is given by = (0, 0, )T . Since for a stationary

Stokes ow = 0, we must have
() = 0.
In other words the stream function satises the biharmonic equation.
In cylindrical polar coordinates assuming a stationary axisymmetric ow with no
swirl, i.e. no dependence and u = 0, the stream function = (r, z) is given by
ur =
1
r z
and
uz =
1
.
r r
Direct computation reveals that the vorticity is given by = (0, 1 D2 , 0)T , where D2
r
is the second order partial dierential operator dened by
D2 := r
1
r r r
2
.
z 2
Again, for a stationary Stokes ow = 0. We use that for a divergence-free vector

eld we have = to show that the stream function satises
D2 (D2 ) = 0.
In spherical polar coordinates assuming a stationary axisymmetric ow with no
swirl, i.e. no dependence and u = 0, the stream function = (r, ) is given by
ur =
r2 sin
and
u =
1
.
r sin r
43
1
Direct computation implies that the vorticity is = (0, 0, r sin D2 )T , where D2 is
now the second order partial dierential operator given by
D2 :=
2
sin
1
+ 2
.
r2
r sin
For a stationary Stokes ow = 0. We again use that = to show that

1
the only non-zero component of is its third component given by r sin D2 (D2 ),
2
where D in both cases is the operator just quoted. Hence the stream function satises
D2 (D2 ) = 0.
Example (Viscous drag on sphere) Consider a uniform incompressible viscous ow

of velocity U , into which we place a spherical obstacle, radius a. The physical set up is
similar to that shown in Fig. 11. We assume that the ow around the sphere is steady.
Use spherical polar coordinates (r, , ) to represent the ow with the south-north pole
axis passing through the centre of the sphere and aligned with the uniform ow U at
innity. We assume that the ow is axisymmetric, i.e. independent of the azimuthal
angle , and there is no swirl so that u = 0. Further, take the ow around the sphere
to be a Stokes ow, i.e. we have
p = u,
where p = p(r, ) is the pressure eld, u the velocity eld and is the viscosity (a
constant parameter). Also assume the stream function is given by = (r, ). For
such a ow, as we have seen, the stream function satises
D2 (D2 ) = 0,
where D2 is the second order partial dierential operator
D2 :=
sin
2
+ 2
r2
r
1
.
sin
Step 1: Determine the boundary conditions. The stream function for the ow
prescribed is given by
ur =
r2 sin
and
u =
1
.
r sin r
First let us consider the boundary conditions as r . Decomposing the far-eld
axial directed velocity eld of speed U into components along r and we get U cos
and U sin , respectively. Hence in the far-eld limit we have
1
U cos
r2 sin
and
1
U sin .
r sin r
The solution to this pair of rst order partial dierential equations is

1
2 U r2 sin2 ,
generating the far-eld boundary condition in terms of .
44
Simon J.A. Malham
Second consider the boundary conditions on the surface of the sphere. The no-slip
condition on r = a =
1
=0
a2 sin
1
= 0.
a sin r
and
Hence is independent of r and along the boundary r = a; we can therefore take

= 0 and /r = 0 to be the boundary conditions on r = a. Thus, to summarize,
the boundary conditions for this problem are
1
2 U r2 sin2 as r
and
= 0 on r = a.
r
Step 2: Solve the modied biharmonic equation. Motivated by the boundary conditions, we look for a solution to the modied biharmonic equation D2 (D2 ) = 0 of the
form = U f (r) sin2 . First computing D2 gives
D2 = U f (r)
2
f (r) sin2 ,
r2
in which we set
F (r) := f (r)
2
f (r).
r2
Now compute D2 (D2 ) which gives

D2 (D2 ) = D2 (U F (r) sin2 )
2
= U F (r) 2 F (r) sin2 .
r
Hence D2 (D2 ) = 0 if and only if
F (r)
2
F (r) = 0.
r2
This is a linear second order ordinary dierential equation whose two independent
solutions are r2 and 1/r; thus F (r) is a linear combination of these two solutions.
However we require f (r) which satises
f (r)
2
f (r) = F (r).
r2
This is a non-homogeneous linear second order ordinary dierential equation, again

whose two independent homogeneous solutions are r2 and 1/r; while the particular
integral for the non-homogeneous component F (r), which is a linear combination of
r2 and 1/r, has the form Ar4 + Cr for some constants A and C. Hence f = f (r)
necessarily has the form
f (r) = Ar4 + Br2 + Cr +
D
,
r
where B and D are two further constants.

Step 3: Substitute the boundary conditions. First we see as r we must have
U Ar4 + Br2 + Cr +
D
sin2 = 1 U r2 sin2
2
r
45
1
so that A = 0 and B = 2 . Second we see that on r = a we must have
a2
D
+ Ca +
sin2 = 0
2
a
U a+C
and
D
sin2 = 0.
a2
Hence we get two simultaneous equations for C and D, namely a2 /2 + Ca + D/a = 0

and a + C D/a2 = 0, whose solution is C = 3a/4 and D = a3 /4. We thus have
2r2
3r
a
+
a
r
a2
1
= 4 U a2
sin2 .
Step 4: Compute the pressure. Using the stationary Stokes equations p = u

we rst compute u, i.e. the components ur and u from the stream function . Using
the relations given
1
ur = 4 U a2 2
2
a
3
+ 3 cos
ar
a2
r
and
u = 1 U a2
4
Then second, using the hint given, compute =

only one non-zero component, namely
1 4r
a
3
2 sin .
r a2
a
r
u and then
. Note has
1
1
ru
ur
r r
r
3U a
= 2 sin ,
2r
with a lot of terms cancelling. Then we have
2 cos
3U a
=
sin .
2r3
0
Thus to nd the pressure p we must solve the pair of rst order partial dierential
equations:
p
r
1 p
r
3U a
2r3
2 cos
sin
These give (here p is the constant ambient pressure as r )

p = p
3U a
cos .
2r2
Step 5: Compute the axial force. For a small patch of area dS on the surface of the
sphere the force is given by dF = (x)n dS. Since the stress tensor = pI + where
p is the pressure and is the deviatoric stress tensor, the total force on the surface of
the sphere r = a is given by

F =
(pI + )n dS.

46
Simon J.A. Malham
Now use that the deviatoric stress is a linear function of the deformation tensor D,
indeed = 2D. Further the normal n to the surface r = a is simply r , the unit
normal in the r coordinate direction, and so

n = 2D
Drr Dr Dr
1
= 2 Dr D D 0
Dr D D
0
Drr
= 2 Dr
0
= 2 Drr r + Dr ,
where is the unit vector in the coordinate direction. Now note that the area integral
over the sphere surface r = a can be split into a single integral of concentric rings of
radius a sin on the sphere surface of area 2a a sin d. Thus we get
F = 2a2
p + 2Drr r + 2Dr sin d,

r
0
The axial component of the force F ax (i.e. along the direction of the far-eld ow),
since the components of r and in the axial direction are given by r cos and sin ,
respectively, is thus given by
F ax = 2a2
(p + 2Drr ) cos 2Dr sin sin d.

0
From the formulae sheet we nd

Drr =
ur
r
and
Dr =
1
2
u
u
1 ur
+
.
r
r
r
The no-slip boundary conditions on the sphere surface r = a implies ur = u = 0.

Further from our expressions for ur and u in part (d) above we see that ur /r =
ur / = 0 and u /r = (3U/2a) sin on r = a. Substituting these into the
expressions for the total force above, as well as using the expression for the pressure p
from Step 4, we get
F ax = 2a2
0
3U
2a
cos2 sin + sin3 d
= 3U a
sin d
0
= 6U a.
Hence the axial component of the force which corresponds to the drag on the sphere
is given by 6U a.
Example (Viscous corner ow) Consider a steady incompressible viscous corner
ow as shown in Fig. 18. The uid is trapped between two plates, one is horizontal,
while the other plate lies above the horizontal plate at an acute angle . The at edge
of the upper plate almost touches the horizontal plate; there is a small gap between
the two. The horizontal plate moves with a speed U to the left perpendicular to the
47
imaginary line of intersection between the two plates; the upper plate remains xed.
We assume the trapped ow between the two plates to be a Stokes ow, i.e. we have
p = u,
where p is the pressure eld, u the velocity eld and is the viscosity. Further we
assume the ow is uniform in the direction given by the imaginary line of intersection
between the platesdenote this the z-axis. Hence in cylindrical polar coordinates there
exists a stream function = (r, ) such that
0
0 .
(r, )
u=
Taking the curl of the Stokes ow equation we see that

u
0=
0
0
(r, )
0
= () 0 .
(r, )
Hence the the stream function for this Stokes ow satises the biharmonic equation
() = 0.
Next we determine the boundary conditions. Explicitly in plane polar coordinates
(we henceforth drop the third z coordinate with respect to which the corner ow is
uniform), the relations between the velocity components ur and u and the stream
function are
1
ur =
and
u =
.
r
r
First, using the no-slip boundary conditions on the plate along = 0 which is moving
towards the origin at speed U we immediately see that
=0
r
1
= U
r
and
on
= 0.
Second, for the plate at the angle = , the no-slip boundary conditions imply
=0
r
and
=0
on
= .
Given the form of the boundary conditions, we look for a solution to the biharmonic
equation () = 0 of the form = U rf (). Directly computing, we have
U rf () = U
=
1
2
1 2
+
+ 2 2
r r
r2
r
U
f () + f () .
r
rf ()
48
Simon J.A. Malham
fluid
Fig. 18 Corner ow: An incompressible viscous uid is trapped between two plates, one is
horizontal, while the other plate lies above at an acute angle . The at edge of the upper
plate almost touches the horizontal plate; there is a small gap between the two. The horizontal
plate moves with a speed U to the left perpendicular to the imaginary line of intersection
between the two plates; the upper plate remains xed.
Then we directly compute

U rf ()
U
f () + f ()
r
2
1
1
1 2
=U
+
+ 2 2
f () + f ()
2
r r
r
r
r
U
= 3 f () + 2f () + f () .
r
Hence () = 0 if and only if f = f () satises

f
+ 2f + f = 0.
This is a linear homogeneous constant coecient fourth order ordinary dierential

equation. Looking for solutions of the form f () = exp() we obtain the polynomial
auxiliary equation (2 + 1)2 = 0 whose roots are i (each repeated). Hence the general
solution has the form
f () = A sin + B cos + C sin + D cos ,
for some constants A, B, C and D.
Now for the boundary conditions, note that the form of the solution assumed implies
ur = U f () and u = U f (). First consider the boundary conditions on = 0:
ur = U
f (0) = 1
u = 0
f (0) = 0.
u = 0
and
f () = 0.
Second for the boundary conditions on = :

ur = 0
f () = 0
and
In other words f = f () must satisfy

f (0) = f () = f () = 0
and
f (0) = 1.
We can use these four boundary conditions to determine the constants A, B, C and
D, indeed we get
sin sin( ) ( ) sin
f () =
.
2 sin2
49
Finally we should ask ourselves, what is the distance from the origin within which
the solution we sought is consistent with our assumption of Stokes ow? Since the
inertia terms u u scale like U 2 /r while the viscous terms u scale like U/r2 . Our
Stokes ow assumption was that
U 2 /r
U/r2
Hence our solution is valid provided r
Ur
1.
/U .
20 Lubrication theory
In lubrication theory we consider the following scenario. Suppose an incompressible
homogeneous viscous uid occupies a shallow layer whose typical depth is H and
horizontal extent is L; see Fig. 19 for the set up. Let x and y denote horizontal Cartesian
coordinates and z is the vertical coordinate. Suppose (u, v, w) are the uid velocity
components in the three coordinate directions x, y and z, respectively. As usual, p
denotes the pressure, while denotes the constant density, of the uid. We also assume
a typical horizontal velocity scale for (u, v) is U and a typical vertical velocity scale for
w is W . As we have already hinted, we assume H
L. Our goal is to systematically
derive a reduced set of equations from the incompressible NavierStokes equations
that provide a very accurate approximation under the conditions stated. Note that the
natural time scale for this problem is T = O(L/U ).
First consider the continuity equation (incompressibility condition) which reveals
the scaling
w
u
v
+
+
= 0.
x
y
z
U/L
U/L
W/H
To maintain incompressibility we deduce we must have W = O U (H/L) . Since H

L
we conclude W
U.
Second consider the NavierStokes equation for the velocity components u and v
and the corresponding scaling of each of the terms therein:
u
u
u
u
2u
2u
2u
+u
+v
+w
=
+
+
2
2
t
x
y
z
x
y
z 2
U 2 /L
U/L2
U/L2
1 p
,
x
1 p
.
y
U/H 2
2v
2v
v
v
v
v
2v
+u
+v
+w
=
+
+
2
2
t
x
y
z
x
y
z 2
U 2 /L
U/H 2
In the rst equation, comparing the viscous terms 2 u/x2 and 2 uy 2 to 2 uz 2 we

see that the ratio of their scaling is
U/L2
H2
= 2
2
U/H
L
1.
Thus we omit the 2 u/x2 and 2 u/y 2 terms. Now compare the inertia terms (those
on the left-hand side) to the viscosity term 2 uz 2 , the ratio of their scaling is
U 2 /L
U H2
UH H
=
=
2
L
L
U/H
1,
50
Simon J.A. Malham
z=h(x,y,t)
fluid
Fig. 19 Lubrication theory: an incompressible homogeneous viscous uid occupies a shallow

layer whose typical depth is H and horizontal extent is L. In the asymptotic limit H
L, the
NavierStokes equations reduce to the shallow layer equations.
and hence we omit all the inertia termsunder the assumption that the modied
Reynolds number Rm = U H/ is small or order one. Finally to keep the pressure term
(which mediates the incompressibility condition) it must have the scaling
x P/ = O U/H 2
P = O U L/H 2 .
An exactly analogous scaling argument applies to the second equation above and thus
we see that the nal equations for these two components are (with = )
2u
p
= 2,
x
z
p
2v
= 2.
y
z
Third consider the NavierStokes equation for the velocity component w. The scaling of the individual terms reveals:
w
w
w
w
+u
+v
+w
=
t
x
y
z
U W/L=U 2 H/L2
2w
2w
+
2
x
y 2
2w
z 2
W/L2 =U H/L3
W/H 2 =U/LH
1 p
.
z
U L/H 3
The largest scaling term appears to be the pressure term so we compare the other
terms to that. The ratio of the inertia terms to the pressure term in terms of their
scaling is
U 2 H/L2
U H4
UH H
=
=
3
L
U L/H
L3
1,
again under the assumption the modied Reynolds number Rm = U H/ is small or

order one. The ratio of the viscous terms 2 w/x2 and 2 w/y 2 to the pressure
term in terms of their scaling is
U H/L3
=
U L/H 3
H
L
1.
Finally the ratio of the viscous term 2 w/z 2 to the pressure term is
U/LH
=
U L/H 3
H
L
1.
51
Hence the third equation reduces to p/z = 0. The complete shallow layer equations
are thus
p
2u
= 2,
x
z
p
2v
= 2,
y
z
p
= 0,
z
together with the incompressibility condition. Note that from the third equation we
deduce the pressure p = p(x, y, t) only.
As shown in Fig. 19 suppose that the uid occupies a region between a lower rigid
plate at z = 0 and a top surface at z = h(x, y, t). We assume here for the moment that
the top surface is free, so perhaps it represents a uid-air boundary, however we can
easily specialize our subsequent analysis to a rigid lid upper boundary. Importantly
though, we assume no-slip boundary conditions at z = 0 while we suppose the velocity
components at the surface height z = h(x, y, t) are u(x, y, h, t) = U and v(x, y, h, t) = V
with U and V given. For this scenario, we can derive a closed form equation for how
the surface height z = h(x, y, t) evolves in time as follows. Since p = p(x, y, t) only
we deduce that from the rst two shallow layer equations that u and v are quadratic
functions of z. Indeed if we integrate the equations for u and v with respect to z twice
and apply the boundary conditions we nd
Uz
1
+
z(z h)
h
2
1
Vz
+
z(z h)
v=
h
2
u=
p
x
p
.
y
However incompressibility implies we have

h
w(x, y, h, t) w(x, y, 0, t) =
0
u
v
+
dz.
x
y
Recall we have no-slip boundary conditions at z = 0. Further note that at z = h the

vertical velocity component is
w=
h
h
h
+U
+V
.
t
x
y
Also we have the calculus identity
u dz =
0
u
u
d + U
x
x
,
z=h
with a similar result for the v/y component in the incompressibility constraint just
above. Putting all this together we nd that
h
=
t
x
u dz
0
v dz.
0
If we substitute our expressions for u and v above into the right-hand side and integrate
we arrive at the following closed form evolution equation for h:
h
h3 p
h3 p
1
1
+
2Uh +
2 V h = 0,
t
x 12 x
y 12 y
52
Simon J.A. Malham
or equivalently
12
(U h)
(V h)
h
+ 6
+
t
x
y
p
h3
+
h3
.
x
x
y
y
Example (Squeeze lm) An incompressible homogeneous viscous uid occupies a

region between two parallel plates that are very close togethera squeeze lm. The
upper plate is a circular disc with radius L. Assume that the volume occupied by the
liquid is a at cylindrical shape with circular cross-section of radius L and height H
see Fig. 20 for the set up. Suppose (r, , z) are cylindrical polar coordinates relative to
the origin which is on the lower plate at the centre of the disc of uid. Let (ur , u , uz ) be
the uid velocity components in the three coordinate directions r, and z, respectively.
We assume throughout that the ow is axisymmetric (independent of ) and there is no
swirl (the velocity component u = 0). Further we assume that the lower plate remains
xed at z = 0 while the height of the parallel upper disc plate given by z = h(t) changes
with time. No-slip boundary conditions apply on both plates, i.e. ur = 0 on z = 0 and
z = h(t), while uz = 0 on z = 0 and uz = h (t) on z = h(t). Our goal in this example is
to compute the total force on the upper disc plate. Since the relative scaling of the terms
in the incompressible NavierStokes equations here concern the separation of scales
between the vertical (z, uz ) and horizontal (r, , ur , u ) coordinates and velocities in
cylindrical polar coordinates, we can equally carry out the scale analysis in Cartesian
vertical (z, w) and horizontal (x, y, u, v) coordinates and velocities, as we did above,
assuming the same scaling for the two, namely (H, W ) and (L, U ) for the vertical and
horizontal coordinates and velocities. Hence in this example, the scale analysis above
assuming H
L generates the shallow layer equations:
p
2 ur
=
,
r
z 2
p
= 0,
z
together with incompressibility (note u 0). We deduce p = p(r) only and
p
2 ur
.
=
r
z 2
Integrating and using the no-slip boundary conditions on both plates, we get
ur =
1 p
z(z h).
2 r
The ow out of the volume of uid (cylinder) is the integral quantity

h
ur r d dz,
0
since r d dz is a small patch of area on the sides of the cylinder and ur is the normal
velocity there. Hence the rate of change of total volume of the uid between the parallel
plates is
h
2r
ur dz,
0
53
upper disc plate

fluid
L
111111111111111111111
000000000000000000000
h
lower plate
Fig. 20 Lubrication theory: an incompressible homogeneous viscous uid occupies a region

between two plates that are very close togethera squeeze lm. The upper plate is a circular
disc with radius L. We assume that the volume occupied by the liquid is a at cylindrical
shape with circular cross-section of radius L and height H.
Substituting the expression for the velocity eld ur above we nd

h
2r
ur dz =
0
r p
r
z(z h) dz
0
r p h3
.
r 6
If the upper disc plate moves with velocity h (t) then the rate of change of total volume
of the uid between the parallel plates is also given by the cross-sectional area times
that velocity, i.e. r2 h (t). Equating this expression with that above we get
r2 h (t) =
r p h3
r 6
p
6r
= 3 h.
r
h
Assuming that the pressure at the r = L boundary of the cylindrical volume of uid
is zero, the total force on the disc plate is given by
L
6
h
r dr
h3
r
3
= 3 h (L2 r2 ).
h
p(r) =
Hence the total force on the disc is given by

3
h
h3
L
0
(L2 r2 ) 2r dr =
3L4
h.
2h3
Finally, we can approximate time for a constant force F to pull the parallel plates apart
if the initial separation is h0 . Assuming the mass of the upper disc plate is negligible,
we have
3L4
4F t
F =
h
h2 = h2
.
0
2h3
3L4
Hence h when t 3L4 /4F h2 ; which is the time it takes to pull the parallel
0
plates apart.
54
Simon J.A. Malham
UE
UE
boundary layer
Fig. 21 Boundary layer theory: the ow over the wing is well approximated by an Euler ow
as it is a high Reynolds number ow. However the uid is ultimately viscous and uid particles
at the wing surface must adhere to it. Hence there is a boundary layer between the wing and
bulk ow across which the velocity eld rapidly changes from zero velocity relative to the wing
to the bulk velocity UE past the wing.
21 Boundary layer theory

The Reynolds numbers associated with ows past aircraft or ships are typically large,
indeed of the order 108 of 109 recall the remark at the end of Section 18 on Dynamical
similarity and Reynolds numbers. For individual wings or ns the Reynolds number
may be an order of magnitude or two smaller. However such Reynolds numbers are
still large and the ow around wings for example would be well approximated by Euler
ow. We can imagine the ow over the top of the wing of an aircraft has a high relative
velocity tangential to the surface wing directed towards the rear edge of the wing.
This would appear to be consistent with no ux boundary conditions we apply for
the Euler equationsin particular there is no boundary restriction on the tangential
component of the uid velocity eld. However air or water ow is viscous. The uid
particles on the wing must satisfy viscous boundary conditions, i.e. exactly at the
surface they must adhere to the wing and thus have zero velocity relative to the wing.
The reconciliation of this conundrum is that there must be a boundary layer on the
surface of the wing. By this we mean a special thin uid layer exists between the wing
and the fast moving Euler ow past the wing. The velocity prole of the ow past the
wing across the boundary layer as one measures continuously from the wing surface to
the top of the boundary layer must change extremely rapidly. Indeed it must change
from zero velocity relative to the wing surface to fast relative velocity (the speed of the
aircraft) towards the rear of the wingsee Fig. 21.
We shall now derive an accurate model, reduced (or even deduced!) from the full
NavierStokes equations, for this scenario. We will assume a two dimensional ow
around an object of typical size L (for example the length from wing tip to rear edge).
We assume that the bulk ow is governed by the Euler equations, which is a good
approximation given it is a very high Reynolds number ow (as discussed above). We
further assume it is one-dimensional and given by a horizontal velocity eld to the right
UE = UE (x, t) which depends on the horizontal parameter x which is positive towards
the rightsee Fig. 21. This is consistent with no normal ow close to the boundary
layer. We return to discussing the bulk ow once we have derived the boundary layer
equations. If U represents the typical bulk uid velocity (relative speed of the object)
then the underlying Reynolds number is
Re =
UL
.
55
A typical time scale is thus T = O(L/U ).

Let us now focus on the boundary layer itself. We assume the ow is two-dimensional
incompressible homogeneous NavierStokes ow. Suppose the velocity components
u = u(x, y, t) and v = v(x, y, t) in the horizontal and vertical directions, respectively,
with typical scale U and V . Let denote the typical width of the boundary layer. The
continuity equation reveals the scaling
u
v
+
= 0.
x
y
U/
V /
To maintain incompressibility we deduce that V = O(U / ). We now consider the

NavierStokes equation for the rst velocity component u which is given by
u
u
u
2 u 1 p
2u
+u
+v
= 2 + 2
,
t
x
y
x
x
y
U 2 /L
V U/
U/L2
U/ 2
P/L
where we suppose the pressure has typical scale P . Note since V = O(U / ), all the
inertia terms have the same scaling. Our goal in the boundary layer is to keep the inertia
terms and a viscous term. In particular the horizontal velocity eld in the boundary
layer varies rapidly with y. We thus want to retain the 2 u/y 2 term. This means we
must have
U2
U
= 2
L
L
U
1/2
L
.
(Re)1/2
The other viscous term 2 u/x2 thus has typical scaling U/L2 = (Re)1 U 2 /L which
is thus asymptotically small compared to the inertia terms since we are assuming the
Reynolds number Re is very large. We thus neglect this viscous term. If we now consider
the second velocity component v, we nd
v
v
v
2v
2 v 1 p
+u
+v
= 2 + 2
.
t
x
y
y
x
y
U V /L
V 2 /
V /L2
V / 2
P/
Using the equivalent scaling we already established above we observe that

UV
U 2
U2
= 2 = (Re)1/2
L
L
L
and
V2
U 2
U2
= 2 = (Re)1/2
.
L
L
We further observe that

V
U
U
U2
= 3 = (Re)1/2 2 = (Re)1
2
L
L
L
L
and
U
V
U
U2
=
= (Re)1/2 2 = (Re)1/2
.
L
L
2
L
Hence we see the viscous term 2 v/y 2 is very small and we immediately neglect
it. Further we observe that the pressure term (1/)p/y, to balance the remaining
terms in the equation for the eld v = v(x, y, t), must have the scaling (Re)1/2 U 2 /L.
56
Simon J.A. Malham
Since these terms are asymptotically small compared to the terms we have retained in
the evolution equation for horizontal velocity eld u = u(x, y, t), we deduce that
p
= 0.
y
Consequently we only retain the evolution equation for the u = u(x, y, t) with the viscous term 2 u/x2 neglected. Let us now non-dimensionalize our variables as follows
x =
x
,
L
y =
y
,
u =
u
,
U
v =
v
,
V
t =
tU
L
and
p =
p
.
U 2
Using the natural scaling above, in these non-dimensional variables after dropping the
primes, we obtain the Prandtl boundary layer equations (derived in 1904):
u
u
u
p
2u
+u
+v
=
+
,
t
x
y
x
y 2
p
= 0,
y
u
v
+
= 0.
x
y
These equations are supplemented with the boundary conditions
u=0
and
v=0
when y = 0. The second equation above implies that the pressure eld in the boundary
layer is independent of y so that p = p(x, t) only. Hence if the pressure eld or more
particularly p/x can be determined at the top boundary of the boundary layer,
then it is determined inside the boundary layer. With this knowledge, we note that the
system of equations represented by the rst and third Prandtl equations above, is third
order with respect to ythe rst equation is a second order partial dierential equation
with respect to y while the third equation is rst order. We thus require an additional
boundary condition in the y direction. This is naturally provided by matching the
boundary layer ow with the bulk Euler ow outside the boundary layer. See Chorin
and Marsden [?, Section 2.2] for an in depth discussion of possible matching strategies.
Here we will simply match horizontal boundary layer velocity eld u = u(x, y, t) with
the far eld Euler ow UE = UE (x, t). In terms of the non-dimensionalized y coordinate,
temporarily reverting back to the primed notation for them, we have
y =
y
y
= (Re)1/2 .
In the limit of large Reynolds number, the top boundary corresponds to y . Thus,
dropping primes again, the third boundary condition we require is, as y , that
u UE (x, t).
Thus we can in principle solve the Prandtl boundary layer equations if UE = UE (x, t)
is known, in which case
UE
U
p
=
+ UE E .
x
t
x
This can be directly substituted into the rst Prandtl boundary layer equation above.
57
(x)
U
x
leading edge
semiinfinite flat plate
Fig. 22 Blasius problem: steady boundary layer ow over a semi-innite at plate, uniform in
the z direction.
Remark 14 We have implicitly assumed that the lower surface is at. If lower surface
has curvature then p/y is not zero, corresponding to some centripetal acceleration.
Example (Blasius problem, 1908) Consider a steady boundary layer ow on a semiinnite at plate as shown in Fig. 22. The plate and ow is assumed to be uniform
in the z direction. The leading edge of the plate coincides the origin while the plate
itself lies along the positive x-axis. The y-axis is orthogonal to the plate as shown in
Fig. 22. We will focus in the ow in the x
0 and y
0 region. We suppose that a
uniform horizontal ow of velocity U in the positive x direction washes over the plate.
We assume that a steady boundary layer ow develops on the plate as shown. Since
the Euler ow outside the boundary is uniformly U there is no pressure gradient with
respect to x so that p/x is zero outside the boundary, as thus by the arguments in
the general theory above, also zero inside the boundary. Hence the Prandtl boundary
layer equations in dimensional form are
u
u
u
2u
+v
= 2,
x
y
y
v
u
+
= 0.
x
y
The boundary conditions for u = u(x, y) and v = v(x, y) on the at plate are
u(x, 0) = 0
and
v(x, 0) = 0,
for all x > 0. Since the plate is semi-innite there is no imposed horizontal scale.
Following the arguments in the general theory above, but with L replace by x we
deduce that V = U /x and thus also that = (x) where
(x) :=
x
U
1/2
In the boundary layer, the natural non-dimensional vertical coordinate is

:=
y
.
(x)
As the ow is incompressible and two dimensional there is a stream function = (x, y)

satisfying /y = u and /y = v. We seek a similarity solution of the form
= U (x) f ().
58
Simon J.A. Malham
Directly computing the partial derivatives we nd that

u = U f ()
v = 1
2
and
U
x
1/2
(f f ).
Substituting these forms for u and v into the rst Prandtl boundary layer equation
above we nd that f = f () satises the third order ordinary dierential equation
f
1
+ 2 f f = 0.
This is supplemented with the boundary conditions on the at plate that correspond
to f (0) = 0 and f (0) = 0. The nal boundary condition should be that u U as
y . Indeed we see that this boundary condition corresponds to f 1 as .
This boundary value problem can be numerically computed.
22 Exercises
Exercise (trajectories and streamlines: expanding jet) Find the trajectories and streamlines when (u, v, w)T = (xe2tz , ye2tz , 2e2tz )T . What is the track of the particle
passing through (1, 1, 0)T at time t = 0?
Exercise (trajectories and streamlines: three dimensions) Suppose a velocity eld
u(x, t) = (u, v, w)T is given for t > 1 by
u=
x
,
1+t
v=
y
1 + 1t
2
and
w = z.
Find the particle paths and streamlines for a particle starting at (x0 , y0 , z0 )T .
Exercise (streamlines: plane/cylindrical polar coordinates) Sketch streamlines for the
steady ow eld (u, v, w)T = (t) (x y, x + y, 0)T show that the streamlines are
exponential spirals. Here = (t) is an arbitrary function of t. (Hint: convert to
cylindrical polar coordinates (r, , z) rst. Note that in these coordinates the equations
for trajectories are
dr
= ur ,
dt
d
= u ,
dt
and
dz
= uz ,
dt
where ur , u and uz are the velocity components in the corresponding coordinate

directions.)
Exercise (steady oscillating channel ow) An incompressible uid is in steady twodimensional ow in the channel < x < , /2 < y < /2, with velocity
u = (1 + x sin y, cos y)T . Find the equation of the streamlines and sketch them. Show
that the ow has stagnation points at (1, /2) and (1, /2).
Exercise (channel shear ow) Consider the two-dimensional channel ow (with U a
given constant)
0
u = U (1 x2 /a2 ) ,
0
between the two walls x = a. Show that there is a stream function and nd it. (Hint:
a stream function exists for a velocity eld u = (u, v, w)T when
u = 0 and we
59
Region of flow is :
fluid lies between the
two walls of the cylinders
R2
R1
Fig. 23 Couette ow between two concentric cylinders of radii R1 < R2 .
have an additional symmetry. Here the additional symmetry is uniformity with respect
to z. You thus need to verify that if u = /y and v = /x, then u = 0 and
then solve this system of equations to nd .)
Show that approximately 91% of the volume ux across y = y0 for some constant
3
y0 ows through the central part of the channel |x|
4 a.
Exercise (ow inside and around a disc) Calculate the stream function for the ow
T
eld u = U cos (1a2 /r2 ), U sin (1+a2 /r2 )/2r in plane polar coordinates,
where U, a, are constants.
Exercise (Couette ow) (From Chorin and Marsden, p. 31.) Let be the region
between two concentric cylinders of radii R1 and R2 , where R1 < R2 . Suppose the
velocity eld in cylindrical coordinates u = (ur , u , uz )T of the uid ow inside , is
given by ur = 0, uz = 0 and
A
+ Br,
u =
r
where
A=
2 2
R1 R2 (2 1 )
2
2
R2 R1
and
B=
2
2
R1 1 R2 2
.
2
2
R2 R1
This is known as a Couette ow see Fig. 23. Show that the:

(a) velocity eld u = (ur , u , uz )T is a stationary solution of Eulers equations of
motion for an ideal uid with density 1 (hint: you need to nd a pressure eld
p that is consistent with the velocity eld given. Indeed the pressure eld should
be p = A2 /2r2 + 2AB log r + B 2 r2 /2 + C for some arbitrary constant C.);
(b) angular velocity of the ow (i.e. the quantity u /r) is 1 on the cylinder r = R1
and 2 on the cylinder r = R2 .
(c) the vorticity eld u = (0, 0, 2B).
Exercise (hurricane) We devise a simple model for a hurricane.
(a) Using the Euler equations for an ideal incompressible ow in cylindrical coordinates
(see the bath or sink drain problem in the main text) show that at position (r, , z),
60
Simon J.A. Malham
for a ow which is independent of with ur = uz = 0, we have

u2
1
=
r
0
1
0=
0
p
,
r
p
+ g,
z
where p = p(r, z) is the pressure and g is the acceleration due to gravity (assume
this to be the body force per unit mass). Verify that any such ow is indeed incompressible.
(b) In a simple model for a hurricane the air is taken to have uniform constant density
0 and each uid particle traverses a horizontal circle whose centre is on the xed
vertical z-axis. The (angular) speed u at a distance r from the axis is
u =
r,
for 0
3/2
a1/2 ,
r
a,
for r > a,
where and a are known constants.

(i) Now consider the ow given above in the inner region 0
r
a. Using the
equations in part (a) above, show that the pressure in this region is given by
1
p = c0 + 2 0 2 r2 g0 z,
where c0 is a constant. A free surface of the uid is one for which the pressure
is constant. Show that the shape of a free surface for 0 r a is a paraboloid
of revolution, i.e. it has the form
z = Ar2 + B,
for some constants A and B. Specify the exact form of A and B.
(ii) Now consider the ow given above in the outer region r > a. Again using the
equations in part (a) above, and that the pressure must be continuous at r = a,
show that the pressure in this region is given by
p = c0
0 2 3
a g0 z + 3 0 2 a2 ,
2
r
where c0 is the same constant (reference pressure) as that in part (i) above.
Exercise (Venturi tube) Consider the Venturi tube shown in Fig. 24. Assume that the
ideal uid ow through the construction is homogeneous, incompressible and steady.
The ow in the wider section of cross-sectional area A1 , has velocity u1 and pressure
p1 , while that in the narrower section of cross-sectional area A2 , has velocity u2 and
pressure p2 . Separately within the uniform wide and narrow sections, we assume the
velocity and pressure are uniform themselves.
(a) Why does the relation A1 u1 = A2 u2 hold? Why is the ow faster in the narrower
region of the tube compared to the wider region of the tube?
(b) Use Bernoullis theorem to show that
1 2
2 u1
p1
p
= 1 u2 + 2 ,
2 2
0
0
where 0 is the constant uniform density of the uid.
61
A 1 , u1, p 1
A 2 , u2 , p 2
A 1 , u1, p 1
streamline
Fig. 24 Venturi tube: the ow in the wider section of cross-sectional area A1 has velocity
u1 and pressure p1 , while that in the narrower section of cross-sectional area A2 has velocity
u2 and pressure p2 . Separately within the uniform wide and narrow sections, we assume the
velocity and pressure are uniform themselves.
(c) Using the results in parts (a) and (b), compare the pressure in the narrow and wide
regions of the tube.
(d) Give a practical application where the principles of the Venturi tube is used or
might be useful.
Exercise (Clepsydra or water clock) A clepsydra has the form of a surface of revolution
containing water and the level of the free surface of the water falls at a constant rate,
as the water ows out through a small hole in the base. The basic setup is shown in
Fig. 25.
(a) Apply Bernoullis theorem to one of the typical streamlines shown in Fig. 25 to
show that
1
2
dz
dt
2
1
= 2 U 2 gz
where z is the height of the free surface above the small hole in the base, U is the
velocity of the water coming out of the small hole and g is the acceleration due to
gravity.
(b) If S is the cross-sectional area of the hole in the bottom, and A is the cross-sectional
area of the free surface, explain why we must have
A
(c) Assuming that S
dz
= S U.
dt
A, combine parts (a) and (b) to explain why we can deduce

U
2gz.
(d) Now combine the results from (b) and (c) above, to show that the shape of the
container that guarantees that the free surface of the water drops at a constant
rate must have the form z = C r4 in cylindrical polars, where C is a constant.
Exercise (coee in a mug) A coee mug in the form of a right circular cylinder (diam1
eter 2a, height h), closed at one end, is initially lled to a depth d > 2 h with static
inviscid coee. Suppose the coee is then made to rotate inside the mugsee Fig. 26.
62
Simon J.A. Malham

free surface
z
P
0
r
typical streamline
z
P = air pressure
0
U
Fig. 25 Clepsydra (water clock).
(a) Using the Euler equations for an ideal incompressible homogeneous ow in cylindrical coordinates show that at position (r, , z), for a ow which is independent of
with ur = uz = 0, the Euler equations reduce to
u2
1 p
=
,
r
r
1 p
0=
+ g,
z
where p = p(r, z) is the pressure, is the constant uniform uid density and g is
the acceleration due to gravity (assume this to be the body force per unit mass).
Verify that any such ow is indeed incompressible.
(b) Assume that the coee in the mug is rotating as a solid body with constant angular
velocity so that the velocity component u at a distance r from the axis of
symmetry for 0 r a is
u = r.
Use the equations in part (a), to show that the pressure in this region is given by
p = 1 2 r2 gz + C,
2
where C is an arbitrary constant. At the free surface between the coee and air,
the pressure is constant and equal to the atmospheric pressure P0 . Use this to show
that the shape of the free surface has the form
z=
2 2 C P0
r +
.
2g
g
(c) Note the we are free to choose C = P0 in the equation of the free surface so that
it is described by z = 2 r2 /2g. This is equivalent to choosing the origin of our
cylindrical coordinates to be the centre of the dip in the free surface. Suppose that
this origin is a distance z0 from the bottom of the mug.
(i) Explain why the total volume of coee is a2 d. Then by using incompressibility,
explain why the following constraint must be satised:
a
a2 z0 +
0
2 r2
2r dr = a2 d.
2g
63
free surface
P
0
coffee
2a
Fig. 26 Coee mug: we consider a mug of coee of diameter 2a and height h, which is initially
lled with coee to a depth d. The coee is then made to rotate about the axis of symmetry
of the mug. The free surface between the coee and the air takes up the characteristic shape
shown, dipping down towards the middle (axis of symmetry). The goal is to specify the shape
of the free surface.
(ii) By computing the integral in the constraint in part (i), show that some coee
will be spilled out of the mug if 2 > 4g(h d)/a2 . Explain briey why this
formula does not apply when the mug is initially less than half full.
Exercise (Channel ow: Froude number) Recall the scenario of the steady channel
ow over a gently undulating bed given in Section 11. Consider the case when the
maximum permissible height y0 compatible with the upstream conditions, and the
actual maximum height ymax of the undulation are exactly equal, i.e. ymax = y0 . Show
that the ow becomes locally critical immediately above ymax and, by a local expansion
about that position, show that there are subcritical and supercritical ows downstream
consistent with the continuity and Bernoulli equations (friction in a real ow leads to
the latter being preferred).
Exercise (Bernoullis Theorem for irrotational unsteady ow) Consider Eulers equations of motion for an ideal homogeneous incompressible uid, with u = u(x, t) denoting the uid velocity at position x and time t, the uniform constant density,
p = p(x, t) the pressure, and f denoting the body force per unit mass. Suppose that
the ow is unsteady, but irrotational, i.e. we know that u = 0 throughout the ow.
This means that we know there exists a scalar potential function = (x, t) such that
u = . Also suppose that the body force is conservative so that f = for some
potential function = (x, t).
(a) Using the identity
u
u=
1
2
(|u|2 ) u (
u),
show from Eulers equations of motion that the Bernoulli quantity

H :=
1
+ 2 |u|2 + +
t
satises H = 0 throughout the ow.

(b) From part (a) above we can deduce that H can only be a function of t throughout
the ow, say H = f (t) for some function f . By setting
t
V :=
f ( ) d,
0
64
Simon J.A. Malham
show that the Bernoulli quantity

V
p
1
+ 2 |u|2 + +
t
is constant throughout the ow.

Exercise (rigid body rotation) An ideal uid of constant uniform density 0 is contained
within a xed right-circular cylinder (with symmetry axis the z-axis). The uid moves
under the inuence of a body force eld f = (x + y, x + y, 0)T per unit mass,
where , , and are independent of the space coordinates. Use Eulers equations of
motion to show that a rigid body rotation of the uid about the z-axis, with angular
1
velocity (t) given by = 2 ( ) is a possible solution of the equation and boundary
conditions. Show that the pressure is given by

p = p0 +
1
0 ( 2 + )x2 + ( + )xy + ( 2 + )y 2 ,
2
where p0 is the pressure at the origin.

Exercise (vorticity and streamlines) An inviscid incompressible uid of uniform density
is in steady two-dimensional horizontal motion. Show that the Euler equations are
equivalent to
H
H
= v
and
= u,
x
y
where H = p/ + 1 (u2 + v 2 ), where p is the dynamical pressure, (u, v)T is the velocity
2
eld and is the vorticity. Deduce that is constant along streamlines and that this
is in accord with Kelvins theorem.
Exercise (vorticity, streamlines with gravity) An incompressible inviscid uid, under
the inuence of gravity, has the velocity eld u = (2y, x, 0)T with the z-axis
vertically upwards; and is constant. Also the density is constant. Verify that u
satises the governing equations and nd the pressure p. Show that the Bernoulli
function H = p/ + 1 |u|2 + is constant on streamlines and vortex lines, where is
2
the gravitational potential.
Exercise (Flow in an innite pipe: Poiseuille ow) (From Chorin and Marsden, pp. 456.) Consider an innite pipe with circular cross-section of radius a, whose centre line
is aligned along the z-axis. Assume no-slip boundary conditions at r = a, for all z,
i.e. on the inside surface of the cylinder. Using cylindrical polar coordinates, look for a
stationary solution to the uid ow in the pipe of the following form. Assume there is
no radial ow, ur = 0, and no swirl, u = 0. Further assume there is a constant pressure
gradient down the pipe, i.e. that p = Cz for some constant C. Lastly, suppose that
the ow down the pipe, i.e. the velocity component uz , has the form uz = uz (r) (it is
a function of r only).
(a) Using the NavierStokes equations, show that
C = uz =
1
uz
r
r r
r
65
(b) Integrating the equation above yields

uz =
C 2
r + A log r + B,
4
where A and B are constants. We naturally require that the solution be bounded.
Explain why this implies A = 0. Now use the no-slip boundary condition to determine B. Hence show that
C
(a2 r2 ).
uz =
4
(c) Show that the mass-ow rate across any cross section of the pipe is given by
uz dS = Ca4 /8.
This is known as the fourth power law.
Exercise (Elliptic pipe ow) Consider an innite horizontal pipe with constant elliptical
cross-section, whose centre line is aligned along the z-axis. Assume no-slip boundary
conditions at
x2
y2
+ 2 = 1,
a2
b
where a and b are the semi-axis lengths of the elliptical cross-section. Using the incompressible NavierStokes equations for a homogeneous uid in Cartesian coordinates,
look for a stationary solution to the uid ow in the pipe of the following form. Assume there is no ow transverse to the axial direction of the pipe, so that if u and v are
the velocity components in the coordinate x and y directions, respectively, then u = 0
and v = 0. Further assume there is a constant pressure gradient down the pipe, i.e. the
pressure p = Gz for some constant G, and there is no body force. Lastly, suppose
that for the ow down the pipe the velocity component w = w(x, y) only.
(a) Using the NavierStokes equations, show that w must satisfy
2w
2w
G
+ 2 = .
2x
y
(b) Show that, assuming A, B and C are constants,
w = Ax2 + By 2 + C
is a solution to the partial dierential equation for w in part (a) above provided
A+B =
G
.
2
(c) Use the no-slip boundary condition to show that

A=
G
b2
2
,
2 a + b2
B=
G
a2
2
2 a + b2
and
C=
G
a2 b2
2
.
2 a + b2
66
Simon J.A. Malham
(d) Explain why the volume ux across any cross section of the pipe is given by
w dS
and then show that it is given by
a3 b3 G
.
4(a2 + b2 )
(Hint: to compute the integral you may nd the substitutions x = ar cos and
y = br sin , together with the fact that the innitesimal area element dx dy
transforms to ab rdr d, useful.)
(e) Explain why for a given elliptical cross-sectional area, the optimal choice of a and
b to maximize the volume ow-rate is a = b.
Exercise (Wind blowing across a lake) Wind blowing across the surface of a lake of
uniform depth d exerts a constant and uniform tangential stress S. The water is initially
at rest. Find the water velocity at the surface as a function of time for t
d2 . (Hint:
solve for the vorticity using the vorticity equation for a very deep lake.)
Suppose now that the wind has been blowing for a suciently long time to establish a steady state. Assuming that the water velocity can be taken to be almost
uni-directional and that the horizontal dimensions of the lake are large compared with
d, show that the water velocity at the surface is Sd/4. (Hint: A pressure gradient
would be needed to ensure no net ux (why?) in the steady state, and this pressure
gradient leads to a small rise in the surface elevation of the lake in the direction of the
wind.)
Exercise (Stokes ow: between hinged plates) Starting with the continuity and Navier
Stokes equations for a steady incompressible two-dimensional ow, show that for Stokes
ow the stream function satises 4 = 0. Two identical rigid plates are hinged
together along their line of intersection O, and have a relative angular velocity .
Find the stream function representing the (two-dimensional) Stokes ow near O, and
estimate the distance within O within which the solution is self-consistent.
Exercise (Stokes ow: rotating sphere) A rigid sphere of radius a is rotating with
angular velocity in a uid at rest at innity. Show that when a2 /
1 the
couple exerted on the uid by the sphere is 8a3 . (Use that in spherical polar
coordinates (r, , ) for a velocity eld u = (0, 0, u )T , the relevant component of the
1
deformation matrix is Dr = 2 r(u /r)/r.)
Exercise (Lubrication theory: shear stress) Incompressible uid of viscosity is contained between y = 0 and y = h(x) for 0 x a where h
a. The uid pressure at
x = 0 exceeds that at x = a by an amount p. Using lubrication theory show that
dp
A
=
dx
h(x)
where A is a constant. For the special case where h = CeBx , determine dp/dx, where
B and C are positive constants. Hence show that the maximum shear stress on the
plane is
BCe2Ba
3
p 3Ba
.
2
e
1
67
Exercise (Lubrication theory: shallow layer) An incompressible homogeneous viscous

uid occupies a shallow layer whose typical depth is H and horizontal extent is L; see
Fig. 19 for the set up. Suppose that for this question x and y are horizontal Cartesian
coordinates and z is the vertical coordinate. Let u, v and w be the uid velocity
components in the three coordinate directions x, y and z, respectively. Suppose p is
the pressure. Assume throughout that a typical horizontal velocity scale for u and v is
U and a typical vertical velocity scale for w is W . We denote by , the constant density
of the uid.
(a) Assume throughout that H
L.
(i) Using that the uid is incompressible, explain how we can deduce that W is
asymptotically smaller than U .
(ii) The body force g in this example is due to gravity g which we suppose acts
in the negative z direction (i.e. downwards). Using part (i), show the three
dimensional NavierStokes equations reduce to the shallow layer equations:
P
2u
= 2,
x
z
P
2v
= 2,
y
z
P
= 0,
z
where P := p + gz is the modied pressure and = is the rst coecient
of viscosity
(b) For the shallow uid layer shown in Fig. 19, assume no-slip boundary conditions on
the rigid lower layer. Further suppose that the pressure p is constant (indeed take
it to be zero) along the top surface of the layer at z = h(x, y, t). Further suppose
that surface stress forces are applied to the surface z = h(x, y, t) so that
u
=
,
z
x
v
=
,
z
y
where = (x, y) is a given function. Using the shallow layer equations in part
(a) and the boundary conditions, show that the height function h(x, y, t) satises
the partial dierential equation
2
+
h2
1 gh2
3
t
x
x
h2
1 gh2
3
y
y
= 0.
Exercise (Lubrication theory: HeleShaw cell) An incompressible homogeneous uid

occupies the region between two horizontal rigid parallel planes, which are a distance
h apart, and outside a rigid cylinder which intersects the planes normally; see Fig. 27
for the set up. Suppose that for this question x and y are horizontal coordinates and z
is the vertical coordinate. Assume throughout that a typical horizontal velocity scale
for u and v is U and a typical vertical velocity scale for w is W .
(a) Explain very briey why a is a typical horizontal scale for (x, y) and h a typical
vertical scale for z.
68
Simon J.A. Malham
fluid
cylinder
a
h
cylinder
Side view
fluid
Above view
Fig. 27 HeleShaw cell: an incompressible homogeneous uid occupies the region between two
parallel planes (a distance h apart) and outside the cylinder of radius a (normal to the planes).
(b) Hereafter further assume that h
a and further that

U h2
where = is the rst coecient of viscosity. Using these assumptions, show

that the NavierStokes equations for an incompressible homogeneous uid reduce
to the system of equations
2u
p
,
=
x
z 2
2v
p
2 =
,
y
z
p
= 0,
z
for a steady ow, where p = p(x, y, z) is the pressure.

(c) We dene the vertically averaged velocity components (u, v)T = u(x, y), v(x, y)
for the ow in part (b) by
u(x, y), v(x, y)
:=
1
h
u(x, y), v(x, y)
dz.
Show that the vertically averaged velocity eld u = (u, v)T is both incompressible
and irrotational.
(d) What is an appropriate boundary condition for u on the cylinder?
Exercise (Boundary layer theory: axisymmetric ow) In an axisymmetric ow the velocity components corresponding to cylindrical polar coordinates (r, , z) are (ur , u , uz )T .
If ur = r/2 and uz = z, where is a constant, verify that the continuity equation
is satised. If the swirl velocity u is assumed independent of z, show that the vorticity
T
has the form = 0, 0, (r, t) .
(a) At t = 0 the vorticity is given by = 0 f (r), where 0 is a constant. Verify from
the dynamical inviscid vorticity equation that at a later time t,
= 0 exp(t)f r exp(t/2) ,
and interpret this result in terms of stretching of material lines.
69
(b) Consider now steady viscous ow. Write down the governing equation for and
show that it is satised by
= 0 exp(r2 /4).
Why is a steady state possible in this case but not in (a)? What is the dominant
physical balance in the ow?
Exercise (Boundary layer theory: rigid wall) Seek similarity solutions of the boundary
layer equation
x xy x yy = U Ux + yyy
in the form = U (x)(x)f () where = y/(x). Show that f satises the equation
f
+ f f + 1 (f )2 = 0,
and explain why and must be constants. Give and in terms of U (x) and
(x), and hence determine the possible forms of U (x) and (x). State the boundary
conditions on f if there is a rigid wall at y = 0 and an outer ow with velocity U (x)
as y .
A Multivariable calculus identities

We provide here some useful multivariable calculus identities. Here and are generic scalars,
and u and v are generic vectors.
w
v
i
j
k
y
z
1.
u = det /x /y /z = u w .
z
x
v
u
v
w
u
x
y
2
2
2
2
+
+
.
2.
( ) =
=
x2
y 2
z 2
3.
( ) 0.
4.
( u) 0.
5.
( u) = ( u) 2 u.
6. () = + .
7. (u v) = (u )v + (v )u + u ( v) + v ( u).
8.
(u) = ( u) + u .
9.
(u v) = v ( u) u ( v).
10.
(u) = u + u.
11.
(u v) = u ( v) v ( u) + (v )u (u )v.
B NavierStokes equations in cylindrical polar coordinates

The incompressible NavierStokes equations in cylindrical polar coordinates (r, , z) with the
velocity eld u = (ur , u , uz )T are
u2
ur
1 p
ur
2 u
+ (u )ur =
+ ur 2 2
+ fr ,
t
r
r
r
r
u
ur u
1 p
2 ur
u
+ (u )u +
=
+ u + 2
2 + f ,
t
r
r
r
r
uz
1 p
+ (u )uz =
+ uz + fz ,
t
z
70
Simon J.A. Malham
where p = p(r, , z, t) is the pressure, is the mass density and f = (fr , f , fz )T is the body
force per unit mass. Here we also have
u
= ur
u
+
+ uz
r
r
z
and
1 2
2
r
+ 2
+
r r
r
r 2
z 2
Further the gradient operator and the divergence of a vector eld u are given in cylindrical
coordinates, respectively, by

=
1
= r
and
1
1 u
uz
(rur ) +
+
.
r r
r
z
u is given by
1 uz
u
r
r
z
ur
u = =
uz
.
z
r
1
1 ur
z
(ru ) r
r r
u=
In cylindrical coordinates
Lastly the diagonal components of the deformation matrix D are

Drr =
ur
,
r
D =
1 u
ur
+
r
r
and
Dzz =
uz
,
z
while the o-diagonal components are given by

2Dr = r
u
r r
1 ur
,
r
2Drz =
ur
uz
+
z
r
and
2Dz =
1 uz
u
+
.
r
z
C NavierStokes equations in spherical polar coordinates

The incompressible NavierStokes equations in spherical polar coordinates (r, , ) with the
velocity eld u = (ur , u , u )T are (note is the angle to the south-north pole axis and is
the azimuthal angle)
ur
+ (u
t
u2
u2
1 p
r
ur
2
2
u
+ ur 2 2 2
(u sin ) 2
r
r sin
r sin
)ur
+ fr ,
u2 cos
ur u
1 p
=
r
r sin
r
2 ur
u
cos u
+ u + 2
2
2 2
+ f ,
r
r sin2
r sin2
u
ur u
u u cos
1
p
+ (u )u +
+
=
t
r
r sin
r sin
2
ur
2 cos u
u
+ u + 2
+ 2
2
+ fz ,
r sin
r sin2
r sin2
u
+ (u
t
)u +
where p = p(r, , , t) is the pressure, is the mass density and f = (fr , f , f )T is the body
force per unit mass. Here we also have
u
= ur
u
u
+
+
r
r
r sin
71
and
=
r2
r2 r
r
sin
r2 sin
2
1
.
r2 sin2 2
Further the gradient operator and the divergence of a vector eld u are given in spherical
coordinates, respectively, by
r
1
r sin
and
u=
In spherical coordinates
1 2
1
1 u
(r ur ) +
(sin u ) +
.
r2 r
r sin
r sin
u is given by
1
(sin u ) u
r
r sin r
1
1
u = = r sin u r r (ru ) .
1
(ru ) 1 ur
r r
Lastly the diagonal components of the deformation matrix D are

Drr =
ur
,
r
D =
1 u
ur
+
r
r
and
D =
1 u
ur
u cot
+
+
,
r sin
r
r
while the o-diagonal components are given by

2Dr = r
u
r r
1 ur
,
r
2Dr =
1 ur
u
+r
r sin
r r
and
2D =
sin
u
r sin
1 u
.
r sin
D Ideal uid ow and conservation of energy

We show that an ideal ow that conserves energy is necessarily incompressible. We have derived
two conservation laws thus far, rst, conservation of mass,
+
t
(u) = 0,
and second, balance of momentum,
Du
= p + f.
Dt
If we are in three dimensional space so d = 3, we have four equations, but ve unknowns

namely u, p and . We cannot specify the uid motion completely without specifying one more
condition.
Denition 10 (Kinetic energy) The kinetic energy of the uid in the region D is
E :=
1
2
|u|2 dV.
72
Simon J.A. Malham
The rate of change of the kinetic energy, using the transport theorem, is given by
dE
d
=
dt
dt
|u|2 dV
1
2
D|u|2
dV
Dt
t
D
= 1
(u u) dV
2
t Dt
Du
=
u
dV
Dt
t
1
2
Du
Dt
dV.
Here we assume that all the energy is kinetic. The principal of conservation of energy states
(from Chorin and Marsden, page 13):
the rate of change of kinetic energy in a portion of uid equals the rate at which the
pressure and body forces do work.
In other words we have
dE
=
dt
u f dV.
p u n dS +
t
We compare this with our expression above for the rate of change of the kinetic energy. Equating the two expressions, using Eulers equation of motion, and noticing that the body force
term immediately cancels, we get
u
(up) dV =
u
p dV
p dV
p dV
p u n dS =
t
p+(
u) p dV =
u) p dV = 0.
Since and therefore t is arbitrary we see that the assumption that all the energy is kinetic
implies
u = 0.
Hence our third conservation law, conservation of energy has lead to the equation of state,
u = 0, i.e. that an ideal ow is incompressible.
Hence the Euler equations for a homogeneous incompressible ow in D are
u
+u
t
u=
p + f,
u = 0,
together with the boundary condition on D which is u n = 0.
E Isentropic ows
A compressible ow is isentropic if there is a function , called the enthalpy, such that
=
p.
73
The Euler equations for an isentropic ow are thus

u
+u u= +f
t
+ (u) = 0,
t
in D, and on D, u n = 0 (or matching normal velocities if the boundary is moving).
For compressible ideal gas ow, the pressure is often proportional to , for some constant
1, i.e.
p = C ,
for some constant C. This is a special case of an isentropic ow, and is an example of an
equation of state. In fact we can actually compute
p (z)
C
dz =
,
z
1
and the internal energy (see Chorin and Marsden, pages 14 and 15)
C
.
1
= (p/) =
F Selected detailed solutions

Solution (Trajectories and streamlines: three dimensions)
The velocity eld u(x, t) = (u, v, w)T is given for t > 1 by
u=
x
,
1+t
v=
y
1
1 + 2t
and
w = z.
To nd the particle paths or trajectories, we must solve the system of equations

dx
= u,
dt
dy
=v
dt
dz
= w,
dt
and
and then eliminate the time variable t between them. Hence for the particle paths we have
x
dx
=
,
dt
1+t
dy
y
=
dt
1 + 1t
2
and
dz
= z.
dt
Using the method of separation of variables and integrating in time from t0 to t, in each of
the three equations, we get
ln
x
x0
= ln
1+t
,
1 + t0
ln
y
y0
= 2 ln
1 + 1t
2
1 + 1 t0
2
and
ln
z
z0
= t t0 ,
where we have assumed that at time t0 the particle is at position (x0 , y0 , z0 )T . Exponentiating
the rst two equations and solving the last one for t, we get
x
1+t
=
,
x0
1 + t0
1
(1 + 2 t)2
y
=
1
y0
(1 + 2 t0 )2
and
t = t0 + ln(z/z0 ).
We can use the last equation to eliminate t so the particle path/trajectory through (x0 , y0 , z0 )T
is the curve in three dimensional space given by
x = x0
1 + t0 + ln(z/z0 )
(1 + t0 )
and
y = y0
1
1 + 2 t0 +
(1 +
2
1
ln(z/z0 )
2
.
1
t )2
2 0
74
Simon J.A. Malham
To nd the streamlines, we x time t. We must then solve the system of equations

dx
= u,
ds
dy
=v
ds
dz
= w,
ds
and
with t xed, and then eliminate s between them. Hence for streamlines we have
dx
x
=
,
ds
1+t
dy
y
=
ds
1 + 1t
2
and
dz
= z.
ds
Assuming that we are interested in the streamline that passes through the point (x0 , y0 , z0 )T ,
we again use the method of separation of variables and integrate with respect to s from s0 to
s, for each of the three equations. This gives
ln
x
x0
s s0
,
1+t
ln
y
y0
s s0
1 + 1t
2
and
ln
z
z0
= s s0 .
Using the last equation, we can substitute for ss0 into the rst equations. If we then multiply
the rst equation by 1 + t and the second by 1 + 1 t, and use the usual log law ln ab = b ln a,
2
then exponentiation reveals that
x
x0
1+t
y
y0
1+ 1 t
2
z
,
z0
which are the equations for the streamline through (x0 , y0 , z0 )T .

Solution (Venturi tube)
(a) Fluid incompressible, so volume ux (= cross-sectional area uniform velocity) through
wide region equals that in narrow
A1 u1 = A2 u2 .
Since A2 < A1 =
u2 =
A1
u1 > u1 .
A2
(b) Bernoullis Theorem = quantity H same in wide and narrow regions on same
streamline passing through tube
1 2
u
2 1
p1
=
1 2
u
2 2
p2
.
(Assume potential dierence along any streamline negligible.)

(c) Bernoulli result
p1 p2 =
1
(u2
2
2
u2 ) > 0
1
since u2 > u1 . Pressure in narrow region less!

(d) Examples are: device for measuring ow speeds (measure p1 p2 , A1 /A2 known =
u1 , u2 ); carburetor, pressure drop draws in another uid from side channel; or lift of an aircraft
wing.
Solution (Clepsydra or water clock)
(a) Use Bernoullis Theorem for a streamline starting at the free surface and going through
the orice. The potential energy of any uid particle at the orice outlet compared to that at
the free surface is g. This we have
1
2
dz
dt
+
1
2
dz
dt
P0
=
P0
g
1 2
U
2
1 2
U
2
g.
75
(Note that inside the container at the level at which water is pushed through the orice,
the pressure is P = P0 + gz. The dierence P P0 accelerates the water through the orice.)
(b) Using incompressibility, we simply equate the volume ux through the orice to the
rate of volume drop at the free surface (as it descends). Thus, since S is the cross-sectional
area of the orice and A = r2 is that of the clepsydra at height z, we get
SU = r2
dz
.
dt
(c) Rearranging the result from part (b) we see that

1 dz
S
= .
U dt
A
Thus using the result from part (a) we nd that
A so that S/A
1 2
U
2
=
We assume S
dz 2
dt
1 dz 2
1
U dt
S 2
.
1
A
U2
1
2
gz =
1 2
U
2
1. Hence we see that

U
2gz.
with an error of order (S/A)2 .

(d) We want dz/dt to be constant, so combining the results of parts (b) and (c) =
dz
dt
dz
2g S dt
2gz = r2
z=
r4 ,
as required.
Solution (Coee mug)
(a) We are given that ur = uz = 0, /t 0, / 0 and fr = f = 0 with fz = g.
Substituting all these into Eulers equations in cylindrical polar coordinates =
u2
1 p
,
r
1 p
0=
,
r
1 p
0=
g.
z
Note
1 (rur )
1 u
uz
+
+
=0
r r
r
z
trivially, and that u , which simplies to (u /r)(/), also acts trivially on the velocity
eld (which independent of ). The second equation in the system above implies p = p(r, z).
u=
(b) Using that u = r =

p
= 2 r
r
p(r, z) =
1
2 r2
2
+ G(z),
76
Simon J.A. Malham
where G an arbitrary function. Substitute this into other equation

1 p
= g
z
G (z) = g.
Hence G(z) = gz + C where C is an arbitrary constant. =

p(r, z) =
1
2 r2
2
gz + C.
At the free surface p = P0 :

P0 = 1 2 r2 gz + C
2
z =( 2 /2g) r2 (C P0 )/g.
(c) Choose C = P0 z = ( 2 /2g) r2 . Relative to chosen origin O, the bottom of the mug
is at z = z0 .
(i) Initially coee depth d = total volume is a2 d.
Incompressibility volume constraint =
a2 z0
+
0
vol under O
2 r2
2r dr = a2 d.
2g
vol O and surface
(ii) Constraint in part (i)

z0 +
2 a2
= d.
4g
Coee at edge of mug is at height

z0 +
2 a2
2 a2
2 a2
2 a2
=d
+
=d+
.
2g
4g
2g
4g
Spillage when this exceeds h, i.e. when 2 > 4g(h d)/a2 .

Solution (Bernoullis Theorem for irrotational unsteady ow)
(a) Since the ow is incompressible and homogeneous, is uniform and constant. The ow
is also irrotational so that u = . Eulers equations =
( ) + u
t
u=
Using the given identity we get
1
2
|u|2 u (
u) =
+ 1 |u|2 + + = u (
2
t
H =u0
H=0
for the quantity H given in the question.
u)
77
(b) Hence H = f (t) for some function f = f (t). Using the suggested redened potential
for u given by
t
V =
f ( ) d
=V +
f ( ) d
0
and substituting this into

H
= f (t)
t
we get

p
1
+ 2 |u|2 + + = f (t)
t t
p
V
+ f (t) + 1 |u|2 + + = f (t)
2
t t
p
V
1
+ 2 |u|2 + + = 0
t t
which gives the result.
Solution (Elliptical pipe ow)

(a) Using the NavierStokes equations in three dimensional Cartesian coordinates, given
that u = 0, v = 0 and w = w(x, y) only, and assuming no body force, we are left with
p
,
x
p
0= ,
y
0=
0=
p
2w
2w
+
+ 2
,
2x
z
where note that u w = 0 as w is independent of z. Since we are given p = Gz, the rst
two equations are consistent and substituting this form for p into the nal equation gives the
required result
2w
2w
G
+ 2 = .
2x
(b) Substituting w = Ax2 + By 2 + C in to the partial dierential equation in part (a), we

immediately deduce
G
A+B = .
2
(c) Using the no-slip boundary condition, i.e. that w = 0 on

x2
y2
+ 2 =1
2
a
b
x2
a2
y 2 = b2 1
x2
,
a2
generates the equation

Ax2 + Bb2 1
+C =0
AB
b2
x2 + Bb2 + C = 0
a2
which must hold for all x [a, a]. Hence equating coecients of x0 and x2 , we arrive at the
following three equations (including the result from part (b) above) for the three unknowns A,
B and C:
b2
G
A B 2 = 0,
Bb2 + C = 0
and
A+B = .
a
2
78
Simon J.A. Malham
Solving the rst equation for A in terms of B and substituting this into the third equation we
nd
B 1+
b2
a2
B=
G
a2
.
2 a2 + b2
G
2
A=
G
b2
.
2 a2 + b2
Substituting this into the rst equation reveals

A+
G
a2
b2
=0
2 a2 + b2 a2
Finally substituting these two answers for A and B into the second equation =
G
a2
b2 + C = 0
2 a2 + b2
C=
G
a2 b2
.
2 a2 + b2
(d) For a small patch of area dS of an elliptical cross section of the pipe, the volume of uid
passing through dS per unit time is equivalent to the volume of the cylinder of cross sectional
area dS and length w (the orthogonal ow rate through dS), i.e. w dS. Summing over all such
small patches of areas to make up the complete cross section generates the integral
w dS
over the whole elliptical cross sectional area.
Computing the integral using the substitutions x = ar cos and y = br sin we see:
Ax2 + By 2 + C dxdy
w dS =
=
0
=
=
=
Aa2 r2 cos2 + Bb2 r2 sin2 + C abr dr d
a3 b3
G
2 a2 + b2
2Ga3 b3
2(a2 + b2 )
2
0
1
(1 r2 ) r dr d
r r3 dr
a3 b3 G
.
4(a2 + b2 )
(e) Denote the ow rate from part (d) above by

Q :=
a3 b3 G
.
4(a2 + b2 )
We wish to nd the maximum of Q subject to the constraint that the cross sectional area of the
elliptic pipe is given, say by K, i.e. ab = K with K xed. Simply substitute that b = K/a
into Q and dierentiate with respect to a =
Q
(K/)3 G
=
2 + K 2 2 a2 )
a
a 4(a
=
(K/)3 G (2a 2K 2 2 a3 )
2
.
4
(a + K 2 2 a2 )2
Thus Q/a is zero and Q maximized if and only if

2a 2K 2 2 a3 = 0
a=
K/.

Note that when a =
79
K/, then
b = K/a = K/( K 1/2 / 1/2 ) = K 1/2 / 1/2 = a.
Hence the ow rate Q is maximized when a = b.
Acknowledgements These lecture notes have to a large extent grown out of a merging of,
lectures on Ideal Fluid Mechanics given by Dr. Frank Berkshire [2] in the Spring of 1989,
lectures on Viscous Fluid Mechanics given by Prof. Trevor Stuart [17] in the Autumn of
1989 (both at Imperial College) and the style and content of the excellent text by Chorin
and Marsden [3]. They also beneted from lecture notes by Prof. Frank Leppington [11] on
Electromagnetism. SJAM would like to thank Prof. Andrew Lacey, Dr Daniel Coutand and
Callum Thompson for their suggestions and input. Lastly, thanks to all the students who
pointed out typos to me along the way!
References
1. Batchelor, G.K. 1967 An introduction to uid mechanics, CUP.
2. Berkshire, F. 1989 Lecture notes on ideal uid dynamics, Imperial College Mathematics
Department.
3. Chorin, A.J. and Marsden, J.E. 1990 A mathematical introducton to uid mechanics,
Third edition, SpringerVerlag, New York.
4. Doering, C.R. and Gibbon, J.D. 1995 Applied analysis of the NavierStokes equations,
Cambridge Texts in Applied Mathematics, Cambridge University Press.
5. Evans, L.C. 1998 Partial dierential equations, Graduate Studies in Mathematics, Volume
19, American Mathematical Society.
6. Fulton, W. and Harris, J. 2004 Representation theory: A rst course, Graduate Texts in
Mathematics 129, Springer.
7. Gurtin, M.E. 1981 An introduction to continuum mechanics, Mathematics in Science and
Engineering, Volume 158, Academic Press.
8. Keener, J.P. 2000 Principles of applied mathematics: transformation and approximation,
Perseus Books.
9. Krantz, S.G. 1999 How to teach mathematics, Second Edition, American Mathematical
Society.
10. Lamb, H. 1932 Hydrodynamics, 6th Edition, CUP.
11. Leppington, F. 1989 Electromagnetism, Imperial College Mathematics Department.
12. McCallum, W.G. et. al. 1997 Multivariable calculus, Wiley.
13. Majda, A.J. and Bertozzi, A.L. 2002 Vorticity and incompressible ow, Cambridge Texts
in Applied Mathematics, Cambridge University Press.
14. Marsden, J.E. and Ratiu, T.S. 1999 Introduction to mechanics and symmetry, Second
edition, Springer.
15. Meyer, C.D. 2000 Matrix analysis and applied linear algebra, SIAM.
16. Saman, P.G. 1992 Vortex dynamics, Cambridge Monographs in Mechanics and Applied
Mathematics, Cambridge University Press.
17. Stuart, J.T. 1989 Lecture notes on Viscous Fluid Mechanics, Imperial College Mathematics
Department.

Malham-Fluid Mechanics Notes

Uploaded by

Copyright:

Available Formats

Malham-Fluid Mechanics Notes

Uploaded by

Document Information

Original Title

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

Malham-Fluid Mechanics Notes

Uploaded by

Copyright:

Available Formats

Lecture notes

Introductory uid mechanics

Simon J.A. Malham (17th March 2014)

Simon J.A. Malham

2 Fluid ow, the Continuum Hypothesis and conservation principles

Introductory uid mechanics

3 Trajectories and streamlines

Simon J.A. Malham

Using that x(0) = x0 and y(0) = y0 we nd that A = x0 and B = y0 so the particle

y(t) = x0 sin t + y0 cos t

Introductory uid mechanics

We can solve the rst dierential equation which tells us

If we eliminate the parameter s between x = x(s) and y = y(s) above, we nd the

Simon J.A. Malham

Introductory uid mechanics

Returning to the principle of conservation of mass, this is now equivalent to the

Using these two relations, the law of conservation of mass is equivalent to

This is the rst of our three conservation laws.

Simon J.A. Malham

Denition 3 (Material derivative) If the velocity eld components are

then we dene the material derivative following the uid to be

Introductory uid mechanics

u((x, t), t) J(x, t).

Using the chain rule we have

Simon J.A. Malham

Combining the last two relations we see that

Abels Theorem then tells us that J = det

((x, t), t) J(x, t) dV

Introductory uid mechanics

u((x, t), t) J(x, t) dV (x)

Further, noting that by denition J(x, 0) = 1, establishes the result.

The continuity equation and the identity,

Hence since > 0, a ow is incompressible if and only if

i.e. the uid density is constant following the uid.

This is equivalent to the statement

(x, t), t J(x, t) dV =

where we made a change of variables. Since is arbitrary, we deduce

Simon J.A. Malham

Example Suppose that in Cartesian coordinates we have the two dimensional ow

u = 0 so there exists a stream function satisfying

We deduce C (x) = 0 and therefore C is an arbitrary constant. Since a stream function

Introductory uid mechanics

Now suppose we used plane polar coordinates instead. The corresponding ow

First note that

8 Rate of strain tensor

where ( u) h is simply matrix multiplication of the 3 3 matrix

u/x u/y u/z

Simon J.A. Malham

Note that D = D(x) is a 3 3 symmetric matrix, while R = R(x) is the 3 3

Note that if we set

then R is more simply expressed as

Further by direct computation we see that

The symmetric matrix D is the deformation tensor. Since it is symmetric, there is

Let us consider in turn each of the eects on the right shown:

Introductory uid mechanics

Making a local change of coordinates so that h = X h we get

We see that we have pure expansion or contraction (depending on whether di

is positive or negative, respectively) in each of the characteristic directions hi ,

9 Internal uid forces