Fluid Dynamics 2

Dr. J. M. Rallison
Michælmas 1997

These notes are maintained by Paul Metcalfe.

Comments and corrections to pdm23@cam.ac.uk.
Revision: 2.7
Date: 2004/08/23 07:15:40

The following people have maintained these notes.

– date Paul Metcalfe

I have included some of EJH’s updates to these notes and will include more as time
progresses.
Contents

Introduction v

1 Review of inviscid fluids 1

1.1 Continuum hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Time derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Mass conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.4 Kinematic boundary condition . . . . . . . . . . . . . . . . . . . . . 1
1.5 Momentum conservation . . . . . . . . . . . . . . . . . . . . . . . . 2
1.6 Dynamic boundary condition . . . . . . . . . . . . . . . . . . . . . . 2
1.7 Steady flow past a circular cylinder . . . . . . . . . . . . . . . . . . . 2

2 The governing equations 3

2.1 Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Rate of strain tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 The stress tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 The constitutive equation . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6 The energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.7 Scaling estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7.1 Compressibility . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7.2 Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Low Reynolds number flows 9

3.1 Properties of the Stokes equations . . . . . . . . . . . . . . . . . . . 9
3.1.1 Instantaneity . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.2 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.3 Reversibility . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.4 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1.5 Minimum dissipation . . . . . . . . . . . . . . . . . . . . . . 10
3.1.6 Solving the Stokes equations . . . . . . . . . . . . . . . . . . 11
3.2 Stokes flow due to a translating sphere . . . . . . . . . . . . . . . . . 12
3.2.1 Dimensional analysis . . . . . . . . . . . . . . . . . . . . . . 12
3.2.2 Brute force . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2.3 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Reciprocal Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Movement of rigid particles in viscous fluids . . . . . . . . . . . . . . 15
3.4.1 Special cases . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5 Stokes flow in a corner . . . . . . . . . . . . . . . . . . . . . . . . . 17

iii
iv CONTENTS

4 Flow in a thin layer 19

4.1 Lubrication theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Thrust bearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Flow in a Hele-Shaw cell . . . . . . . . . . . . . . . . . . . . . . . . 21
4.4 Saffman-Taylor instability . . . . . . . . . . . . . . . . . . . . . . . 21
4.5 Gravitational spreading . . . . . . . . . . . . . . . . . . . . . . . . . 23

5 Vorticity generation and confinement 27

5.1 Vorticity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1.1 Planar flows . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2 Vorticity generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3 Vorticity confinement on a flat plate . . . . . . . . . . . . . . . . . . 29
5.4 Stagnation point flow . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.5 The bathtub vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

6 Boundary layers 31
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2 Steady boundary layers . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.3 Blasius boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.4 Similarity solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.5 High Reynolds number flow past a wedge . . . . . . . . . . . . . . . 34
6.5.1 Outer problem . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.5.2 The boundary layer . . . . . . . . . . . . . . . . . . . . . . . 35
6.5.3 Numerical solution . . . . . . . . . . . . . . . . . . . . . . . 35
6.5.4 Separation of the boundary layer . . . . . . . . . . . . . . . . 36
6.6 Converging and diverging flow in a wedge . . . . . . . . . . . . . . . 37

7 Aerodynamics 39
7.1 Complex potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.2 Conformal mappings . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.2.1 Flow past an ellipse with circulation . . . . . . . . . . . . . . 41
7.3 Forces, drag and lift . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.4 The Kutta-Joukowski condition . . . . . . . . . . . . . . . . . . . . . 43
7.5 Physical mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

8 Kelvin-Helmholtz instability 45

9 Rising bubbles 49
9.1 Dimensional analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9.2 Low Reynolds number, low capillary number . . . . . . . . . . . . . 49
9.3 High Reynolds number, low capillary number . . . . . . . . . . . . . 51
9.4 The oblate spheroidal bubble . . . . . . . . . . . . . . . . . . . . . . 52
9.5 Spherical cap bubble . . . . . . . . . . . . . . . . . . . . . . . . . . 52
9.6 The skirted bubble . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

Stress tensor: EJH approach 55

Introduction

These notes are based on the course “Fluid Dynamics 2” given by Dr. J.M. Rallison in
Cambridge in the Michælmas Term 1997. These typeset notes are totally unconnected
with Dr. Rallison. Recommended books will be discussed at the end.
Other sets of notes are available for different courses. At the time of typing these
courses were:
Probability Discrete Mathematics
Analysis Further Analysis
Methods Quantum Mechanics
Fluid Dynamics 1 Quadratic Mathematics
Geometry Dynamics of D.E.’s
Foundations of QM Electrodynamics
Methods of Math. Phys Fluid Dynamics 2
Waves (etc.) Statistical Physics
General Relativity Dynamical Systems
Combinatorics Bifurcations in Nonlinear Convection

They may be downloaded from

http://www.istari.ucam.org/maths/.

v
vi INTRODUCTION
Chapter 1

Review of inviscid fluids

1.1 Continuum hypothesis

We assume that at each point x of the fluid we can define, by averaging over a small
volume, properties like density ρ(x, t), velocity u(x, t) and pressure p(x, t) and that
these vary smoothly over the fluid. We do not deal with the dynamics of individual
molecules.

1.2 Time derivatives

A fluid particle, sometimes called a material element or Lagrangian point, is one that
moves with the fluid, so that its position x(t) satisfies ẋ = u(x, t).
D
The rate of change of some quantity moving with the fluid is written Dt ; the chain
rule gives
D ∂
= + u · ∇. (1.1)
Dt ∂t
Du ∂u
In particular, the acceleration of a fluid particle is Dt = ∂t + u · ∇u.

1.3 Mass conservation

Since mass is conserved the mass density ρ satisfies ∂ρ∂t + ∇ · (ρu) = 0. The quantity
ρu is called the mass flux.
For an incompressible fluid the density of each material element is constant, and so
Dρ
Dt = 0. Thus ∇ · u = 0.
In this course we will restrict to fluids which are incompressible and have uniform
density so that ρ is independent of both x and t.
For planar flows, the condition ∇ · u = 0 is automatically satisfied if we have u =
∇ × (0, 0, ψ(x, y)), so that u = (ψy , −ψx , 0). ψ(x, y) is called the streamfunction.

1.4 Kinematic boundary condition

Applying mass conservation to a region close to a boundary S we get n · u+ = n · u−
at S.

1
2 CHAPTER 1. REVIEW OF INVISCID FLUIDS

This states that the normal component of velocity is continuous across S. In par-
ticular, if S is fixed we have n.u = 0 at S.
The kinematic boundary condition can be written a different way. Suppose the
boundary of a fluid is given by F (x, t) = 0. Then since the surface consists of material
points DFDt = 0. This is sometimes more convenient for free surface problems.

1.5 Momentum conservation

Assuming that the only force acting across a material surface n dS is given via a pres-
sure p(x, t) as −pn dS then we obtain Euler’s equation:
Du
ρ = −∇p + ρF(x, t), (1.2)
Dt
where F(x, t) is the body force per unit mass (for instance gravity) that acts on the
fluid.

1.6 Dynamic boundary condition

On the same assumption, applying momentum conservation to a region close to a
boundary S gives −p− n = −p+ n in the absence of surface tension.
In this course we will abandon the assumptions of §1.5 and §1.6 and include tan-
gential frictional forces across material surfaces.

1.7 Steady flow past a circular cylinder

2
The steady Euler equation with F = 0 is satisfied if u = ∇φ and p + 21 ρ |u| = const.
The incompressibility condition ∇ · u = 0 becomes ∇2 φ = 0.
A solution with φ ∼ U r cos θ as r → ∞ (uniform stream with velocity U ) and
u.n = 0 on r = L is

L2
 
φ=U r+ cos θ, (1.3)
r
with streamfunction

L2
 
ψ=U r− sin θ. (1.4)
r
The tangential velocity on r = L is 2U sin θ.
Chapter 2

The governing equations for a

Newtonian fluid

2.1 Viscosity
Suppose we have two parallel plates a distance h apart, and we put fluid between them.

What force per unit area on the top plate is needed to keep it moving at a velocity
U ? Experiments show that it is proportional to Uh , and measuring the flow profile shows
that Uh = ∂u
∂y .
The coefficient of proportionality is the viscosity, µ. It has dimensions M L−1 T −1 .

2.2 Rate of strain tensor

Consider the fluid motion near a point 0. Then
∂ 2 ui 
 
∂ui  1
ui (x) = ui (0) + xj + x x
j k + ....
∂xj 0 2 ∂xj xk 0
Thus 
∂ui 
ui (x) − ui (0) ≈ xj .
∂xj 0
∂ui
∂xjis called the velocity gradient, and is sometimes written (∇u)ji .
The symmetric part of the velocity gradient is the rate of strain tensor,
 
1 ∂ui ∂uj
eij = + (2.1)
2 ∂xj ∂xi
and the antisymmetric part is the vorticity tensor
 
1 ∂ui ∂uj 1
Ωij = − = − ijk ωk , (2.2)
2 ∂xj ∂xi 2

3
4 CHAPTER 2. THE GOVERNING EQUATIONS

where the vector ω is the vorticity, ω = ∇ × u.

Thus
1
ui (x) − ui (0) ≈ xj eij + (ω × x)i .
2
The ω × x part of this is a solid body rotation, which we hope causes no stress.
eij is symmetric and so can be diagonalised. Its eigenvalues α, β and γ are the
principal rates of strain. Note that α + β + γ = eii = ∇ · u = 0.

2.3 The stress tensor

The forces acting on a fluid are of two kinds.
1. Volume or body forces. These have a long range and are proportional to the
volume of a fluid element. (gravity)
2. Surface tractions. These have a short range and are proportional to the surface
area of a fluid element.
Let ndS be an arbitrary element of area drawn in the fluid at (x, t). We write the
force exerted by the fluid on the + side of dS on the fluid on the − side as τ dS. Here
we establish our convention: normals point out of the fluid.
τ is called the surface traction and depends on x, t and n.
Theorem. We claim that τ is linearly related to n, that is
τi = σij nj .
σij (x, t) is a second rank tensor called the Cauchy stress tensor.
Proof. Let V (t) be an arbitrary material volume with surface S(t). The momentum of
the fluid in V (t) is thus Z
ρudV,
V (t)

and so the equation of motion for the fluid in V (t) is

Z Z Z
d
ρudV = ρFdV + τ dS. (2.3)
dt V (t) V (t) S(t)

Now suppose that V (t) is small, with linear dimension . As volume integrals are
O(3 ) and surface integrals are O(2 ), and in the limit  → 0 the equation of motion
must balance at leading order, we have
Z
lim τ dS = 0.
→0 S(t)

Now let V be instantaneously a small tetrahedron (as sketched), with a sloping face
having area dS and normal n. The areas of the other faces are therefore dSi · n, dSj · n
and dSk · n, where i, j and k are the usual unit vectors.
2.3. THE STRESS TENSOR 5

Since the surface forces on the this tetrahedron must balance,

τ (n)dS + τ (−i)i · ndS + τ (−j)j · ndS + τ (−k)k · ndS = 0.
Eliminating dS we obtain a linear relationship between τ and n.
With this expression for the surface tractions, the equation of motion (2.3) is
Z Z Z
d
ρui dV = ρFi dV + σij nj dS.
dt V (t) V (t) S(t)

Using the divergence theorem and the fact that V (t) is a material volume,
Z Z Z
Dui ∂σij
ρ dV = ρFi dV + dV.
V (t) Dt V (t) V (t) ∂xj

Since V is arbitrary, we obtain the Cauchy momentum equation

Dui ∂σij
= ρFi +ρ . (2.4)
Dt ∂xj
If F = 0 then momentum must be conserved. We can rewrite the Cauchy momen-
tum equation in the form
∂ ∂
(ρui ) + (ρui uj − σij ) = 0.
∂t ∂xj
This is the the form of a conservation equation for the momentum ρu, and we
identify the momentum flux tensor as ρui uj − σij .
Theorem. Provided no body couples act on the fluid, σij = σji .
Proof. Taking the origin to lie instantaneously within V (t), the angular momentum of
the fluid in V is Z
ρx × u dV,
V (t)
4
which is O( ). Conservation of angular momentum implies that in the absence of
body couples1
Z Z Z
d
ρx × u dV = ρx × F dV + x × τ dS.
dt V (t) V (t) S(t)

The last term here is O(3 ) and is therefore of lower order than the other two terms.
Therefore at leading order it must vanish:
Z
lim x × τ dS = 0.
→0 S(t)
th
Using our result for τ , the i component of this can be written
Z Z
∂
ijk xj σkm nm dS = ijk (xj σkm ) dV
S(t) V (t) ∂x m
Z Z
∂σkm
= ijk xj dV + ijk σkj dV.
V (t) ∂xm V (t)

The integrals here are O(4 ) and O(3 ) respectively, so letting  → 0 we obtain
ijk σkj = 0 and so the stress tensor is symmetric.
1 As would arise for a suspension of orientable magnetic particles in an external magnetic field.
6 CHAPTER 2. THE GOVERNING EQUATIONS

2.4 The constitutive equation for a Newtonian fluid

We define the pressure p = − 13 σii , and so
0
σij = −pδij + σij ,
0
where σij , the deviatoric stress, is tracefree.
We assume that σ 0 is linearly related to the instantaneous value of ∇u and that the
fluid is isotropic. Thus we have

0 ∂uk
σij = Aijkl ,
∂xl
with A a property of the fluid, and so isotropic. Therefore

Aijkl = λδij δkl + µδik δjk + µ0 δil δjk

and so
0 ∂ui ∂uj
σij = λδij ∇ · u + µ + µ0 .
∂xj ∂xi
As the stress tensor is symmetric, µ = µ0 . µ is the (shear) viscosity. Thus
0
σij = 2µeij and we obtain the constitutive equation for a Newtonian fluid,

σij = −pδij + 2µeij . (2.5)

As we hinted on on page 4, this does not depend on vorticity. In general µ depends
on temperature and so can depend on position.
We can now substitute (2.5) into (2.4) to obtain the Navier-Stokes equations

Dui ∂p
ρ =− + ρFi + µ∇2 ui or in vector notation
Dt ∂xi
(2.6)
Du
ρ = −∇p + ρF + µ∇2 u.
Dt

2.5 Boundary conditions

We can keep the kinematic boundary condition (that only depended on mass conserva-
tion) and so the normal component of u is continuous at a boundary.
At a boundary S no net force can be applied to a pillbox; therefore the surface
tractions must balance. Thus [σ · n]S = 0. More generally, if surface tension acts at
S, a net force parallel to n and proportional to the curvature R1−1 + R2−1 (where R1

and R2 are the principal radii of curvature) appears and we have

 
1 1
[σ · n]S = γ + n.
R1 R2

γ is the surface tension coefficient. We can see that in the inviscid case µ = 0 we
get back the dynamic boundary condition.
We need an extra boundary condition as a ∇2 term has appeared. We assume that
n × u is continuous — this is the no-slip condition.
2.6. THE ENERGY EQUATION 7

2.6 The energy equation

Taking (as before) V (t) to be an arbitrary material volume and S(t) its surface, the
kinetic energy of the fluid inside V (t) is
Z
ρ 2
E= |u| dV
2 V (t)
and the rate of change of this energy is given by the Cauchy equation (2.4) as
Z Z Z
dE Dui ∂σij
=ρ ui dV = ρ ui Fi dV + ui dV.
dt V (t) Dt V (t) V (t) ∂xj

The final term here may be written

Z Z  
∂σij ∂ ∂ui
ui dV = (ui σij ) − σij dV
V (t) ∂xj V (t) ∂xj ∂xj
Z Z
= ui σij nj dS − σij eij dV
S(t) V (t)
Z Z
= ui τi dS − 2µ eij eij dV.
S(t) V (t)

We have therefore shown that

Z Z Z
dE
=ρ ui Fi dV + ui τi dS − 2µ eij eij dV.
dt V (t) S(t) V (t)

The first two terms on the right represent the rate of working by body and surface
forces respectively, so the final term must be the rate of energy dissipation due to vis-
cosity. The rate of viscous heating per unit volume Φ = 2µeij eij , and the second law
of thermodynamics demands that Φ and therefore µ must be positive.
This heating can change the temperature in the fluid. If then the density or viscosity
depend on temperature then a further equation involving the convection and diffusion
of heat is needed to determine the temperature. We shall not persue this (interesting)
complication in this course.
Using the momentum flux equation we may alternatively write the energy equation
in the form
∂ 1 2
 ∂
2 ρ |u| + ui (ρui uj − σij ) = ρui Fi .
∂t ∂xj
Hence
∂ 1 2
 ∂  1 2

ρ |u| + u j ρ |u| − u i σ ij = ρui Fi − Φ.
∂t 2 ∂xj 2

We can thus identify the energy flux vector q by,

2
qi = 12 ρ |u| ui − σij uj ,

and the energy flux equation takes the canonical form

∂
(energy) + ∇ · (energy flux) =
∂t
(rate of doing work by body forces) − (loss of energy due to viscous heating) .
8 CHAPTER 2. THE GOVERNING EQUATIONS

2.7 Scaling estimates

We would like to know when it is appropriate to treat a real fluid as if it were incom-
pressible or inviscid. We can get crude estimates as follows.
Suppose the flow has a typical velocity U , lengthscale L and timescale T . For a
sphere in a uniform steady stream, U is the far-field velocity, L is the radius of the
sphere and T is infinite.
If we now wobble this sphere with angular frequency ω we would keep U and L
the same, but T becomes ω −1 .

2.7.1 Compressibility
The pressure p depends on the mass density ρ, and it may be shown that for small
variations in density about some ambient level ρ we have

dp
= c2 ,
dρ
where c is the speed of sound in the fluid. Thus
∆ρ p
∼ 2.
ρ ρc
The Navier-Stokes equations (2.6) give

ρU ρU 2
   
p ∂u
∼ ∇p ∼ max ρ , ρu · ∇u, µ∇2 u ∼ max , , µf racU 2 L2 .
L ∂t T L

Thus
LU U 2 µ U
 
∆ρ
∼ max , , .
ρ c2 T L2 ρ c2 L
∆ρ U
The fluid is incompressible if ρ  1, and in practice the Mach number, c  1.

2.7.2 Viscosity
2
For a steady flow ρU
L ∼
µU
L2 , and the ratio of inertial forces over viscous forces (the
Reynolds number) is
ρU L
Re = .
µ
The ratio ν = µρ appears. This is the kinematic viscosity.
If Re  1 then the viscosity is negligible and inertia dominates. If Re  1 then
inertia is negligible.
2
If we have rectilinear flow, u = (U (y, z), 0, 0) then u · ∇u = 0 and UL is a bad
estimate of this...
For some unsteady flows the inertial term ρ ∂u
∂t wins, in this case we get an unsteady
L2
Reynolds number νT . This is also called the Stokes number.
Chapter 3

Low Reynolds number flows

In this case Re  1 and we neglect inertial terms in the Navier-Stokes equations (2.6)
to obtain

µ∇2 u = ∇p − ρF ∇ · u = 0. (3.1)
If F ≡ 0 then these are the Stokes equations. Natural boundary conditions are that
at each point of S either u or σ · n is given.
(3.1) can also be written
∂σij
= −ρFi σij = −pδij + 2µeij ∇ · u = 0. (3.2)
∂xj

3.1 Properties of the Stokes equations

3.1.1 Instantaneity
There are no time derivatives in (3.1). Thus u responds instantaneously to the boundary
motion (and the force F). There is thus an infinite propagation speed; this situation is
sometimes called “quasi-static”.
For instance, in a sphere falling in an unbounded fluid then the terminal velocity is
acheived at once. For a sphere falling towards a wall then the change in velocity is due
only to the change in the fluid domain.

3.1.2 Linearity
There is no u · ∇u term in (3.1); therefore u, p and σ are linearly forced by any
boundary motion (or body force).
For instance; if we have a falling sphere, doubling the velocity will double σ and
thus double the drag. More generally, force ∝ velocity (as opposed to acceleration).
Another example; if we have a moving ellipsoid the problem can be solved by
superimposing the solutions when the ellipsoid moves along its principal axes.

3.1.3 Reversibility
If the velocity on the boundary of a Stokes flow is reversed then so is the velocity
everywhere in the fluid. If a prescribed boundary motion is reversed over time then

9
10 CHAPTER 3. LOW REYNOLDS NUMBER FLOWS

each material point retraces its history.

Does a sphere falling by a wall migrate towards/away from the wall? No — on

reversal of g, u must reverse and so if the sphere migrates to the wall under g then it
must migrate away from the wall under −g (and similarly for the other case).

3.1.4 Uniqueness
There exists at mosts one Stokes flow in a volume V for which u is specified on the
boundary.
Proof. Suppose u1 and u2 are two such flows. Let u∗ = u1 − u2 , so ∇2 u∗ = 0. Also,
let σ ∗ = σ 1 − σ 2 and e∗ =∗ e1 − e2 .
∂σ ∂u∗
Then (3.2) gives that ∂xijj = 0 and ∂xii = 0. Now consider

∂u∗i
Z Z
2µ e∗ij e∗ij dV = ∗
σij dV
V ∂xj
ZV
∂
σ ∗ u∗ dV


=
∂xj ij i
ZV
∗ ∗
= σij ui nj dS = 0.
S

Thus since e∗ij e∗ij

≥ 0 we must have e∗ij
= 0. Now the most general motion having
no rate of strain is a rigid body motion and so u∗ = u∗ + Ω∗ × x for constant u∗ and
Ω∗ . But since u∗ = 0 on S we have u∗ = 0 everywhere.
√
π 3
A more sophisticated argument proves uniqueness if Re < 3 .

3.1.5 Minimum dissipation

Suppose u(x) is the unique Stokes flow in V satisfying u = u on S. Let ū(x) be
another flow in V such that ∇ · ū = 0 and ū = u on S (kinematically admissible).
Then Z Z
2µ ēij ēij dV ≥ 2µ eij eij dV,
V V
with equality only if u = ū.
Proof. Let u∗ = u − ū and e∗ = e − ē. Then
Z Z
2µ e∗ij eij dV = σij e∗ij dV
V V
∂u∗i
Z
= σij dV
∂xj
ZV
= u∗i σij nj dS = 0.
S
3.1. PROPERTIES OF THE STOKES EQUATIONS 11

Now consider

Z Z
(ēij ēij − eij eij ) dV = − e∗ij (ēij + eij ) dV
V V
Z Z
= e∗ij e∗ij dV − 2 e∗ij eij dV
V V
≥0 as required.

As an example of this we consider the drag on a rigid particle in unbounded fluid.

Z
F= σ · n dS
S

is the force exerted by the particle on the fluid (the drag is −F). We have to solve
the Stokes equations in V with the boundary conditions that u = u on S and u → 0 as
x → ∞.
Now the rate of working by the particle on the fluid is
Z Z
u·F= u · σ · n dS = 2µ eij eij dV ≥ 0.
S V

For a “bar” problem we choose

with ū satisfying the Stokes equations in V̂ , ū → 0 at infinity and ū = u on S.

Then ū is kinematically admissable and ē = 0 between S and Ŝ. Therefore
Z Z
u · F̂ = 2µ ēij ēij dV ≥ 2µ eij eij dV = u · F,
V̂ V

where F̂ is the force exerted by the sphere on the fluid. The magnitude of the drag
in the direction of motion is therefore less than the drag on the circumscribing sphere.

3.1.6 Solving the Stokes equations

Taking the divergence of the Stokes equations (3.1) (with F = 0) we see that p is
harmonic. Taking the curl we see similarly that vorticity is harmonic, and finally taking
the (vector) Laplacian we see that ∇4 u = 0 — u is biharmonic.
12 CHAPTER 3. LOW REYNOLDS NUMBER FLOWS

For a planar flow u
= ∇ × (0, 0, ψ) and ω = (0, 0, −∇2 ψ). Thus ∇4 ψ = 0.

We can solve ∇2 ∇2 ψ = 0 for ∇2 ψ = f (x) (Laplace equation) and then solve
∇2 ψ = f (x) (Poisson equation).
Alternatively we could use the method of sheet 2, question 11 and write down a
solution. This is OK in nice geometries. The final method is of course numerical
solution.

3.2 Stokes flow due to a translating sphere

We consider the inertialess flow generated by a sphere of radius a and velocity u im-
mersed in unbounded fluid of viscosity µ which is at rest at infinity. In particular we
want to calculate the force F exerted by the sphere on the fluid.

3.2.1 Dimensional analysis

The linearity of the Stokes equations requires that F is proportional to both U and µ.
Dimensional considerations therefore give F = αµaU , where α is a positive dimen-
sionless constant. The isotropy of the sphere shape then implies that F = αµau.

3.2.2 Brute force

As you may have guessed from the title of this section this is an algebraically unpleas-
ant calculation. It’s probably good for your soul though...
We take spherical polars (r, θ, φ) with θ = 0 parallel to u. The flow is then ax-
isymmetric with no φ dependence and so admits a streamfunction ψ(r, θ) such that the
components of u are
1 ∂ψ 1 ∂ψ
ur = uθ = − .
r2 sin θ ∂θ r sin θ ∂r


It follows from the Stokes equations that D2 D2 ψ = 0, where

∂2
 
2 sin θ ∂ 1 ∂
D ≡ 2+ 2 .
∂r r ∂θ sin θ ∂θ

The no-slip condition on the sphere surface gives:

1 2 2 ∂ψ
ψ= U a sin θ and = U a sin2 θ on r = a.
2 ∂r
Finally, for r → ∞, ψ = o(r2 ). We look for a solution ψ = f (r) sin2 θ where
1 2
f (a) = Ua , f 0 (a) = U a, f = o(r2 ) as r → ∞.
2
2f
We then obtain D2 ψ = F (r) sin2 θ where F (r) = f 00 − r2 so that

2F
D4 ψ = 0 ⇔ F 00 − = 0.
r2
D
Integrating these equations we have f = Ar4 + Br2 + Cr + r , and the boundary
conditions give A = B = 0, C = 34 U a and D = − 14 U a3 .
3.2. STOKES FLOW DUE TO A TRANSLATING SPHERE 13

Substituting back we obtain

   
C D C D
ur = 2 + 3 cos θ and uθ = − + 3 sin θ.
r r r r
2C
The vorticity ω is necessarily in the φ direction with magnitude r2 sin θ. We can
also obtain the pressure (to within an arbitrary constant) as
2Cµ cos θ
p − p∞ = .
r2
The stress can now be determined from
 
T
σ = −pI + µ ∇u + (∇u) ,

although care must be taken in evaluating ∇u in this curvilinear co-ordinate system.

Recalling that n is the normal out of the fluid the force exerted by the sphere on the
fluid is finally given as
Z
F= σ · n dS = 6πµau, (3.3)
r=a
a result known as Stokes’ law.

3.2.3 Comments
Note the fore and aft symmetry in the streamline pattern, unlike higher Re.

u ∼ 1r as r → ∞, so far field effects are important and distant boundaries and

other particles affect the flow.
We can calculate F more easily by moving the integral to a sphere at infinity using
the divergence theorem; Z
F=− σ · n dS.
S∞
Since F is parallel to u we only need to calculate
Z
F =− σrr cos θ − σrθ sin θ dS.
r=∞

Only terms of order r12 in σ (r−1 in u or r in ψ) matter here. In the far field
ψ ∼ Cr sin2 θ and p − p∞ ∼ 2µCr2cos θ . Thus
∂ur cos θ
σrr ∼ −p + 2µ = −p∞ − 6Cµ 2
∂r r
and  
∂  uθ  1 ∂ur
σrθ ∼ ν r + = 0.
∂r r r ∂θ
This simplifies the calculation of F .
14 CHAPTER 3. LOW REYNOLDS NUMBER FLOWS

Corollary

For general shapes of particle in unbounded fluid exerting a force F on the fluid, mea-
suring θ = 0 from the direction of F, then at large distances

Fr F F
ψ∼ sin2 θ p − p∞ ∼ cos θ u∼ (2 cos θ er − sin θ eθ ) ,
8πµ 4πr2 8πµr

where er and eθ are the usual unit vectors.

This solution for u and p satisfies the Stokes equations everywhere except at r = 0
and corresponds to a point force F acting at r = 0. It is called a Stokeslet velocity field.

Failure of neglect of inertia

Near infinity,
2
 
 ∂u 
ρ  ∼ |ρu · ∇u| ∼ ρU
 2  µU
and µ∇ u ∼ .
 ∂t  r2 r3

Thus the Reynolds number (near infinity) is ρUµ a ar and even if Re = ρUµ a  1
inertial effects still matter at infinity. This is the Stokes-Oseen paradox. We need to
use the technique of matched asymptotic expansions, but at leading order the result is
unaffected.

More elegant techniques

There are more elegant (and algebraically less complicated) techniques for solving
Stokes’ equations. The Papkovich-Neuber method (as covered in Part III Slow Viscous
Flow) is probably the easiest.

3.3 Reciprocal Theorem

If (u, σ) and (u0 , σ 0 ) are two Stokes flows in V (with different boundary conditions)
then Z Z
0
ui σij nj dS = u0i σij nj dS.
S S

Proof.
Z Z
0 0

∂ 0 0


ui σij − uj σij nj dS = ui σij − uj σij dV
S V ∂x j
∂u0
Z
0 ∂ui
= σij − σij i dV
∂xj ∂xj
ZV
= eij 2µe0ij − 2µe0ij eij dV = 0.
V
3.4. MOVEMENT OF RIGID PARTICLES IN VISCOUS FLUIDS 15

3.4 Movement of rigid particles in viscous fluids

Many useful fluids are suspensions (for instance paints, inks, abrasive cleaners, settling
tanks for removal of pollutants). We want to know how fast particles sediment under
gravity.
Consider a rigid particle of arbitrary shape in an unbounded viscous fluid.

R
Let ρp (x) be the particle density, so that the mass of the particle is M = Vp
ρp dV .
Take the origin to be at the centre of buoyancy, that is such that
Z
x dV = 0.
Vp

The Archimidean upthrust −ρVp g acts at the origin (ρ is the fluid density). Let xg
be the centre of mass: Z
1
xg = ρp x dV.
M Vp

Then the total external force acting on the particle is F = (M − ρVp )g and the total
external couple is G = M xg × g.
What velocity u and angular velocity Ω are generated? It is easier in practice to
solve the inverse problem. First, though:
2
Suppose that ρUµ a  1 and ρΩa µ  1 so that the fluid inertia is negligible. Unless
ρp  ρ the particle inertia is also negligible and therefore the external force/couple
applied to the particle equals the external force/couple applied to the fluid.

We want to solve the problem u = u + Ω × x on S, u → 0 at infinity and u

satisfies the Stokes equations in V . The force and couple exerted by the body on the
fluid are
Z Z
F= σ · n dS G= x × σ · n dS.
S S

This has a unique solution. By linearity, u is linear in u and Ω, so σ is linear is u

and Ω, and also proportional to µ. Thus F and G are linear in u and Ω and proportional
to µ. Therefore

Fi = µ aAij Uj + a2 Bij Ωj

Gi = µ a2 Cij Uj + a3 Dij Ωj ,


16 CHAPTER 3. LOW REYNOLDS NUMBER FLOWS

where A, B, C and D are dimensionless second rank “resistance” tensors that

depend on the particle shape. (In fact, B and C are pseudotensors.)

Now Z Z
(1) (2)
u ·σ · n dS = u(2) · σ (1) · n dS
S S

(1) (2) (2) (1)

and so u(1) · F(2) = u(2) · F(1) . Thus = Ui Aij Uj for all U (1)
Ui Aij Uj
(2)
and U , so A is symmetric.
The rate of working by external forces is F · u = µu · A · u > 0 if u 6= 0. Thus
A has positive eigenvalues and so is invertible. It can be proved (see example sheet 2)
that the matrix
aA a2 B
 
a2 C a3 D
is symmetric and positive definite.

3.4.1 Special cases

Sphere
Take a to be the radius of the sphere. A, B, C and D are all isotropic. There exists
no force on a rotating sphere (by reflection symmetry) and so B = C = 0. Thus
Aij = αδij and Dij = βδij .
α and β are 6π and 8π respectively. Thus

2 a2
u= (ρp − ρ) g.
9 µ

is the sedimentation rate under gravity.

Cube
Take a as half the side length. The principal axes of A, B, C and D must coincide
with the axes of the cube and the eigenvalues of each of these matrices must be equal
(by symmetry). Thus A, B, C and D are isotropic. B = C = 0 still, and α and β are
unknown.
Thus a falling cube does not rotate whatever its orientation and also falls straight
down.

Ellipsoid
The best choice for a is the semi-major axis. The principal axes of A and D must
coincide with those of the ellipsoid.
A and D are known and are not isotropic. B = C = 0 still.
3.5. STOKES FLOW IN A CORNER 17

Helix
B and C are nonzero. Note that the helix is asymmetric on reflection of axes.

3.5 Stokes flow in a corner

We try a local solution ψ(r, θ) = rλ f (θ) where λ and f are to be determined. λ > 1
for u → 0 as r → 0.
We look for solutions even in θ of ∇4 ψ = 0 in 0 < θ < α, f = f 0 = 0 on θ = ±α
(no slip condition) and f 0 = f 000 = 0 at θ = 0 (even function).
Now ∇2 ψ = rλ−2 F (θ), where F = f 00 + λ2 f and ∇4 ψ = 0 implies F 00 +
2
(λ − 2) F = 0.
Thus f = B cos λθ + A cos (λ − 2) θ (restricting to the even solution and taking
λ 6= 1). Applying the conditions at θ = α we get sin 2 (λ − 1) α = (1 − λ) sin 2α.

If the angle of the wedge is too small there are no real solutions. However, there
are complex solutions. We need <λ > 1 to get u → 0 as r → 0. If λ − 1 = p + ıq and
2α = π6 then a numerical solution is 2αp = 4.2 and

2αq = 2.2.
We can evaluate uθ |θ=0 = −< λrλ−1 f (θ) = Crp cos (q log r + ) with C and
 real. We see an infinite sequence of counter-rotating Moffatt eddies.

πp
The eddies are geometrically similar and decrease in intensity by a factor e q (here
about 400) as r → 0.
18 CHAPTER 3. LOW REYNOLDS NUMBER FLOWS
Chapter 4

Flow in a thin layer

For a rectilinear flow u · ∇u = 0. What if the flow is nearly rectilinear?

4.1 Lubrication theory

Suppose fluid is confined in a gap 0 ≤ y ≤ h(x).

Dy−h
The gap is thin, so HL  1. We know that Dt
= 0 as y = h(x) is a material
surface. Thus we can either specify (u, v) or u, ∂h ∂t on y = h.
u L
We put u = ψy and v = −ψx . Then v ∼ H  1 and v is negligible compared
with u.
If we can neglect inertia then
   2 
 ∂p 
  ∼ µ ∇2 u  ∼ µ  ∂ u 
   
 ∂x   ∂y 2 
   2 
 ∂p 
  ∼ µ ∂ v .
 
 ∂y   ∂y 2 
   
 ∂p   ∂p 
There are therefore large pressure gradients in the x direction, so  ∂x    ∂y  and
∂p
at leading order p is a function of x only. Put ∂x = G(x, t). Then
∂2u G G Uy
=− ⇒ u=− y (y − h) + .
∂y 2 µ 2µ h
Taking ψ = 0 at y = 0 we obtain
y3 hy 2 U y2
 
G
ψ=− − + .
2µ 3 2 2h
The total flux in the layer is
U h Gh3
ψ|y=h = + = Q(x).
2 12µ

19
20 CHAPTER 4. FLOW IN A THIN LAYER

∂Q
Using mass conservation (directly), ∂x = − ∂h
∂t . This gives the Reynolds lubrica-
tion equation:
   
∂ ∂G ∂u ∂h ∂h
h3 = 6µ h +u +2 . (4.1)
∂x ∂x ∂x ∂x ∂t

With two boundary conditions we can find the pressure. For flow in a sheet 0 ≤
z ≤ h(x, y, t) we have u, v  w, p = p(x, y, t) and u and v are parabolic in z. The
equation is
 
3

∂h
∇ · h ∇p = 6µ h∇ · u + u · ∇h + 2 , (4.2)
∂t
 
∂ ∂
where u = (u, v) and ∇ = ∂x , ∂y .
We must now see if we were justified in neglecting inertia:

ρU 2
|ρu · ∇u| ∼
L
 2  µU
µ∇ u ∼ .
H2
UH H
The ratio of inertial forces to viscous forces is thus ν L and we need this effective
Reynolds number to be small.

4.2 Thrust bearing

Assume that the flow is axisymmetric. We can immediately apply (4.2) to get

1 ∂ ∂p 12µ ∂h
∇2 p = r = 3 .
r ∂r ∂r h ∂t

Thus
3µr2 ∂h
p= + C log r + D.
h3 ∂t
We know that C = 0 (to avoid a singularity at the origin), and we put p = p∞ at
r = a. We obtain
3µ ∂h 2
r − a2 .


p − p∞ = 3
h ∂t
To obtain the force on the hammer we need σzz = −p + 2µ ∂w ∂z . The
∂w
∂z term is
H2
O( L2 ) and is smaller than p so we neglect this. Thus, allowing for p∞ ,
4.3. FLOW IN A HELE-SHAW CELL 21

Z a
F = (p − p∞ ) 2πr dr
0
3µ ∂h a
Z
r r2 − a2 dr


= 2π 3
h ∂t 0
3 πµ ∂h 4
=− a .
2 h3 ∂t
For fixed V0 = πa2 h, and if the force applied to the hammer is fixed, then
 14
3µV02 ∂ 1 3µV02

− 41
F = ⇒ h(t) = (t + t0 )
8π ∂t h4 8πF
1 1
Thus a(t) ∝ (t + t0 ) 8 and as t → ∞, h ∼ t− 4 when a is less than the radius of
9
the hammer. h00 ∼ t− 4 and so the inertia of the hammer is negligible.
Lubrication forces are in general big — hence Sellotape.

4.3 Flow in a Hele-Shaw cell

Suppose we have a flow confined between two accurately parallel sheets at z = 0 and
z = h. Then the Reynolds equation (4.2) gives ∇2 p = 0 (∇ is two dimensional) and
1
(u, v) = z (z − h) ∇p
2µ
and the depth averaged velocity ū is
h2 p
 
ū = ∇ − .
12µ
Thus the velocity field is the gradient of a harmonic potential — ū is an inviscid,
irrotational velocity field.
The no-slip condition is accomodated in a small region of size h near the cylinder
surface. We can’t get a circulation as p (and therefore φ) is single-valued.

4.4 Saffman-Taylor instability for a planar interface

We attempt to analyse the fingering instability observed when a less viscous fluid ad-
vances under a pressure gradient into a more viscous fluid. For simplicity suppose that
one of the fluids is inviscid and consider a planar interface in a Hele-Shaw cell. The
velocity u(x, y) is therefore

h2 p
u = ∇φ and φ = − ,
12µ
where p is the fluid pressure.
Suppose first that the interface at x = V t is planar. Then the basic state is
Water Air
φ0 = V x + const
p0 = − 12µV h(x−V
2
t)
+ p∞ p = p∞ .
22 CHAPTER 4. FLOW IN A THIN LAYER

Now imagine that the interface suffers an infinitesimal perturbation with wavenum-
ber k and amplitude . We anticipate that this perturbation will grow or decay exponen-
tially in time and check this assumption later. The position of the interface becomes

x = V t + eıky+σt ,

where  is arbitrarily small. At leading order in  all the perturbation quantities inherit
this dependence on y and t, so that the velocity potential in the water becomes

φ = φ0 + f (x)eıky+σt .

As φ is harmonic we can find f (x) (picking the solution decaying as x → −∞

φ = V x + const + Aeıky+σt+|k|(x−V t) ,

for some constant A. The corresponding pressure is

12µ ıky+σt+|k|(x−V t)
p = p0 + p 1 where p1 = − Ae .
h2
We therefore have
Water Air
φ = φ0 + φ1
p = p 0 + p1 p = p∞ .
Our aim is to find σ. If a surface tension γ acts between the fluids then the interface
2
curvature is just ∂∂yx2 at leading order, so that

∂2x
[p] = γ = −k 2 γeıky+σt ,
∂y 2
where the jump is across the position of the perturbed interface. At leading order
in  this gives
 
12µV ıky+σt 12µ ıky+σt
p∞ − − 2 e + p∞ − 2 Ae = −k 2 γeıky+σt .
h h
The first term is the pressure in the air and the second is the pressure in the water.
Simplifying we find
γh2 k 2
A = −V − .
12µ
We can now determine the velocity of the interface as ∂φ ∂x , which may again be
evaluated (at leading order) at x = V t. This must correspond to ∂x ∂t = V + σe
ıky+σt
.
This gives
γh2 k 2
 
σ = A |k| = −V |k| 1 + .
12µV
The time dependence cancels out, which justifies our initial assumption. In the
absence of surface tension we see that σ > 0 whenever V < 0, thus the interface
perturbation grows (according to linear theory) if the air moves into the water. On the
other hand σ < 0 when V > 0 and the water moves into the air. In the unstable case,
the fastest growing modes are those short waves for which k → ∞.
If γ > 0 surface tension is predicted to stabilise the shortest waves. The interface
is still unstable, but the fastest growing mode has a finite value for k and σ.
4.5. GRAVITATIONAL SPREADING 23

Physical mechanism

Where the interface lags the pressure gradient is bigger, and the disturbance reduces.
The other way is unstable.

4.5 Gravitational spreading on a horizontal surface

Assume axisymmetry and ∂h ∂r  1. We will neglect surface tension (and the contact
line).
As this is a free surface problem we can’t use the Reynolds equation directly.
In the layer there is a hydrostatic pressure p = p∞ + ρg (h − z). The radial mo-
mentum equation is
∂2u ∂p ∂h
µ 2 = = ρg .
∂z ∂r ∂r
∂u
As u = 0 at z = 0 and ∂z = 0 at z = h we have

ρg ∂h
u= z (z − 2h) .
2µ ∂r
The flux out of a cylinder of radius r is
Z h
2πρgrh3 ∂h
Q(r) = 2πr u dz = − .
0 3µ ∂r
Mass conservation gives

∂h ∂Q
2πr =− and so
∂t ∂r  
∂h g 1 ∂ 3 ∂h
= rh .
∂t 3ν r ∂r ∂r
The volume of the drop is fixed, so
Z a(t)
2πrh dr = V.
0
1
We now rescale the variables. The natural lengthscale is V 3 so rescale h, r and a
1 1
by V 3 and the natural timescale is ν 1 , so we rescale t by this. Letting r∗ = rV − 3
gV 3
24 CHAPTER 4. FLOW IN A THIN LAYER

and so on, we get

∂h∗ ∂h∗
 
1 1 ∂
= r∗ h∗3 ∗
∂t∗ 3 r∗ ∂r∗ ∂r
and Z a∗ (t∗ )
2πh∗ r∗ dr∗ = 1.
0

We now drop the ∗’s. We need h(r, 0) to get a unique solution.

We look for a similarity solution

1. To aid computer programs,

2. For physical insight,
3. Often the long-time solution, independent of initial conditions.

Try h = t−α H(η) where η = r

tβ
and so a = Atβ . We get
Z A
1= 2πηH(η) dηt2β−α ,
0

and so α = 2β. Substituting in the differential equation

1 −4α−2β d
(−αH + βηH 0 ) t−α−1 = ηH 3 H 0 .


t
3η dη

So −α − 1 = −4α − 2β. We can solve for α and β to obtain α = 41 , β = 18 . The

ODE for H is
d
3
ηH 3 H 0 + η (ηH 0 + 2H) = 0.
dη 8
This can be integrated to give
3
ηH 3 H 0 + η 2 H = 0,
8
We can integrate this equation to give
  13
9
13
H= A2 − η 2 .
16
The constant volume condition gives
  31
3π 9 8
A 3 = 1.
4 16
Putting all the dimensions back in, we get
 18 1
! 18
210

1 gV 3 t
a(t) = V 3 .
35 π 3 ν

If surface tension is included then we must consider the curvature. We get

∂2h
p = p∞ + ρg (h − z) − γ .
∂r2
4.5. GRAVITATIONAL SPREADING 25

The governing equation is

γ ∂3h
  
∂h g 1 ∂ ∂h
= rh3 − .
∂t 3ν r ∂r ∂r ρg ∂r3

This has no similarity solutions. We need extra boundary conditions on h(r) as it

is now a fourth order equation. This is an open problem.
26 CHAPTER 4. FLOW IN A THIN LAYER
Chapter 5

Vorticity generation and

confinement

5.1 Vorticity equation

Our point of departure is the Navier-Stokes equations with no force term
 
∂u 
2
 Du
ρ 1
+ ∇ 2 |u| − u × ω = ρ = −∇p + µ∇2 u.
∂t Dt
We take the curl of this to get
∂ω
ρ = ρ∇ × (u × ω) + µ∇2 ω.
∂t
Re-arranged, this gives the vorticity equation
Dω
= ω · ∇u + ν∇2 ω. (5.1)
Dt
Physical meaning: vorticity moves with fluid particles, is diffused by viscosity and
stretched by ∇u.
There are no source terms, so ω originates on boundaries.
For an inviscid fluid (ν = 0) we have Dω Dt = ω · ∇u. Recall that for a material
line element Ddl
Dt = dl · ∇u. Therefore vortex lines stretch and rotate like material line
elements.
We can give an integral form for (5.1) in the case ν = 0. Define the circulation
around C(t) as I
κ= u · dl.
C(t)
Then
I
dκ Du Ddl
= · dl + u ·
dt C(t) Dt Dt
I  
p 1 2
= dl · ∇ − + 2 |u| = 0 as p and u are single valued.
C(t) ρ
Thus (in the inviscid case), κ is constant. This is Kelvin’s circulation theorem.
If κ = 0 at t = 0 for all contours C then κ = 0 for all time. In this case u = ∇φ
and we have an irrotational inviscid problem.

27
28 CHAPTER 5. VORTICITY GENERATION AND CONFINEMENT

5.1.1 Planar flows

In the ν 6= 0 case, if we have a planar flow we can write u = (ψy , −ψx , 0) and then
ω = (0, 0, −∇2 ψ).
Thus ω · ∇u = 0 and the vorticity equation becomes

∂ 2 ∂(ψ, ∇2 ψ)
∇ ψ− = ν∇4 ψ. (5.2)
∂t ∂(x, y)
This is the two dimensional vorticity equation.

5.2 Vorticity generation

Suppose that we have the flow u = (u, v) with

(
U t>0
u|y=0 =
0 t < 0.

What is u(y, t) for t > 0? We try a solution u = (u(y, t), 0) and p = p(y, t).
∂p
The y momentum equation gives ∂y = 0 and the x momentum equation gives

∂u ∂2u
= µ 2.
∂t ∂y

This is a diffusion equation for u (or ω, since ω = − ∂u

∂y ). We have the boundary
conditions u → 0 as y → ∞ and u = 0 for all y at t = 0. Now u ∝ U by linearity, so
u = U f (y, t, ν) for dimensionless f .
As f is dimensionless it depends only on 2√ytν = η — we have a similarity solution.
The chain rule gives
f 00 + 2ηf 0 = 0,
which has the solution Z η
2
f =A e−ξ dξ + B.
0

Now f (0) = 1 and f → 0 as η → ∞, so A = √2 and B = 1. Thus

π
 
y
u(y, t) = U 1 − erf √ .
2 νt
y 2
The vorticity ω = √U e− 4νt . As t → 0, ω(y) → U δ(y), and as t increases
2 πνt
1
ω spreads into a boundary layer of thickness δ ∝ (νt) 2 . This is characteristic of a
diffusion process.
5.3. VORTICITY CONFINEMENT ON A FLAT PLATE 29

R∞
Note that 0
ω dy = U ∀t.

5.3 Vorticity confinement on a flat plate

Consider a steady flow past a flat plate with suction.

u = (0, −V ) at y = 0, u → (U, −V ) as y → ∞. We try a solution u =

(u(y), v(y)). Incompressibility gives v = −V ∀y. The x component of the momentum
equation is
∂u ∂2u
−V = ν 2.
∂y ∂y
Vy
−
Thus the general solution is u = A +  Be ν . If V > 0 Vwe can apply the boundary
− Vνy y
conditions to get u(y) = U 1 − e and ω = − UνV e− ν .
The vorticity is confined in a boundary layer near the wall of thickness Vν . Outside
the boundary layer the flow is irrotational.
If V < 0 we cannot apply the boundary conditions consistently.

5.4 Stagnation point flow

The irrotational flow u = α(x, −y) with streamfunction ψ = αxy has a stagnation
point at the origin. What does the flow become if a rigid wall is placed at y = 0? (We
look at the α > 0 case first.)
We look for a streamfunction ψ such that ψ ∼ αxy as y → ∞, and propose the
solution ψ = αxf (y) with f (y) ∼ y as y → p∞.
We nondimensionalise x and y by δ = αν , and put
x
ψ = αδ 2 f ( yδ ).
δ
Letting η = yδ , for a steady flow we have u · ∇ω = ν∇2 ω, ω = −∇2 ψ we get

f 0 f 00 − f f 000 = f (iv)

with boundary conditions f 0 → 1 as η → ∞, f = f 0 = 0 at η = 0.

We can integrate this equation once to get

f 02 − f f 00 = f 000 + 1.
30 CHAPTER 5. VORTICITY GENERATION AND CONFINEMENT

This equation must be solved numerically; it appears there is a unique solution.

Far from the plate ψ ∼ αx (y − 0.65) — so the flow is irrotational, with a per-
turbation as if the stagnation streamline was at y = 0.65δ (this is the displacement
thickness).
The vorticity is confined to a layer of thickness proportional to δ. q
ν
If α < 0 the above analysis carries through, provided we let δ = |α| and ψ =
|α| δxf ( yδ ). The ODE is unchanged but the boundary condition at infinity becomes
f 0 → −1 as η → ∞. No solution to this equation exists.

5.5 The bathtub vortex

1 ∂
The axisymmetric flow u = v(r, t)eθ has vorticity ω = r ∂r (rv) ez . The vorticity
diffuses according to the equation

∂ω
= ν∇2 ω,
∂t
and the local vorticity intensity falls. To maintain the vorticity distribution add in a flow
u = (−αr, 0, 2αz). This will advect and stretch ω (for α > 0). What is the steady
vorticity?

u · ∇ω − ω · ∇u = ν∇2 ω
∂ω ν ∂ ∂ω
−αr − 2αω = r
∂r r ∂t ∂r
∂ω
⇒ −αr2 ω = νr + C.
∂r
If ω → 0 as r → ∞ then C = 0 and
αr 2 νω0  αr 2

ω = ω0 e− 2ν ⇒ v= 1 − e− 2ν .
αr
pν
The vorticity is confined to r . α.
dp ρν 2 2
Now dr = r − 2α r and p has a minimum near r = 0 — so if we have a free
surface a dip will appear.
Chapter 6

Boundary layer theory at high

Reynolds number

6.1 Introduction
Suppose we have a steady flow past a circular cylinder.

UL
Suppose also that the Reynolds number, Re = ν  1. Near the front stagna-
q
tion point we expect ψ ∝ xy U
L which implies a boundary layer of thickness νL
U =
√L  1 where the viscosity is important. Outside the layer the flow remains irrota-
Re
1
tional, although it is slightly modified by a displacement effect of size Re− 2 , which we
ignore. Note that x = Lθ and y = (r − L) are co-ordinates parallel and perpendicular
to the surface.
We expect the boundary layer to continue around the cylinder, and would like to
know what happens in this layer and at the rear stagnation point.

6.2 Steady boundary layer theory at rigid surface

Recall the steady planar vorticity equation (5.2),

∂(ψ, ∇2 ψ)
− = ν∇4 ψ.
∂(x, y)

In the Euler limit we suppose that x and y have the same scale L and let Re → ∞.
Let ψ = U Lψ̃, x = Lx̃ and y = Lỹ. Then the steady vorticity equation becomes

˜ 2 ψ̃)
∂(ψ̃, ∇ 1 ˜4
− = ∇ ψ̃
∂(x̃, ỹ) Re

31
32 CHAPTER 6. BOUNDARY LAYERS

and as Re → ∞ viscosity disappears and we recover the Euler equation ρu · ∇u =

−∇p.
In the Euler limit we lose the no-slip boundary condition; setting µ = 0 in the
Navier-Stokes equations reduces the order and in general fewer boundary conditions
can be satisfied.
We expect that in the boundary layer viscosity is always important. In the Prandtl
limit we suppose x = Lx̄ but y = δ ȳ where δ = √LRe and ψ = U δ ψ̄. Thus u is scaled
by U but v is scaled by √URe .
For any fixed x, limRe→∞ u(x) = u|inviscid (x). This convergence is not uniform;
for any fixed Re, however big, we can always find x at which u(x) 6= u|inviscid (x).
Now ω̄ = −∇ ¯ 2 ψ̄ = − ∂ 2 ψ̄2 which implies that
∂ ȳ
2
∂(ψ̄, ∂∂ ȳψ̄2 ) ∂ 4 ψ̄
− =−
∂(x̄, ȳ) ∂ ȳ 4
to get a balance between inertia and viscosity in the boundary layer. We can inte-
grate this once with respect to ȳ to get

ψ̄ȳ ψ̄x̄ȳ − ψ̄x̄ ψ̄ȳȳ = ψ̄ȳȳȳ + G(x̄)

which may be alternatively written

ūūx̄ + v̄ūȳ = ūȳȳ + G(x̄).

G is a pressure gradient which does not depend on ȳ in the layer.

On the wall at ȳ = 0 we have ψ̄ = ψ̄ȳ = 0 which gives the no-slip condition.

Matching
As ȳ → ∞ (many distances δ, but for large Re still near the boundary) the inner
solution must match the outer solution. Now ν = 0 already matches. We also need

lim ψ̄ȳ = lim ψ̃ỹ .

ȳ→∞ ỹ→0

The pressure must also match. If limȳ→∞ ψ̄ȳ = U (x̄) then by Bernoulli’s equation,
2
p + 12 ρ |u| = const and so G(x̄) = U ∂U
∂ x̄ .
Putting this back into dimensional form we obtain the boundary layer equation:

uux + vuy = U U 0 + νuyy (6.1)

with boundary conditions u = v = 0 at y = 0 and u → U as y → ∞.

Notes
1. This is a parabolic equation. We need upstream conditions on u, for instance the
value at x = 0.
∂2
2. The boundary layer equation is the x momentum equation with ∇2 = ∂y 2 and
∂p
∂x a function of x only.
3. It is a nonlinear equation. Very few analytic solutions are known, and those that
are known are similarity solutions.
6.3. BLASIUS BOUNDARY LAYER 33

6.3 Flow past a flat plate: Blasius boundary layer

The Euler problem is u = (U, 0), and the boundary layer equation becomes

ψy ψxy − ψx ψyy = νψyyy

with ψ = ψy = 0 at y = 0 and ψy → U as y → ∞. √
This is like a spatial version of vorticity diffusion. Vorticity spreads a distance νt
where t = Ux is the time taken for fluid to reach x starting from x = 0.

12 y
We try δ(x) = νx U and a similarity solution η = δ(x) , ψ = U δ(x)f (η). Substi-
tuting into the boundary layer equation we get

f 000 + 21 f f 00 = 0,

with f = f 0 = 0 at η = 0 and f 0 → 1 as η → ∞. This must be solved numerically, to

give a flow profile:


∂u  µU 00
The traction on the plate y = 0+ is µ ∂y  = δ f (0), numerical solution
y=0
gives 0.3ρU 2 Uνx .
p

Notes
1. The displacement thickness is the lateral displacement of streamlines outside the
boundary layer:
Z ∞ r
u νx
δ1 = 1− dy = 1.7 .
0 U U

2. We could get an improved result by modifying the outer Euler flow to account for
1
the displacement effect (flow past a parabola). This is an O(Re− 2 ) correction.

3. What is the Reynolds number? The q only available lengthscale is x, so the effec-
U δ1 Ux
tive Reynolds number is ν ∝ ν . There is therefore a small nose region
ν
near x = 0 of size U where the theory breaks down.

4. It is found experimentally that if the Reynolds number is big enough (far enough
downstream) the flow becomes unstable. (At Re ≈ 1000.) Disturbance to the
34 CHAPTER 6. BOUNDARY LAYERS

boundary layer grow, flow becomes unsteady and ultimately turbulent. The drag
on the plate increases. In practice, this Blasius boundary layer agrees with ex-
periment for 1  Re . 1000.

6.4 Similarity solutions of the boundary layer equation

For geometries with no intrinsic lengthscale it is sensible to try a solution ψ(x, y) =
y
U (x)δ(x)f (η), where η = δ(x) , U ∝ xp and δ ∝ xq .
Then u = ψy = U f 0 (η) and so f 0 → 1 as η → ∞. To get a balance between
inertia and viscosity in the boundary layer we must have |uux | ∼ ν |uyy | (in practice
|uux | ∼ |vuy |). Thus
νU
U Ux ∼
δ2

and p + 2q = 1. To fix p and q we need extra information.

In the Blasius layer we had u → U independently of x and so p = 0. In the
stagnation point flow we had u → αx and so p = 1.
Dimensional arguments give the rest.

6.5 High Reynolds number flow past a wedge

6.5.1 Outer problem

Symmetric flow

In the outer inviscid, irrotational region we have u = ∇φ and ∇2 φ = 0 with boundary

conditions ∂φ πβ
∂θ = 0 on θ = 0 and θ = 2π − 2 . The last condition is a symmetry
requirement.
2
The trial solutions φ = Crλ cos λθ will work if λ = 2−β . The outer flow velocity
for the anticipated boundary layer on θ = 0 then has magnitude U (x) = Axm , where
β
m = λ − 1 = 2−β ≥ 0.
The case β = 0 (m = 0) gives the Blasius boundary layer and the case β = 1
(m = 1) gives stagnation point flow.
6.5. HIGH REYNOLDS NUMBER FLOW PAST A WEDGE 35

Antisymmetric flow

This is the same problem as above, but with the symmetry condition replaced by ∂φ
∂r =
0 on θ = π2 + πβ
2 .
1
The same trial solution works, but now with λ = 1+β and so U (x) = Axm , with
β
m = − 1+β ≤ 0.

6.5.2 The boundary layer

The boundary layer equation along x > 0 (leaving the apex of the wedge) becomes
ψy ψxy − ψx ψyy = mA2 x2m−1 + νψyyy .
The only lengthscale for growth of the boundary layer thickness is then provided
by x, and since U (x) = Axm we try a similarity solution with p = m and q = 1−m 2 .
Dimensional considerations dictate a structure of the form
√
r
y νx1−m
ψ = νAxm+1 f (η) η= with δ(x) = .
δ(x) A
√
If m < 0 the boundary layer thickness increases faster than the x behaviour that
would arise from diffusion alone.
Substituting in the boundary layer equation we obtain the Falkner-Skan equation
m + 1 00
f 000 + f f + m 1 − f 02 = 0,


(6.2)
2
with boundary conditions f = f 0 = 0 on η = 0 (no slip) and f 0 → 1 as η → ∞
(to match the outer Euler flow).

6.5.3 Numerical solution

The ordinary differential equation may be solved numerically using a shooting tech-
nique. We find that for m > 0 there is a unique for f qualitatively similar to the
Blasius profile having f 0 > 0 for all η. Thus symmetric flows that accelerate away
from the apex of the wedge pose no difficulties for the boundary layer equation.
For antisymmetric flows with m < 0 the position is more complicated.

For m < −0.904 there is again a unique solution for f , but now f 00 (0) < 0. There
36 CHAPTER 6. BOUNDARY LAYERS

is flow reversal near the wall and the flow must separate at the apex of the wedge on the
downstream side. This is unacceptable — “upstream infinity” in the parabolic bound-
ary layer equation for u is now at x = ∞. Furthermore, for large η, f 0 approaches its
asymptotic value of unity from above, so the presence of the boundary layer apparently
speeds up the outer flow over its inviscid value — this is unphysical.
For −0.904 < m < 0 there are two solutions for f , one having reversed flow and
the other not. The solution without reversed flow is acceptable.

6.5.4 Separation of the boundary layer

High Reynolds number steady boundary layers on rigid surfaces are commonly found
in experiments, but are not normally observed to contain regions of reversed flow.1
The wedge example above suggests that non-reversed boundary layers will arise on the
rigid boundary x > 0 provided the “imposed” pressure gradient U U 0 is positive — that
is if the external stream accelerates.
If U 0 > 0 then ∂V∂y < 0 by mass conservation and since V = 0 at the boundary, V <
0 in the interior of the fluid. In this case convection tends to confine the vorticity near
the boundary. If U 0 < 0 vorticity confinement to a thin boundary may be impossible.
If U U 0 is sufficiently (in fact only slightly) negative, called an adverse pressure gra-
dient, then the boundary layer thickness grows more rapidly and flow reversal occurs.
This phenomenon is called boundary layer separation. Separation brings into question
our entire method of solution, in particular the imposition of upstream boundary data
on u at or near x = 0.
Sometimes, worse still, it implies that the outer irrotational Euler solution is it-
self incorrect because gross separation occurs. The classic example here is flow past
a circular cylinder, for which we noted that the outse inviscid irrotational flow has
U U 0 = 2 sin 2x, where x measures distance from the front stagnation point. This pres-
sure gradient becomes adverse at x = 90◦ , and the boundary layer equation shows a
singularity at x = 104.5◦ . Experimentally, the boundary layer is observed to separate
and to introduce vorticity into the wake of the cylinder, changing the leading order
outer flow as sketched below (this happens at Re ≈ 20). At higher flow rates still the
flow becomes unsteady.2

This gross separation is characteristic of high Reynolds number flow past any bluff
body and the only way to prevent the separation is to reduce the adverse pressure gra-
dient by streamlining the body into an aerofoil shape.
Without separation the magnitude
√ of the “skin friction” boundary layer drag on
the body scales as Lµ ∂u∂y = µU Re and the contribution from the pressure ρU 2 in
the outer inviscid flow is zero. With separation the modified √ external pressure gives a
“form drag” of magnitude ρU 2 L, which is a (large) factor of Re bigger than the skin
friction.
1 For an exception, see Van Dyke page 26.
2 See Van Dyke pages 28 – 31.
6.6. CONVERGING AND DIVERGING FLOW IN A WEDGE 37

From a mathematical perspective, note that the limit Re → ∞ is in general singular.

The steady flow field for Re → ∞, if it exists, may be completely different from that
for an inviscid fluid with Re = ∞.

6.6 Converging and diverging flow in a wedge

We will do the source problem first; consider a source with strength Q. The outer
Q Q
problem has a solution ur = πβr and so U (x) = A
x , A = πβ > 0.
We seek a similarity solution with p = −1 and q = 1. Thus
√
r
y ν
ψ = νAf (η) η= with δ = .
δ A
Substituting into the boundary layer equations we get

f 000 + f 02 − 1 = 0, Falkner-Skan with m = −1.

The boundary conditions are f = f 0 = 0 at η = 0 and f 0 → 1 as η → ∞. We can

integrate this once to get

1 002 2
2f + 13 f 03 − f 0 = const = − using ∞.
3
At η = 0, f 0 = 0 and so f 002 = − 43 — giving a contradiction. There is no steady
2
boundary layer. Thus the pressure gradient U U 0 = − A x3 is too adverse and vorticity
must diffuse into the interior.
In fact there is an exact solution of the full Navier-Stokes equations (Jeffrey-Hamel
flow). We get rapid oscillations and so viscosity matters everywhere. In practice this is
very unstable.

We can do the sink problem by sending Q 7→ −Q in the above. We obtain the same
differential equation, but the boundary condition at infinity is f 0 → −1 as η → ∞. We
integrate the differential equation once to get

1 002 2
2f + 31 f 03 − f 0 = ,
3
  q
which implies f 0 = 2 − 3 tanh2 √η2 + C , where tanh C = ± 23 , one of which has
reversed flow and is no good. The other is OK.
38 CHAPTER 6. BOUNDARY LAYERS
Chapter 7

Aerodynamics

We are interested in flows that do not separate.

We hope (require!) that boundary layers do not separate, are passive and are dic-
tated by the external inviscid flow, u = (ψy , −ψx , 0) = ∇φ.

7.1 Complex potential

For an inviscid irrotational flow we have both a streamfunction ψ and a velocity poten-
tial φ, such that
(ψy , −ψx , 0) = u = (φx , φy , 0).
Thus ψy = φx and −ψx = φy and if we set the complex variable z = x + ıy, the
function w(z) = φ + ıψ is analytic except at singularities.1
We can find the velocity from the complex potential w as dw
dz = φx + ıψx = u − ıv.

Examples
1. w = U z — uniform stream.

2. w = 21 αz 2 , giving ψ = αxy — stagnation point flow.

3. w = Az λ , giving φ = Arλ cos λθ, — flow past a wedge.

ıκ
4. w = − 2π log z for κ ∈ R — line vortex.

1
5. w = U z 2 + a2 2 . This is multivalued and has branch points at ±ıa. Put
the branch
√ cut along [−ıa, ıa] and choose the square root such that if x > 0
then x2 + a2 > 0. As |z| → ∞, w(z) ∼ U z. If z = ıy and |y| < a then
p 1 1
w = U a2 − y 2 ∈ R and ψ = 0. Near z = ıa, w ∼ (2ıa) 2 (z − ıa) 2 , like
1 You know what I mean — meromorphic, or something like that...

39
40 CHAPTER 7. AERODYNAMICS

flow past a wedge. This is flow past a plate. We find u − ıv = (z2U+a z

2) → ∞

as z → ±ıa and infinite velocities are predicted at the tip of the plate. This flow
could be impulsively generated, but viscosity will act to generate vorticity on the
plate and cause the flow to separate.

Theorem (Milne-Thomson circle theorem). If f (z) is a complex potential with no

singularities in |z| < a and a cylinder |z| = a is introduced into the flow then the new
potential is
2
w(z) = f (z) + f¯( az̄ ).

2
Proof. If f is analytic then so is f¯( az̄ ) (use the Cauchy-Riemann equations). On the
cylinder z = aeıθ and
w = f (aeıθ ) + f¯(aeıθ ),

which is real. Thus ψ = 0 on the surface of the cylinder and so the surface of the
cylinder is a streamline.
2
Outside the cylinder f¯( a ) introduces no new singularities.
z̄

We can use this result to get flow past a cylinder without circulation. If u =
(−U, V ), then the complex potential for a uniform stream is f (z) = −(U + ıV )z.
Thus on inserting a cylinder, we get

a2
w(z) = − (U + ıV ) z − (U − ıV ) . (7.1)
z

It is easy to bolt a circulation on to this to get

a2 ıκ
w(z) = − (U + ıV ) z − (U − ıV ) − log z. (7.2)
z 2π

7.2 Conformal mappings

If w(ζ) is analytic in ζ and ζ = f (z) with f analytic then W (z) = w(f (z)) is an
analytic function of z.
By judicious choice of w, f can generate lots of flows. At points z0 where f is
analytic and f 0 (z0 ) 6= 0, f is a conformal mapping and a closed curve C in the z plane
that doesn’t pass through a singular point of f will become a closed curve C 0 in the ζ
plane.
For flow past an aerofoil C in the z plane we choose f to make C 0 a circle. If in
addition f (z) ∼ z as |z| → ∞ then the flow at ∞ is the same in both planes.
Note that if we have w(ζ) ∼ m−ıκ 2π log (ζ − ζ0 ) as ζ → ζ0 then since ζ − ζ0 =
f (z) − f (z0 ) ∼ (z − z0 )f 0 if f 0 6= 0, ∞ we have W (z) ∼ m−ıκ
2π log (z − z0 ) + const
— sources and line vortices are the same in both planes.
7.3. FORCES, DRAG AND LIFT 41

7.2.1 Flow past an ellipse with circulation

2
Consider the inverse map z = ζ + λζ (a Joukowski map). A point on ζ = c is ceıφ ,
which is mapped to
λ2 λ2
   
x= c+ cos φ y = c− sin φ,
c c
2 2
a −b
which will be the ellipse provided c = a+b 2
2 and λ = 4 .
√
z± z 2 −4λ2
Solving for ζ we find ζ = 2 . We want ζ ∼ z as z → ∞, and so we
choose the + sign. Thus the complex potential for flow past an ellipse with circulation
is (using (7.2))

c2 ıκ
w(z) = − (U + ıV ) ζ − (U − ıV ) − log ζ
ζ 2π
√ (7.3)
z + z 2 − 4λ2
where ζ = .
2

Flow past a flat plate

The special case of a flat plate has b = 0, so λ = c = a2 .

κ = 0, A and B are stagnation points. L and T are the leading and trailing edges
respectively.
As κ increases, A and B move to the left until, at κ = κc , A coincides with T .

7.3 Forces, drag and lift

To avoid a crisis of notation, we let q = |u| (vector norm).
Starting from the Euler equation
 
∂u 1 2
ρ + 2 ∇q − u × ω = −∇p
∂t
42 CHAPTER 7. AERODYNAMICS

we can, if the flow is steady and irrotational, derive (a form of) Bernoulli’s equation,

p = p∞ − 12 ρq 2 . (7.4)
Consider a body in the fluid.

Now dl = (dx, dy) and ndl = (−dy, dx). The force exerted by the body on the
fluid is
I I I
1 2
−pn dl = 2 ρ |u| n dl − p∞ n dl.
C c C

The last term vanishes by the divergence theorem and we see that
Z
1 2
Fx − ıFy = − 2 ρ |u| (dy + ıdx) .
C

On C, the flow is tangential and so dz = dl (u+ıv)

q and dy + ıdx = ıdl (u−ıv)
q .
Thus

u − ıv
I
Fx − ıFy = − 12 ıρ q2 dz
u + ıv
IC
2
= − 12 ıρ (v − ıv) dz
C
I  2
dw
= − 21 ıρ dz,
C dz

and we have derived Blasius’ formula:

I  2
dw
Fx − ıFy = − 12 ıρ dz. (7.5)
C dz
0
Note that by Cauchy’s theorem, if dw
dz is analytic between C and C we may deform
0
the contour C onto C . In particular, if there are no singularities in the fluid, we can
deform C to C∞ and then use the calculus of residues to evaluate the integral.

Example
ıκ
If w(z) ∼ −U z − 2π log z as z → ∞ then
I 
ıκ 2
Fx − ıFy = − 12 ıρ U+ dz.
C 2πz
The residue is ıκU
π and so Fx − ıFy = ıρU κ. If U is real (WLOG) we see that the
drag Fx = 0 and the lift on the body, −Fy = ρU κ.
7.4. THE KUTTA-JOUKOWSKI CONDITION 43

7.4 The Kutta-Joukowski condition

For an aerofoil with a sharp trailing edge, viscosity will cause the decel-
erating boundary layer to separate at the edge, modifying the external
Euler flow. The outer flow will adjust its circulation so as to streamline
the flow at the edge.

This was proved in 1970 for steady flows.

We wish to find this critical value of the circulation (κc ) for a flat plate (of length
2a).
Recall that
c2
  
dW dW dζ ıκ 1 z
= = − (U + ıV ) + (U − ıV ) 2 − + √ .
dz dζ dz ζ 2πζ 2 2 z 2 − a2

There is a singularity at z = −a and so the first bracket must vanish at z = −a to

make the velocity
√ √ the resulting equation to get κc = 2πaV and so
finite. We can solve
the lift is ρ U 2 + V 2 κc = 2πρa U 2 + V 2 V

2
Equivalently, the lift is 2πρa |U | sin α. This result suggests that for a wing of area
A, the total lift is proportional to ρU 2 A sin α.
We have ignored separation at the leading edge. This can be delayed by rounding
it.

Even in this case, if α is big enough & 10◦ , flow will separate at the leading edge
(stall) with a catastrophic decrease in lift and increase in drag.

7.5 Physical mechanism

How is a circulation established from rest? At t = 0, the picture looks like:

At t = 0+ , the flow between T and A decelerates rapidly and there is a severe

adverse pressure gradient on T A. The boundary layer therefore separates to give a
small region of reversed flow in the boundary layer. This gives a small eddy with
circulation −κc .
44 CHAPTER 7. AERODYNAMICS

The eddy is then convected away from the plate (“starting vortex left at the air-
port”). As total circulation at infinity is conserved there must be a circulation κc around
the wing.
Chapter 8

Kelvin-Helmholtz instability

At high Reynolds number many flow profiles are unstable. We will consider the easiest
case, steady inviscid flow with a discontinuity in velocity.

We have u = (− 12 U sgn y, 0, 0) and hence ω = (0, 0, U δ(y)), a vortex sheet. Sup-

pose the vortex
 sheet is perturbed to y = η(x, t) = f (t)eıkx and that the disturbance is
 ∂η 
small:  ∂x  = |kf |  1.
Now vorticity moves with the fluid, so y = η is a material surface and we have
D
the kinematic boundary condition Dt (y − η)y=η = 0. For y ≷ η the flow remains
irrotational, so
(
− 1 U ex + ∇φ> y > η
u= 1 2
2 U ex + ∇φ< y < η,

with ∇2 φ≷ = 0 and φ≷ → 0 as y → ±∞. φ≷ must inherit the eıkx dependence

on x as the perturbation is linear; so

φ≷ = g≷ (t)eıky∓|k|y .
D

We now apply the kinematic boundary condition Dt (y − η)y=η = 0 to get

∂φ  ∂η ∂η ∂η
− −u −v =0 at y = η.
∂y y=η ∂t ∂x ∂y
We use Taylor’s theorem to evaluate this from information at y = 0 and neglect
quadratic terms to get the linearised boundary condition

∂f
∓ 12 U ıkf = ∓ |k| g≷ .
∂t
We still need the pressure to be continuous at y = η. To do this we derive the
unsteady form of Bernoulli.

45
46 CHAPTER 8. KELVIN-HELMHOLTZ INSTABILITY

For an inviscid irrotational flow we have

∂u 
2

= − ρ1 ∇ p + 21 ρ |u|
∂t
and so

∂φ 1 2
p = ρF (t) − ρ − 2 ρ |u| . (8.1)
∂t
Therefore
 

2 ∂φ>  2
p> = p∞ + 12 ρ 12 U − ρ + 21 − 12 U ex + ∇φ> 
∂t
 
∂φ> ∂φ >
=C −ρ − 12 U + O(η 2 ).
∂t ∂x

We now apply p> = p< at y = 0 to get

ġ> − 21 ıkU g> = ġ< + 12 ıkU g< .

We have three linear equations with constant coefficients for g> , g< and f , so
2 2
each is proportational to eσt . Plugging this solution in gives σ 2 = U 4k and so σ =
± U2k . Thus there exists a growing mode with σ = 12 U |k| and the sheet is unstable to
disturbances of all wavelengths.

Notes

1. As k → ∞, σ → ∞ and thus short waves grow infinitely fast.

2. The disturbance rapidly grows out of the linear régime. We get roll-up of vor-
tices.1

3. Physical mechanism when σ = 12 U |k|. We get

g≷ = 12 U {∓ sgn k + ı} f

+ +
and so [u]− = − 21 ıU |k| f − U and if η = η0 cos kxeσt we have [u]− = −U +
η0 U |k| sin kxeσt . Thus the vortex sheet is stronger at x = 3π
2k and weaker at
π
z = 2k .

1 See van Dyke page 85.

 47
√
4. If viscous effects are included then the vorticity diffuses over a distance δ ∝ νt
in a time t. We expect that if k −1 . δ the inviscid theory will be wrong (and
viscosity damps short waves), but long waves with k −1  δ should not be
−1
affected by vorticity diffusion. Since the growth time σ −1 ∝ (U k) we must
U
have k  ν for the inviscid theory to work.
If we guess (on dimensional grounds) that short waves are damped at a rate νk 2
U
then σ = 21 U |k| − νk 2 and there is a most unstable wavelength k = 4ν .

5. A long-wave inviscid mechanism will also apply to inviscid profiles with inflex-
ion points (eg tanh y).
2 2
6. The relationship σ 2 = U 4k is called a dispersion relation. In waves, σ = ıω is
pure imaginary and ωk = c is a wavespeed; η = η0 eık(x−ct) .
In some ways this calculation is artificial; in practice we can’t establish a fully-
developed unstable steady state to perturb. (This is temporal instability.) It is
more natural to introduce a perturbation eıωt with ω ∈ R at x = 0 and to observe
the growth or decay in x. We thus have η = η0 eıωt−ıkx with k ∈ C. This is a
spatial instability problem. We find that k = ±2ı  Uω 
Temporal and spatial analyses are identical at or near marginal stability. In gen-
eral, spatial analysis is harder and so temporal analysis is more common.
48 CHAPTER 8. KELVIN-HELMHOLTZ INSTABILITY
Chapter 9

Rising bubbles

9.1 Dimensional analysis

We want to know both the rise velocity and the shape of the bubble.
1
13
The natural lengthscale is V 3 and we set a = 3V 4π , the radius of a sphere of
volume V . 2
For a steady rise, buoyancy is presumably balanced by viscosity. Define U0 = aνg
and then the Reynolds number is
U0 a a3 g inertia
Re = = 2 = .
ν ν viscosity
We need a second dimensionless group to indicate the importance of surface ten-
sion. This is the capillary number

viscous stresses µ U0 µU0

Ca = = γa = .
surface tension a γ
hydrostatic pressure
Note that in this case, Ca = ρga
γ = surface tension pressure , a quantity which is usually
a
called the Bond number which in this case happens to be equal to the capillary number.
The actual rise speed of the bubble, U = U0 f (Re, Ca). The shape must also
depend on Re and Ca.
If Ca  1 then surface tension is very large and the shape remains almost spherical.
This is theoretically tractable, we have U = U0 fˆ(Re). If Ca ∼ 1 other shapes are
possible — this is a hard problem.

9.2 Low Reynolds number, low capillary number

This is the Re, Ca  1 case. The bubble shape is a near sphere r = a (1 + O(Ca))
and viscous forces dominate.

49
50 CHAPTER 9. RISING BUBBLES

We must solve

µ∇2 u = ∇P r>a
∇·u=0

where P = p − ρgz is the modified pressure.

The kinematic boundary condition is u · n = u · n on r = a. We also have u → 0
as r → ∞.
The tangential stresses must be continuous: n × [σ · n] = 0 on r = a. This
simplifies to n × [e · n] = 0 and thus erθ = 0 on r = a (in spherical polars).
Finally, [σ · n] = γκn on r = a, where κ is the surface curvature. Taking the
normal component we get pint − p + 2µenn = γκ and therefore

1
κ= (const − (P + ρgz) + 2µenn ) on S.
γ

Thus at Ca = 0 the drop is spherical, κ = a2 and surface tension increases the

internal pressure to 2γ
a .
If Ca  1 then at leading order the drop is spherical, but the non-zero right hand
side causes an O(Ca) modification to κ and we can use this to find an O(Ca) modifi-
cation to the shape at the end.
We can either solve D4 ψ = 0 (see page 12) and apply the boundary conditions to
get C = U2a and D = 0 or use the method of sheet 2:

u = E · x + 2φ − ∇ (φ · x)

with ∇2 φ = 0. We have E = 0 and the harmonic potential φ must be linear in u,

so
αu 1
φ= + βu · ∇∇ .
r r
u evaluates to
 
u u · x x u 3u · x x
u=α + + 4β − + .
r r3 r3 r5

Applying the boundary conditions we get α = a2 and β = 0. We also have P =

−2µ∇ · φ = −µau · ∇ 1r .
For r > a this is a pure Stokeslet field |u| ∝ 1r . Integrating over a sphere gives
a2 g
F = 4πµau, the force of the bubble on the fluid. Thus u = 3ν (by equating this to
the Archimidean uplift).
To get the shape change, we have

u · x 
−p + ρgz + 2µerr = ρg · x − 3µ + const = const on r = a.
r3 r=a


Thus, surprisingly, a spherical drop at low Reynolds number has no tendency to

deform even if Ca is not small. Our solution works for Re  1 and Ca arbitrary.
9.3. HIGH REYNOLDS NUMBER, LOW CAPILLARY NUMBER 51

9.3 High Reynolds number, low capillary number

2
Our first guess is u = ∇φ and ∇2 φ = 0. We have φ = a ru·x3 , which satisfies the Euler
equation and u · n = u · n on r = a.
The tangential stress on r = a is σrθ = 2µerθ = − 3µU a sin θ 6= 0. Also, the drag
on the sphere is zero.
We expect that the outer flow will be modified (at O(Re−1 )) and a boundary layer
will arise near r = a, sweeping vorticity into the wake, perhaps modifying the outer
1
flow at O(Re− 2 ).
Without gross separation (which is not observed in experiments), we know the flow
almost everywhere and can calculate
Z
u · F = 2µ eij eij dV
r>a
∂2φ ∂2φ
Z
= 2µ dV
r>a ∂xi ∂xj ∂xi ∂xj
∂φ ∂ 2 φ
Z  
∂
= 2µ dV
∂xi ∂xj ∂xi ∂xj
Zr>a
n · ∇ 12 u2 dS


= 2µ
r=a
= 12πµaU 2 .
2
Thus U = 1a g
9 ν = 19 U0 and fˆ(Re) → 1
9 as Re → ∞.

As for the shape change, we see that pressure variations over r = a scale as 21 ρU 2
 2 2
a g
and so ∆p = 2·91
2ρ ν . The drop will remain spherical if ∆p
γ  1, or alternatively
a
Re Ca  160.

Free surface boundary layers

∂u
Near r = a there is a boundary layer across which not uk but ∂nk jumps. There is thus
a jump in Ω across the boundary layer.
1
The thickness of the boundary layer is still δ = aRe− 2 and the boundary layer
equation still applies, but the velocity gradient in the layer scales as Ua and not Uδ .
2 2
The energy dissipation in the layer scales as 4πa2 δ Ua2 = 4πaU√
Re
Thus
 1

u · F = 12πµaU 2 1 + βRe− 2 .
52 CHAPTER 9. RISING BUBBLES

In fact, β = 2.2, but we need the dissipation in the wake of the bubble to find this.
Observation suggests that the boundary layer does not separate unless the curvature
is very high.

9.4 The oblate spheroidal bubble

This is the Re  1 and Ca ∼ 1 case. It is tough to make progress.

The first correction will be spheroidal and will modify the rise speed. As the capil-
lary number increases we will eventually get boundary layer separation.

9.5 Spherical cap bubble

This is the Re, Ca  1 case.

π 3
ā 2 − 3 cos α + cos3 α .

V =
3
 
ā3
The outer flow is flow past a sphere of radius ā, φ = −U r + 2r 2 cos θ. The
3
tangential velocity on r = ā is uθ = 2U sin θ. As Ca  1 we have continuity of
pressure, and so 12 ρu2θ − ρgz = const. Near θ = 0, u2θ ≈ 94 U 2 θ2 and z ≈ 12 āθ2 . Thus
√
U 2 = 94 gā and so the rise velocity U = 23 gā independent of ν.
Note that the Re → ∞ limit differs from the Re = ∞ limit.
There are turbulent dissipative processes in the wake that give a rise velocity inde-
pendent of ν as ν → 0. Note that this rise velocity agrees with experiment. Experi-
mentally, α is found to be in the range 40◦ < α < 60◦ .

9.6 The skirted bubble

This has Re and Ca “largish”. If we decrease surface tension from the spherical cap
bubble a discontinuity appears near the sharp edge. For suitable parameter values a
9.6. THE SKIRTED BUBBLE 53

cusped shaped edge appears and forms a skirt around the bubble.

If surface tension is decreased a little from this, Kelvin-Helmholtz instability occurs

near the cusped edge.

This is poorly understood.

54 CHAPTER 9. RISING BUBBLES
Stress tensor: EJH approach

Unfortunately, the (standard) derivations given earlier for the surface traction and sym-
metry of σ do not work (by dimensional arguments). The following can be inserted in
the appropriate places in §2.3. Discovering the precise places is left as a challenge to
the reader.

Linearity
Consider the force balance on this small tetrahedron. The volume and acceleration
forces are O(ρgL3 ), where L is the linear size of the tetrahedron. The surface forces
are O(pL2 ), with a typical pressure ρgH, where H is the height of the atmosphere.
Hence for small tetrahedra with L  H the surface forces are O( H L ) larger than the
volume and acceleration forces and so must balance amongst themselves.

Symmetry
The moment of the surface forces is O(pL3 ) and the moments of the volume and
acceleration forces are O(ρgL4 ). So again the surface forces must balance amongst
themselves.

55
56 STRESS TENSOR: EJH APPROACH
References

◦ D.J. Acheson, Elementary Fluid Dynamics, OUP, 1990.

This is an excellent book, easy to read and with almost everything in. Highly recom-
mended.

◦ G.K. Batchelor, An Introduction to Fluid Dynamics, CUP, 1967.

A good reference but somewhat intimidating. It looks like the book of an older version
of this course, although the content is roughly similiar. Probably not worth spending
money on...

◦ Slow Viscous Flow, unpublished, 1998.

These are the course notes for the Part III Slow Viscous Flow course, which essen-
tially covers Chapter 3 in more detail. Worth a look — if you can find them!

◦ M. van Dyke, An Album of Fluid Motion, The Parabolic Press, 1982.

Lots of pictures of flows. An excellent book. Go out and buy it. Now.

There are many other books covering this course. I have heard good things about
Landau and Lifschitz, and Lamb is a standard text for inviscid fluid mechanics. I
haven’t looked at them so I can’t personally recommend them. If you can, please send
me a brief review and I will include it above.

57