Sie sind auf Seite 1von 368

Elementary Fluid Dynamics

D.J.Acheson, Clarendon Press (90)


Notes by CKWong
1 Introduction
Sources:
D.J.Acheson, Elementary Fluid Dynamics, Clarendon Press (90)
R.E.Meyer, Introduction to Mathematical Fluid Dynamics, Wiley (71).
Experiment:
Aerofoil & Starting Vortex
Applications:
Waves (3.1)
Flow Instability (9.1)
Hydraulic Jumps (3.10)
Smoke Ring Interactions (5.4)
Atmospheric Jet Stream (9.8)
Quantum Vortices (5.8)
Sperm Movement (7.5)
Spindown of Stirred Tea (8.5)
1.1 Preliminary
Flow velocity: , ) , ) , , , t u v w = = u u x (1.1)
Steady Flow:
c
c
=
u
t
0 (1.2)
2-D Flow: , ) , ) , , , , , , 0 u x y t v x y t = 1
]
u (1.3)
2-D Steady Flow: , ) , ) , , , , 0 u x y v x y = 1
]
u (1.4)
Streamline: Curve with tangent equal to u everywhere.
Let curve be: ( ) x x s = , ( ) y y s = , ( ) z z s =
Then
dx ds
u
dy ds
v
dz ds
w
/ / /
= = (1.5)
Streak photographs: illuminated polystyrene beads of neutral buoyancy.
Material derivative, or, rate of change following the fluid:
Df
Dt
d
dt
f x t y t z t t = [ ( ), ( ), ( ), ]
rc
fvv
f f dx f dy f dz
t x dt y dt z dt
c c c c
= + + +
c c c c
f f f f
u v w
t x y z
c c c c
= + + +
c c c c
( )
f
f
t
c
= + V
c
u (1.6)
Acceleration:
D
Dt t
u
u u
u
= V +
c
c
( ) (1.7)
Uniform rotation: ( , , 0) y x = O O u
Steady flow: 0
t
c
=
c
u
So that
, )
D
Dt
= V
u
u u
, ) , , 0 y x y x
x y
| | c c
= O + O O O
|
c c
\ .
, )
2
, , 0 x y = O
For steady flow:
, )
Df f
f f
Dt s
c
= V = V =
c
u u s u

where s = s s

is the tangent vector to a streamline.


, ) 0 f V = u means f is constant along a streamline but it can vary between different
streamlines.
0
Df
Dt
= means f is constant for a fluid particle but it can differ between different
particles.
A material (dyed) volume is a collection of designated fluid particles.
It may deform during the course of motion.
1.2 Ideal Fluids
An ideal fluid is
Incompressible: = constant, or 0
D
Dt

= , which implies 0 V = u .
Inviscid (no viscosity): p t = n .
Mass of a volume V fixed in space & bounded by surface S is:
V
m dV =
}
Rate of change is
V V
dm d
dV dV
dt dt t

c
= =
c
} }
Net rate of mass leaving the volume is
, )
S V
dS dV = V
} }
u n u

Conservation of mass means


, )
V V
dV dV
t


c
= V
c
} }
u (A)
Actually, one can show that eq.(A) is true even if V denotes a volume moving along
with the fluid. [ see R.E.Meyer, Chap 1 ]. More generally, the convection theorem
( Meyer, p.10) states that
V V
D Df
fdV f dV
Dt Dt
| |
= + V
|
\ .
} }
u
Going back to eq.(A), since it is true for all V, we have
, ) 0
t


c
+ V =
c
u
which is called the equation of continuity.
Using
, ) V = V + V u u u
and
D
Dt t


c
= + V
c
u
We can writing the equation of continuity as
0
D
Dt

+ V = u
For an incompressible fluid, = constant, or 0
D
Dt

= , so that
0 V = u (incompressible condition)
Force on dyed blob in inviscid fluid:
S V
p dS pdV = V
} }
n

which means the force density is p V .


1.2.1 Eulers Equation
Newtons law applied to a fluid volume:
, )
V V
D
dV p dV
Dt
= V +
} }
u g
Using the convection theorem, the left side is
, )
V V
D D D
dV dV
Dt Dt Dt


| |
| |
+ V = + + V
| |
\ .
\ .
} }
u u
u u u u u
V
D
dV
Dt
=
}
u
where the last equality was obtained with the help of the eq. of continuity.
We thus obtain Eulers eqs.:
D
p
Dt
= V +
u
g (1.12)
where g is the gravitational acceleration.
Writing
_ = V g (1.13)
we have
D
p
Dt t
_
c
| |
= + V = V V
|
c
\ .
u
u u
or
p
t
_

| | c
| |
+ V = V +
| |
c
\ .
\ .
u u
Using
, ) , )
2
1
2
V = V V u u u u u
we have
, )
2
1
2
p
t
_

| | c
+ V = V + +
|
c
\ .
u
u u u (1.14)
1.2.2 Bernoulli Streamline Theorem
For steady flow, (1.14) becomes
, ) H V = V u u
where
2
1
2
p
H _

= + + u (1.15)
Using
, ) 0 V = 1
]
u u u
we have
, ) 0 H V = u (1.16)
ie, H = const along streamline of an ideal fluid in steady flow.
Irrotational flow means
0 V = u (1.17)
Bernoulli Streamline Theorem for irrotational flow:
H = const everywhere in a steady irrotational flow of an ideal fluid.
1.3 Irrotational Flow
Vorticity: = V u (1.18)
2-D flow:
[ ( , , ), ( , , ), 0] u x y t v x y t = u
, ) 0, 0, 0, 0,
0
v u
x y z x y
u v
e
| | c c c c c
= = =
|
c c c c c
\ .
i j k

v u
x y
e
c c
=
c c
(1.19)
Consider 2 mutually perpendicular line segments AB & AC (fig.1.3):
( , , ) ( , , )
B A
v
v v v x x y t v x y t x
x
o o
c
= + ~
c
= CCW angular velocity of AB
( , , ) ( , , )
C A
u
u u u x y y t u x y t y
y
o o
c
= + ~
c
= CW angular velocity AC
Hence
1 1
2 2
v u
x y
e
| | c c
=
|
c c
\ .
= average angular velocity of AB & AC.
Note: donotes local spin & has nothing to do with global rotations. (Fig.1.6)
Eg. Shear flow (fig.1.4)
, ) , 0, 0 y | = u (1.20)
has no global rotation but
0
v u
x y
e |
c c
= = =
c c
Eg. Line vortex flow

k
r
= u

(1.21)
has
1 1
0
0 0
r z
r r
r r z r r z
u ru u k
u
u u
c c c c c c
= V = = =
c c c c c c
r
z r z
u


but is obviously rotating as a whole.
Eg. Flow

r = O u

spin has same rate O as global rotation.


Eg. Rankine Vortex (fig.1.7)
2
r
u
a
r
u
O

=
O

for
r a
r a
<
>
(1.23)
0
r z
u u = =
1.4 Vorticity Equation
Eulers eq. (1.14):
, ) H
t
c
+ V = V
c
u
u u
can be written as
H
t
c
+ = V
c
u
u
V both sides gives
, ) 0
t
c
+ V =
c

u (1.24)
Using
, ) , ) , ) , ) , ) V = V V + V V a b b a a b a b b a
we have
, ) , ) , ) , ) , ) V = V V + V V u u u u u
, ) , ) = V V u
u
where
0 V = u for incompressible fluid
, ) 0 V = V V = u for any vector u.
Thus (1.24) becomes
, ) , ) 0
t
c
+ V V =
c

u
u
or
, )
D
Dt
= V

u (1.25)
which is called the vorticity equation.
For 2-D flow
, ) , ) , , , , , , 0 u x y t v x y t = 1
]
u , ) 0, 0,e =
we have
, ) 0
z
e
c
V = =
c
u u
so that
0
D
Dt
=

(1.27)
Thus, for a 2-D flow of ideal fluid under conservative forces, is conserved for each
fluid element.
For steady flow, we have
, ) 0 V = u

so that is conserved along each streamline.


Thus, steady 2-D flow pass an aerofoil is irrotational since 0 e = for x .
1.5 Steady Flow Past Fixed Wing
Lift of wing in steady flow is due to (fig.1.9; also demonstrated by Wright program)
upper
p p

<
lower
p p

>
Flow is irrotational.
Bernoulli theorem means
2
1
2
p + u is constant everywhere.
Thus lift means
upper lower
> u u .
1.5.1 Circulation
Circulation around closed curve C is
C
d I =
}
u x

(1.31)
Stokes theorem:
, )
C S
d dS = V
} }
u x u n

(1.32)
S
dS =
}
n
where C is the boundary of S.
For the irrotational flow past an aerofoil,
0 I = for any C not enclosing the aerofoil.
0 I = for any C that encloses the aerofoil
(surface of foil is another boundary of S).
Furthermore, for fig.1.8, lift means 0 I < .
1.5.2 Kutta-Joukowski Hypothesis
For wing with sharp trailing edge, u can be finite everywhere only for one value of
circulation
K
I . Otherwise, u at the sharp edge.
That
K
I describes the actual flow is the Kutta-Joukowski hypothesis.
For thin, symmetric wing with small , (see Chap 4)
sin
K
UL t o I (1.34)
where
U u

= L = length of wing
1.5.3 Lift
For ideal flow,
Drag = 0
Lift U = I (1.35)
For
K
I = I , we have
2
sin U L t o = (1.36)
For large angles, 1.36 overestimates badly (fig.1.11).
1.6 Conclusion
Near wing tip, flow is not 2-D & cant be irrotational.
This leads to trailing vortices & hence drag. (fig.1.12)
Generation of is a response to starting vortices (Chap 5), which in turn requires
viscosity.
2 Viscous Flow
Source: D.J.Acheson, Elementary Fluid Dynamics, Chapter 2, Clarendon Press (90)
2.1 Introduction
Theoretically, inviscid, and in particular, irrotational (potential) flow is easiest to deal
with. Naively, we may expect it to describe flows with low viscosity.
In practice, we do find the theory useful in describing phenomena such as water
waves, sound waves, and slow flow past streamlined objects.
For flow past a general object, the down-stream side becomes turbulent.
The root cause of this may be traced to the viscosity related no-slip boundary
condition which a potential flow usually cannot satisfy. Thus, flows from the inviscid
theory are valid only if they can be made to reconcile with the no-slip condition
within a sufficiently thin transitional unseparated boundary layer.
If the inviscid theory predicts an increase of pressure along the flow direction on the
boundary, separation of boundary layer is expected and turbulence results. (see
chapters 4 & 8).
On the other hand, theory of very viscous fluid is very successful (see chapter 7).
Newtonian fluid:
du
dy
t = (2.1)
where is the shear stress, and is the coefficient of viscosity.
Navier-Stokes Eqs (for incompressible Newtonian fluids):
, )
2
1
p
t
v

c
+ V = V + V +
c
u
u u u g (2.3)
0 V = u
See Landau for a derivation.
The kinematic viscosity is defined as

v

= (2.2)
No-slip condition:
0 = u at stationary boundary.
Reynolds Number:
UL
R
v
= (2.4)
where U and L are the characteristic flow velocity and dimension of the system,
respectively. And is the kinematic viscosity.
Using estimates
, ) O U = u
, )
U
O
L
V = u
on the Navier-Stokes eqs. (2.3), we have
Inertia term: , )
, )
2
U
O
L
V = u u
Viscous term:
, )
2
2
U
O
L
v
vV = u (2.5)
so that
, )
2
2
inertia term
viscous term
U
L
O O R
U
L
v
| |
|
= =
|
\ .
(2.6)
High R flows:
Corresponds to flows with small viscosity.
Under favorable circumstances, eg. flow past streamlined bodies, it corresponds
to the flow of ideal fluid except for a thin boundary layer (required for the no-slip
condition) and a narrow trailing wake.
Typical thickness of such a boundary layer is
, )
1
2
O R
L
o

= (2.7)
In general, high R flows are unstable & easily become turbulent.
Low R flows:
0.01 R s .
Flow is well-ordered & reversible.
2.2 Navier-Stokes Eqs.
Consider the Euler eq
, ) D D
p
Dt Dt

+ V = = V
u u
u u
Using
, ) , )
, )
D
Dt t


c
+ V = + V + V
c
u u
u u u u u u
, )
, )
t

c
= + V
c
u
uu
we can re-write it as
, )
, ) 0 p
t

c
+ V + V =
c
u
uu
or
, )
0
t
c
+ V H =
c
u
(A)
where the momentum flux density tensor is
ij i j ij
u u p o H = +
Eq(A) is just the eq of continuity for momentum which expresses the conservation
of momentum.
Since we expect the principle of the conservation of momentum to be valid even in
viscous fluids, eq(A) should apply there with a suitable modification of . We thus set
ij i j ij
u u o H =
where the stress tensor
ij
o is related to the viscous stress tensor '
ij
o by
'
ij ij ij
p o o o = +
[Note: Acheson used
ij
T instead of
ij
o (see chap 6) ]
What this means is that the generalization of the Eulers eq to viscous fluids can be
done by replacing p V with ' p o o V = V + V so that, for example,
D
p
Dt
= V
u
becomes
'
D
p
Dt
o = V + V
u
or
' p
t
o
c
| |
+ V = V + V
|
c
\ .
u
u u (B)
Next, we need to determine the form of ' o .
Now, viscous effects are due to friction between adjacent fluid elements moving
with different velocities. Hence ' o should depend on
i
j
u
x
c
c
but not on u itself.
Furthermore, theres no friction if the fluid rotates as a whole with uniform angular
velocity so that = u
r
or
i ijk j k
u x c = O .
Now:
i m
ijk j km ijm j mji j
m i
u u
x x
c o c c
c c
= O = O = O =
c c
and
0
i
iji j
i
u
x
c
c
= O =
c
Hence, the choice
' '
j
i k
ij ij ji
j i k
u
u u
a b
x x x
o o o
| | c
c c
= + + =
|
|
c c c
\ .
where a, b are constants, will satisfy the above requirements.
A more common form is
2
'
3
j
i k k
ij ij ij
j i k k
u
u u u
x x x x
o o , o
| | c
c c c
= + +
|
|
c c c c
\ .
(C)
where the positive constants
a =
2 2
3 3
b a b , = + = +
are called coefficients of viscosity. (Landau used instead of )
This also defines the Newtonian fluid. [cf eq(2.1)].
Using (C), (B) becomes
'
ji
i i
j
j i j
u u p
u
t x x x
o

| | c
c c c
+ = +
|
|
c c c c
\ .
2
3
j
i k k
ij ij
i j j i k k
u
p u u u
x x x x x x
o , o
(
| | c
c c c c c
= + + +
( |
|
c c c c c c
(
\ .

2
3
j
k i
i k j j i
u
u u
p
x x x x x
,
| | c ( c c c c
| |
= + + + +
|
| (
|
c c c c c
\ .

\ .
1
3
k i
i k j j
u u
p
x x x x
,
( c c c c
| |
= + + +
| (
c c c c
\ .

or in vector form:
' p
t
o
c
| |
+ V = V + V
|
c
\ .
u
u u
2
1
3
p ,
(
| |
= V + V + V
|
(
\ .

u u
, )
2
1
3
p ,
| |
= V + + V V + V
|
\ .
u u (D)
For incompressible fluids, we have the Navier-Stokes eqs.:
2
p
t

c
| |
+ V = V + V
|
c
\ .
u
u u u (2.3)
0 V = u
Note also that for incompressible fluids,
'
j
i
ij
j i
u
u
x x
o
| | c
c
= +
|
|
c c
\ .
[cf (2.1)]
2.3 Simple Flows
2.3.1 Plane Parallel Shear Flow
Plane parallel shear flow is defined as
, ) , , 0, 0 u y t = (

u (2.8)
Thus
0
u
x
c
V = =
c
u (incompressible)
Navier-Stokes eqs becomes
2
2
1 u p u
t x y
v

c c c
= +
c c c
0
p p
y z
c c
= =
c c
(2.9)
The last eq means p and hence
p dp
x dx
c
=
c
is a function of , ) , x t only.
Hence, the right and left hand side of
2
2
1 p u u
x t y
v

c c c
= +
c c c
is a function of , ) , x t and , ) , y t , respectively. Thus, either can only be function of t
only.
2.3.2 Impulsively Moved Boundary
Let fluid lies at rest in region 0 y < < at 0 t < .
At 0 t = , boundary at 0 y = is set in motion in x-direction with constant speed U.
No-slip condition means , ) , 0, 0 U = u at boundary for 0 t > .
In general, we expect , ) , , 0, 0 u y t = (

u so that it is a plane parallel shear flow.
According to the last section, , )
p
f t
x
c
=
c
so that , ) p f t x C = + .
Let there be no externally applied pressure gradient, ie., p at x = are equal.
This means we must have , ) 0 f t = and p C = .
Eq(2.9) thus becomes
2
2
u u
t y
v
c c
=
c c
(2.12)
which is simply the 1-D diffusion eq.
This must be solved with the initial condition
, ) , 0 0 u y = for 0 y >
and boundary conditions
, ) 0, u t U = for 0 t >
, ) , 0 u t = for 0 t >
Now, (2.12) is invariant under the scale transformation
y y o
2
t t o
where is a constant.
This means we can attempt a similarity solution:
, ) u f q = where
y
t
q
v
= (2.13)
so that
3
2
y
t
t
q
v
c
=
c
1
y t
q
v
c
=
c
3
' ' '
2
2
u y
f f f
t t t
t
q q
v
c c
= = =
c c
1
' '
u
f f
y y t
q
v
c c
= =
c c
2
2
1 1
'' ''
u
f f
y y t t
q
v v
c c
= =
c c
where '
df
f
dq
= . Hence (2.12) becomes
1
' ''
2
f f
t t
q
v
v
=
or
'' 1
' 2
f
f
q =
so that
2
1
ln '
4
f const q = +
or
2
4
' f Be
q
=
whence
, )
2
4
0
s
f A B e ds
q
q

= +
}
where A, B are constants.
Using
q for 0 t = or y
0 q = for 0 y
the boundary & initial conditions become
, ) 0 f = and , ) 0 f U =
Hence
2
4
0
1
0
s
A B e ds A B
t

= + = +
}
U A =
whence
, )
2
4
0
1
1
s
f U e ds
q
q
t

| |
=
|
\ .
}
(2.14)
or
, )
2
4
0
1
, 1
y
s
t
u y t U e ds
v
t

| |
=
|
\ .
}
[see fig 2.8]
The vorticity is
2 2
4 4
y
t
u U U
e e
y y t
q
v
q
e
t tv

c c
= = =
c c
(2.15)
which indicates a diffusion with standard deviation 2 t v . In other words,
viscous diffusion time
, )
2
L
O
v
= (2.16)
where L is a distance.
2.3.3 Flow Down Inclined Plane
With reference to fig. 2.9, the Navier-Stokes eqs
, )
2
1
p
t
v

c
+ V = V + V +
c
u
u u u g
becomes
2 2
2 2
1
sin
u p
u v u u g
t x y x x y
v o

| | | | c c c c c c
+ + = + + +
| |
c c c c c c
\ . \ .
2 2
2 2
1
cos
v p
u v v v g
t x y y x y
v o

| | | | c c c c c c
+ + = + +
| |
c c c c c c
\ . \ .
where , ) , , 0 u v = u .
The no-slip condition means 0 = u for 0 y = .
Hence u must vary with y if its not identically 0.
The simplest solution possible is therefore a steady flow of the form
, ) , ) , , 0 u y v y = (

u
The incompressibility condition
0
u v
x y
c c
+ =
c c
becomes
0
v
y
c
=
c
so that v const = .
However, , ) 0 0 v = so that 0 v = for all y.
The steady state Navier-Stokes eqs. thus become
2
2
1
0 sin
p d u
g
x dy
v o

c
= + +
c
1
0 cos
p
g
y
o

c
=
c
(2.17)
The 2
nd
eq gives
, ) cos p gy f x o = + (A)
Since all streamlines are in the x direction, the free surface must be y h = where h is
a constant. For a flat interface, p must be equal on both sides and the tangential stress
must vanish. Let the atmospheric pressure be
0
p , we have
0
p p = and 0
du
dy
= at y h = (2.18)
Putting these into (A) gives
, )
0
cos p gh f x o = +
so that (A) becomes
0
cos cos p gy p gh o o = + +
or
, )
0
cos p p g h y o =
whence
0
p
x
c
=
c
and the remaining Navier-Stokes eq reduces to
2
2
0 sin
d u
g
dy
v o = +
which gives
sin du g
y A
dy
o
v
= +
2
sin
2
g
u y Ay B
o
v
= + +
Putting in the boundary conditions
0 u = at 0 y =
and
0
du
dy
= at y h =
we have
0 B =
sin
0
g
h A
o
v
= +
so that
2
sin
2
g y
u hy
o
v
| |
= +
|
\ .
(2.19)
The volume flux per unit length in the z-direction is
2 3 2
3
0 0
sin sin sin
2 6 2 3
h h
g y g h h g
Q udy hy dy h h
o o o
v v v
| | | |
= = + = + =
| |
\ . \ .
} }
2.3.4 Another Example
Consider the impulsively moved plane boundary problem with an added stationary
plane boundary a distance h above it. (see fig 2.10)
The problem is to solve , ) , , 0, 0 u y t = (

u from
2
2
u u
t y
v
c c
=
c c
(2.20)
with the initial condition
, ) , 0 0 u y = for 0 h y > >
and boundary conditions
, ) 0, u t U = and , ) , 0 u h t = for 0 t >
The form of (2.20) clearly suggests the use of the method of separation of variables.
Let
, ) , ) u Y y T t =
(2.20) becomes
2
2
dT d Y
Y T
dt dy
v =
or
2
2
dT d Y
Tdt Ydy
v o = = const o =
Thus, for 0 o = ,
t
T Ae
o
=
y y
Y Be Ce
| |
= +
where A, B, C are constants and
o
|
v
=
For 0 o = , we have
T A = ' ' Y B y C = +
where ', ' B C are constants.
Hence,
, ) , )
0 0
0
,
t y y
u y t a y b e a e b e
o | |
o o
o

=
= + + +
_
(A)
is a general solution of (2.20).
The next task is to adjust the constants to satisfy the initial & boundary conditions:
, ) , )
0 0
0
, 0 0
y y
u y a y b a e b e
| |
o o
o

=
= + + + =
_
for 0 h y > >
, ) , )
0
0
0,
t
u t b e a b U
o
o o
o=
= + + =
_
for 0 t >
, ) , )
0 0
0
, 0
t h h
u h t a h b e a e b e
o | |
o o
o

=
= + + + =
_
for 0 t >
The last 2 eqs can be satisfied only if the time dependent sums vanish identically.
Hence, we have
0
b U =
0 0
0 a h b + = so that
0
U
a
h
=
and
, )
0
0
t
e a b
o
o o
o=
+ =
_
, )
0
0
t h h
e a e b e
o | |
o o
o

=
+ =
_
for 0 t >
Since the
t
e
o
terms are independent, each term in these sums must vanish
individually so that
0 a b
o o
+ =
0
h h
a e b e
| |
o o

+ =
which gives
b a
o o
=
2sinh 0
h h
e e h
| |
|

= =
so that
h n i | t =
and
2 2
2
2
0
n
h
t
o v| v = = <
Eq(A) becomes
, )
2 2
2
1
, sin
n
t
h
n
n
U n
u y t y U e A y
h h
t
v
t

=
| |
= + +
|
\ .
_
where 2
n
A ia
o
= , ) 0 o < are to be determined by the initial condition, namely,
, )
1
, 0 sin 0
n
n
U n
u y y U A y
h h
t
=
| |
= + + =
|
\ .
_
for 0 h y > >
Using
0
sin sin
2
nm
nx mxdx
t
t
o =
}
we have
0
sin 0
2
h
n
U n
dy y U y A
h h h
t t t
| | | |
+ + =
| |
\ . \ .
}
or
0
2
1 sin
h
n
y n
A U dy y
h h h
t
| | | |
=
| |
\ . \ .
}
0
2
1 sin
n
z
U dz z
n n
t
t t
| |
=
|
\ .
}
where
n
z y
h
t
=
, )
0
2 1
cos cos sin
n
U
z z z z
n n
t
t t
(
= +
(

2U
nt
=
so that, finally
, )
2 2
2
1
2 1
, 1 sin
n
t
h
n
y n
u y t U e y
h n h
t
v
t
t

=
(
| |
= (
|
\ .
(

_
(2.21)
2.4 Flow with Circular Streamlines
In cylindrical coordinates , ) , , r z u , (see Appendix A.5)
, ) , ,
r z r r z z
u u u u u u
u u u
= = + + u e e e
1 2
cos sin
r
u u = + e e e
1 2
sin cos
u
u u = + e e e
3 z
= e e
r
u
u
c
=
c
e
e
r
u
u
c
=
c
e
e (all other partials vanishes) (2.23)
r z
f f
r r z
u
u
c c c
| |
V = + +
|
c c c
\ .
e e e
r z
f u u u f
r r z
u
u
c c c
| |
V = + +
|
c c c
\ .
u
, )
z
r
V V
rV
r r r z
u
u
c c c
V = + +
c c c
V
1
r z
r z
r
r r z
V rV V
u
u
u
c c c
V =
c c c
e e e
V
2 2
2
2 2 2
f r f
r r r r z u
( c c c c
| |
V = + +
| (
c c c c
\ .

To write the Navier-Stokes eqs
, )
2
1
p
t
v

c
+ V = V + V
c
u
u u u
in cylindrical coordinates, we 1
st
consider the term , ) V u u :
, ) , ) , ) , )
r r r r r r
u u u V = V + V u e u e e u
, )
r r z r r r
u u u u u
r r z
u
u
c c c
| |
= + + + V
|
c c c
\ .
e e u
, )
r r r r
u u u
r
u
u
c
= + V
c
e e u
, )
r
r r
u u
u
r
u
u
= + V e e u
, ) , ) , ) , ) u u u
u u u u u u
V = V + V u e u e e u
, )
r z
u u u u u
r r z
u u u u u
u
c c c
| |
= + + + V
|
c c c
\ .
e e u
, ) u u u
r
u u u u u
u
c
= + V
c
e e u
, )
2
r
u
u
r
u
u u
= + V e e u
, ) , ) , ) , )
z z z z z z
u u u V = V + V u e u e e u
, )
z z
u = V e u
so that
, ) , ) , ) , )
2
r
r r r z z
u u u
u u u
r r
u u
u u u
V = + V + V + V u u e e u e e u e u
, ) , ) , )
2
r
r r z z
u u u
u u u
r r
u u
u u
(
(
= V + V + + V
(
(


e u e u e u
Next, we turn to
2
V u :
, ) , )
2 2
2
2 2 2 r r r r
u r u
r r r r z u
( c c c c
| |
V = + +
| (
c c c c
\ .

e e
, )
2 2
2 2 2 r r r r
r u u
r r r z r u
( c c c c
| |
= + +
| (
c c c c
\ .

e e
Now
, )
2
2
r r
r r r r
u
u u
u u u u
c c c c
(
= +
(
c c c c

e
e e
2 2
2 2
2
r r r r
r r
u u
u
u u u u
c c c c
= + +
c c c c
e e
e
2
2
2
r r
r r r
u u
u
u
u u
c c
= + +
c c
e e e
so that
, )
2 2
2
2 2 2
1
2
r r
r r r r r r r
u u
u r u u
r r r z r
u
u u
( ( c c c c c
| |
V = + + + +
| ( (
c c c c c
\ .

e e e e e
2
2
1
2
r
r r r r
u
u u
r
u
u
c
(
= V + +
(
c

e e e
2
2 2
2
r r
r r
u u
u
r r
u
u
c
| |
= V +
|
c
\ .
e e
Similarly
, )
2
2
u
u u
u u
u u u u
u u u u
c c c c
(
= +
(
c c c c

e
e e
2 2
2 2
2
u u
u
u u u u
u u
u u u u
c c c c
= + +
c c c c
e e
e
2
2
2
r
u u
u
u u
u u u
u u
c c
= +
c c
e e e
so that
, ) , )
2 2
2
2 2 2
u r u u
r r r z r
u u u u u u
u
( c c c c
| |
V = + +
| (
c c c c
\ .

e e e
2 2
2 2 2
1
2
r
u u
r u u
r r r z r
u u
u u u u u
u u
( | | c c c c c
| |
= + + +
| | (
c c c c c
\ .
\ .
e e e e
2
2
1
2
r
u
u u
r
u
u u u u
u
c
| |
= V +
|
c
\ .
e e e
Finally, with
, )
2 2
z z z z
u u V = V e e
we have
2 2 2 2
2 2 2
1
2 2
r r
r r r z z
u u u
u u u u
r r r
u
u u u u u
u u
c c
| | | |
V = V + + V + + V
| |
c c
\ . \ .
u e e e e e e
2 2 2
2 2 2 2
2 1
2
r r
r r z z
u u u
u u u u
r r r r
u
u u u
u u
c c
| | | |
= V + V + + V
| |
c c
\ . \ .
e e e
Hence
, )
2
2
2 2
1 2
r r
r r
u u p u u
u u
t r r r r
u u
v
u
c c c
| |
+ V = + V
|
c c c
\ .
u
, )
2
2 2
1 1
2
r r
u u u p u
u u u
t r r r r
u u
u u u
v
u u
c c c
| |
+ V + = + V +
|
c c c
\ .
u (2.22)
, )
2
1
z
z z
u p
u u
t z
v

c c
+ V = + V
c c
u
where
r z
f u u u f
r r z
u
u
c c c
| |
V = + +
|
c c c
\ .
u
2 2
2
2 2 2
f r f
r r r r z u
( c c c c
| |
V = + +
| (
c c c c
\ .

The incompressibility condition is simply
, ) 0
z
r
u u
ru
r r r z
u
u
c c c
V = + + =
c c c
u
2.4.1 Differential Eq.
Consider a 2-D circular flow
, ) , u r t
u u
= u e (2.27)
The incompressibility condition
, ) 0
z
r
u u
ru
r r r z
u
u
c c c
V = + + =
c c c
u
is automatically satisfied.
The Navier-Stokes eqs (2.22) becomes
r
u :
2
1 u p
r r
u

c
=
c
u
u
:
2
1 1 u p u
r u
t r r r r r
u u
u
v
u
c c c c (
| |
= +
|
(
c c c c
\ .

z
u :
1
0
p
z
c
=
c
Now, u
u
is a function of , ) , r t only.
The u
u
eq can therefore be written as
, ) ,
p
f r t
u
c
=
c
Integrating, we have, since p is independent of z:
, ) , ) , ) , , , , p r t f r t C r t u u = +
where C is another arbitrary function of , ) , r t .
Now, p must be single valued. This means
, ) , ) , , , 2 , p r t p r n t u u t = +
This is possible only if 0 f = , ie, p is a function of , ) , r t only.
The Navier-Stokes eqs can be further simplify to
, ) , u r t
u u
= u e , ) , p p r t =
r
u :
2
1 u p
r r
u

c
=
c
u
u
:
2
1 u u
r u
t r r r r
u u
u
v
c c c (
| |
=
|
(
c c c
\ .

(2.30)
2.4.2 Steady Flow Between Cylinders
For steady flow, (2.30) becomes
0
d du
r r u
dr dr
u
u
| |
=
|
\ .
Setting ln y r = , it becomes
0
d du
u
dy dy
u
u
| |
=
|
\ .
so that
y y
B
u Ae Be Ar
r
u

= + = + (2.31)
If the fluid occupies region
1 2
r r r s s between 2 cylinders rotating with angular
velocities
1
O and
2
O , respectively, the no-slip condition gives
1 1 1
1
B
r Ar
r
O = +
2 2 2
2
B
r Ar
r
O = +
Solving for A and B gives
2 2
1 1 2 2
2 2
1 2
r r
A
r r
O O
=

, )
2 2
1 2 1 2
1 2 2 2
2 1
2 2
1 2
1 1
r r
B
r r
r r
O O
= = O O

(2.32)
This formula will be of use in the study of Taylor vortices. (see Fig 9.8)
2.4.3 Spin-Down
Let fluid occupies the interior of an infinitely long cylinder of radius a.
Initially, the whole system is rotating with uniform angular velocity so that
u r
u
= O for r a s 0 t s
The cylinder is suddenly stopped at 0 t = .
We thus need to solve
2
1 u u
r u
t r r r r
u u
u
v
c c c (
| |
=
|
(
c c c
\ .

with the above initial condition as well as the (no-slip) boundary condition
0 u
u
= at r a = for all 0 t > .
Using the separation of variables method:
, ) , ) , ) , u r t R r T t
u
=
we have
2
1 dT d dR
r R
Tdt R rdr r r
v
o
(
| |
= =
|
(
c
\ .

so that for 0 o = ,
, )
t
T t e
o
=
, )
1
R r J r
o
v
| |
=
|
\ .
where
n
J is the Bessel function defined by
, )
2
2
2
0
n
d d n
r k J kr
rdr dr r
(
| |
| |
+ =
( | |
\ .
\ .

[see M.Abramowitz, I.A.Stegun, Handbook of Mathematical Functions]
Hence
, )
1
,
t
u r t c e J r
o
u o
o
o
v

| |
=
|
\ .
_
To satisfy the no-slip condition at r a = , we need
n
a
o
i
v
= where
n
i are roots of
1
J , ie., , )
1
0
n
J i = .
Therefore
2
n
a
i
o v
| |
=
|
\ .
and
, )
2
1
1
,
n
t
a
n n
n
r
u r t c e J
a
i
v
u
i
| |

|
\ .
=
| |
=
|
\ .
_
The initial condition
u r
u
= O for r a s 0 t =
means
1
1
n n
n
r
r c J
a
i
=
| |
O =
|
\ .
_
Using the orthogonality of the Bessel functions:
, )
2
2
1
0
2
a
n m nm n
r r a
J J rdr J
a a
v v v v v v
i i o i
+
| | | |
= (
| |

\ . \ .
}
where
n v
i is the nth root of J
v
, we have
, )
2
2
2
1 2
0
2
a
n n n
r a
J r dr c J
a
i i
| |
O = (
|

\ .
}
Using
, ) , )
1
1
1
0
1
J ar r dr J a
a
v
v v
+
+
=
}
( see I.S.Gradshteyn, I.M.Ryzhik, Table of Integrals, Series, and Products, formula
6.561.5)
we have
, ) , )
2
2
3
2 2
1
2
n n n
n
a
a J c J i i
i
O = (

so that
, )
2
2
n
n n
a
c
J i i
O
=
whereupon
, )
, )
2
1
1 2
2
,
n
t
a
n
n n n
a r
u r t e J
J a
i
v
u
i
i i
| |

|
\ .
=
O
| |
=
|
\ .
_
Finally, using
, ) , ) , )
1 1
2
J x J x J x
x
v v v
v
+
+ =
we have
, ) , ) , )
0 2 1
2
0
n n n
n
J J J i i i
i
+ = = (
n
are roots of
1
J )
so that
, )
, )
2
1
1 0
2
,
n
t
a
n
n n n
a r
u r t e J
J a
i
v
u
i
i i
| |

|
\ .
=
O
| |
=
|
\ .
_
(2.33)
This equation overestimate the spin down time of a cup of tea by an order of
magnitude since theres no account for the effect of the bottom (see chapter 5).
2.4.4 Viscous Decay of Line Vortex
A line vortex
0
2 r
u
t
I
= u e (2.34)
The vorticity singularity at 0 r = is expected to diffuse in a viscous fluid.
Consider the circulation of a circular path C of radius r,
, ) , ) , )
2
0
, , 2 ,
C
r t d u r t rd ru r t
t
u u
u t I = = =
} }
u r

(2.35)
Eq(2.30)
2
1 u u
r u
t r r r r
u u
u
v
c c c (
| |
=
|
(
c c c
\ .

can be rewritten in terms of I as follows:
1
2
u
r
u
t
= I
1
2
u
t r t
u
t
c cI
=
c c
2
1 1 1
2
u
r r r r
u
t
c cI
| |
= I +
|
c c
\ .
2 2
2 3 2 2
1 2 2 1
2
u
r r r r r r
u
t
| | c cI c I
= I +
|
c c c
\ .
2
2
1 u u u
r
r r r r r r
u u u
c c c c
| |
= +
|
c c c c
\ .
2
3 2 2 2
1 2 2 1 1 1 1
2 2 r r r r r r r r r t t
| | cI c I cI
| |
= I + + I +
| |
c c c
\ .
\ .
2
3 2 2
1 1 1 1
2 r r r r r t
| | cI c I
= I +
|
c c
\ .
so that
2
3 2 2 3
1 1 1 1 1 1
2 2 2 r t r r r r r r
v
t t t
( | | cI cI c I
= I + I
( |
c c c
\ .
or
2
2
1
t r r r
v
| | cI cI c I
= +
|
c c c
\ .
(2.36)
For the line vortex decay, the initial condition is
, )
0
, 0 r I = I
The requirement of u
u
be finite at 0 r = afterwards means
, ) 0, 0 t I = for 0 t >
Eq(2.36) is invariant under the same scale transformation as (2.12), namely,
r r o
2
t t o
where is a constant. We therefore seek a similarity solution:
, ) , ) , r t f q I = where
r
t
q
v
=
As in the case of (2.13), we have
3
2
r
t
t
q
v
c
=
c
1
r t
q
v
c
=
c
3
' ' '
2
2
r
f f f
t t t
t
q q
v
cI c
= = =
c c
1
' ' f f
r r t
q
v
cI c
= =
c c
2
2
1 1
'' '' f f
r r t t
q
v v
c I c
= =
c c
so that (2.36) becomes
1 1 1
' ' ''
2
f f f
t r t t
q
v
v v
| |
= +
|
\ .
or
1
' ' ''
2
f f f
q
q
= +
' 1
' 2
df
d
f
q
q
q
| |
=
|
\ .
2
ln ' ln
4
f C
q
q = +
or
2
4
' f
Ce
q
q

=
2 2
4 4
f C e d ae b
q q
q q

= = +
}
Hence
2
4
r
t
ae b
v

I = +
Putting in the initial & boundary conditions, we have
0
b I =
0 a b = + so that
0
a b = = I
Hence
2
4
0
1
r
t
e
v

| |
I = I +
|
\ .
(2.37)
The circulation thus diffuses with a standard deviation of 2 t o v = . (see fig.2.12)
For very small r, ie., 2 r t v , we have
2
0
4
r
t v
I I
so that
0
8
r
u
t
u
tv
I (2.38)
which corresponds to uniform rotation with angular velocity
0
8 t tv
I
.
2.5 Convection & Diffusion of Vorticity
Applying the same technique that begets vorticity eq (1.25) to the Navier-Stokes eq
2
p
t

c
| |
+ V = V + V
|
c
\ .
u
u u u
we obtain the viscous vorticity eq.
, )
2
D
Dt
v = V + V

u (2.39)
For 2-D flow
, ) , ) , , , , , , 0 u x y t v x y t = (

u
, ) 0, 0,e =
we have (cf 1.27),
2
D
Dt
v = V

or
, )
2 2
2 2
t x y
e e e
e v
| | c c c
+ V = +
|
c c c
\ .
u (2.40)
Obviously, the term , ) V u

describes the vorticity convection while


2
vV its
diffusion.
2.5.1 2-D Flow Near Stagnation Point
Consider the irrotational flow pattern near a stagnation point as shown in fig.2.13 with
u x o = v y o = (2.41)
where 0 o > is a constant.
Obviously, the no-slip condition is not satisfied at the boundary 0 y = .
However, the mainstream flow speed | | | | u x o = increases in the flow direction
along the boundary. According to the Bernoullis theorem, the mainstream pressure
decreases in the flow direction along the boundary. Therefore, we expect a thin,
unseparated boundary layer with thickness
1
O
L
R
o
| |
=
|
\ .
.
Using
2
UL L
R
o
v v
= = , we have
, )
O
v
o
o
= .
The no-slip condition creates a vortex sheet at the boundary which diffuses into the
fluid. The existence of an unseparated boundary layer implies the convective effects
are sufficient to balance this diffusion and confine it within the layer. (Do Ex.2.14)
2.5.2 High R Flow Past Flat Plate
3. Waves
Ref: D.J.Acheson, Elementary Fluid Dynamics, Clarendon Press (90)
3.1 Introduction
Consider a simple harmonic surface wave:
, ) cos A kx t q e = (3.1)
Wave speed is dispersive:
g
c
k k
e
= (3.2)
so that waves with longer wavelengths
2
k
t
= travel faster.
The group velocity is defined as
g
d
c
dk
e
= (3.3)
For a dispersion given by (3.2), we have
1 1
2 2
g
d g
c gk c
dk k
= = = (3.5)
Hence, individual wave crests travel twice as fast as the group so that they continually
appear at the back of the group & disappear at the front.
Dispersion is also responsible for wake pattern behind moving boat.
Stationary wave patterns of flow past object contains both up & down stream
disturbances because of surface tension effects.
Waves at interface between 2 fluids caused by buoyancy effects have speed
1 2
1 2
g
c
k

=
+
(3.6)
where
1
and
2
are the densities of the fluids.
A variant of this is internal gravity waves that travel in a fluid with density that
varies with height (eg. the atmosphere).
Sound waves are due to compressibility and have speed
0
0
0
p
a

= (3.7)
which is non-dispersive.
3.2 Surface Waves on Deep Water
2-D water waves:
u = [ ( , , ), ( , , ), ] u x y t v x y t 0
Irrotational motion:
V =
c
c
c
c
c
c
= u
i j k
x y z
u v 0
0
c
c

c
c
=
v
x
u
y
0
Thus, there exists a velocity potential :
u
x
=
c
c

v
y
=
c
c

Incompressibility:
V = u 0
c
c
+
c
c
=
2
2
2
2
0

x y
Let wave on free surface be described by
y t = q( , ) x
3.2.1 Condition at Free Surface
Surface condition: Fluid particles on surface remain there.
Let
, ) , ) , , , F x y t y x t q =
the surface condition means that
, ) , , , 0 F x x t t q = 1
]
ie, 0 F = on surface. Thus
0
DF F
F
Dt t
c
= + V =
c
u on , ) , y x t q = (3.17)
Using
F
t t
q c c
=
c c
F
x x
q c c
=
c c
1
F
y
c
=
c
(3.17) becomes
0 u v
t x
q q c c
+ =
c c
on , ) , y x t q = (3.18)
3.2.2 Bernoullis Eq. for Unsteady Irrotational Flow
Consider the Euler eq. for incompressible, irrotational flow:
2
1
2
p
t
_

| | c
= V + +
|
c
\ .
u
u where gy _ =
Writing | = V u and integrate, we have
, )
2
1
2
p
G t
t
|
_

c
+ + + =
c
u (3.19)
which is called the Bernoullis eq. for unsteady irrotational flow.
Note that the arbitrary function G is doesnt affect the value of u.
The incompressibility condition in Eq(3.19) can be removed as follows. (see Landau)
Consider the specific (per unit mass) enthalpy h,
1
dh Tds vdp Tds dp

= + = +
we have, for adiabatic processes ( 0 ds = ),
1
p h

V = V
Therefore, (3.19) becomes
, )
2
1
2
h G t
t
|
_
c
+ + + =
c
u
3.2.3 Pressure Condition at Free Surface
For inviscid fluid, there is no surface tension.
Hence, the condition at a free surface is simply
0
p p = on , ) , y x t q =
where
0
p is the (usually constant) atmospheric pressure.
The unsteady Bernoulli eq thus becomes, with a proper choice of G,
, ) , )
2 2 0
0
1
0
2
p
u v g G t
t
|
q

c
+ + + =
c
on , ) , y x t q = (3.20)
3.2.4 Linearization of Surface Conditions
To 1
st
order in u, v, and , (3.20) becomes
0 g
t
|
q
c
+ =
c
on , ) , y x t q =
Now
, ) , ) , )
2
2
0 0
, , , , , ,
y y y
x y t x y t x y t
t t t
q
| | |
q
= = =
c c c
= + +
c c c

Since
t
| c
c
is already of order u, the true 1
st
order approximation of (3.20) is
0 g
t
|
q
c
+ =
c
on 0 y = (3.22)
Similarly, for (3.18):
0 v
t
q c
+ =
c
on 0 y =
Using v
y
| c
=
c
, we have
0
t y
q | c c
+ =
c c
on 0 y = (3.21)
3.2.5 Dispersion Relation
Eliminating from (3.21-22) gives
2
2
1
0
g t y
| | c c
+ =
c c
on 0 y =
which is clearly separable. Setting
, ) , ) , ) , , , x y t f y x t | = u
we have
, ) , )
0
' 0 ,
y
f x t
y
|
=
c
= u
c
, )
2 2
2 2
0
0
y
f
t t
|
=
c c u
=
c c
so that
, ) , )
2
2
1
0 ' 0 0 f f
g t
c u
+ u =
c
or
2
2
2
0
t
e
c u
+ u =
c
where
0
'
y
f
g
f
e
=
=
On the other hand, incompressibility means | satisfies the Laplace eq. inside the
fluid for all t:
2 2
2 2
0
x y
| | c c
+ =
c c
ie
2
2
'' 0 f f
x
c u
+ u =
c
Since f is a function of y only, we must have
2
2
2
''
const
f
k
x f
c u
= = =
uc
which, combining with the t eq., gives
2 2
2 2 2 2
1 1
t k x e
c u c u
= = u
c c
Thus, both the time and spatial components of are harmonic oscillators while
together, they obey the 1-D wave eq. with propagation speed c
k
e
= .
A general solution is of the form
, ) , ) , ) , sin sin x t C kx t C kx t e o e o
+ +
u = + + + +
where C, are arbitrary constants.
If we consider a simple wave propagating in the x + direction, we have
, ) , ) , sin x t C kx t e
+
u =
Now, f itself satisfies
2
'' 0 f k f =
so that
, ) , ) , ) exp exp f y B ky D ky = + (3.A)
Since the wave should die out towards the bottom, we must have
, ) 0 f y = (3.B)
so that 0 D = and
, ) , ) exp f y B ky =
whereupon
0
'
y
f
g gk
f
e
=
= = (3.26)
giving a wave speed
g
c
k k
e
= =
The potential itself is
, ) , ) , ) , , exp sin x y t C ky kx t | e = (3.25)
From (3.22), we have
, ) , ) , )
0
1
, cos cos
y
x t C kx t A kx t
g t g
| e
q e e
=
c
= = =
c
(3.23)
where
k
A C C
g
e
e
= =
3.2.6 Meaning of Small Amplitude
From (3.23,5), we have
, ) , ) , ) , ) exp cos exp cos u Ck ky kx t A ky kx t
x
|
e e e
c
= = =
c
, ) , ) , ) , ) exp sin exp sin v Ck ky kx t A ky kx t
y
|
e e e
c
= = =
c
, ) , ) sin sin Ck kx t Ak kx t
x g
q e
e e
c
= =
c
, ) , ) , ) , )
2
sin sin exp sin C kx t Ck kx t A ky kx t
t g
q e
e e e e
c
= = =
c
Thus, the linearization condition that leads to (3.21)
u v
x
q c
c

means
1
x
q c
c

ie.
1
2
A
k

t
=
the wave amplitude of the surface wave is much small than its wavelength .
For (3.22), we need
2 2
u v gq +
which means
2 2
A gA e
ie
2
1
2
g
A
k

e t
= =
3.2.7 Particle Path
In the Eulerian picture, the equations
, ) , ) exp cos u A ky kx t e e =
, ) , ) exp sin v A ky kx t e e =
give the velocity at each point , ) , x y at time t.
In the Lagrangian picture, the corresponding velocity of a fluid particle whose
trajectory is , ) , ) , x t y t 1
]
is described by
, )
, ) , ) exp cos
dx t
A ky t kx t t
dt
e e = 1 1
] ]
, )
, ) , ) exp sin
dy t
A ky t kx t t
dt
e e = 1 1
] ]
which can be integrated to give the particle path.
It is straightforward to prove that the ansatz
, ) , ) , ) exp sin x t x A ky kx t e =
, ) , ) , ) exp cos y t y A ky kx t e = +
is a solution to 1
st
order in A.
It describes a circular motion of radius , ) exp A ky about the average position
, ) , x y .
3.3 Dispersion
Given a dispersion relation
, ) k e e = (3.28)
the group velocity is defined as
g
d
c
dk
e
= (3.29)
Properties:
1. Waves of a single wavelength travels with phase velocity c
k
e
= .
2. Whenever superpositions of nearby wavelengths are substantial, the packet
travels with the group velocity
g
d
c
dk
e
= . This is true whether the packet is
isolated or just a Fourier component of a complicated excitation.
3. Energy is transferred at group velocity. (see Ex.3.12)
3.3.1 Motion of a Wave Packet
A general disturbance can be written as
, ) , ) , ) , exp x t a k i kx t dk q e

= 1
] }
(3.30)
where , ) a k is the Fourier amplitude and it is understood that only real quantities are
physically meaningful.
When , ) a k is narrowly centered at
0
k , the disturbance is called a wave packet.
Writing
, )
0 0
'
g
k k c e e + (3.31)
where
0
' k k k = + , , )
0 0
k e e = ,
0
0 g
k k
d
c
dk
e
=
= (3.32)
we have
, ) , )
0 0 0
' '
g
kx t k k x k c t e e + +
, )
0 0 0
'
g
k x t k x c t e +
so that (3.30) becomes
, ) , ) , ) , )
0 0 0 0
, exp ' exp ' '
g
x t i k x t a k k ik x c t dk q e

1
+ 1
]
]
}
(3.33)
3.3.2 Gaussian Packet
Consider a Gaussian wave packet with standard deviation :
, )
, )
2
0
1
exp
2 2
k k
a k
o to
1

= (
(
]
The integral in (3.33) becomes
, )
2
0
1 '
exp ' '
2 2
g
k
I ik x c t dk
o to

1
= +
(
]
}
, ) , )
2
2
0 0
1 1
exp exp ' '
2 2 2
g g
x c t k i x c t dk
o
o
o to

1
1
= +
`
( ]
) ]
}
, )
2
0
exp
2
g
x c t
o
1
=
(
]
so that
, ) , ) , )
2
0 0 0
, exp
2
g
x t i k x t x c t
o
q e
1

(
]

which shows a wave packet with an envelope moving with velocity


0 g
c .
3.3.3 Large Time Response to Localized
Disturbance
After a sufficiently long time, different Fourier components of a local disturbance will
be greatly dispersed. Whats left is a sinusoidal wavetrain with both k and slowly
varying functions of , ) , x t .
For surface waves on deep water,
g
c
k
= so that waves with longer wavelengths
travel faster, one expects the wavelength to increase with x. (see fig 3.8)
Let the slowly varying wave train be
, ) , ) , ) , , exp , x t A x t i x t q u = 1
]
(3.34)
where , ) , x t u is a phase function.
The local wavenumber k and frequency are defined by
k
x
u c
=
c t
u
e
c
=
c
(3.35)
Now
2 2
k
t t x x t x
u u e c c c c
= = =
c c c c c c
or
0
k
t x
e c c
+ =
c c
(3.36)
Hence
0
k d k
t dk x
e c c
+ =
c c
or
0
g
c k
t x
c c
| |
+ =
|
c c
\ .
(3.37)
ie., k is constant for an observer moving with velocity
g
c .
This can also be seen from the fact that
, )
g
k f x c t = (3.38)
satisfies (3.37) and is obviously constant for const
g
x c t = .
3.3.4 Surface Waves on Deep Water
gk e = (3.39)
1
2
g
g
c
k
=
After a long time, any local disturbance becomes locally sinusoidal with wavenumber
k in the neighborhood of distance
g
x c t = (3.40)
from the initial disturbance region so that
2
2
4 4
g
g g t
k
c x
| |
= =
|
\ .
(3.41)
2
gt
x
e =
and hence
2
2
4
gt
x x
u c
=
c 2
gt
t x
u c
=
c
which admits a solution
2
4
gt
x
u c = +
where is a constant. Eq(3.34) thus becomes
, ) , )
2
, , exp
4
gt
x t A x t i
x
q c
1
| |
= +
( |
\ .
]
(3.42)
3.4a Laplaces Formula
Source: Landau
Consider 1
st
the interface between 2 inviscid fluids.
Let the principal radii of curvature at a point on the interface be
1
R and
2
R . ( 0 R >
if it is drawn into fluid 1)
The corresponding line elements there would be
1 1 1
dl R du = and
2 2 2
dl R du = where
d denotes the angle subtended by the line element. A surface element there is
1 2
dS dl dl = .
Let the interface be moved by an infinitesimal amount or .
Both
1
R and
2
R would be changed by an amount n o r , where n the interface
normal pointing from fluid 1 to fluid 2.
Line element
1
dl would become

, )

, )
1
1 1 1 1
1 1
1
1
dl
R n d R n dl n
R R
o u o o
| |
+ = + = +
|
\ .
r r r
and similarly for
2
dl
The surface element dS then becomes

1 2
1 2
1 1
1 1 dl n dl n
R R
o o
| | | |
+ +
| |
\ . \ .
r r
The change induced in dS is therefore

1 2
1 1
ndS
R R
o
| |
+
|
\ .
r
This will change the surface energy by

1 2
1 1
ndS
R R
o o
| |
+
|
\ .
}
r
where is the surface tension coefficient.
On the other hand, the work required to be done against the pressure difference is
, )
2 1

S
p p ndS o
}
r
In equilibrium, the total change of energy must be zero so that we have

2 1
1 2
1 1
0 p p ndS
R R
o o
1
| |
+ + =
( |
\ .
]
}
r
ie.
1 2
1 2
1 1
p p
R R
o
| |
= +
|
\ .
For viscous fluids, this is easily generalized to
, )

2 1
1 2
1 1
n n
R R
o o o
| |
= +
|
\ .
where the stress tensor is defined by
' pI o o = +
3.4b Curves
Source: T.M.Apostol, Calculus, vol I.
Consider a curve , ) C t with parameter t in an n-dimensional space.
If C is the path of a particle, it is usually denoted in vector notation as , ) t r .
The tangent (velocity) of C is defined by

d
vv
dt
= =
r
v
where v = v is the speed (magnitude of v) and v is the unit tangent.
Assuming Euclidean metric for all vectors, the path length can be written as
, ) , )
0
t
s t v t dt =
}
so that
ds
v
dt
=
The principal normal is defined by
dv
dt
= n
or
2
1 1

dv d d dv dv
v nn
dt dt v v dt v dt v dt
| | | |
= = = = =
| |
\ . \ .
v v v
n a

where n is the unit normal and a the acceleration.
A more familiar way to write this is
, )

d vv d dv dv
v v
dt dt dt dt
= = = +
v
a
Consider the rate of v with respect to s:

dv dt dv n
n n
ds ds dt v v
k = = = = n

where
1 n dv
v v dt
k = =
is called the curvature.
The radius of curvature is
1
R
k
=
3.4b.1 Circle
Consider curve , ) , ) cos , sin t a t a t e e = r which describes a circular path of radius a
in the x-y plane.
The tangent (velocity) is , ) , ) sin , cos t a t a t e e e e = v .
The unit tangent is , ) , )
sin , cos v t t t e e = .
The speed is v ae = .
The path length is , )
0
t
s t a dt a t e e = =
}
.
The normal is , ) , )

cos , sin
dv
t t t
dt
e e e e = = n .
The unit normal is , ) , )
cos , sin n t t t e e =
, )
1 1 1
cos , sin
dv
t t n
ds v a a
e e = = = n

The curvature is
1
a
k = .
The radius of curvature is
1
a
k
= .
3.4b.2 Graph
Consider a graph of , ) f t vs t.
The foregoing analysis can be applied here using
, ) , ) , t t f t = 1
]
r
Hence
The tangent (velocity) is , ) , ) 1, ' t f = v where '
df
f
dt
= .
The unit tangent is , ) , )
2
1
1, '
1 '
v t f
f
=
+
.
The speed is
2
1 ' v f = + .
The path length is , )
2
1 ' s t f dt = +
}
.
The normal is , ) , )
2
1
1, '
1 '
dv d
t f
dt dt
f
1
= = (
+ (
]
n .
, ) , )
2
3 3
2
2 2 2 2
' '' ' '' ''
,
1 '
1 ' 1 '
f f f f f
f
f f
| |
|
= +
|
+ |
+ +
\ .
, )
, )
3
2 2
''
',1
1 '
f
f
f
=
+
The unit normal is , ) , )
2
1
',1
1 '
n t f
f
=
+
, )
, )
, )
2 3
2
2 2
1 '' ''
',1
1 '
1 '
dv f f
f n
ds v
f
f
= = =
+
+
n

The curvature is
, )
3
2 2
1 ''
1 '
f
R
f
k = =
+
.
The radius of curvature is
, )
3
2 2
1 '
1
''
f
R
f k
+
= = .
3.4 Surface Tension & Capillary Waves
For our 2-D wave problem for an inviscid fluid, the boundary condition on the free
surface when surface tension effects are included is (see 3.4a and b)
2
0 2
1
p p
R x
q
o o
c
=
c
(
2
x
q c
| |
|
c
\ .
neglected)
where is the coefficient of surface tension,
0
p the (constant) atmospheric pressure
and R the radius of curvature of the surface ( 0 R > if it is drawn inside the fluid).
Eq(3.22) thus generalized to
2
2
0 g
t x
| o q
q

c c
+ =
c c
on 0 y = (3.44)
while the other eqs remain unchanged, ie.
0
t y
q | c c
+ =
c c
on 0 y = (3.21)
2 2
2 2
0
x y
| | c c
+ =
c c
Eliminating gives
2 3
2 2
0 g
t y x y
| | o |

c c c
+ =
c c c c
on 0 y =
Setting
, ) , ) , ) , , , x y t f y x t | = u
gives
, ) , )
2 2
2 2
0 ' 0 0 f f g
t x
o

| | c u c u
+ u =
|
c c
\ .
(A)
while the Laplace eq. still gives
2
2
2
''
const
f
k
x f
c u
= = =
uc
with solution , ) , ) exp f y B ky = as before.
Hence (A) becomes
2 2 2
2
2 2 2 2
0
g
k g k k
t t k x
o o

| | | | c u c u c u
+ + u = = +
| |
c c c
\ . \ .
and the wave speed is
g
c k
k
o

= + (3.46)
so that
3
kc k gk
o
e

= = + (3.45)
2
3
3
2
g
k g
d
c
dk
k gk
o
e
o

+
= =
+
(3.47)
For large k (short wavelengths), we have
k
c
o

3 3
2 2
g
k
c c
o

=
Such waves are called capillary waves or ripples. In contrast with the gravity waves,
they travel faster if their wavelengths are shorter. Furthermore, the group velocity is
greater than the corresponding phase velocity. Thus, wave crests in a wave packet
travel backwards.
In general, we have capillary-gravity waves. (see fig 3.11 for dispersion)
Salient points are:
1. there is a minimum phase velocity
1
4
min
4g
c
o

| |
=
|
\ .
at
g
k

o
= .
2. there is both a capillary & a gravity wave for each phase velocity.
3.5 Finite Depth Effects
The effect of a finite depth h is to change the boundary condition satisfied by
, ) , ) , ) exp exp f y B ky D ky = + (3.A)
to
, ) 0
y h
df
v h
dy
=
= = (3.C)
ie
, ) , ) exp exp 0 B kh D kh =
, ) exp 2
D
kh
B
=
, ) , ) , ) exp exp 2 f y B ky kh ky = + 1
]
, ) , )
, }
exp exp
kh
Be k y h k y h

= + + + 1 1
] ]
, ) 2 cosh
kh
Be k y h

= + 1
]
so that (3.25) is replaced by
, ) , ) , ) , , cosh sin x y t C k y h kx t | e = + 1
]
(3.D)
The dispersion relation (3.26) becomes
, )
, )
, )
0
sinh '
tanh
cosh
y
k kh f
g g gk kh
f kh
e
=
= = = (3.51)
so that the phase speed is (see fig 3.12)
, ) tanh
g
c kh
k k
e
= = (3.52)
which, for h large, becomes
g
c
k
(3.53)
as it should.
For small h, we have
g
c kh gh
k
= (3.54)
which is non-dispersive.
3.6a Thermodynamics of Ideal Gas
An ideal gas is defined by 2 eqs.
A B
Pv RT N k T = =
0 v
u c T =
where v is the molar volume, u the molar energy. The molar heat capacity at constant
volume
0 v
c is a function of T only. (Note: specific heat is heat capacity per unit
mass)
Joules
8.3144 universal gas constant
mole Kelvin
R = =

23
6.023 10 Avogardro's number
A
N = =
23
Joules
1.381 10 Boltzmann's constant
Kelvin
B
k

= =
From 1
st
law:
du Tds Pdv =
or
du P
ds dv
T T
= +
which, for an ideal gas, becomes
0 v
c R
ds dT dv
T v
= +
and integrated to
0
0
0
0
ln
T
v
T
c v
s s dT R
T v
| |
= +
|
\ .
}
The T integral cannot be done until the functional form of
0 v
c is known.
Assuming
0 v
c to be a constant, we have
0
0
0 0
ln
v
c
R
T v
s s R
T v
1
| |
(
=
|
(
\ .
(
]
Next, consider the molar enthalpy
h u Pv = +
which, for an ideal gas can be written as
0 p
h c T =
where the molar heat capcity at constant pressure
0 p
c is a function of T.
Thus
0 0 p v
c T c T RT = +
so that
0 0 p v
c c R =
or
0
1
v
R
c
= where
0
0
p
v
c
c
=
Hence, for
0 v
c independent of T:
1
1
1
0
0 0 0 0
ln ln
1
T v R T v
s s R
T v T v

1
1
| | | |
(
= = (
| |
(

( \ . \ .
]
(
]
For an adiabatic process, 0 s A = so that
1
Tv const

=
Other forms of this are
Pv P const

= =
3.6 Sound Wave
For an inviscid, compressible fluid, we have
Eulers eq.: p
t

c
| |
+ V = V
|
c
\ .
u
u u (3.55)
Eq. of continuity: , ) 0
t


c
+ V =
c
u (3.56)
Note that an implicit assumption for inviscid fluid is that all processes are adiabatic.
(see Landau, 2).
Let the fluid be an ideal gas. Adiabatic processes satisfy
p const

=
0
0
p
v
c
c
=
or
, )
0
D
p
Dt

= (3.57)
3.6.1 Small Amplitude Waves
Let the unperturbed state be a homogeneous one at rest.
For small perturbations, we can write
1
= u u
0 1
p p p = +
0 1
= + (3.58)
where subscript 0 denotes unperturbed quantities and 1 the (small) perturbations.
Since the unperturbed state is homogeneous,
0 0
p


is independent of position.
For adiabatic perturbations, p

remains
0 0
p


for each fluid element and is
therefore also independent of position. Hence
, ) , )
0 1 0 1 0 0
p p p


+ + =
ie
1 1
0 0
1 1 1
p
p

| || |
+ + =
| |
\ .\ .
which, to 1
st
order of perturbation, gives
1 1 1 1
0 0 0 0
1 1 1 1
p p
p p



| || |
+ + + =
| |
\ .\ .

ie
1 1
0 0
0
p
p

=
which, in terms of the sound velocity (see later),
0
0
0
p
a

= (3.59)
can be written as
2
1 0 1
p a = (3.60)
To 1
st
order of perturbation, the Euler eq (3.55) becomes
1
0 1
p
t

c
= V
c
u
(3.61)
and the eq of continuity (3.56),
1
0 1
0
t


c
+ V =
c
u (3.62)
Eliminating
1
u from (3.61-62), we have
, )
2
2
1 0 1 1 2
p
t t

c c
V = V =
c c
u
which, with the help of (3.60), becomes
2
2
1 1 2 2
0
1
0 p p
a t
c
V =
c
(3.63)
or
2
2
1 1 2 2
0
1
0
a t

c
V =
c
Thus, both the pressure and density variation are governed by the wellknown wave
equation.
2
2
2 2
1
0
c t

| | c
V =
|
c
\ .
whose general solution is of the form
, ) , ) , ) , t t = = x x v

so that
i
i
i i
v
t t


c c c c
= =
c c c c
2 2
2 i j i j
i j i j
v v v v
t


| |
c c c c
= =
|
|
c c c c c
\ .
j
ij
i i j j i
x x


o

c
c c c c
= = =
c c c c c
2 2
i i i i
x x


c c
=
c c c c
, )
1 1
p p t = x v with
0
a = v
so that the perturbation, called sound, travels with speed
0
a .
3.6.2 Spherical Waves
For spherically symmetric waves, , ) , r t = . The wave eq. becomes
2
2
2 2 2
1 1
0 r
r r r c t
c c c
| |
=
|
c c c
\ .
(3.65)
Setting
, ) , )
1
, , r t h r t
r
=
we have
2
1 1
h
r r r r
c c
| |
=
|
c c
\ .
2 2
2 2 2 2 3
1 1 1 2 2
h h
r r r r r r r r r r
| | c c c c c 1
| |
= = +
| |
(
c c c c c
\ .
] \ .
2 2
2
2 2 2
1 2 1
r h
r r r r r r r r
c c c c c
| |
= + =
|
c c c c c
\ .
so that (3.65) simplifies to a 1-D wave eq. for h:
2 2
2 2 2
1
0
h h
r c t
c c
=
c c
with solution
, ) , ) , ) , h r t h r ct h r ct
+
= + +
so that
, ) , ) , )
1
, r t h r ct h r ct
r

+
= + + 1
]
(3.66)
The h
+
and h

obviously denote outgoing and incoming waves, respectively.


The radiation condition simply sets 0 h

= .
3.7 Supersonic Flow Past a Thin Aerofoil
Consider a 2-D inviscid, compressible, adiabatic flow past an aerofoil of the form
, )
1 1
, , 0 U u v = + u
0 1
p p p = +
0 1
= + (3.67)
As in 3.6.1, we consider only relations up to 1
st
order in perturbed quantities.
The adiabatic relation is the same as in 3.6.1:
2
1 0 1
p a = (3.68)
The Euler eqs.
, ) , ) , ) , )
0 1 1 1 1 0 1
U u v U u p p
t x y x

1 c c c c
+ + + + + = +
(
c c c c
]
, ) , ) , )
0 1 1 1 1 0 1
U u v v p p
t x y y

1 c c c c
+ + + + = +
(
c c c c
]
simplify to
1
0 1
p
U u
t x x

c c c
| |
+ =
|
c c c
\ .
1
0 1
p
U v
t x y

c c c
| |
+ =
|
c c c
\ .
which, for steady flow, become
1 1
0
u p
U
x x

c c
=
c c
1 1
0
v p
U
x y

c c
=
c c
Combining these 2 eqs gives
2 2 2
1 1 1
0 0 2
u p v
U U
x y x y x

c c c
= =
c c c c c
ie.
1 1
0
0
u v
U
x y x

| | c c c
=
|
c c c
\ .
which says the vorticity
1 1
u v
y x
e
c c
=
c c
is independent of x.
On the other hand, 0 e = for the uniform flow far away from the aerofoil. Hence,
0 e = everywhere. The flow is irrotational (potential) so that we can write
1
u
x
| c
=
c
1
v
y
| c
=
c
(3.69)
The Euler eqs thus become
0 1
0 U p
x x
|

c c
| |
+ =
|
c c
\ .
0 1
0 U p
y x
|

c c
| |
+ =
|
c c
\ .
which means
1 0
p U C
x
|

c
= +
c
(3.70)
where C is a constant. Since
1
0 p = when
1
0 u
x
| c
= =
c
, we must have 0 C = .
Similarly, the eq of continuity
, ) 0
t


c
+ V =
c
u
ie
, )
, ) , ) , )
0 1
0 1 1 0 1 1
0 U u v
t x y


c + c c
+ + + + + = 1 1
] ]
c c c
becomes, to 1
st
order perturbation,
1 1 1 1
0
0
u v
U
t x x y


| | c c c c
+ + + =
|
c c c c
\ .
which, for steady state, simplifies to
1 1 1
0
0
u v
U
x x y


| | c c c
+ + =
|
c c c
\ .
which, in terms of | , becomes
2 2
1
0 2 2
0 U
x x y
| |

| | c c c
+ + =
|
c c c
\ .
Putting in the adiabatic condition (3.68), we have
2 2
1
0 2 2 2
0
0
U p
a x x y
| |

| | c c c
+ + =
|
c c c
\ .
which, by (3.70), becomes
2 2 2 2
0
0 2 2 2 2
0
0
U
a x x y
| | |

| | c c c
+ + =
|
c c c
\ .
ie
, )
2 2
2
2 2
1 0 M
x y
| | c c
+ =
c c
(3.71)
where
0
U
M
a
=
is called the Mach number for the flow.
3.8 Internal Gravity Waves
Incompressible, inviscid, fluids with stratified density variation can support internal
gravity waves which are driven by buoyancy forces.
One example is salty water with vertical salt concentration variation.
Consider the 2-D case where
, ) , ) , , , , , , 0 u x y t v x y t = 1
]
u
In the static equilibrium state, we assume
, )
0
y = with
0
' 0 <
so that the density increases as one goes down toward the bottom.
The corresponding pressure distribution is given by the steady state Euler eqs.,
0
0
p
x
c
=
c
0
0
0
p
g
y

c
=
c
so that by a suitable choice of the origin, we have
, )
0 0
p y gy =
Since the fluid is incompressible & inviscid, the relevant eqs are
p
t

c
1
+ V = V +
(
c
]
u u g
0 V = u (3.80)
, ) , ) 0
t t


c c
+ V = + V =
c c
u u
which, in our choice of coordinates, becomes
, )
, )
0 1
0 1 1 1 1
p p
u v u
t x y x

c + | | c c c
+ + + =
|
c c c c
\ .
, )
, )
, )
0 1
0 1 1 1 1 0 1
p p
u v v g
t x y y

c + | | c c c
+ + + = +
|
c c c c
\ .
1 1
0
u v
x y
c c
+ =
c c
, )
1 1 0 1
0 u v
t x y

1 c c c
+ + + =
(
c c c
]
For 1
st
order perturbations, these simplify to
1 1
0
u p
t x

c c
=
c c
1 1
0 1
v p
g
t y

c c
=
c c
1 1
0
u v
x y
c c
+ =
c c
(3.81)
1 0
1
0 v
t y
c c
+ =
c c
Consider a Fourier mode of the form
, ) , ) , ) , , exp a x y t A y i kx t e = 1
]
(3.82)
where a can be
1 1 1
, , u v p , or
1
. Eq(3.81) becomes
0 1 1
i U ikP e =
1
0 1 1
dP
i V R g
dy
e =
1
1
0
dV
ikU
dy
+ = (3.83)
0
1 1
0
d
i R V
dy

e + =
Using ' to denote
d
dy
, the 3
rd
, 1
st
& last eqs give
1 1
'
i
U V
k
=
0 0
1 1 1 2
' P U i V
k k
e e
= =
1
1 0
'
V
R
i

e
=
so that the 2
nd
become
, )
1
0 1 0 1 0 2
' ' '
V
i V i V ig
k
e
e
e
=
which simplifies to
, )
0
1 0 1 0 1 1 2 2
0 0
1 '
' ' '' V V V g V
k


e
= +
or
2 0 0
1 1 1 2
0 0
' '
'' ' 1 0 V V k g V

e
| |
+ + =
|
\ .
2 2
2
1 1 1 2
'' ' 1 0
N N
V V k V
g e
| |
+ =
|
\ .
(3.84)
where the buoyancy number N is defined by
2
0
0
' 0
g
N

= > (
0
' 0 < )
The simplest case is when
0
is exponential, ie
, )
0
exp
y
y A
H

| |
=
|
\ .
so that
2
g
N
H
=
is a constant. So are the coefficients Eq(3.84), the solution of which is simply
, )
1
exp V B y =
where satisfies
2 2
2 2
2
1 0
N N
k
g

e
| |
+ =
|
\ .
or
2 2
2
1
1 0
g
k
H H

e
| |
+ =
|
\ .
ie.
2
2
1 1
1 1 4
2 2
g
Hk H il
H H

e

| |
= = +
`
|
\ .

)
where
2
2
1
4 1
2
g
l Hk H
H e
| |
=
|
\ .
(3.84A)
The solution (3.82) thus become
, ) , )
1
1
, , exp
2
v x y t B il y i kx t
H
e

| |
= + +
`
|
\ . )
, ) exp exp
2
y
B i kx ly t
H
e
| |
= + 1
| ]
\ .
(3.86)
which is a plane wave with wavevector , ) , , 0 k l = k and amplitude that decays
exponentially as it rises.
Using
1 1
1 0
' exp
V V y
R i
i H H

e e
| |
= =
|
\ .
, we have
, ) , )
1
, , exp exp
2
B y
x y t i i kx ly t
H H
e
e
| |
= + 1
| ]
\ .
(3.88)
Note that
1
is usually more accessible to experimental observation.
Solving for from (3.84A), we get the dispersion relation:
2
2 2 2
1
4
g l H
H
Hk k e
= + +
2
2 2
2
2
2 2 2 2
2 2
2 2
1 1
1
4 4
4
gk
g N k
H
l H
k l k l
H
H H
Hk k
e = = =
+ + + +
+ +
(3.87)
which is clearly anisotropic.
If the wavelength
2 2
2 2
k l
t t
= =
+
k
is small compared with the scale height H, (3.87) simplifies to
2 2
2
2 2
N k
k l
e
+
(3.89)
2 2
Nk
k l
e
+

which gives a group velocity


, ,
g k
k l m
e e e
e
c c c
| |
= V =
|
c c c
\ .
c , ) , , k l m = k (3.90)
, ) , )
2
2 2 3 3
2 2 2 2
1
, , 0
k kl
N
k l
k l k l
| |
|
=
|
+
|
+ +
\ .
, )
, )
3
2 2
, , 0
Nl
l k
k l
=
+
, )
, )
2 2
, , 0
l
l k
k l k
e
=
+
(3.91)
Now, the phase velocity is
, )
2 2

, , 0
p
k k l
k
k k l
e e
= =
+
c
so that
0
g p
= c c
ie.,
g p
c c . (see fig.3.13)
3.9a Partial Differential Eqs.
Source: F.B.Hildebrand, Advanced Calculus for Applications, 2
nd
ed, Chapter 8,
Prentice Hall (1976)
nth order ODE:
n independent arbitrary constants.
linear ODE: General solution = linear combination of n independent functions.
Non- linear ODE: may exist singular solution with no arbitrary constants.
nth order PDE:
General solution = arbitrary functions of specific functions (arguments).
In case of specific boundary conditions, its usually easier to determine a set of
particular solutions & combine them to satisfy the BCs.
3.9a1 Quasi-Linear Eq. of 1
st
Order
Most general quasi-linear eq. of 1
st
order:
, ) , ) , ) , , , , , ,
z z
P x y z Q x y z R x y z
x y
c c
+ =
c c
(15)
Linear case:
, ) , ) , ) , )
1 2
, , , ,
z z
P x y Q x y R x y z R x y
x y
c c
+ = +
c c
(16)
Let
, ) , , u x y z c = (17)
defines a solution (integral surface) of (15). Treating z as , ) , z x y , the chain rule
gives:
0
u u z
x z x
c c c
+ =
c c c
0
u u z
y z y
c c c
+ =
c c c
(18)
or
u
z
x
u
x
z
c
c
c
=
c
c
c
u
z
y
u
y
z
c
c
c
=
c
c
c
(19)
which, when substituted into (15), gives
0
u u u
P Q R
x y z
c c c
+ + =
c c c
(20)
Introducing a vector , ) , , P Q R = V in the , ) , , x y z space, (20) becomes
0 u V = V (21)
Since u V is normal to the integral surface (17), V is tangent to a curve on the latter.
Conversely, a curve with tangent parallel to V everywhere must be on the integral
surface. Such curves are called characteristic curves of the differential eq.
Any curve can be denoted by the position vector , ) s = r r where s is the arc length.
The unit tangent to the curve is
d dx dy dz
ds ds ds ds
= + +
r
i j k
If it is a characteristic curve, we must have
dx
P
ds
=
dy
Q
ds
=
dz
R
ds
= (22)
where can be any function of , ) , , x y z .
Eq(22) can be written as
dx dy dz
P Q R
= = (23)
Let the independent solutions to (23) be
, )
1 1
, , u x y z c = , )
2 2
, , u x y z c = (24)
where
1 2
, c c are constants. Any surface specified as
, )
1 2
, 0 F u u = (25a)
or
j
2 1
u f u = (25b)
will be an integral surface of (15).
3.9a2 Initial Conditions
For the linear case
, ) , ) , ) , )
1 2
, , , ,
z z
P x y Q x y R x y z R x y
x y
c c
+ = +
c c
(35)
One of the eqs in (23), namely,
dx dy
P Q
=
is an ordinary differential eq.
Let its solution be
, )
1 1
, u x y c = (35a)
which is called a characteristic cylinder. (The intersection of this cylinder with the
xy plane is called a characteristic base curve.)
Eq(35a) can be used to express y in terms of x and
1
c so that another eq of (23), say,
1 2
dx dz
P R z R
=
+
ie.
1 2
dz R R
z
dx P P
= +
is just an ODE involving , ) z x .
The solution to this can be written as
, ) , )
2 1 1 2 1 2
, , u z x c x c c o o = + =
or
, ) , )
2 1 2 2
, , u z x y x y c | | = + =
The general solution to (35) is therefore
, )
, )
, )
, )
2
1
1 1
, 1
,
, ,
x y
z f u x y
x y x y
|
| |
= 1
]
or
, ) , ) , )
1 2 3
, , , z s x y f s x y s x y = + 1
]
(36)
For a 1
st
order ODE, the arbitrary integration constant can be determined by requiring
the integral curve to go through a specific point in the xy plane.
For a 1
st
order PDE, the arbitrary integration function in (25) can be determined by
requiring the integral surface to include a specific curve in the xyz space.
Let this curve be given by the intersection of 2 surfaces
, )
1
, , 0 x y z = , )
2
, , 0 x y z = (42)
and the independent solutions to (23) are
, )
1 1
, , u x y z c = , )
2 2
, , u x y z c = (43)
The elimination of , , x y z then results in an eq of the form , )
1 2
, 0 F c c = .
The required integral surface is then simply , )
1 2
, 0 F u u = .
3.9a3 Characteristics of Linear 1
st
Order Eqs.
Any curve curve in the xyz space can be specified as
C: , ) x x = , , ) y y = , , ) z z = (90)
The projection of this onto the xy plane is
0
C : , ) x x = , , ) y y = , 0 z = (91)
The prescribed curve (42) can alternatively be given as a specification of z on
0
C in
the form of (90).
The question is whether the PDE has an integral surface that actually contains C.
Now, on
0
C , everything, including z, is a known function of . Hence
dz z dx z dy
d x d y d
c c
= +
c c
(94)
Also, the PDE itself becomes an eq of independent variable .
, ) , ) , ) , )
1 2
z z
P Q R z R
x y

c c
+ = +
c c
(93)
These 2 eqs can be considered as simultaneous eqs for
z
x
c
c
and
z
y
c
c
on
0
C .
The condition for the existence of a unique solution is
0
P Q
dy dx
P Q
dx dy
d d
d d


= = (95)
ie.
dx dy
P Q
= on
0
C (96)
In other words, a unique solution that satisfies the boundary condition prescribed on a
curve
0
C exists only if
0
C is no where tangent to a characteristic base curve.
Now, in case
0
C is itself a characteristic base curve, the determinant in (95) vanishes
identically:
0
P Q
dx dy
d d
= (96a)
Therell either be no or infinitely many solutions to (93) & (94). For the latter case,
we need, according to Cramers rule,
1 2
0
R z R Q
dz dy
d d
+
= or
1 2
0
P R z R
dx dz
d d
+
= (96b)
This means a curve C that satisfies eqs (96a,b) must be a characteristic curve.
Indeed, this provides a way to find characteristic curves for more complicated PDEs.
3.9 Finite-Amplitude Waves in Shallow Water
Let the bottom of the fluid be at 0 y = ; its free surface, , ) , y h x t = .
Let
0
h be some typical value of , ) , h x t , and L be a typical horizontal length scale.
The shallow water approximation means
0
h L (3.92)
The full 2-D eqs for ideal fluid are
1 u u u p
u v
t x y x
c c c c
+ + =
c c c c
(3.93)
1 v v v p
u v g
t x y y
c c c c
+ + =
c c c c
(3.94)
0
u v
x y
c c
+ =
c c
(3.95)
The shallow water approximation means (see 3.9.1 for justification)
Dv
g
Dt

so that (3.94) simplifies to
1
0
p
g
y
c
=
c
which can be integrated immediately to give
, ) , p gy f x t =
The integration constant , ) , f x t can be determined by the boundary condition that
we have atmospheric pressure
0
p at the free surface , ) , y h x t = . Hence
, )
0
, p p g y h x t = 1
]
With this, (3.93) becomes
Du h
g
Dt x
c
=
c
which means the change of u for each fluid element is independent of y.
Hence, if u is independent of y at some time, it will be so for all time.
In which case, (3.93) simplifies to
u u h
u g
t x x
c c c
+ =
c c c
(3.96)
and (3.95) can be integrated to give
, ) ,
u
v y f x t
x
c
= +
c
The integration constant , ) , f x t can be determined by the boundary condition at the
bottom, ie.,
0 v = at 0 y =
so that 0 f = and
u
v y
x
c
=
c
(3.96a)
As shown in 3.2.1, the kinematic condition at the free surface is
h h
u v
t x
c c
+ =
c c
on , ) , y h x t = (3.18)
Combining with (3.96a), we have
0
h h u
u h
t x x
c c c
+ + =
c c c
on , ) , y h x t = (3.97)
Eqs(3.96-7) are known as the shallow water equations.
3.9.1 Shallow Water Equations
The shallow water eqs are (see Acheson 3.9)
0
t x x
u uu gh + + = (3.96)
0
x t x
hu h uh + + = (3.97)
where subscripts denote partial derivatives.
Let u, h be specified on curve , )
0
C .
We have
' ' '
t x
u u t u x = +
' ' '
t x
h h t h x = +
where ' denotes derivative with respect to .
Rewriting these as a set of simultaneous eqs for , , ,
t x t x
u u h h , we have
1 0 0
0 1 0
' ' 0 0 '
0 0 ' ' '
t
x
t
x
u u g
u h u
h t x u
h t x h
| | | | | |
| | |
| | |
=
| | |
| | |
\ . \ . \ .
The determinant of coefficients is
1 0
' 0 0 ' 1
0 ' ' 0 ' '
h u u g
D x t h u
t x t x
= +
, ) , ) ' ' ' ' ' ' ' x x ut t u x ut ght = + + 1
]
, )
2 2 2
' 2 ' ' ' x x t u t gh u = + +
Setting 0 D = gives
, )
2 2 2 2
' ' ' ' x ut u t t gh u = +
, )
' ' x t u gh =
dx
u gh
dt
= (a)
which give us the characteristic base curves
0
C .
According to Cramers rule, the condition for infinitely many solutions for
x
u is
1 0 0
0 0 1
0
' ' 0 0
0 ' ' '
g
u
t u
h t x
=
ie
0 1 0 0
' 0 0 ' 0 1 ' ' ' ' ' ' 0
' ' ' ' ' '
u g
u t u u x uu t t gh
h t x h t x
+ = + =
'
' '
'
h
x t u g
u
| |
=
|
\ .
dx dh
u g
dt du
= (b)
Combining (a) and (b) gives
dh
gh g
du
=
ie
2
dh
du g d gh
h
= =
2 u gh const = (c)
which are satisfied on the characteristic base curves.
Following Acheson, we define
c gh = (3.98)
so that the characteristic base curves (a) become
dx
u c
dt
= (3.103)
on which 2 u c const = .
Consider the functions
2 u c = + and 2 u c q =
so that
, )
1
2
u q = + , )
1
4
c q =
, )
2
2
1 1
16
h c
g g
q = =
and
, ) , )
1
8
t t t
h
g
q q =
The shallow water eq (3.96) thus become
, ) , ) , ) , ) , )
1 1 1
0
2 4 8
t t x x x x
q q q q q + + + + + =
which simplifies to
, )
1
3 3 0
4
t t x x x x
q qq q q + + + + + =
, ) , ) , ) 4 3 3 0
t t x x
q q q q + + + + + = (d)
Similarly, (3.97) becomes
, ) , ) , ) , ) , ) , ) , )
2 1 1 1
0
32 8 16
x x t t x x
g g g
q q q q q q q + + + + =
, ) , ) , ) , ) , ) 4 2 0
x x t t x x
q q q q q + + + + =
, ) 4 3 3 0
t t x x x x
q qq q q + + =
, ) , ) , ) 4 3 3 0
t t x x
q q q q + + + = (e)
(d)+(e): , ) 4 3 0
t x
q + + =
, )
1
3 0
4 t x
q
c c
1
+ + =
(
c c
]
, ) , ) 2 0 u c u c
t x
c c
1
+ + + =
(
c c
]
(3.99)
(d)-(e): , ) 4 3 0
t x
q q q + + =
, )
1
3 0
4 t x
q q
c c
1
+ + =
(
c c
]
, ) , ) 2 0 u c u c
t x
c c
1
+ =
(
c c
]
(3.100)
3.9.2 Flow Caused By Dam Break
Let static 2-D water be contained between
0
0 y h < < for 0 x < by a dam at 0 x = .
(see fig.3.15a)
The problem is to find the water flow after the dam breaks at 0 t = .
Consider 1
st
the initial conditions.
For 0 x < , 0 t < , we have 0 u = ,
0
h h = , so that
0 0
c c gh = = .
The characteristic base curves
dx
u c
dt
= (3.103)
become
0
dx
c
dt
=
ie.,
0 0
x c t x = + (3.104a)
which are straight lines that intercept the x axis at
0
x . (see fig.3.15c)
After the dam breaks, these characteristic base curves will extend continuously,
wherever possible, into the 0 t > region. On such extensions,
0
2 2 u c c = .
Now, some points may be reachable by extensions of both types of base curves. (Point
P in fig.3.15c) Both eqs
0
2 2 u c c = must then be satisfied, which means
0 u = and
0
c c =
ie., the water there are not yet disturbed and the characteristic base curves are still
straightlines given by (3.104a). As is obvious from fig.3.15d, the upper boundary of
this undisturbed region is the line
0
x c t = .
Beyond the line
0
x c t = , a point (eg., Q in fig.3.15.d) can be reached only by the
extension of characteristic base curves of the positive type
dx
u c
dt
= + on which
0
2 2 u c c + = (3.104)
The other, negative type, characteristic base curve that pass through Q , ie.
dx
u c
dt
= (3.104b)
requires
2 u c k = (3.104c)
where
0
2 k c = is a constant along the negative base curve (3.104b).
The solution of (3.104) & (3.104c) is
0
2
k
u c = + and
0
1
2 2
k
c c
| |
=
|
\ .
(3.105a)
which are both constants on the negative base curve (3.104b) which itself becomes a
straight line
0
1 3
2 2
dx
c k
dt
| |
= +
|
\ .
ie.
0 0
1 3
2 2
x c k t x
| |
= + +
|
\ .
where
0
x is the value of x at 0 t = .
Now, different negative base curves are characterized by different k so that they have
different slopes. To avoid crossing of such curves, we need
0
0 x = so that the
negative base curves all eminate from the origin:
0
1 3
2 2
x c k t
| |
= +
|
\ .
(3.105b)
The positive base curves are given by
0
1 1
3
2 2
dx
c k
dt
| |
= +
|
\ .
(3.105c)
which are not straightlines since k is not constant on such curves.
From (3.105b), we have
0
2
2
3
x
k c
t
| |
= +
|
\ .
so that the positive base curve (3.105c) becomes
0 0 0
1 1 1
3 2 4
2 3 3
dx x x
c c c
dt t t
1 | | | |
= + + = +
| |
(
\ . \ . ]
while (3.105a) becomes
0
2
3
x
u c
t
| |
= +
|
\ .
and
0
1
2
3
x
c c
t
| |
=
|
\ .
(3.105,6)
Since 0 c gh = > , these relations are valid only for
0
2 x c t s
Using (3.106) we have
2
2
0
1
2
9
c x
h gh
g g t
| |
= =
|
\ .
for
0 0
2 c t x c t < <
At
0
2 x c t = , we have 0 h = , which should remain so for
0
2 x c t > on physical
grounds. (see fig.3.15b).
3.9.3 Non-Linear Wave Distortion
The dam break flow may be taken as a smoothing out of the initial discontinuity in
, ) , h x t through non-linear mechanism.
Let us return to the shallow water eqs
, ) , ) 2 0 u c u c
t x
c c
1
+ + + =
(
c c
]
(3.99)
, ) , ) 2 0 u c u c
t x
c c
1
+ =
(
c c
]
(3.100)
Consider the region
0 0
2 c t x c t < < where
0
2 2 u c c + = .
Eq (3.99) is automatically satisfied.
Eliminating u from (3.100) gives
, )
0
2 3 0 c c c
t x
c c
1
+ =
(
c c
]
which can be rewritten as
0
z z
z
t x
c c
=
c c
(3.109)
with
0
3 2 z c c = (3.108)
The characteristics eqs of (3.109) are (see 3.9a1)
1 0
dt dx dz
z
= =

so that
1
z c =
and
dx
z
dt
= ie.
2
x zt c + =
The general solution is therefore
, ) z F x zt = + (3.110)
where F is an arbitrary differentiable function.
Note that (3.110) is an implicit solution.
To show that (3.110) is indeed a solution of (3.109), we need to evaluate the various
partials of (3.110) with x and t taken as the independent variables. Setting
x zt = +
the partials on (3.110) are
'
z dF z
z t F
t t d t

c c c
| |
= = +
|
c c c
\ .
1 '
z dF z
t F
x x d x

c c c
| |
= = +
|
c c c
\ .
where '
dF
F
d
= .
Solving for the partials of z, we have
'
1 '
z zF
t tF
c
=
c
(3.111)
'
1 '
z F
x tF
c
=
c
(3.112)
which obviously satisfies (3.109) identically for arbitrary differentiable F.
Now, (3.110) is a wave of z that travels with velocity z. Assuming positive amplitude
( 0 z > ), the wave travels to the left with speed that is proportional to its amplitude,
thus providing the mechanism for smoothing out the discontinuity in the dam break
problem.
Consider now the eq
0
z z
z
t x
c c
+ =
c c
(3.113)
Applying the foregoing analysis, we see that its solution is
, ) z F x zt = (3.114)
which describes a wave of z that travels with velocity z. Assuming positive amplitude
( 0 z > ), the wave travels to the right with speed that is proportional to its amplitude,
thus steeping the wave profile as shown in fig.3.16.
From the analog of (3.112):
, )
, )
'
1 '
F X z
x tF X
c
=
c +
X x zt = (3.115)
we see that
z
x
c
c
goes to infinity at time
c
t defined by
, ) 1 ' 0
c c
t F x zt + = (3.116)
In case of multiple solutions, only the minimum value will be physically meaningful
since the wave will collapse after that. (see fig.3.16c)
3.9.4 The Formation of a Bore
Let fluid be contained by a vertical plate in the region 0 x > ,
0
0 y h < < for 0 t < .
At 0 t = , the plate at 0 x = begins to move in the x + direction with speed U t o = ,
where is a constant. (see fig.3.17)
Once again, we need to solve the shallow water eqs (3.96-7) or (3.99-100).
The treatment is analogous to that of the dam break problem so well simply highlight
the salient points.
For 0 t < , the fluid occupies 0 x > with 0 u = and
0
h h = ,
0 0
c c gh = = .
The corresponding characteristic base curves
0 0
x c t x = + (3.104a)
are straight lines in the 4
th
quadrant that intercept the x axis at
0
x . (see fig.3.18)
The undisturbed region ( 0 u = and
0
c c = ) is found by extending the above straight
lines into the 1
st
quadrant. The upper boundary of this is obviously the line
0
x c t = .
(see fig.3.18)
As is clear from fig.3.18, only negative type of base curves
dx
u c
dt
= can be
extended across the boundary
0
x c t = . On such base curves,
0
2 2 u c c = (3.117)
so that (3.100) is automatically satisfied. Eliminate c from (3.99) gives
0
3
0
2
u c u
t x
c c 1 | |
+ + =
|
(
c c
\ . ]
(3.118)
whose characteristics are given by
0
3
1 0
2
dt dx du
u c
= =
+
so that
1
u c =
and
0
3
2
dx
u c
dt
= + ie.
1 0 2
3
2
x c c t c
| |
+ =
|
\ .
The general solution of (3.118) is therefore
0
3
2
u F x u c t
1 | |
= +
|
(
\ . ]
(3.119)
Now, we expect u to have a maximum at the plate and decrease to zero at
0
x c t = .
Eq(3.118) is then of the profile steepening type (3.113) discussed in 3.9.3.
The actual form of F is of course determined by the boundary conditions.
At time t, the vertical plate is at
2
1
2
x t o = moving with speed u t o = .
Eq(3.119) becomes
, )
2 2
0 0
1 3
2 2
t F t t c t F t c t o o o o
1 | |
= + =
|
(
\ . ]
0 t > (3.120)
with
, ) 0 0 F =
Writing
2
0
t c t o =
we have
, )
2
0 0
1
4
2
t c c o
o
=
so that (3.120) becomes
, )
, )
2
0 0
1
4
2
F c c o = + 0 < (3.120)
where the + root is chosen so that , ) 0 0 F = .
Eq(3.119) is evaluated by putting
0
3
2
x u c t
| |
= +
|
\ .
in (3.120) so that
2
0 0 0
1 3
4
2 2
u c c x u c t o
| |
1 | |
= + + |
|
(
|
\ . ]
\ .
Solving fo u, we have
, )
2
2
0 0 0
3
2 4
2
u c c x u c t o
1 | |
+ = +
|
(
\ . ]
, ) , )
2
0 0
4 4 6 4 0 u c t u x c t o o + + =
, )
2
0 0
3
0
2
u c t u x c t o o
| |
+ + =
|
\ .
so that
, )
2
0 0 0
1 3 3
4
2 2 2
u c t c t x c t o o o
1
| | | |
( = +
| |
\ . \ . (
]
(3.121)
where the + root is chosen to make 0 u = at the undisturbed boundary
0
x c t = .
Eq(3.117) can be used to find c:
, )
2
0 0 0 0 0
1 3 3
4
2 4 2 2
u
c c c c t c t x c t o o o
1
| | | |
( = + = + +
| |
\ . \ . (
]
(3.122)
Hence
, )
1
2 2
0 0
3
4
2 2
c
c t x c t
x
o
o o

1
c
| |
=
(
|
c
\ .
(
]
At the boundary
0
x c t = ,
0
3
2
2
c
x
c t
o
o
c
=
c | |

|
\ .
which explodes at time
0
2
3
c
c
t
o
= (3.123)
At this moment, the vertical plate is at position
2
0
1 1
2 3
c c
x t c t o = =
whereupon (3.122) becomes
, )
0 0 0 0 0
1 1 8 4
4
4 4 3 3
c c
c c x c t c c t c o o = + = + =
which means the water level at the gate at time
c
t is
2
0
1 16
9
h c h
g
= =
3.10 Hydraulic Jumps & Shock Waves
3.10.1 Hydraulic Jumps
Abore (3.9.4) that is stationary to the observer is called a hydraulic jump.
Both are shallow water phenomena which exhibit an abrupt jump of water level.
One way to produce it is to turn on the kitchen tap. Water in the sink splays outward
radially in a thin layer which exhibits a sudden increase in height at a certain radius
that depends on the flow rate.
With reference to fig.3.19 for notations, we see that
1. Conservation of mass means
1 1 2 2
U h U h = (3.124)
2. Conservation of momentum means
2 2 2 2
1 1 1 2 2 2
1 1
2 2
gh hU gh h U + = + (3.125)
where terms proportional to g come from pressure, and those to U from
convective contributions to the stress tensor.
3. Conservation of energy means turbulence at the jump will dissipate energy in the
form of heat. The energy loss at the jump is (Ex.3.20)
, )
3
1
2 1
2
4
gU
h h
h

(3.128)
where
2 1
h h > (3.129)
Introduce now the Froude number
p
U U
F
c gh
= = (3.126)
where
p
c gh = is the phase speed of water waves in shallow water.
It plays a similar role as the Mach number in compressible flow.
Flows with 1 F < ( 1 F > ) are called sub- (super-) critical.
Now, eliminating
2
U using (3.124), (3.125) becomes
2 2
2 2 2 1 1
1 1 1 2
2
1 1
2 2
h U
gh hU gh
h
+ = +
so that
, )
, )
, )
2 2
2 2 1
2 2 2 1
1
1 2 1 1
2 2
gh h h
gh h h
U
h h h h

+
= =

Interchanging indices 1 2 gives


, )
2 1 2 1
2
2
2
gh h h
U
h
+
=
Hence, the corresponding Froude numbers are
, )
2 1 2 1
1 2
1 1
h h h U
F
h gh
+
=
, )
1 1 2 2
2 2
2 2
h h h U
F
h gh
+
= (3.127)
Since
2 1
h h > , we have
1
1 F > and
2
1 F < (3.130)
ie., the jump changes a supercritical flow to a subcritical one.
3.10.2 Unsteady 1-D Gas Dynamics
Consider the sound wave problem (3.6) without the small amplitude approximation.
The relevant eqs are
Euler eq.: p
t

c
| |
+ V = V
|
c
\ .
u
u u (3.55)
Eq. of continuity: , ) 0
t


c
+ V =
c
u (3.56)
Adiabatic (homentropic) condition:
, )
0
D
p
Dt

= (3.57)
Using the Leibniz rule, (3.57) becomes
1
0
Dp D
Dt Dt




=
or
2
Dp p D D
a
Dt Dt Dt

= = (3.133a)
where
2
p
a

= (3.133)
Taking the material derivative of (3.133) gives
2
1
2
Da Dp p D
a
Dt Dt Dt


| |
=
|
\ .
2 2
a a D
Dt


| |
=
|
\ .
[from (3.133a)]
, )
2
1
a D
Dt

= (3.133b)
Rewriting (3.56) as
D
Dt

= V u
(3.133b) becomes
, ) 2 1
Da
a
Dt
= V u
or
2
0
1
a
a a
t
c
| |
+ V + V =
|
c
\ .
u u (3.133c)
Now, if the gas is initially at rest, (3.57) becomes
p const

=
so that, for example,
1
0 p p



V V =
ie
2
p
p a

V = V = V
Also, analogous to (3.133b), we have
, )
2
2 1
a
a a

V = V
so that
2
1
a
p a

V = V

The Euler eq. then becomes


2
0
1
a
a
t
c
+ V + V =
c
u
u u (3.133d)
For 1-D gas flow with , ) , , 0, 0 u x t = 1
]
u , (3.133c,d) simplify to [cf. Ex.3.22]
2
0
1
a a u
u a
t x x
c c c
| |
+ + =
|
c c c
\ .
2
0
1
u u a a
u
t x x
c c c
+ + =
c c c
(3.133e)
These are similar to, but not the same as, the shallow water eqs(3.96-7).
The characteristics are obtained by considering
, ) 1 0 1
0
2
2
0
0 1
1
'
' ' 0 0
'
0 0 ' '
t
x
t
x
a
u
a
a
a
u
u a
t x
u u
t x

| |

|
| | | |
|
| |
|
| |
=
|

| |
|
| |
|
\ . \ .
|
\ .
This set of eqs is indeterminate if 2 conditions hold:
I.
, ) 1 0 1
2
2
0 1
0
1
' ' 0 0
0 0 ' '
a
u
a
u
t x
t x

ie.,
, )
2
0 1
1
2
1
2
' 0 0 ' 1
1
0 ' '
0 ' '
a
a
u
u
a
x t u
t x
t x

, ) , )
2
' ' ' ' ' ' ' x x ut t u x ut a t 1 = + +
]
, )
2
2 2
' ' ' 0 x ut a t = + =
which requires
, ) ' ' x u a t =
dx
u a
dt
= (3.133f)
II.
, ) 1 0 0 1
2
0 0 1
0
' ' 0 0
0 ' ' '
a
u
t a
u t x

=
ie.
, ) 0 0 1
0 1
2
' 0 0 ' 0 1
' ' ' ' ' '
a
u
a t u
u t x u t x

+
, ) , ) ' ' ' ' ' 1
2
a
a x ut t u =
, ) , ) ' ' ' ' ' 1 0
2
a
a x ut t u = =
so that
, )
2
1
du dx
u
da a dt
| |
=
|

\ .
Combining with (3.133f), we have
2
1
du
da
=

ie.
2
1
a
u c


= +

To summarize. The characteristic base curves are


dx
u a
dt
= (3.133f)
on which
2
1
a
u c

(3.133g)
where c

are constants.
These can be used to simplify (3.133e) as follows. [cf. 3.9.1]
Consider the functions
2
1
a
b

u b = + and u b q =
so that
, )
1
2
u q = + , )
1
2
b q =
, )
1 1
2 4
a b

q

= =
The 1st eq in (3.133e) thus become
, ) , ) , ) , )
1 1 1
0
2 2 8
t t x x x x

q q q q q

| |
+ + + + =
|
\ .
which simplifies to
, ) , ) , ) , ) , ) 4 1 3 3 1 0
t t x x
q q q q + + + + = 1 1
] ]
(d)
Similarly, the 2
nd
one becomes
, ) , ) , ) , )
1 1 1
0
2 2 8
t t x x x x

q q q q q

| |
+ + + + + =
|
\ .
, ) , ) , ) , ) , ) 4 1 3 3 1 0
t t x x
q q q q + + + + = 1 1
] ]
(e)
(d)+(e): , ) , ) 4 1 3 0
t x
q + + = 1
]
Now:
, ) , ) , ) , )
2 2
1 3 1 3
1 1
a a
u u q

| | | |
+ = + +
| |

\ . \ .
, ) 4 u a = +
so that we have
, ) 0
t x
u a + + =
, )
2
0
1
a
u a u
t x
| | c c
1
+ + + =
|
(
c c
]
\ .
(3.131)
Similarly, (d)-(e): , ) , ) 4 3 1 0
t x
q q q + = 1
]
Now:
, ) , ) , ) , )
2 2
3 1 3 1
1 1
a a
u u q

| | | |
+ = + +
| |

\ . \ .
, ) 4 u a =
so that we have
, ) 0
t x
u a q q + =
, )
2
0
1
a
u a u
t x
| | c c
1
+ =
|
(
c c
]
\ .
(3.132)
Consider the case of a piston moving with speed U t o = into a long tube containing
gas at rest with sound speed
0
0
0
p
a

= .
The situation is analogous to the formation of a bore (3.9.4).
For 0 t < , the gas occupies 0 x > with
0 u = and
0
a a =
The characteristic base curves are straightlines in the 4
th
quadrant:
0
dx
a
dt
= ie.
0
0
1
t t x
a
=
where
0
t is the intercept with the t axis.
The + type lines must have
0
0 t s . The line
0
1
t x
a
= ie.,
0
x a t =
thus form an upper boundary for them.
For the type lines, only those with
0
0 t > can be extended across this boundary.
Thus, inside the bore region, we have, along the type characteristic base curves
0
2 2
1 1
a a
u

=

(3.134)
and (3.132) is automatically satisfied.
For the + type characteristic base curves, eliminating a using (3.134) turns (3.131)
into
, )
0
1
1 0
2
u a u
t x

c c 1 | |
+ + + =
|
(
c c
\ . ]
(3.135)
The characteristics of this are given by
, )
0
1
1 0
1
2
dt dx du
u a
= =
+ +
The general solution of (3.135) is therefore
, )
0
1
1
2
u F x u a t

1
= + +
`
(
] )
(3.136)
where F is an arbitrary differentiable function.
The function F is determined by the boundary condition at the piston:
u t o = at
2
1
2
x t o =
ie.,
0
1
2
t F t a t o o

1
= +
`
(
] )
0 t >
Setting
0
1
2
t a t o
| |
= +
|
\ .
we have
, )
2
0 0
1
2 t a a o
o
=
so that
, )
, )
2
0 0
1
2 a a F o

+ =
where the + root was chosen to make , ) 0 0 F = .
Thus
, )
2
0 0 0
1 1
2 1
2
u a a x u a t o

| |

1
= + + + |
`
(
|
] )
\ .
Solving for u, we have
, )
2 2 2 2
0 0 0 0
1
2 2 1
2
u a u a a x u a t o

1
+ + = + +
`
(
] )
, ) , )
2
0 0
2 1 2 0 u a t u x a t o o + + + = 1
]
, ) , ) , )
, }
2
0 0 0
1
2 1 2 1 8
2
u a t a t x a t o o o

= + + + 1 1
] ]
where the + root is chosen to make 0 u = at
0
x a t = .
Hence
, ) , )
2
0 0
2
2 1 8
u
x
a t x a t
o
o o
c
=
c
+ 1
]
Now, (3.136) is of the wave steepening type,
u
x
c
c
will eventually become singular,
which happens when
, ) , )
2
0 0
2 1 8 0 a t x a t o o + = 1
]
At
0
x a t = , this occurs at time
, )
0
2
1
c
a
t
o
=
+
whence (3.135) breaks down and a shock propagates down the tube.
3.10.3 Normal Shock Waves
A shock normal to the flow is similar to a hydraulic jump (3.10.1).
With reference to fig.3.20 [cf. fig.3.19], we have the Rankine-Hugoniot equations:
Conservation of mass:
1 1 2 2
U U =
Conservation of momentum:
2 2
1 1 1 2 2 2
p U p U + = + (3.137)
Conservation of energy:
2 2
2 2 1 2
1 2
1 1
2 1 2 1
a a
U U

+ = +

[The last is different from the hydraulic jump case which substains an energy loss
across the jump.]
Furthermore, the 2
nd
law of thermodynamics
2 1
0 S S >
means that
, ) , )
2 2 1 1
log log p p



>
ie.,
2 2 1 1
p p



> or
2 2
1 1
p
p

| |
>
|
\ .
(3.138)
We leave as an exercise to show that
1.
2 0 1
U a U s s , ie.,
2 1
1 M M s <
2.
2 1
p p > ,
2 1
>
3.11 Viscous Shocks & Solitary Waves
3.11.1 Weak Viscous Shocks
For a weak viscous shock propagating in x + direction into a gas at rest, the velocity
, ) , u x t satisfies Burgers equation
, )
2
0 2
1 2
1
2 3
u
u a u
t x x
v
c c c
1
+ + + =
`
(
c c c
] )
(3.140)
where
0

if both and A are small.


For 0 v = , (3.140) reverts to (3.135).
Consider the ansatz
, ) u f x Vt = (3.141)
such that
, )
1
f U = , ) 0 f =
Now:
'
u
Vf
t
c
=
c
'
u
f
x
c
=
c
2
2
''
u
f
x
c
=
c
where
'
df
f
d
= x Vt =
Eq(3.140) becomes
, )
0
1 2
1 ' ''
2 3
V f a f f v

1
+ + + =
`
(
] )
which, using
' ' '
'' '
df df df df
f f
d d df df
= = =
becomes
, )
0
2 1
' 1
3 2
df V f a df v

1
= + + +
`
(
] )
ie
, ) , )
2
0
2 1
' 1
3 4
f f V a f Const v = + + + +
or
, )
, )
2 0
4 8
'
3 1 1
a V
f f f C
v


= + +
+ +
where C is a constant.
Hence
, )
, )
2 0
3 1
4 8
1
df
a V
f f C

+
=

+ +
+
}
In order to the formula
, )
2
1
ln
dx x p
ax bx c a p q x q

=
+ +
}
where p, q are the roots of
2
ax bx c + + , we set
1 a =
, )
0
4
1
a V
b


=
+
c C =
2
1
, 4
2
p q b b C
1
=
]
so that
, )
2
3 1 1
ln
8
4
f p
f q
b C

v
+
=

Now, the points = require , f p q = .


The condition , ) 0 f = thus require either
2
4 0 b b C + = or
2
4 0 b b C =
Only the 1
st
eq has a solution with 0 C = . Therefore, we have
, ) 3 1 1
ln
8
f b
b f

v
+ +
=
The condition , )
1
f U = then means
1
b U =
so that
, )
1
1
3 1 1
ln
8
f U
U f

v
+
=
which can be solved for u f = to give
, )
1 1
3 1
exp
8
U U u
u

v
+ 1
=
(
]
(
1
u U s )
ie.,
, )
1 1
1
3 1
1 exp
1 exp
8
U U
u
x Vt
U

v
= =

+ 1
+
+
(
A
]
(3.142)
where the shock thickness is given by
, )
1
8
3 1 U
v

A =
+
(3.144)
Thus, u decays from
1
U to 0 within a distance of order .
Finally, the condition on b means
, )
0
1
4
1
a V
U


=
+
Solving for the shock speed V gives
, )
0 1
1
1
4
V a U = + + (3.143)
3.11.2 Solitary Waves in Shallow Water
We now look for a nonlinear shallow water wave of permanent form whose
steepening effects are caused by weak dispersion.
Let
0
q be typical magnitude of , the vertical displacement of water surface.
0
h be that for h, the depth of water.
L be that for horizontal length scale.
If
0
0
h
q
c = and
2
0
2
h
L
o = (3.145)
are small & of the same order of magnitude, then satisfies the Korteweg- de Vries
equation:
3
2 0
0 0 0 3
0
3 1
0
2 6
c
c c h
t h x x
q q q
q
| | c c c
+ + + =
|
c c c
\ .
(3.146)
where the last term describes dispersive effects.
We seek a traveling solution of the form
, ) , ) f x Vt f q = =
with boundary conditions
, )
( )
0
n
f = 0,1, 2, n = (3.145a)
Thus, (3.146) becomes
2 0
0 0 0
0
3 1
' ' ''' 0
2 6
c
Vf c f f c h f
h
| |
+ + + =
|
\ .
or
, ) , )
2 2 0
0 0 0
0
3 1
' ' ''' 0
4 6
c
c V f f c h f
h
+ + =
which, on integrating, becomes
, )
2 2 0
0 0 0
0
3 1
''
4 6
c
c V f f c h f C
h
+ + = (3.146a)
where C is a constant.
Putting in the boundary conditions (3.145a) gives 0 C = .
Writing
2
1
'' '
2
d
f f
df
| |
=
|
\ .
(3.146a) becomes
, )
2 2 2 0
0 0 0
0
1 1 3
'
6 2 4
c
c h d f c V f f df
h
1
| |
= +
| (
\ .
]
, )
2 2 2 3 0
0 0 0
0
1 1
'
12 2 4
c
c h f c V f f C
h
= + (3.146b)
where C is another constant.
Putting in the boundary conditions (3.145a) gives 0 C = .
Rearranging terms in (3.146b) gives
, )
3 2 2 2
0 0
0
1
' 2 1
3
V
h f h f f a f f
c
1 | |
= =
( |
\ . ]
(3.146c)
where
0
0
2 1
V
a h
c
| |
=
|
\ .
(3.146d)
Taking the square root of (3.146c):
, )
3
0
3
' f f a f
h
=
so that
, )
3
0
3 df
C
h
f a f
= +

}
which, upon the use of the formula
1
1 2
ln tanh
dx a bx a a bx
a x a bx a a bx a a

+ +
= =
+ + +
}
gives
1
3
0
3 2
tanh
a f
h a a



=
where 0 C = according to the boundary conditions (3.145a).
Solving for f gives
3
0
3
tanh
4
a f a
a h

=
2 2
3 3
0 0
3 3
1 tanh sech
4 4
a a
f a a
h h

| |
= =
|
\ .
ie.,
, )
2
3
0
3
sech
4
a
a x Vt
h
q
1
=
(
]
(3.147)
Eq(3.146d) can also be rewritten as
0
0
1
2
a
V c
h
| |
= +
|
\ .
(3.148)
Since sech ranges between 0 and 1, a is the maximum of the wave height 0 q > . [see
fig.3.23]. Owing to the assumptions (3.145), these eqs are valid only for
0
a h .
According to (3.148), waves with larger amplitude a travels with greater velocity V.
Thus, a higher wave can overtake a lower one. What seems astonishing is that after
such a collision, both waves emerge unscathed. [see fig.3.24] Such waves are called
solitons.
4. Classical Aerofoil Theory
Source: D.J.Acheson, Elementary Fluid Dynamics, Chapter 4, Clarendon Press (90)
4.1 Introduction
4.2 Velocity Potential & Stream Function
4.2.1 Velocity Potential
For irrotational flow,
0 V = u
so that one can define a velocity potential | at point P by
( )
P
O
P d | =
}
u x (4.2)
| = V u (4.3)
where O is some reference point.
In a simply connected region, | is path independent & hence single- valued.
In a multiply connected region, | can be path dependent & hence multi- valued.
The circulation around a closed contour C is
j
C
C C
d d | | I = = V =
} }
u x x

(4.4)
where j
C
| is the change of | in going around C.
Examples:
1. Uniform flow , ) , 0, 0 U = u has Ux const | = + .
2. Irrotational flow , ) , , 0 x y o o = u has stagnation point at the origin.
To solve
x
x
|
o
c
=
c
y
y
|
o
c
=
c
0
z
| c
=
c
we see that the 3
rd
eq means | is a function of x and y only.
Next, we integrate the 1
st
eq to get
2
1
( )
2
x f y | o = +
Taking
y
c
c
of this & equating the result to the 2
nd
eq gives
df
y
dy
o =
which gives
2
1
2
f y C o = + C const =
so that
, )
2 2
1
2
x y C | o = +
which is single valued so that 0 I = for any circuit.
3. Line vortex flow
k
r
u
= u e
is irrotational except at origin.
Since u is not defined at origin, any domain that covers the origin is multiply
connected.
In cylindrical coordinates,


r r z
| | |
|
u
c c c
V = + +
c c c
r
z
so that we need to solve
0
r z
| | c c
= =
c c
and
k
r r
|
u
c
=
c
This gives
k C | u = +
which is multi-valued as expected.
Circulation for a circuit that goes round the origin n times is
2 nk t I =
4.2.2 Stream Function
For 2-D incompressible irrotational flow, we can write
, ) , , u v
y x
| | c c
= =
|
c c
\ .
u (4.5)
so that
0
u v
x y
c c
V = + =
c c
u (4.6)
is automatically satisfied.
Now,
, ) 0 u v
x y y x x y


| | | | c c c c c c
V = + = =
| |
c c c c c c
\ . \ .
u (4.7)
so that is constant along a streamline.
is called a stream function since streamlines are simply curves with const = .
Another way to write (4.5) is
, ) ,
0 0
x y z y x

| | c c c c c
= V = =
|
c c c c c
\ .
i j k
u k (4.8)
In cylindrical coordinates, we have
, )

0 0
r
r r z r r

u u

c c c c c
= V = =
c c c c c
r
k
u k r

or
r
u
r

u
c
=
c
u
r
u
c
=
c
(4.9)
which obviously satisfies the incompressibility condition trivially
, ) 0
r
ru u
r r r
u
u
c c
+ =
c c
(4.10)
4.3 Complex Potential
Consider a 2-D incompressible flow.
0
u v
x y
c c
+ =
c c
we have
u
y
c
=
c
v
x
c
=
c
(4.11)
If it is also irrotational,
0
u v
y x
c c
=
c c
we have
u
x
| c
=
c
v
y
| c
=
c
(4.11a)
Thus, and | satisfy the Cauchy- Riemann conditions
x y
| c c
=
c c y x
| c c
=
c c
(4.11b)
so that they are the real & imaginary parts of an analytic function, ie.,
w i | = + (4.12)
where , ) w z , with z x iy = + , is known as the complex potential.
By definition
, ) , )
0
lim
z
f z z f z df
dz z
A
+ A
=
A
We say f is analytic at z if
df
dz
is independent of the path z A approaches 0.
Thus, if f a ib = + is analytic, we have
df a b a b
i i i
dz x x y y
| | c c c c
= + = +
|
c c c c
\ .
for the 2 cases z x A = A and z i y A = A .
Another useful property for f to be analytic at z is that it has a Taylor expansion there.
Since w is analytic, we have
dw
i u iv
dz x x
| c c
= + =
c c
(4.15)
The flow speed is therefore
2 2
dw
q u v
dz
= + = (4.16)
Either of the real & imaginary parts of an analytic function satisfies the Laplace eq.
individually. Hence
2 2
2 2
0
x y
| | c c
+ =
c c
(4.13)
2 2
2 2
0
x y
c c
+ =
c c
(4.14)
4.3.1 Uniform Flow At Angle to x-Axis
Here
cos u U o = sin v U o =
so
, ) cos sin
i
dw
u iv U i Ue
dz
o
o o

= = =
and, integrating gives
i
w Uze
o
= (4.17)
4.3.2 Line Vortex
2 r
u
t
I
= u e (4.18)
From 2.4.4, we have
2
u
|
t
I
= (4.19)
From (4.9), we have
0
r
u
r

u
c
= =
c 2
u
r r
u

t
c I
= =
c
so that
, ) ln
2
f r r
t
I
= =
Hence
, ) , ) ln ln ln
2 2 2
w i i r i r i i z | u u
t t t
I I I
= + = = + = (4.20)
where
i
z re
u
=
This can immediately generalized to a line vortex at
0
z z = :
, )
0
ln
2
w i z z
t
I
= (4.21)
4.3.3 Flow Near Stagnation Point
The Taylor expansion of w near
0
z is
, ) , ) , ) , ) , ) , )
2
0 0 0 0 0
1
' ''
2
w z w z z z w z z z w z = + + +
The constant term , )
0
w z is inconsequential.
If
0
z is a stagnation point, , )
0
' 0 w z = .
Hence, the flow near a stagnation point is determined, to the lowest order, by the
quadratic term.
Writing
, )
0
''
i
w z e
|
o = and
0 1
z z z =
we have
2
1
1
2
i
w const z e
|
o = + +
2
2
1
2
const z o = + +
where
2
2 1
i
z z e
|
=
Dropping the constant term & subscripts, the complex potential near a stagnation
point is therefore
, )
2 2 2
1 1
2
2 2
w z x y ixy o o = = + (4.22)
where is real.
The velocity potential is
, )
2 2
1
Re
2
w x y | o = =
so the equipotential lines are rectangular hyperbolae.
The stream function is
Imw xy o = = (4.24)
so the streamlines (constant ) are rectangular hyperbolae perpendicular to the
equipotentials [see fig.4.1]
The corresponding flow is
, ) Re Re
dw
u z x
dz
o o
| |
= = =
|
\ .
, ) Im Im
dw
v z y
dz
o o
| |
= = =
|
\ .
(4.23)
4.4 Method of Images
Let there be a rigid plane wall at 0 x = .
Consider a line vortex of strength at , ) , 0 d = r [see fig.4.2].
The boundary condition
, ) 0, 0 u y = or
0
0
x
x
|
=
c
=
c
can be satisfied by replacing the wall with a mirror image of the vortex.
As in electrostatics, the image should be at , ) , 0 d = r with strength I.
According to (4.21), the complex potential is therefore
, ) , ) ln ln
2 2
w i z d i z d
t t
I I
= + + (4.25)
ie.,
ln
2
z d
i i
z d
|
t
I
+ =
+
(4.26)
ln
2
z d
i i
z d
u
t
I (
= +
(
+

where
i
z d z d
e
z d z d
u

=
+ +
Hence
ln
2
z d
z d

t
I
=
+
so that the streamlines are
ln
2
z d
const
z d

t
I
= =
+
or simply
z d
R
z d

=
+
(4.27)
where R is a constant.
Now, (4.27) can be written as
, ) , )
2 2
2 2 2
x d y R x d y
(
+ = + +

or, on collecting terms:
, ) , ) , )
2 2 2 2 2
1 2 1 0 R x d y R dx + + + =
2
2 2 2
2
1
2 0
1
R
x dx d y
R
| | +
+ + =
|

\ .
2 2
2
2 2
2 2
2 2 2
1 1 2
1
1 1 1
R R R
x d y d d
R R R
(
( | | | | + +
| |
+ = = (
( | | |

\ .
( \ . \ .

which is a circle centered at
2
2
1
, 0
1
R
z d
R
( | | +
=
( |

\ .
with radius
2
2
1
R
a d
R
=

.
Such circles are known as coaxial circles.
The distances between the center of a coaxial circle and those of the vortices at
x d = are
2 2
2 2
1 2
1
1 1 1
R R d
c d
R R


+
= =



so that
2
2 2
2
2
1
d
c c R a
R
+
| |
= =
|

\ .
(4.27a)
Consider now the flow inside a circular cylinder z a = due to a line vortex of
strength at , ) , 0 z c a = < . The boundary condition of no flow in the normal
direction is satisfied if z a = is a streamline. This is easily accomplished by placing
an image vortex of strength I at a point which according to (4.27a) is
2
, 0
a
z a
c
| |
= >
|
\ .
. The corresponding complex potential is [cf. (4.25)]:
, )
2
ln ln
2 2
a
w i z c i z
c t t
| | I I
= + +
|
\ .
(4.28)
4.4.1 Milne-Thomsons Circle Theorem
Consider ( ) w f z = where all singularities of f lie in z a > . Then
, )
2
a
w f z f
z
| |
= +
|
\ .
(4.29)
is the complex potential of a flow with
1. same singularities as , ) f z in z a > .
2. z a = as a streamline.
Proof:
1. Since all singularities of , ) f z are in z a > , those of
2
a
f
z
| |
|
\ .
and hence
2
a
f
z
| |
|
\ .
are in
2
a
a
z
> ; ie., a z > . Therefore,
2
a
f
z
| |
|
\ .
is regular z a > and
w has the same singularities as , ) f z there.
2. On z a = , we have
i
z ae
u
= ( real)
2
i
a
ae z
z
u
= =
so that
, ) , ) , ) , ) 2Re w z f z f z w z = + = (

Hence,
0 = on z a =
which means z a = is a streamline.
4.5 Irrotational Flow Past A Circular Cylinder
IR flow in x direction, with speed U at infinity, past fixed cylinder z a = .
Complex potential for unperturbed flow is: , ) f z Uz = (singular only at infinity).
Applying the circle theorem (4.4.1), we have
2 2
a a
f U Uz
z z
| |
= =
|
\ .
2 2
a a
f U
z z
| |
=
|
\ .
so
, )
2
a
w z U z
z
| |
= +
|
\ .
(4.31)
is the complex potential with z a = as a streamline.
Any superposition with a line vortex of strength is also a solution with z a = as a
streamline so that a more general form of (4.31) is
, )
2
ln
2
a
w z U z i z
z t
| | I
= +
|
\ .
(4.32)
Consider first (4.31). Writing
i
z re
u
= we have
, )
2
i i
a
w z U re e i
r
u u
|

| |
= + = +
|
\ .
so that
2
cos
a
U r
r
| u
| |
= +
|
\ .
(4.33)
2
sin
a
U r
r
u
| |
=
|
\ .
(4.34)
Hence
2
2
1 cos
r
a
u U
r r
|
u
| | c
= =
|
c
\ .
2
2
1 sin
a
u U
r r
u
|
u
u
| | c
= = +
|
c
\ .
(4.35)
The flow is symmetric fore & aft (see fig4.4a)
At r a = , 2 sin 0 u U
u
u = = for 0, u t = , ie, there is slip everywhere except for 2
points.
Define the slip velocity as 0
s
u u
u
= > at r a = .
The arc length along the top of the cylinder from the forward stagnation point is
, ) s a t u = .
Eq(4.35) then gives (with r a = ),
2 sin 2 sin
s
s s
u U U
a a
t
| |
= =
|
\ .
(4.37)
2
cos
s
du U s
ds a a
=
The slip velocity therefore rises from 0 at front stagnation point to a maximum at
2
t
u = , then falls to 0 at rear stagnation point.
For finite , we deal with (4.32), which becomes
, ) , )
2
ln
2
i i
a
w z U re e i r i
r
u u
u
t

| | I
= + +
|
\ .
so that
2
cos
2
a
U r
r
| u u
t
| | I
= + +
|
\ .
2
sin ln
2
a
U r r
r
u
t
| | I
=
|
\ .
and
2
2
1 cos
r
a
u U
r r
|
u
| | c
= =
|
c
\ .
2
2
1 sin
2
a
u U
r r r
u
|
u
u t
| | c I
= = + +
|
c
\ .
(4.38)
For aerofoil problems, one usually deal with 0 I < ; in which case, let
0
2
B
Ua t
I
= > (4.39)
Now we search for the stagnation points at which 0 = u .
On r a = , (4.38) gives
0
r
u =
, ) 2 sin 2sin
2
s
u u U U B
a
u
u u
t
I
= = = +
Thus, for 2 B < , there are 2 stagnation points at [see fig.4.4.b],
1
sin
2
B
u

= and
1
sin
2
B
u t

= (4.39a)
For 2 B = , these 2 stagnation points coalesce into 1 at
1
3
sin 1 or
2 2
t t
u

= = .
For 2 B > , there is no real solution to (4.39a), ie., the stagnation points, if exist, are
off the cylinder. This means 0
r
u = must now be satisfied by setting cos 0 u = in
the 1
st
eq in (4.38). We put
3
2
t
u = so that our result may converge properly to that
of the 2 B = case. The 2nd eq in (4.38) thus becomes
2 2
2 2
1 1 0
2
a a Ba
u U U
r r r r
u
t
| | | | I
= + + = + =
| |
\ . \ .
ie.,
2 2
0 r Bar a + =
with solutions
2
4
2
a
r B B
(
=

However, the root gives 0 r = at 2 B = so that it must be discarded.
Therefore, there is only 1 stagnation point at
2
4
2
a
r B B
(
= +

,
3
2
t
u = (4.40)
Since we have irrotational steady flow, the Bernoulli theorem applies and we have
2
1
2
p
const

+ = u everywhere.
Note: if the flow is not irrotational, the above still applies to the surface of the
cylinder since it is a streamline.
On the cylinder, r a = so that
2
2 2 2 2
2 sin
2
r
u u u U
a
u u
u
t
I
| |
= + = = +
|
\ .
u
2 2
4 sin 2 sin U U const
a
u u
t
I
= +
The Bernoulli eq. thus becomes
2 2
2 sin sin
p U
const U
a
u u
t
I
= + on r a = (4.40a)
The net force F on the cylinder is
2
0
pndS a p d
t
u = =
} }
F r

where
cos sin u u = + r i j
Now, (4.40a) is symmetric fore & aft, ie., invariant under transformation u t u .
Since cosu changes sign under the same transform, the x-component of F vanishes.
For the y-component, we have
2
0
sin
y
F a p d
t
u u =
}
2
2 2
0
2 sin sin sin
U
a const U d
a
t
u u u u
t
I
(
= +
(

}
U = I (4.41)
where only the 3
rd
term in the integrand contributes.
Thus, we have no drag but an upward lift L U = I for 0 I < , as mentioned in
1.6.3.
If the oncoming stream makes an angle with the x-axis, the unperturbed complex
potential is given by (4.17) as
i
w Uze
o
= . Eqs(4.31,2) are accordingly modified to
, )
2
i i
a
w z U ze e
z
o o
| |
= +
|
\ .
, )
2
ln
2
i i
a
w z U ze e i z
z
o o
t

| | I
= +
|
\ .
(4.42)
Fig.4.4 should likewise be turned anticlockwise through angle to describe the
present situation.
4.6 Conformal Mapping
Let , ) w z be the complex potential of some 2-D irrotational flow in the z-plane with
w i | = + .
Consider an analytic function
, ) Z f z = (4.43)
with inverse
, ) z F Z = (4.44)
which is also analytic. Then
, ) , ) , ) W Z w F Z w z = = (

(4.45)
is analytic.
Let
Z X iY = + (4.46)
and write
, ) , ) , ) , , W Z X Y i X Y = u + + (4.47)
where X, Y , , are real.
Since , ) W Z is analytic, the Cauchy-Riemann conditions give
, )
*
, u X Y
X Y
cu c+
=
c c
, )
*
, v X Y
Y X
cu c+
=
c c
(4.48)
where
* *
and u v are the flow velocity components in the Z-plane.
Thus,
* *
dW
i u iv
dZ X X
cu c+
= + =
c c
Similarly,
dw
u iv
dz
=
Writing
dw
dW
dz
dZ
dZ
dz
= (4.49)
we have
, )
* *
'
u iv
u iv
f z

= (4.50)
where, from (4.43), weve used ' ' Z f = .
Thus, if we want the mapping to preserve the uniform flow at infinity, we must have
, ) ' 1 f z . as z
Consider a point
0
z with , )
0 0
Z f z = .
Let , )
( )
0
n
f z be the 1
st
non-vanishing derivative at
0
z .
The Taylor expansion of Z around
0
z gives
, )
, )
, )
, ) , )
1
0 0 0
!
n
n
n
z
Z z z Z f z O z
n
o
o o
+
+ = + +
so that
, )
, )
, )
, ) , )
1
0 0 0
!
n
n
n
z
Z Z z z Z f z O z
n
o
o o o
+
= + = + (4.50a)
Any complex function f can be written as
j exp arg f f i f =
For the product of 2 complex functions,
, ) exp arg arg fg fg i f g = + (

so that
, ) arg arg arg fg f g = +
Applying this to (4.50a) gives
, ) , )
, )
, )
0
arg arg arg
n
Z n z f z o o
(
= +

where weve dropped the higher order terms and made use of the fact that the arg of a
real function vanishes.
Consider now 2 such expansions around
0
z , namely,
, ) , )
, )
, )
1 1 0
arg arg arg
n
Z n z f z o o
(
= +

, ) , )
, )
, )
2 2 0
arg arg arg
n
Z n z f z o o
(
= +

Subtracting them leads to
, ) , ) , ) , )
2 1 2 1
arg arg arg arg Z Z n z z o o o o = (

(4.51)
Hence, for 1 n = , the angle between small line segments is preserved under the
mapping. Such mappings are called conformal.
The Joukowski transformation is given by
, )
2
c
Z f z z
z
= = + (4.52)
Now,
2
2
' 1
c
f
z
=
2
3
2
''
c
f
z
=
so that at z c = , we have 2 n = .
The inverse of (4.52), ie., , )
1
f Z z

= , is obtained by solving for z from


2 2
0 z Zz c + =
which gives
2 2
1
4
2
z Z Z c
(
=

so that
1
f

has 2 branches with branch points at 2 Z c = . As is the usual practice,
we take the line joining these banch points to be the branch cut.
Furthermore, to ensure
z Z as Z
we choose the + branch so that
2 2
1
4
2
z Z Z c
(
= +

(4.53)
4.7 Irrotational Flow Past An Elliptical Cylinder
We now apply the Joukowski transformation
2
c
Z z
z
= + (4.52)
to the circle
i
z ae
u
=
with 0 c a s s .
Thus
2
i i
c
Z X iY ae e
a
u u
= + = +
2 2
cos sin
c c
a i a
a a
u u
| | | |
= + +
| |
\ . \ .
ie.,
2
cos
c
X a
a
u
| |
= +
|
\ .
2
sin
c
Y a
a
u
| |
=
|
\ .
Eliminating gives
2 2
2 2
2 2
1
X Y
c c
a a
a a
+ =
| | | |
+
| |
\ . \ .
(4.54)
which is an ellipse with semi-axis
2
c
a
a
. [see fig.4.5.b].
The complex potential of a flow passing an elliptical cylinder is therefore obtained by
substituting the inverse
2 2
1
4
2
z Z Z c
(
= +

(4.53)
of (4.52) into the complex potential
, )
2
ln
2
i i
a
w z U ze e i z
z
o o
t

| | I
= +
|
\ .
(4.42)
of a flow past a circular cylinder. Here is the angle that the uniform flow makes
with the x-axis.
Using
2 2
2
2 2
1 2 4
2
4
Z Z c
z c
Z Z c

= =
+
we have
, )
2 2
2 2 2
2
1 4
4
2 2
i i
Z Z c
W Z U Z Z c e a e
c
o o
| |

(
= + + |
|
\ .
2 2
1
ln 4
2 2
i Z Z c
t
I
(
+
`

)
(4.55)
4.8 Irrotational Flow Past A Finite Flat Plate
If we choose c a = so that (4.52) becomes
2
a
Z z
z
= + (4.56)
The ellipse
2
cos
c
X a
a
u
| |
= +
|
\ .
2
sin
c
Y a
a
u
| |
=
|
\ .
collapses to
2 cos X a u = 0 Y =
which describes a flat plate of length 4a on the X-axis.
From
, )
2
ln
2
i i
a
w z U ze e i z
z
o o
t

| | I
= +
|
\ .
(4.42)
we have
2
2
2
i i
dw a
U e e i
dz z z
o o
t

| | I
=
|
\ .
From (4.56), we have
2
2
1
dZ a
dz z
=
Therefore
* *
dw
dW
dz
u iv
dZ
dZ
dz
= =
1
2 2
2 2
1
2
i i
a a
U e e i
z z z
o o
t

( | | | | I
=
( | |
\ . \ .
(4.57)
Using
2 2
1
4
2
z Z Z c
(
= +

(4.53)
eq(4.57) can be rewritten as a function of Z.
At the ends of the plate where 2 Z a = or z a = , the flow speed given in (4.57)
usually becomes infinite. [see, eg., fig.4.6a]
One exception is when the numerator of (4.57) also vanishes, ie.,
, )
0
2
i i
U e e i
a
o o
t

I
= at z a =
or
, )
2 4 sin
i i
aiU e e aU
o o
t t o

I = =
Thus, to remove the singularity at the trailing edge ( z a = ), the circulation must be
4 sin aU t o I = (4.58)
in which case, (4.57) is evaluated by the LHospital rule to be
1
2 2
* * 3 2 3
2 2
lim
2
i
z a
Ua e a
u iv i
z z z
o
t

( | | I
= +
| (
\ .
1
2 2 sin 2
i
Ue U
i
a a a
o
o

(
| |
=
| (
\ .

sin
i
Ue iU
o
o =
cos U o =
so that
*
cos u U o =
*
0 v =
which means the flow leaves the trailing edge horizontally.
Note that the singularity at the leading edge ( z a = ) is still present. [see fig.4.6.b]
4.9 Flow Past A Symmetric Aerofoil
A mapping of a circular cylinder to an aerofoil is obtained by applying
2
a
Z z
z
= + (4.56)
to a circle centered at , ) , 0 z i = , of radius r a i = + , where 0 i > . Note that the
circle goes through the point , ) , 0 z a = but encloses , ) , 0 z a = . [see fig.4.7a]
Points on the circle are given by
, )
i
z a e

i i + = + (4.59a)
where is the phase angle [see fig.4.7a].
Putting it into (4.56) gives
, )
, )
2
i
i
a
Z a e
a e

i i
i i
= + + +
+ +
(4.59)
The positions of the trailing / leading edges are giving by 1
i
e

= , so that
, )
, )
2
a
Z a
a
i i
i i
= + +
+
2
2
2
2
a
a
a
a
i
i

+
Turning to the potentials, we must 1
st
adapt
, )
2
ln
2
i i
a
w z U ze e i z
z
o o
t

| | I
= +
|
\ .
(4.42)
to the present situation. This is easily done by the substitution
z z i + a a i +
whereupon (4.42) becomes
, ) , )
, )
, )
2
ln
2
i i
a
w z U z e e i z
z
o o
i
i i
i t

| |
+ I
= + + + |
|
+
\ .
Doing the same to
2
2
2
i i
dw a
U e e i
dz z z
o o
t

| | I
=
|
\ .
gives
, )
2
2
i i
dw a
U e e i
dz z z
o o
i
i t i

| |
+ I
| |
=
|
|
|
+ +
\ .
\ .
Hence,
* *
dw
dW
dz
u iv
dZ
dZ
dz
= =
, )
1
2
2
2
1
2
i i
a a
U e e i
z z z
o o
i
i t i

( | |
| | + I
| |
=
( |
| |
|
+ +
\ .
\ . (
\ .
(4.60)
The most important difference from the flat plate case is that the singularity at
z a = is inconsequential since it now resides inside the aerofoil. The othe
singularity at the trailing edge, z a = , is eliminated as before by setting the numerator
to zero, ie.,
, )
, )
0
2
i i
U e e i
a
o o
t i

I
=
+
which gives
, ) 4 sin a U t i o I = + (4.61)
When a i , the aerofoil given by (4.59) is thin & symmetric.
Keeping only 1
st
order terms in , (4.59) becomes
, ) , )
1 1
i i i
Z a e ae e
a

i
i i
(
+ + +
(

, ) , )
2 i i i
a e a e e

i i i i

= + + + +
, ) , ) 2 cos 1 cos2 2sin sin2 a i i i = + +
The length of the aerofoil is
j
0
Re Re 2 2 2 2 4 L Z Z a a a
t
i i
= =
= = = (4.61a)
The thickness of the foil is
, ) Im Im 2 2sin sin2 T Z Z

i

= =
The extremum point satisfies
, ) 2 2cos 2cos2 0
dT
d
i

= =
ie.,
2
cos 2cos 1 cos 2 = =
which gives
1
1
cos 1 1 8
1
4
2

(
= + =


or
0
2
3


0
3
2 3 3 3
2
T
i i

= | |

+ =
|

\ .
Hence, the maximum thickness is 3 3i .
Using (4.61a) and dropping the term, (4.61) becomes
sin LU t o I
4.10 Blasiuss Theorem
Consider a steady irrotational 2-D flow about a fixed body with boundary C.
Let w be the complex potential of the flow, ,
x y
F F be components of net force on
body, then
2
1
2
x y
C
dw
F iF i dz
dz

| |
=
|
\ .
}
(4.62)
This is known as the Blasiuss theorem.
Proof:
Reminder: a complex number z x iy = + is equivalent to a 2-D vector , ) , x y .
Consider an infinitesimal segment z o on the boundary C.
We can write
i i
z z e e s
u u
o o o = =
where s o is the length, and the angle the segment made with the x-axis.
As 0 s o , the segment, which points in the direction , ) cos , sin u u , becomes
tangent to C.
The (outward) normal to C is therefore in the direction , ) sin , cos u u .
This means the pressure force on the segment is , ) sin , cos p s u u o . [see fig.4.8]
In other words,
, ) sin
x
F p s o u o = , ) cos
y
F p s o u o =
so that
, ) sin cos
i
x y
F i F p i s pie s
u
o o u u o o

= + = (4.62a)
For inviscid fluid, C is a streamline so that
, ) , ) , cos , sin u v q u u = = u on C
where
2 2
q u v = = + u is the speed.
Hence,
, ) cos sin
i
dw
u iv q i qe
dz
u
u u

= = = on C. (4.62b)
Since the flow is steady & irrotational, the Bernoulli theorem applies so that
2
1
2
p q k + = where k const =
Eq(4.62a) thus becomes
2
1
2
i
x y
F i F q k ie s
u
o o o
| |
=
|
\ .
which, with the help of (4.62b) to eliminate q, becomes
2
2
1
2
i i
x y
dw
F i F e k ie s
dz
u u
o o o

(
| |
=
(
|
\ .
(

2
1
2
i i
dw
i e s ike s
dz
u u
o o
| |
=
|
\ .
Now
i
e s z x i y
u
o o o o

= =
so that
, )
2
1
2
i
x y
dw
F i F i e s ik x i y
dz
u
o o o o o
| |
=
|
\ .
2
1
2
dw
i z ik z
dz
o o
| |
=
|
\ .
Integrating over C then gives the theorem (4.62) since the integration of the 2
nd
term
depends only on the end points and hence vanishes over any closed contour.
We leave it as an exercise (Ex.4.5) to show that the moment about the origin is
2
1
Re
2
C
dw
z dz
dz


| |
=
`
|
\ .

)
}

(4.63)
4.10.1 Uniform Flow Past A Circular Cylinder
This problem was already treated in 4.5 so it serves as a check on the Blasiuss
theorem.
The complex potential is
, )
2
ln
2
a
w z U z i z
z t
| | I
= +
|
\ .
(4.32)
so that
2
2
1
2
dw a
U i
dz z z t
| | I
=
|
\ .
Applying the theorem gives
2
1
2
x y
C
dw
F iF i dz
dz

| |
=
|
\ .
}
2
Res
dw
dz
t
(
| |
=
(
|
\ .
(

U
i i U t
t
I
| |
= = I
|
\ .
Hence
0
x
F =
y
F U = I (4.64)
as before.
4.10.2 Uniform Flow Past An Elliptical Cylinder
The complex potential , ) W Z is assumed to be obtained by conformal mapping from
the circular cylinder result , ) w z as described in 4.7.
We shall calculate the torque using
2
1
Re
2
ellipse
dW
Z dZ
dZ


| |
=
`
|
\ .

)
}

(4.63)
In terms of the circular cylinder quantities:
2
c
Z z
z
= + (4.52)
dw
dW
dz
dZ
dZ
dz
=
2
2
1
dZ c
dZ dz dz
dz z
| |
= =
|
\ .
we have
2 2 1
dW dw dZ
dZ dz
dZ dz dz

| | | | | |
=
| | |
\ . \ . \ .
Eq(4.63) becomes
2
1
Re
2
z a
dw dz
Z dz
dz dZ

=

| |
=
`
|
\ .

)
}

For 0 I = ,
, )
2
i i
a
w z U ze e
z
o o
| |
= +
|
\ .
(4.42)
so that
2
2
i i
dw a
U e e
dz z
o o
| |
=
|
\ .
2
1
Re
2
U I = (4.65a)
where
2 1
2 2 2
2 2
1
i i
z a
c a c
I z e e dz
z z z
o o

=
| || | | |
= +
| | |
\ .\ . \ .
}
, ) , )
, )
2
2 2 2 2
2 2 3
i i
z a
z c z e a e
dz
z c z
o o
=
+
=

}
so the poles are at 0, z c = . Since the pole at 0 z = is of 3
rd
order, direct
evaluation of the residue there will involve 2
nd
derivatives of the other parts of the
integrand. This is rather tedious. We thus choose an alternative way & evaluate the
residue by Laurents expansion. Using
1
2 2 4
2 2 2 2 2 2 4
1 1 1
1 1
c c c
z c c z c z z

| | | |
= = + + +
| |

\ . \ .
( z c > )
the integrand of I becomes
, ) , )
2 4
2 2 4 2 2 2 4 2
2 3 2 4
1 1
2 1
i i
c c
z c z e a z a e
c z z z
o o
| |
+ + + + +
|
\ .

The coefficient of
1
z

is
, )
2
2 2 2 2 2 2
2 2
1
2 2
i i i
a
c e a c e e
c c
o o o
| |
+ =
|
\ .
Therefore
2 2
2
2 2
4 4 cos2 4 sin2
i
a a
I i e i
c c
o
t t o t o

| | | |
= = +
| |
\ . \ .
so that (4.65a) becomes
2 2
2 sin2 U c t o = (4.65)
which is clockwise & tends to turn the ellipse broadside to the stream.
4.11 The Kutta-Joukowski Lift Theorem
Consider a steady irrotational 2-D flow past a body with contour boundary C.
Let the flow at infinity be uniform with , ) , 0 U = u , and the circulation around the
body be . Then
, ) 0, U = I F (4.66)
This is the Kutta-Joukowski Lift Theorem.
Proof:
1
st
, choose the coordinates so that the body lies entirely inside a circle z R = .
Assuming the absence of any singularities, the flow
dw
dz
is analytic for z R > ,
inclusive of the infinity where lim
z
dw
U
dz

| |
=
|
\ .
. Hence,
dw
dz
has a Taylor series in
R
z
so that
1 2
2
dw a a
U
dz z z
= + + + (4.67)
Consider now the Blasiuss theorem
2
1
2
x y
C
dw
F iF i dz
dz

| |
=
|
\ .
}
(4.62)
Since
dw
dz
is analytic for z R > , the contour integral is path independent there.
Putting in (4.67) gives
2
1 2
2
'
1
2
x y
C
a a
F iF i U dz
z z

| |
= + + +
|
\ .
}

, ) , )
1
1
2 2
2
i i Ua t =
1
2 Ua t = (4.68)
where C satisfies z R > .
Now, (4.67) implies
1
'
2
C
dw
dz ia
dz
t =
}
Thus
j j
1
' '
2
C C
a w i t | = = + (4.68a)
where j
C
F denotes the difference of the value of F going once around C.
By means of analytic continuation, (4.68a) can be extended to C so that
j j
1
2
C C
a w i t | = = +
Now, since C is a streamline, const = there; hence,
j
1
2
C
C
a d t | = = =I
}
u l

(4.69)
Eq(4.68) thus becomes
x y
F iF i U = I
which is just the theorem.
4.12 Lift: The Deflection of Airstream
Weve explained the lift of an aerofoil in terms of the pressure difference due to the
negative circulation around the aerofoil (K-J theorem).
Another way to look at it is to say that the aerofoil imparts a downward momentum to
the oncoming stream & thereby, according to Newtons law of action-reaction, attains
an upward lift.
For a single aerofoil, the deflection of the airstream is not apparent since it becomes
undetectible at infinity.
One thus needs to study the flow past an infinite stack of airfoils for such effects.
5. Vortex Motion
5.1 Kelvins Circulation Theorem
5.2 Persistence Of Irrotational Flow
5.3 Helmholtz Vortex Theorem
5.4 Vortex Rings
5.5 Axisymmetric Flow
5.6 Motion Of A Vortex Pair
5.7 Vortices In Flow Past A Circular Cylinder
5.8 Instability Of Vortex Patterns
5.9 Steady Viscous Vortex Maintained By A
Secondary Flow
5.10 Viscous Vortices: Prandtl-Batchelor
Theorem
5.1 Kelvins Circulation Theorem
Theorem
Circulation is conserved along particle path for an ideal fluid subject to
conservative forces.
Proof
Let the circulation be
, ) C t
d I =
}
u x

(5.1)
To avoid confusion, let us denote derivatives of coordinates by and reserve the
symbol d for time derivatives. Hence [see Ex.5.2 for a more formal approach],
, ) C t
I =
}
u x

, ) , ) C t C t
d d d
dt dt dt

I
= +
} }
u x
x u

(5.1a)
Now
2
1
2
d d
dt dt

= = =
x x
u u u u u
so that
, ) , )
2 2
1 1
0
2 2
C C t C t
d
dt

1 = = =
] } }
x
u u u

since u is single valued.
Hence, (5.1a) becomes
, ) C t
d d
dt dt

I
=
}
u
x

For an ideal fluid under conservative forces = V f , the Euler eq. becomes
d p
dt t

| | c
= + V = V + = V
|
c
\ .
u u
u u
where
p

= +
Hence
, )
j 0
C
C t
d
dt

I
= V = =
}
x

since is single valued.


Notes on the Theorem
1. The theorem only applies to circuits that move with the fluid; not to stationary
ones.
2. The theorem applies whenever we can write
d
dt
= V
u
(a)
Hence, the listed assumptions on the fluid can be relaxed as long as (a) is valid.
3. Similarly, the theorem applies wherever (a) holds on a circuit C; whether it holds
for other regions of the fluid is immaterial.
4. The theorem does not require the fluid domain to be simply connected since we
did not invoke the Stokes theorem.
Generation of Lift on Aerofoil
As mentioned in 1.6 and Chapter 4, the uplift of an aerofoil requires a negative
circulation around the aerofoil.
If the aircraft is initially at rest, 0 I = . If the air is an ideal fluid, will always be
zero according to Kelvins circulation theorem. Hence, no flight is possible.
If air is viscous, a wake of separated boundary layers and vortices will always be
present downstream of the aircraft. In particular, if the starting vortex is positive, its
shedding will give the aerofoil a negative . [see Fig.5.2]
Encarsia Formosa
This tiny wasp employs the Weis-Fogh mechanism of lift generation that works even
for inviscid fluid. [see Fig.5.3 & caption for explanation]
5.2 Persistence Of Irrotational Flow
Cauchy-Lagrange Theorem
Consider an ideal fluid under conservative forces.
Any portion of fluid that is irrotational at some time will remain so afterwards.
Proof
Suppose = V u is not identically 0 for that portion V of fluid at some later time.
Then, by Stokes theorem, there exists a circuit C bounding the surface S of V such
that
0
C S S
d dS dS I = = V = =
} } }
u x u n
n

However, this violates Kelvins theorem since 0 I = at some time before. QED.
Comments
As discussed in 1.5, the vorticity eq.
, )
D
Dt
= V

u (1.25)
simplifies to
0
D
Dt
=

(1.27)
for 2-D flows, so that the above result is rather obvious.
The usefulness of the Cauchy-Lagrange theorem is that it applies also to 3-D flows.
For an irrotational flow, we can write
= V u (5.3)
where is single-valued if the flow region is simply connected.
If the flow is incompressible,
0 V = u ie.,
2
0 V = (5.4)
[See Ex.5.23-29 for applications in this respect]
5.3 Helmholtz Vortex Theorem
Avortex line is, at one instance, a curve
, ) , ) , ) , ) , , s x s y s z s = 1
]
x
whose tangent at each point is parallel to the vorticity vector
, )
, ,
x y z
= = V u (5.5)
ie.,
x y z
dy
dx dz
ds ds ds

= =
or
x y z
dx dy dz

= = (5.5a)
Avortex surface is a surface to which is tangent at each point. [fig.5.4b]
The set of all vortex lines that pass through a closed curve in space forms the
boundary of a vortex tube. [See fig.5.4a]
Theorems
Consider an ideal fluid under conservative forces so that Kelvins circulation theorem
applies. The Helmholtz vortex theorems state that
1. Vortex lines, & hence vortex tubes, move with the fluid.
2. The strength of a vortex tube, as defined by,
S
dS I =
}
n (5.6)
is time independent and the same for all cross section of the tube.
Proof of (1)
Consider a circuit C that lies, at time 0 t = , completely on a vortex surface S.
Let C be bounded by a simply connected subsurface
*
S of S.
The circulation around C is, with the help of Stokes theorem,
, )
* *
C S S
d dS dS I = = V =
} } }
u x u n
n

which is simply the strength of a vortex tube that goes through C.


Since
*
S is a vortex surface, 0 = n there. Hence
0 I = on C
Now, let , ) C t be the circuit composed of, at time t, the same fluid particles as C at
0 t = . [In short, C moves with the fluid flow to , ) C t at time t.]
Let , ) t I the circulation of , ) C t . According to Kelvins theorem, , ) 0 t I = for all
t.
Since C is arbitrary, this means
0 = n on , ) S t
where , ) S t is the surface composed of, at time t, the same fluid particles as S at
0 t = . [In short, S moves with the fluid flow to , ) S t at time t.]
Thus, , ) S t is also a vortex surface.
Another way to say it is that a vortex surface moves with the flow.
A vortex line can be considered as the intersection of 2 vortex surfaces.
Since both surfaces moves with the flow, so is their intersection, the vortex line.
Proof of (2)
Consider the volume V inside a vortex tube bounded by 2 of its cross sections [see
fig.5.4a].
Denote the section of the vortex tube surface involved by
T
S ; the cross sections by
1
S and
2
S ; their bounding circuits,
1
C and
2
C , respectively.
The Gauss theorem gives
1 2 T
S S S V
dS dV
1
+ + = V
(
(
]
} } } }
n
Now, since
T
S is a vortex surface, 0 = n everywhere on it so
0
T
S
dS =
}
n
Next,
, ) 0 V = V V = u
so
0
V
dV V =
}

Hence,
1 2
0
S S
dS
1
+ =
(
(
]
} }
n
Noting that n here are outward normals of the surface bounding V, this becomes
1 2
I = I
where
i
I is the strength of the tube at
i
S , as defined by (5.6). QED.
Thin Vortex Tube
Let the cross section S of a thin vortex tube be small enough so that is
practically constant over it. Thus, S I = .
Since the surface of the tube is a vortex surface, it moves with the flow.
Thus, the volume of fluid inside the tube and bounded by 2 cross sections will remain
thus bounded during the flow. Since the flow is incompressible, this volume is
conserved. Therefore, stretching the tube will decrease S . Since is a constant,
must increase to compensate.
To summarize, stretching a vortex tube will intensify local vorticity.
[see fig.5.5 for application to a tornado; fig.5.6 for that to spin down of a cup of tea].
5.3.1 Vorticity Equation
The Helmholtz theorems were originally proved starting with the vorticity equation
, )
D
Dt
= V

u (5.7)
Some of the conclusions drawn from the Helmholtz theorems are also obtainable
directly through (5.7)
One such example is the local intensification of vorticity due to the stretching of a
vortex tube.
Let the vortex lines be mostly in the z-direction so that k and
D
Dt z

c
c
u
(5.8)
The z-component of which is
D w
Dt z


c
c

Now, stretching the fluid in the z direction makes 0


w
z
c
>
c
so that 0
D
Dt

> as
promised.
Another simple example is that of 2-D flow.
Vortex tubes are all straight & parallel to the z-axis and 0 w = .
There is no stretching of tubes and
0
D
Dt

= (5.9)
Still another example is the axisymmetric flow, which, in terms of cylindrical
coordinates , ) , , R z , is of the form
, ) , ) , , , ,
R R z z
u R z t u R z t = + u e e (5.10)
Since u is independent of , the streamlines all lie in planes const = .
The vorticity is
1
0
R z
R z
R z
R
u u
R R z z R
u u

c c c c c
| |
= =
|
c c c c c
\ .
e e e
e (5.11)
The vortex lines are rings centered around the z axis.
Vortex tubes are surfaces of doughnuts similarly centered.
According to the 1
st
Helmholtz theorem, they move with the fluid, which means they
can only expand or contract about the z axis.
Incompressibility of the fluid means the volume enclosed by each tube is conserved.
Let the radius of a thin tube be R, its cross section S , so that its volume is 2 R S .
As R varies in time with the flow, S also changes to maintain a constant volume.
However, according to the 2
nd
Helmholtz theorem, S const = so that
const R =
In more precise language, we write
0
D
Dt R

| |
=
|
\ .
(5.12)
which can be deduced more formally from (5.7) using (5.10).
When, in axisymmetric flow, an isolated vortex tube surrounded by irrotational flow
is called a vortex ring. [see fig.5.7].
5.3.2 Proof of eq(5.12)
Using
, ) , ,
R R z z
u u R z t = + = u e e u

= e where , ) , ,
R z
u u
R z t
z R

c c
= =
c c
we have
, ) , )
R R z z R
u u u
R R

c
V = + =
c
u e e e
where weve used
R

c
=
c
e e
Thus, the vorticity eq (5.7) becomes
R
D
u
Dt R

e
Writing
, )
D D D
R
Dt Dt Dt R

| |
= =
|
\ .

e e
, )
D D
R R
Dt R R Dt


| |
= +
|
\ .
e e
Now
, ) , )
D
R R
Dt

= V e u e
, )
R z
u u R
R z

c c
| |
= +
|
c c
\ .
e
R
u

= e
so that
R R
D D
R u u
Dt Dt R R R


| |
= + =
|
\ .

e e e
ie.,
0
D
Dt R

| |
=
|
\ .
5.4 Vortex Rings
Collision
Leapfrog
5.5 Axisymmetric Flow
5.5.1 The Stokes Stream Function
5.5.2 Irrotational Flow Past A Sphere
5.5.3 Hills Spherical Vortex
5.5.1 The Stokes Stream Function
For an incompressible 2-D flow,
, )
z
= V u e (4.8)
ensures that the incompressibility condition 0 V = u is automatically satisfied while
the stream function , ) , , x y t = is constant along streamlines, ie.,
, ) 0 V = u (4.7)
The question is whether an analogous form
, )
'

= V u e , with , ) ' ' , , R z t = ,
exists for axisymmetric incompressible flows.
Now
, ) ' '
R z
u u u
R R z

| | c c c
V = + +
|
c c c
\ .
u
'
R z
u u
R z

c c
| |
= +
|
c c
\ .
Using
1 ' ' '
0 ' 0
R z
R z
R
R R z z R R
R

c c c c c
| |
= = + +
|
c c c c c
\ .
e e e
u e e
we have
, )
' ' ' ' ' ' '
'
z R R R z R z

c c c c c
| |
V = + + =
|
c c c c c
\ .
u
so that ' is not generally constant along streamlines.
Now, the offending term originates from the curl of ' in the expression of u. It is
easily removed by setting '
R

+
= so that
R

+
| |
= V
|
\ .
u e (5.13)
1
0 0
R z
R z
R
R R z z R

c c c c+ c+
= = +
c c c c c
+
e e e
e e
whereupon
, ) 0
z R R z
c+ c+ c+ c+
V + = + =
c c c c
u
ie., the Stokes stream function + is constant along streamlines.
In spherical coordinates , ) , , r ,
2
sin
1
sin
sin
r
r
r r
r r
A rA r A

c c c
V =
c c c
e e e
A
The analog (5.13) of is obviously
sin r

+
| |
= V
|
\ .
u e (5.14)
so that
2
sin
1 1
sin sin
0 0
r
r
r r
r r r r r


c c c c+ c+
| |
= =
|
c c c c c
\ .
+
e e e
u e e (5.15)
while , ) , , r t + = + is constant along streamlines:
, )
sin
r
u
u
u
r r r


| | c c c
V + = + + +
|
c c c
\ .
u
r
u
u
r r

c c
| |
= + +
|
c c
\ .
1 1
0
sin r r r r r
c+ c+ c+ c+
| |
= =
|
c c c c
\ .
5.5.2 Irrotational Flow Past A Sphere
In a steady flow, the vorticity eq. for axisymmetric flow,
0
D
Dt R

| |
=
|
\ .
(5.12)
reduces to
, ) 0
sin r

| |
V =
|
\ .
u (5.16)
where sin R r = .
Thus
sin r

is constant along a streamline.


Consider a uniform inviscid flow past a rigid sphere with surface r a = .
If there are no closed streamlines, all streamlines will start and end at the infinities,
where 0 = . Hence, according to (5.16), 0 = everywhere and the flow is
irrotaional.
Now, for an axisymmetric flow in spherical coordinates:
, ) , ) , , , ,
r r
u r t u r t

= + u e e
2
sin
1
sin
0
r
r
r r
r r
u ru


c c c
= V =
c c c
e e e
u
, )
1
r
u
ru
r r

c c
1
= =
(
c c
]
e e
where
, )
1 1
r r
u u u u
ru
r r r r r


c c c c
1
= = +
(
c c c c
]
(5.17)
In terms of the Stokes stream function defined by
1
sin sin
r
r r r r


+ c+ c+
| | | |
= V =
| |
c c
\ . \ .
u e e e (5.14-5)
we have
2
2 2
1 1 1 1
sin sin r r r


1 c + c c+
| |
=
| (
c c c
\ .
]
(5.18)
Thus, for our irrotational flow past a sphere, we need to solve
2
2 2
sin 1
0
sin r r


c + c c+
| |
+ =
|
c c c
\ .
r a > (5.19)
with boundary conditions
0
r
u = on r a = [impenetrable wall] (5.20a)
and
cos
r
u U , sin u U

as r
which, by (5.15), means
2
1
cos
sin
U
r


c+

c
,
1
sin
sin
U
r r

c+

c
as r
or
2
sin cos r U

c+

c
,
2
sin rU
r

c+

c
as r (5.20b)
which, upon integration, becomes
, )
2 2
1
1
sin
2
r U C r + +
, )
2 2
2
1
sin
2
r U C + + as r
In order for these 2 expressions to be compatible, the arbitrary functions
1 2
, C C must
be constants that can be conveniently set to zero. Hence
2 2
1
sin
2
r U + as r (5.20)
Eq (5.19), as well as the boundary conditions (5.20,a,b), suggests that we can look for
a separable solution, ie.,
, ) , ) , ) , r R r + = O
whereupon, (5.19) becomes 2 ODEs
2
2 2
0
d R
R
dr r

+ = (5.21a)
1
sin
sin
d d
d d


O
| |
= O
|
\ .
(5.21b)
where is a constant.
In order to satisfy (5.20), we have
, )
2
1
2
R r r U =
2
sin O = (5.21c)
Substituting into (5.21b), we have
, )
2
. . . 2sin cos 2sin
d
L H S
d

= =
2
. . . sin R H S =
so that (5.21b) is satisfied only if 2 = ; whereupon, (5.21a) becomes
2
2 2
2
0
d R
R
dr r
= (5.21d)
Since this is homogeneous, we look for a power solution
n
R r =
which turns (5.21d) into
, ) 1 2 0 n n =
with solutions
2 n = , or 1
so that
2
B
R Ar
r
= + with A, B constants.
From (5.21c), we have
1
2
A U =
so that
, )
2 2
1
, sin
2
B
r Ur
r

| |
+ = +
|
\ .
Finally, to satisfy (5.20a), we have
2
2 2 3
1 1 1 2
2sin cos cos
sin sin 2
r
B B
u Ur U
r r r r


c+
| | | |
= = + = +
| |
c
\ . \ .
so that at r a =
3
2
cos 0
B
U
a

| |
+ =
|
\ .
which gives
3
1
2
B Ua =
so that
, )
3
2 2
1
, sin
2
a
r U r
r

| |
+ =
|
\ .
for r a > (5.21)
3
3
1 cos
r
a
u U
r

| |
=
|
\ .
3 3
2 3
1 sin
2 1 sin
sin 2 2
U a a
u r U
r r r r r

| | | | c+
= = + = +
| |
c
\ . \ .
The streamlines const + = are shown in Fig.5.12a.
The slip velocity on the sphere surface is
, ) 0
r
u r a = =
, )
3
sin
2
u r a U

= = (5.22)
Thus, the fore & aft points 0, = are stagnation points.
Maxima are at
2

= .
According to the Bernoulli theorem, this implies a severe adverse pressure gradient
over the downstream side of the sphere. ( 0
2

< < ). Alarge wake with separated


boundary layer is expected.
5.5.3 Hills Spherical Vortex
In 5.5.2, we solved the steady axisymmetric flow outside a solid sphere r a s .
Now, let the surface r a = to be an imaginary one in the fluid. We shall show that
there is a flow pattern, called the Hills spherical vortex, inside the sphere that can
match up at r a = with the one given in 5.5.2.
In other words, one can find vortices embedded in a steady, uniform flow. [see
Fig.5.12]
The problem is to solve
, ) 0
sin r

| |
V =
|
\ .
u for r a < (5.16)
with the boundary conditions
, ) 0
r
u r a = =
, )
3
sin
2
u r a U

= = (5.22)
Now, (5.16) means that
sin r

is a constant along each streamline, a property


shared by the Stokes stream function . Hence, (5.16) implies
, )
sin
c
r

= +
where c is an arbitrary function.
Using the spherical coordinate relation (5.18) between and , we have
, )
2
2 2
1 1 1 1
sin
sin sin
c r
r r r


( c + c c+
| |
= = +
| (
c c c
\ .

or
, )
2
2 2
2 2
sin 1
sin
sin
c r
r r



c + c c+
| |
+ = +
|
c c c
\ .
(5.23)
Now, in terms of , (5.22) become
2
1
0
sin
r a
a
=
c+
=
c
1 3
sin
sin 2
r a
U
a r

=
c+
=
c
(5.24)
Now, the easiest way to match (5.24) is to try the same type of ansatz as in the r a >
region, namely,
, ) , )
2
, sin r g r + = for r a <
Plugging into (5.23) gives
, )
2
2 2 2 2
2 2
sin 2 sin sin
d g g
c r
dr r
= +
or
, )
2
2 2 4
1
2
d g g
c
r dr r
= + (5.24a)
Since the R.H.S. is independent of , we must have , ) c const + = .
The homogeneous eq to (5.24a) the same as the radial eq. for r a > . Thus, the
homogeneous solution is
2
B
g Ar
r
= +
To find the particular solution, we again try a power solution
n
g r =
so that (5.24a) becomes
, )
4
1 2
n
n n r c

= (

which gives
4 n = and
10
c
=
Hence, the general solution for (5.24a) is
2 4
10
B c
g Ar r
r
= +
so that
, )
2 4 2
, sin
10
B c
r Ar r
r

| |
+ = +
|
\ .
Now, to keep regular at 0 r = , we must set 0 B = so that
, )
2 4 2
, sin
10
c
r Ar r
| |
+ =
|
\ .
and
2 4 2
2
2
sin cos 2 cos
sin 10 10
r
c c
u Ar r A r
r

| | | |
= =
| |
\ . \ .
3 2 2
1 4 1
2 sin 2 sin
sin 10 5
c
u Ar r A cr
r

| | | |
= =
| |
\ . \ .
Matching the conditions (5.24) at r a = requires
2
2 cos 0
10
c
A a
| |
=
|
\ .
and
2
1 3
2 sin sin
5 2
A ca U
| |
=
|
\ .
ie.,
2
0
10
c
A a =
2
1 3
5 4
A ca U =
with solution
3
4
A U =
2
15
2
U
c
a
=
so that
, )
2
2 2
2
3
, 1 sin
4
r
r Ur
a

| |
+ =
|
\ .
for r a < (5.25)
The corresponding streamlines, which are all closed, can be found in Fig.5.12b.
The circulation around each streamline differs for different streamlines.
The maximum is the streamline that goes around a full hemispherical cross section.
By Stokes theorem,
max
0 0
a
hemi
dS d drr

I = =
} } }
Using
2
15
sin sin
2
U
cr r
a
= =
we have
, )
3
2
max 2 2
0 0
15 15
sin 2 5
2 2 3
a
U U a
d drr Ua
a a


| |
I = = =
|
\ .
} }
Thus, a Hill spherical vortex will travel through stationary fluid with a speed
max
5
U
a
I
=
5.6 Motion Of A Vortex Pair
A line vortex is defined as
2 r

I
= u e
Consider a vortex pair of strength I and a distance 2d apart as shown in
Fig.5.13a. Each will induce a downward flow / 4 d I at the instantaneous position
of the other. Thus, the pair, as well as the accompanying streamline pattern, moves
downward at that speed.
In a reference frame moving with the vortices, an additional uniform upward flow
/ 4 d I is experienced. The full complex potential is therefore
, ) , ) ln ln
4 2 2
i z i i
w z d z d
d
I I I
= + + (5.26)
When calculating the velocity of one of the vortices, the contribution of that vortex to
w must be subtracted. Thus, the velocity , ) , U V of the vortex at z d = is given by
, ) ln
4 2
z d
d i z i
U iV z d
dz d
=
I I
(
= + +
(

, )
0
4 2
z d
i i
d z d
=
(
I I
= + =
(
+

(5.27)
ie., its stationary as stated before.
Separating the real and imaginary parts in (5.26), we have
ln
4 2 4 2
y x z d
w i
d d z d


I I I I (
= + +
(
+

where
i
z d z d
e
z d z d


=
+ +
Thus, the stream function is
2ln
4
x z d
d z d

I (
= +
(
+

Now,
z d x iy d
z d x iy d
+
=
+ + +
, ) , )
, ) , )
, )
, )
2 2
2
2
2
x d iy x d iy x d y z d
z d x d iy x d iy
x d y
+ +
= =
+ + + +
+ +
so that
, )
, )
2
2
2
2
ln
4
x d y x
d
x d y

(
+ I
= + (
`
+ +
(

)

The corresponding streamlines are sketched in Fig.5.13b.
5.7 Vortices In Flow Past A Circular Cylinder
Consider a circular cylinder of radius a initially at rest in a fluid of kinematic
viscosity . Let it be suddenly set in uniform motion with speed U perpendicular to its
axis so that the Reynolds number
2aU
R

(5.28)
is about 200.
5.7.1 The Initial Phase: Almost Irrotational Flow
5.7.2 Flow At Later Stage: The von Karman Vortex
Street
5.7.3 A Simple Model
5.7.1 The Initial Phase: Almost Irrotational Flow
If the fluid were inviscid, our 2-D flow would be governed by
0
D
Dt

(5.9)
Since the fluid is initially at rest, 0 .
According to (5.9), it will remain so for all subsequent time so the flow will always be
irrotational.
Consider now our viscous fluid.
During a very short initial phase before the cylinder has moved a distance comparable
to its radius, the flow is predominantly irrotational (see fig.4.4a).
There is intense vorticity in the rapidly thickening boundary layer.
However, there is not yet time for separation to occur so the mainstream is still free of
vorticity.
At this stage, results of the (inviscid) irrotational flow give the velocity at the edge of
the boundary layer.
Positions at which reversed flow 1
st
occur are those where the velocity decreases most
rapidly with distance along the boundary in the flow direction.
In case of a circular cylinder, this happens at the rear stagnation point. (see Fig.5.14a)
5.7.2 Flow At Later Stage: The von Karman Vortex
Street
Once reversed flow is initiated, the flow begins to differ substantially from the
irrotational results.
1
st
, the 2 attached eddies behind the cylinder grow in size (Fig.5.14b).
Then the flow becomes asymmetric about the centerline (Fig.5.14c).
Finally, it settles into an unsteady but highly structured flow in which vortices are
shed alternatively from the 2 sides of the cylinder, giving rise to the von Karman
vortex street (Fig.5.14d,e).
Note that the wake is entirely turbulent for 2000 R while for 30 R , the 2 eddies
remain symmetrically attached.
5.7.3 Von Karman Vortex Street: A Simple Model
We now model a fully formed vortex street (Fig.5.14d,e) by 2 sets of line vortices as
shown in fig.5.15.
Members of the lower set of vortices, all of strength , are at z na = , with
0, 1, 2, n = .
Those of the upper set, of strength I, are at
1
2
z n a ib
| |
= + +
|
\ .
.
As in 5.6, we assume each line vortex to move with the local velocity due to the
combined flow of everything other than itself.
Consider any vortex.
Only the x-component U of velocity due to vortices in the other row is finite.
All else vanishes since contributions from other vortices cancel in pairs.
By symmetry, U is the same (and points to the left for 0 I > ) for all vortices.
Since the complex potential , ) w z due to a line vortex at
0
z is
, ) , )
0
ln
2
i
w z z z

I
= (4.21)
the total complex potential due to the lower line of vortices is
, ) , ) ln
2
n
i
w z z na


=
I
=
_
, ) , )
1
1
ln ln ln
2
n n
i
z z na z na


= =
I (
= + +
(

_ _
, ) , )
1
1
ln ln ln
2
n n
i
z z na z na


= =
I (
= + + +
(

_ _
, )
2 2 2
1
ln ln
2
n
i
z z n a


=
I (
= +
(

_
, )
2
2 2
2 2
1 1
ln ln 1 ln
2
n n
i z
z n a
n a

= =
( | | I
= + +
( |
\ .
_ _
2
2 2
1
ln 1
2
n
i z
z C
n a

=
( | | I
= +
( |
\ .
[
where
, )
2 2
1
ln
2
n
i
C n a


=
I
=
_
is a constant and can be dropped.
Using the identity (see Carrier et al)
2
2
1
sin 1
n
z
z z
n


=
| |
=
|
\ .
[
we have
ln sin
2
i z
w const
a

I
| |
= +
|
\ .
(5.29)
Hence
cot
2
dw i z
dz a a
I
=
For an upper row vortex at
1
2
z n a ib
| |
= + +
|
\ .
, we have
1
cot tan tanh
2 2 2 2
dw i b i b b
n i i
dz a a a a a a

I I I ( | | | | | |
= + + = =
| | |
(
\ . \ . \ .
Thus, the whole street moves with velocity
tanh , 0
2
b
a a
I ( | |
=
|
(
\ .
u
5.8 Instability Of Vortex Patterns
Consider 2 line vortices of equal strength and distance 2a apart.
Each vortex induces, at the center of the other, a flow of speed
4 a
I
perdendicular
to the line joining them. Hence, both rotates about the mid-point of that line with
angular velocity
2
/ 4 a I .
Next, consider n line vortices of strength placed symmetrically on a circle of radius
a. (see Fig.5.16) The angular separation between adjacent vortices is
2
n

. The
distance between the mth neighbors is therefore 2 sin
2
m
m
d a

= , where
2
m
m
n

=
with
2
n
m s , and the floor a of a is the largest integer that is equal or less than a.
Note that all neighbors occur in pairs except for the
2
n
th one, which is unique, when
n is even.
Consider a vortex and the pair of its mth neighbors. The line joining a vortex to its
mth neighbor makes an angle
2
m

with the radius that passes through the vortex.
The flow at the vortex, induced by that neighbor, then makes an angle
2
m
m
n

=
with the radius. By symmetry, the radial components of the flows induced by a pair
of the neighbors cancels each other. The tangential component is equal to
2 sin
2 2 2
m
m
d a


I I
=
For n even, there is only one
2
n
th neighbor situated at the other end of the diameter
that goes through the vortex. The induced flow is again tangential with magnitude
4 a
I
.
For n odd, there are , ) 1 / 2 n pairs of neighbors so that the total flow is
, )
1
1
2 2 4
n
n
a a
I I
| |
=
|
\ .
.
For n even, there are / 2 n pairs and a unique
2
n
th neighbor. The total flow is
, ) 1
2 2 4 4
n
n
a a a
I I I
| |
+ =
|
\ .
In both case, the vortex rotates with angular velocity
, )
2
1
4
n
a
I
O = (5.32)
Stability of a dynamical system can be investigated by the introduction of a small
perturbation. If the state of the system remains close to the unperturbed one, it is
stable. Otherwise, it is unstable.
There are cases that do not quite fit in this scheme of classification. For example,
some systems are stable under infinitesimal perturbation but unstable under finite
perturbation. Others can be stable under any instantaneous perturbation but unstable
under a prolonged one. These cases are sometimes called metastable.
An introduction to the subject can be found in chapter 9.
Here, we list, without proof, some results of interest concerning the stability of vortex
patterns.
1. The double row vortex pattern in fig.5.15, 5.7.3, is unstable except for the case
cosh 2
b
a

= ie., 0.281
b
a
(5.31)
2. The equally spaced vortices on a circle pattern is stable for 7 n < , unstable for
7 n > and metastable for 7 n = . This is observed in liquid helium.
Finite Patches of Concentrated Vorticity
An elliptical path of uniform vorticity rotates with angular velocity
, )
2
ad
a b
O =
+
(5.33)
where a and b are the semi-axes of the ellipse.
This motion is stable for
1
3
3
b
a
s s .
A circular array of n patches of vorticity is unstable even for 7 n < , if the patches are
big enough. The critical patch size decreases with increasing n.
For the finite patch version of the vortex street, there is again only 1 spacing ratio for
which the pattern is stable. This ration is close to the von Karman value and only
weakly dependent on the patch size.
A patch of 2-D vorticity distribution
v u
x y

c c
=
c c
is still governed by the vorticity eq.
0
D
u v
Dt t x y
c c c
= + + =
c c c
(5.9)
For incompressible flow, we also have
0
u v
x y
c c
+ =
c c
(5.35)
Now, (5.9) implies is conserved for a fluid element while (5.35) implies the
cross-section of that element in the x-y plane is conserved. Thus
dS const =
}
(5.36)
where the integration is over the whole flow plane.
There are other relationships of this kind:
x dS const =
}
y dS const =
}
(5.37)
Vortex Merging
Let 2 circular patches of uniform and equal vorticity, each of radius R, have centers a
distance d apart at 0 t = .
For 3.5
d
R
> , the patches end up deformed and rotating about a common center.
For 3.5
d
R
< , the patches quickly merge. To satisfy the conservation laws, they do
this by wrapping around each other with a layer of irrotational flow between them.
When the vorticity are of opposite sign, a vortex pair may be formed.
5.9 Steady Viscous Vortex Maintained By A
Secondary Flow
In the presence of viscosity, the vorticity eq. becomes
, ) , )
2
t

c
+ V = V + V
c

u
u
(5.38)
The processes controlled by the various terms are roughly:
, ) V u

: convection of vortex lines.
, ) V u: intensification of vorticity (when vortex lines are stretched).
2
V : diffusion of vorticity.
Burgers Vortex
One known situation that involves all 3 processes and still admits an exact solution to
the Navier Stokes eqs is known as the Burgers vortex.
Basically, it is a line vortex whose viscous decay (2.4.4) is countered by a secondary
flow which
1. sweeps the vorticity back towards the axis.
2. intensifies the vorticity by stretching fluid element along the axis direction.
The result is a steady vortex of the form
1
2
r
u r =
, )
2
/ 4
1
2
r
u e
r


I
=
z
u z = (5.39)
where 0 > and are constants.
The velocity profile is shown in fig.5.18.
The vorticity
, )
2
/ 4
1
1
1
2 2
r z
r
r
r r z
r e z


c c c
=
c c c
I

e e e

2
/ 4
4
r
z
e



I
= e (5.40)
is concentrated within a core of radius of order

.
In (5.39), the secondary flow, as parametrized by , and the rotary flow, as
parametrized by are not coupled since both parameters can vary independently.
For a real flow, the secondary and rotary flows are usually coupled due to the
presence of rigid boundaries.
5.10 Viscous Vortices: Prandtl-Batchelor
Theorem
For a 2-D steady inviscid flow, the vorticity eq. reduces to
, ) 0 V = u where = k
ie., is constant along a streamline so that we can write
, ) = (5.41)
since the stream function is, by definition, constant along a streamline.
If, as in fig.1.8, all flow lines can be traced to the infinities where the flow is uniform,
we can deduce immediately that 0 = everywhere.
If closed streamlines are present, no simple conclusion can in general be drawn for
inviscid flows.
For steady viscous flow, the Navier-Stokes eq become
, )
2
p

V = V + V u u u
Using
, ) , )
2
1
2
V = V V u u u u u
, ) , )
2
V V = V V V u u u
we have
, ) , ) , )
2
1
2
p

V V = V + V V V V 1
]
u u u u u
or
, ) , )
2
1
2
p

1
V V = V + V + V
(
]
u u u u u
Integrating over a closed contour C, we have
, ) , )
C C
d d V V = V 1 1
] ] } }
u x u u x

If C is a streamline, dx is parallel to u so that the right hand side vanishes. Thus
, ) 0
C
d V V = 1
] }
u x

if 0 =
ie.,
j 0
C
d V =
}
x

(5.42)
For our 2-D flow,
, , 0
0 0
x y z y x

| | c c c c c
V = =
|
c c c c c
\ .
i j k

, , 0
d
d y x

| | c c
=
|
c c
\ .
, ) , , 0
d d
u v
d d


= = u
so that (5.42) becomes
0
C
d
d
d

=
}
u x

(5.43)
or
0
C
d
d

I =
where
d
d

can be taken outside the integral because is constant on the streamline


C. In case of a finite circulation
C
I , we have
0
d
d

=
which is the Prandtl-Batchelor theorem: in a steady 2-D viscous flow, the vorticity
is constant throughout any region of closed streamline in the limit 0 .
6. The Navier-Stokes Equations
6.1 Introduction
6.2 The Stress Tensor
6.a Navier-Stokes Eqs.
6.B Convection Theorem
6.2 The Stress Tensor
Let x be the position vector of some point fixed in the fluid, and S a small
geometrical surface element, with unit normal n, that passes through x.
Let the force exerted on this surface by the fluid on the side pointed by n be
S t (6.1)
where t is called the stress vector.
For an inviscid fluid, we have
, p t t x n
so that the stress aligns with the unit normal.
For viscous fluid, we expect t to have non-vanishing components tangent to the
surface. This is described by the stress tensor T whose Cartesian component
ij
T is
defined as the i-component of stress on a surface with unit normal
j
n e .
Thus, for a surface with unit normal
j j
n n e , the stress vector is
i ij j
t T n or T t n (6.3)
For a more fastidious proof, see Acheson.
6.3 Cauchys Equation Of Motion
Convection Theorem
Ref: R.E.Meyer, Introduction to Mathematical Fluid Dynamics, Wiley (71)
Theorem 3.1
If , ) , )
1
,
t
f t C e O x ,
D
Dt
fdV
Df
Dt
f dV
t t
O O

= + V

v (3.5)
Corollary 3
Let
1 1
( ) t t O = O = ,
D
Dt
fdV
t
fdV f dS
t
O O

=
c
c
+
cO
1 1
v n (3.6)
Proof
0 t t
H O = O
So that
fdV g t J t dV
t
O O

= ( , ) ( , ) a a
0
0
where
, ) , ) , )
, , , g t f t t = a x a
( , 0) = a x a
det
i
j
x
J
a
c
=
c
Thus
D
Dt
fdV
t
gJdV
t
gJ dV
t
O O O

=
c
c
=
c
c
0 0
0 0

= +
c
c

J
Df
Dt
f
J
t
dV
0
0
O
= +
c
c

Df
Dt
f
J
J t
dV
t
O
= + V

Df
Dt
f dV
t
v
O
if
c
c
= V
J
t
J v
Now:
i n
ik kn nk
n i
x x
J A A
a a

c c
= =
c c
where
kn
A are the cofactors (Caution: our
ij
A is Meyers
ji
A ),
( )
k n
kn kn
A
+
=
where
kn
is the determinant of the minor matrix obtained by striking out the kth
row & nth column of the matrix
i
j
x
a
| |
c
|
|
c
\ .
.
Writing the ith row of J as a vector
i
r with components , )
i
j
r , we have
, ) , ) , ) , )
1 2 3 1 2 3 1 2 3 1 2 3
det , , det , , det , , det , , c = c + c + c r r r r r r r r r r r r
, ) , ) , )
1 1 2 2 3 3 k k k
k k k
A A A = c + c + c r r r
where is the differential operator, ie.,
j
jk
k
x
J A
a
c
c = c
c
[see Determinant.doc for a better derivation]
Hence:
k k k i k
kn kn kn ik
n n i n i
J x v v x v
A A A J J
t t a a x a x

| | | | c c c c c c c
= = = = = V
| |
c c c c c c c
\ . \ .
v QED
For the corollary:
Df
Dt
f
f
t
f + V =
c
c
+ V v v
so that
V =

cO
f dV f dS v v n
O
1 1
6.4 A Newtonian Viscous Fluid: The
Navier-Stokes Equations
6.a Navier-Stokes Eqs.
Consider the Euler eq
, ) D D
p
Dt Dt

+ V = = V
u u
u u
Using
, ) , )
, )
D
Dt t


c
+ V = + V + V
c
u u
u u u u u u
, )
, )
t

c
= + V
c
u
uu
we can re-write it as
, )
, ) 0 p
t

c
+ V + V =
c
u
uu
or
, )
0
t
c
+ V H =
c
u
(A)
where the momentum flux density tensor is
ij i j ij
u u p H = +
Eq(A) is just the eq of continuity for momentum which expresses the conservation
of momentum.
Since we expect the principle of the conservation of momentum to be valid even in
viscous fluids, eq(A) should apply there with a suitable modification of . We thus set
ij i j ij
u u H =
where the stress tensor
ij
is related to the viscous stress tensor '
ij
by
'
ij ij ij
p = +
[Note: Acheson used
ij
T instead of
ij
(see chap 6) ]
What this means is that the generalization of the Eulers eq to viscous fluids can be
done by replacing p V with ' p V = V + V so that, for example,
D
p
Dt
= V
u
becomes
'
D
p
Dt
= V + V
u
or
' p
t

c
| |
+ V = V + V
|
c
\ .
u
u u (B)
Next, we need to determine the form of ' .
Now, viscous effects are due to friction between adjacent fluid elements moving
with different velocities. Hence ' should depend on
i
j
u
x
c
c
but not on u itself.
Furthermore, theres no friction if the fluid rotates as a whole with uniform angular
velocity so that = u
r
or
i ijk j k
u x = O .
Now:
i m
ijk j km ijm j mji j
m i
u u
x x

c c
= O = O = O =
c c
and
0
i
iji j
i
u
x

c
= O =
c
Hence, the choice
' '
j
i k
ij ij ji
j i k
u
u u
a b
x x x

| | c
c c
= + + =
|
|
c c c
\ .
where a, b are constants, will satisfy the above requirements.
A more common form is
2
'
3
j
i k k
ij ij ij
j i k k
u
u u u
x x x x

| | c
c c c
= + +
|
|
c c c c
\ .
(C)
where the positive constants
a =
2 2
3 3
b a b = + = +
are called coefficients of viscosity. (Landau used instead of )
This also defines the Newtonian fluid. [cf eq(2.1)].
Using (C), (B) becomes
'
ji
i i
j
j i j
u u p
u
t x x x

| | c
c c c
+ = +
|
|
c c c c
\ .
2
3
j
i k k
ij ij
i j j i k k
u
p u u u
x x x x x x

(
| | c
c c c c c
= + + +
( |
|
c c c c c c
(
\ .

2
3
j
k i
i k j j i
u
u u
p
x x x x x

| | c ( c c c c
| |
= + + + +
|
| (
|
c c c c c
\ .

\ .
1
3
k i
i k j j
u u
p
x x x x

( c c c c
| |
= + + +
| (
c c c c
\ .

or in vector form:
' p
t

c
| |
+ V = V + V
|
c
\ .
u
u u
2
1
3
p
(
| |
= V + V + V
|
(
\ .

u u
, )
2
1
3
p
| |
= V + + V V + V
|
\ .
u u (D)
For incompressible fluids, we have the Navier-Stokes eqs.:
2
p
t

c
| |
+ V = V + V
|
c
\ .
u
u u u (2.3)
0 V = u
Note also that for incompressible fluids,
'
j
i
ij
j i
u
u
x x

| | c
c
= +
|
|
c c
\ .
[cf (2.1)]
6.B Convection Theorem
Ref: R.E.Meyer, Introduction to Mathematical Fluid Dynamics, Wiley (71)
Theorem 3.1
If , ) , )
1
,
t
f t C e O x ,
D
Dt
fdV
Df
Dt
f dV
t t
O O

= + V

v (3.5)
Corollary 3
Let
1 1
( ) t t O = O = ,
D
Dt
fdV
t
fdV f dS
t
O O

=
c
c
+
cO
1 1
v n (3.6)
Proof
0 t t
H O = O
So that
fdV g t J t dV
t
O O

= ( , ) ( , ) a a
0
0
where
, ) , ) , )
, , , g t f t t = a x a
( , 0) = a x a
det
i
j
x
J
a
c
=
c
Thus
D
Dt
fdV
t
gJdV
t
gJ dV
t
O O O

=
c
c
=
c
c
0 0
0 0

= +
c
c

J
Df
Dt
f
J
t
dV
0
0
O
= +
c
c

Df
Dt
f
J
J t
dV
t
O
= + V

Df
Dt
f dV
t
v
O
if
c
c
= V
J
t
J v
Now:
i n
ik kn nk
n i
x x
J A A
a a

c c
= =
c c
where
kn
A are the cofactors (Caution: our
ij
A is Meyers
ji
A ),
( )
k n
kn kn
A
+
=
where
kn
is the determinant of the minor matrix obtained by striking out the kth
row & nth column of the matrix
i
j
x
a
| |
c
|
|
c
\ .
.
Writing the ith row of J as a vector
i
r with components , )
i
j
r , we have
, ) , ) , ) , )
1 2 3 1 2 3 1 2 3 1 2 3
det , , det , , det , , det , , c = c + c + c r r r r r r r r r r r r
, ) , ) , )
1 1 2 2 3 3 k k k
k k k
A A A = c + c + c r r r
where is the differential operator, ie.,
j
jk
k
x
J A
a
c
c = c
c
[see Determinant.doc for a better derivation]
Hence:
k k k i k
kn kn kn ik
n n i n i
J x v v x v
A A A J J
t t a a x a x

| | | | c c c c c c c
= = = = = V
| |
c c c c c c c
\ . \ .
v QED
For the corollary:
Df
Dt
f
f
t
f + V =
c
c
+ V v v
so that
V =

cO
f dV f dS v v n
O
1 1
7. Very Viscous Flow
7.1 Introduction
7.2 Low R Flow Past A Sphere
7.3 Corner Eddies
7.4 Uniqueness & Reversibility of Slow Flows
7.5 Swimming At Low R
7.6 Flow in a Thin Film
7.7 Flow in a Hele-Shaw Cell
7.8 Adhesive Problem
7.9 Thin Film Flow Down a Slope
7.10 Lubrication Theory
7.1 Introduction
For a steady viscous flow, the Navier-Stokes eqs become
, )
2
1
p

V = V + V + u u u g (7.1)
The term very viscous flow is used to denote situations in which the convective
(inertia) term , ) V u u is negligible, whereupon (7.1) is linearized.
There are 2 rather different ways to achieve this, as described in the following.
Slow Flow Equations
This happens when
1
UL
R

= (7.2)
Since
, )
2
U
O
L
| |
V =
|
\ .
u u
2
2
U
O
L

| |
V =
|
\ .
u
we have
, )
, )
2
1
UL
O O R

V
| |
= =
|
V
\ .
u u
u

Furthermore, if gravity is negligible, (7.1) simplifies to
2
0 p = V + V u (7.3)
which, together with the incompressible condition
0 V = u (7.3a)
are known as the slow flow eqs.
Thin Film Equations
This applies to motion of thin films of fluid.
If the thickness of the film is sufficiently small, the velocity gradient across it may be
enhanced to a level where the convective term is overcome by the viscous term, even
though R is quite large in the conventional sense.
The governing eqs, called the thin-film eqs, willl be derived in 7.6.
7.2 Low R Flow Past A Sphere
We now seek a solution to the slow flow eqs (7.3) for the uniform flow past a sphere
of radius a.
7.2.1 Equation for the Stokes Stream Function
7.2.2 Boundary Conditions
7.2.3 Solutions
7.2.4 Drag
7.2.5 Further Considerations
7.2.6 Navier-Stokes Eqs For
7.2.7 Matched Asymptotic Expansions
7.2.1 Equation for the Stokes Stream Function
Let the uniform flow be in the z direction.
The flow is axisymmetric about the z-axis, so that in terms of spherical coordinates
, ) , , r , it takes the form
, ) , ) , , , , 0
r
u r u r

= 1
]
u
The incompressibility condition 0 V = u is
sin r

+
| |
= V
|
\ .
u e (5.14)
so that
2
sin
1
sin
0 0
r
r r
r r


c c c
=
c c c
+
e e e
u
1
sin
r
r r r


c+ c+
| |
=
|
c c
\ .
e e (5.15)(7.6)
and
2
sin
1
sin
1 1
0
sin sin
r
r r
r r
r r r



c c c
V =
c c c
c+ c+

c c
e e e
u
2
2 2
1 1 1 1
sin sin r r r


1 c + c c+
| |
= +
| (
c c c
\ .
]
e
2
1
sin
E
r

= + e
where
2
2
2 2
sin 1
sin
E
r r


c c c
| |
= +
|
c c c
\ .
2 2
2 2
Q
r r
c
= +
c
and
2
1
sin
sin
Q

c c
| |
=
|
c c
\ .
Hence
, )
2
2
sin
1
sin
0 0
r
r r
r r
E


c c c
V V =
c c c
+
e e e
u
, ) , )
2 2
2
1
sin
r
E r E
r r


c c
1
= + + +
(
c c
]
e e
Using
, ) , )
2 2
V V = V V V = V u u u u
eq(7.3) can be written as
, ) p V = V Vu
, ) , )
2 2
2
sin
r
E r E
r r


c c
1
= + + +
(
c c
]
e e
ie.,
, )
2
2
sin
p
E
r r


c c
= +
c c
, )
2
1
sin
p
E
r r r


c c
= +
c c
Eliminating p gives
, ) , )
2
2 2
2 2
1 1 1
0
sin sin
E E
r r
c c c
1
+ + + =
(
c c c
]
or
2
2
2 2
sin 1
0
sin
E
r r


1 c c c
| |
+ + =
| (
c c c
\ .
]
, )
2 2
0 E E + =
2
2
2 2
sin 1
0
sin r r


1 c c c
| |
+ + =
| (
c c c
\ .
]
(7.7)
7.2.2 Boundary Conditions
The no-slip boundary condition means

2
1
, 0
sin
r
r a
u a
a


1
, 0
sin
r a
u a
a r

ie.,
0
r a r a
r




(BC1)
As the flow becomes uniform at infinity, ie.,
z
U u e for r
we have
, cos
r
u U , sin u U


so that
2
1
lim cos
sin
r
U
r

1
lim sin
sin
r
U
r r

(7.7a)
which is possible only if is of the form

2
r f
whereupon (7.7a) become
1
' cos
sin
f U

and
2
sin
sin
f U


ie.,
2
1
sin
2
f U
2 2
1
sin
2
r U for r (BC1a)
Finally, 0 u on the surface of the sphere so that every curve on it is a streamline
with zero flow velocity. Thus, is a constant on r a and, for convenience, we can
set
, 0 a (BC1b)
7.2.3 Solutions
In view of the form of (7.7) & (BC1), a simple ansatz could be
, )
2
sin f r + =
To see if it works, we calculate:
2
1
sin 2 sin cos 2
sin
Q f

c c+ c
| |
+ = = = +
|
c c c
\ .
2 2 2
2 2 2
2 2 2 2 2
2 2
'' sin sin
Q d
E f f
r r r dr r

1 | | | | c
+ = + + = + =
( | |
c
\ . \ . ]
, )
2
2 2
2 2
2 2
Q
E E
r r
| | c
+ = + +
|
c
\ .
2 2 2
2
2 2 2 2
2
sin
Q d
f
r r dr r

1 | | | | c
= +
( | |
c
\ . \ . ]
2
2
2
2 2
2
sin
d
f
dr r

1
| |
= (
|
( \ .
]
so that (7.7) is satisfied if
2
2
2 2
2
0
d
f
dr r
| |
=
|
\ .
(7.7b)
with the boundary conditions
, ) , ) ' 0 f a f a = =
2
1
lim
2
r
f r U

(BC2)
(BC2) suggests the ansatz
f r

=
so that
, )
2
2
2 2
2
1 2
d
f r
dr r



| |
= 1
| ]
\ .
, ) , ) , )
2
2
4
2 2
2
1 2 2 3 2
d
f r
dr r



| |
= 1 1
| ] ]
\ .
Thus, (7.7b) is satisfied if
, ) , ) , ) 1 2 2 3 2 0 = 1 1
] ]
which means
, ) 1 2 0 = or , ) , ) 2 3 2 0 =
so that
2
1
1 1 8
1 2


1
= + =

]

or
4
1
5 25 16
1 2


1
= =

]

A general solution to (7.7b) is therefore
2 4
A
f Br Cr Dr
r
= + + +
where A, B, C, D are constants.
Putting in (BC2), we have
2 4
0
A
Ba Ca Da
a
= + + +
3
2
0 2 4
A
B Ca Da
a
= + + +
0 D =
1
2
C U =
Substituting the last 2 into the first 2 gives
2
1
2
A
Ba Ua
a
+ =
2
A
B Ua
a
+ =
which gives
3 3
1 1 1
1
2 2 4
A Ua Ua
| |
= + =
|
\ .
2
1 1 3
1
2 2 4
B Ua Ua
a
| |
= =
|
\ .
Hence
3
2
3 2
4
U a
f ar r
r
| |
= +
|
\ .
(7.8a)
3
2 2
3 2 sin
4
U a
ar r
r

| |
+ = +
|
\ .
(7.8)
See Fig.7.2 for the corresponding streamlines.
7.2.4 Drag
Now, from (7.8a),
, ) , ) , )
2 3
2 2 3
2 3 3
2 2 0 2 2 2 2
4 2
d U a a Ua
f
dr r r r r
| | (
= + =
| (
\ .
so that
2 2
3
sin
2
Ua
E
r
+ =
, )
2
3
sin cos
Ua
E
r

c
+ =
c
, )
2 2
2
3
sin
2
Ua
E
r r

c
+ =
c
and
3
3
cos
p Ua
r r

c
=
c
(7.8a)
3
1 3
sin
2
p Ua
r r

c
=
c
(7.8b)
Integrating (7.8a) gives
, )
2
3
cos
2
Ua
p c
r
= + (7.8c)
Taking the partial with respect to :
2
3
sin '
2
p Ua
c
r

c
= +
c
which, on comparison with (7.8b), gives
' 0 c =
ie., c const = .
Thus, (7.8c) becomes
2
3
cos
2
Ua
p p
r


= +
where
r
p p

=
= .
Since the unit normal of the spherical surface is
r
= n e
the stress vector
T = t n
can be written in spherical coordinates as
, ) , )
, , , ,
r rr r r
t t t T T T

= = t
where, according to eq(A.44)
2
r
rr
u
T p
r

c
= +
c
1
r r
r
u u u u u
T r
r r r r r r




c c c c
| | | |
= + = +
| |
c c c c
\ . \ .
0
r
T

=
Now, from
3
2 2
3 2 sin
4
U a
ar r
r

| |
+ = +
|
\ .
(7.8)
we have
3
2
2 2
1
3 2 cos
sin 2
r
U a
u ar r
r r r


| | c+
= = +
|
c
\ .
3
2
1
3 4 sin
sin 4
U a
u a r
r r r r

| | c+
= = +
|
c
\ .
so that, on r a = , 0
r
u u

= = as expected.
Further differentiation gives
3 2
4 2 2 2
3
3 3 cos 1 cos
2 2
r
u U a a Ua a
r r r r r

| | | | c
= + =
| |
c
\ . \ .
3
2
2
3 2 sin
2
r
u U a
ar r
r r

| | c
= +
|
c
\ .
3 2
4 2 2 2
3
3 3 sin 1 sin
4 4
u U a a Ua a
r r r r r


| | | | c
= + = +
| |
c
\ . \ .
which, on r a = , become
0
r r
u u
r
c c
= =
c c
3
sin
2
u U
r a


c
=
c
Hence, on the surface of the sphere, the stress vector is
3
cos
2
r
U
t p p
a


= =
3
sin
2
U
t
a

=
0 t

=
The net force on the sphere should, by symmetry, in the direction of the uniform
stream (z-axis). The corresponding component of t being
cos sin
z r
t t t

=
2 2
3 3
cos cos sin
2 2
U U
p
a a

= +
3
cos
2
U
p
a

=
The drag on the sphere is therefore
2 1
2
0 1
cos
z
D t a d d

=
} }
1
2
1
2 cos
z
a t d

=
}
1
2
1
3
2 cos cos
2
U
a p d
a

| |
=
|
\ .
}
2
3
2 6
U
a aU
a

| |
= =
|
\ .
(7.9)
This formula was confirmed experimentally by measuring the terminal velocity
T
U
of a steel ball dropped into a pot of glycerine. Thus, the viscous drag exactly balances
the buoyancy reduced pull of the gravity so that
, )
3
4
6
3
T shere fluid
U a a g

=
7.2.5 Further Considerations
The above theory was due to Stokes.
It was found to be flawed since, for example, its application to the 2-D case fails in
the sense that the solutions it provides cannot satisfy all the required boundary
conditions (see Ex.7.4).
The trouble lies in the fact that the non-linear term u u dropped to obtain the
slow flow eqs (7.3) is not negligible in the transitional region where the deflection due
to the sphere begins to die out and the flow is more or less, but not quite, uniform.
One way to handle the problem is by means of the mathched asymptotic expansions
due to I. Proudman and J.R.A. Pearson. [J.Fluid.Mech. 2, 237-62 (1957)].
7.2.6 Navier-Stokes Eqs for
7.2.6 Navier-Stokes Eqs for
We now re-write the full steady state Navier-Stokes eqs
, )
2
1
p

V = V + V u u u
in terms of the Stokes stream function .
Using
, ) , )
2
1
2
V = V V u u u u u
, ) , )
2 2
V V = V V V = V u u u u
we have
, ) , )
2
1 1
2
p

| |
V + = V V V
|
\ .
u u u u
Using
2
1
sin
E
r

V = + u e
we have
, )
2
1
0 0
sin
r
r
u u u
E
r

V =
+
e e e
u u
2 2
sin sin
r
r
u u
E E
r r


= + + + e e
2 2
2 2 3 2
1 1
sin sin
r
E E
r r r


c+ c+
= + + +
c c
e e
2
2 2
1 1
sin
r
E
r r r


c+ c+
| |
= + +
|
c c
\ .
e e
From 7.2.1, we have
, ) , ) , )
2 2
2
1
sin
r
E r E
r r


c c
1
V V = + + +
(
c c
]
u e e
Hence
2
1 1
2
p

| |
V +
|
\ .
u
2
2 2
1 1
sin sin
sin
r
r E
r r r r



c+ c c+ c 1 | | | |
= + + +
| |
(
c c c c
\ . \ . ]
e e
ie.,
2 2
2 2
1 1 1
sin
2 sin
p E
r r r


c c+ c
| | | |
+ = + +
| |
c c c
\ . \ .
u
2 2
2 2
1 1 1 1
sin
2 sin
p r E
r r r r


c c+ c
| | | |
+ = +
| |
c c c
\ . \ .
u
Eliminating
2
1
2
p + u by cross differentiation gives
2 2
2 2 2
1 1 1
sin sin
sin sin
E r E
r r r r r r


c c+ c c c+ c 1 1 | | | |
+ + = +
| |
( (
c c c c c c
\ . \ . ] ]
Terms proportional to were already worked out in 7.2.1. Thus
, )
2 2 2 2
2 2 2 2
1 1
sin sin sin
E E E E
r r r r


c c+ c c+ 1 1 | | | |
+ = + + +
| |
( (
c c c c
\ . \ . ] ]
2 2
2 2 2 2
1 1
sin sin
E E
r r r r
c+ c c+ c
| | 1 | | 1
= + + +
| |
( (
c c c c
\ . ] \ . ]
2
2 3 2
1 2cos 1
sin sin
E
r r


c+ c
| | 1
= +
|
(
c c
\ . ]
2
2 3 2
1 2 1
sin
E
r r r
c+ c
| | 1
+ + +
|
(
c c
\ . ]
2
2 2
1 2
2cot
sin
E
r r r r


c+ c c+ c 1 | || | | || |
= +
| | | |
(
c c c c
\ .\ . \ .\ . ]
ie.,
, )
2 2 2
2
1 2
2cot
sin
E E E
r r r r


c+ c c+ c 1 | || | | || |
+ = +
| | | |
(
c c c c
\ .\ . \ .\ . ]
(7.11a)
Shifting to dimensionless variables, we set
'
r
r
a
= '
U
=
u
u
2
'
Ua
+
+ =
so that
, ) , )
2 2 2 2
2
' ' '
U
E E E E
a
+ = +
2
2 2
2 2
1 1 '
' '
' '
U
E E
r r r r a
c+ c+
| | | |
+ = +
| |
c c
\ . \ .
Substituting these into (7.11a) and then leave out the primes for clarity, we have
, )
2 2 2
2
2
2cot
sin
R
E E E
r r r r


c+ c c+ c 1 | || | | || |
+ = +
| | | |
(
c c c c
\ .\ . \ .\ . ]
(7.11)
where the Reynolds number is defined by
Ua
R

= (7.12)
7.2.7 Matched Asymptotic Expansions
The method is rather involved.
Here, we are contended with only a summary of the salient results.
Interested readers should consult the original paper:
I. Proudman and J.R.A. Pearson. [J.Fluid.Mech. 2, 237-62 (1957)],
where they obtained solution to the Navier-Stokes eqs (7.11) in 2 parts.
Near the sphere
They found
, )
2
2
1
1 sin
4
r +
2
3 1 3 1 1
1 2 2 cos
8 8
R R
r r r

1 | || | | |
+ + + +
| | |
(
\ .\ . \ . ]
(7.13)
which is just the sum of the , ) 1 O and , ) O R terms in an asymptotic expansion for
that is exact in the limit 0 R for fixed r.
If we take only the , ) 1 O terms, we have
, )
2
2
1 1
1 2 sin
4
r
r

| |
+ +
|
\ .

2 2
1 1
2 3 sin
4
r r
r

| |
= +
|
\ .
which is just Stokes solution (7.8) in dimensionless quantities.
The , ) O R term improves the accuracy of the solution so that, for example, the drag
(7.9) becomes
3
6 1
8
D Ua R
| |
+
|
\ .

Far from the sphere


With
*
r Rr = (7.15)
they found that
, ) , )
2
2 *
* 2
1 3 1 1
sin 1 cos 1 exp 1 cos
2 2 2
r
r
R R

1 | |
+ +
|
(
\ . ]
(7.14)
which is just the sum of the
, )
2
O R

and
, )
1
O R

terms in an asymptotic expansion


for that is exact in the limit 0 R for fixed
*
r .
Thus, by far away from the sphere, we mean a distance r of order
1
R

or greater as
0 R .
Matching
Solutions (7.13) & (7.14) are matched in the following sense.
Rewriting (7.13) in terms of
*
r , we have
2
2 *
1
1 sin
4
r
R

| |
+
|
\ .

2
2
* * *
3 3
1 2 2 cos
8 8
R R R
R R
r r r

1 | | | |
| |
+ + + +
( | | |
\ .
\ . \ . ]
Keeping only terms of
, )
2
O R

and
, )
1
O R

,
, )
2
2 *
*
1 3 1
1 2 1 cos sin
4 4
r
R
R r

1 | |
| |
+ + +
( | |
\ .
\ . ]

, )
2 2
2 * *
2
*
1 3 3
2 1 cos sin
4 4
r r
R r R

1 | |
+
( |
\ . ]

, )
2
2 *
2
*
1 3 3
2 1 cos sin
4 4
r
R
R r

1 | |
+
( |
\ . ]
(7.15a)
Rewriting (7.14) in terms of r, we have
, ) , )
2 2
1 3 1 1
sin 1 cos 1 exp 1 cos
2 2 2
r rR
R

1 | |
+ +
|
(
\ . ]

Keeping only the , ) 1 O and , ) O R terms,


, ) , ) , )
2
2 2 2 2
1 3 1 1 1
sin 1 cos 1 cos 1 cos
2 2 2 8
r rR r R
R

1
+ +
(
]

, )
2 2 2 2 2
1 3 3
sin sin 1 cos sin
2 4 16
r r r R = +
, )
2 2
1 3 3
sin 2 1 cos
4 4
r R
r

1
= +
(
]
(7.15b)
In view of (7.15), we see that (7.15a) & (7.15b) are identical.
This is what we mean by saying that (7.13) & (7.14) are matched.
7.3 Corner Eddies
In a 2-D slow flow past a symmetric obstacle, eddies may occur symmetrically for
and aft of the body (see Fig.7.1a).
If the shape of the obstacle is a circular arc, eddie-free flow is still possible if the
corner angle is larger than 146.3
C


.
Below
C
, an infinite sequence of eddies of alternating senses occurs (see Fig.7.4).
Here, we follow an elementary approach to the problem due to Moffatt.
7.3.1 Governing Equations
7.3.2 Solutions
7.3.3 Boundary Conditions
7.3.4 Discussions
7.3.1 Governing Equations
We begin with the stream function representation
u
y
c
=
c
v
x
c
=
c
ie.,
, ) = V u k or
3 i ij
j
u
x

c
=
c
so that
, ) V = V V 1
]
u k
or
, )
2
3 ijk kl
i
j l
x x


c
V =
c c
u
, )
2
3 3 il j i jl
j l
x x


c
=
c c
2 2
3
3
i
l j j
x x x x

c c
=
c c c c
Since is a function of x and y only, we have
, )
2
2
3 3 i i
i
j j
x x


c
V = = V
c c
u
or
2
3
V = V u e
Consider now the slow flow equation
2
p V = V u (7.3)
Taking the curl gives
, ) , ) , )
2 2 2 2
3
0 = V V = V V = V V u u e
ie.,
, )
2 2
0 V V = (7.16)
In terms of cylindrical coordinates,
1
r
u
r

c
=
c
u
r

c
=
c
(7.17)
2 2
2
2 2 2
1 1
r r r r
c c c
V = + +
c c c
so that (7.16) becomes
2
2 2
2 2 2
1 1
0
r r r r

| | c c c
+ + =
|
c c c
\ .
(7.18)
7.3.2 Solutions
Seeking for a separable solution of (7.18), we write
, ) , ) g r f =
It is easy to see that separation can be achieved if
2
2
2
d f
const f m f
d
= = (7.18a)
where the particular form of the constant is chosen for later convenience.
Substituting into (7.18) gives
2
2 2
2 2
1
0
d d m
g
dr r dr r
| |
+ =
|
\ .
(7.18b)
which is homogeneous in r. A natural ansatz is therefore
g r

=
Now
, )
2 2
2 2
2 2
1
1
d d m
r m r
dr r dr r



| |
( + = +
|

\ .
, ) , ) , )
2
2 2
2 2 2
2 2
1
1 2 3 2
d d m
r m m r
dr r dr r



| |
( ( + = + +
|

\ .
, ) , )
2
2 2 2 2
2 m m r



(
=

so that (7.18b) is satisfied if
m = or 2 m =
Now, a general solution to (7.18a) is
1 2
cos sin f c m c m = +
For a given , the general solution to (7.18) is therefore
, ) , ) cos sin cos 2 sin 2 r A B C D

= + + + (

(7.18c)
where A, B, C, D are constants.
7.3.3 Boundary Conditions
The system of interest is shown in Fig.7.4.
We shall put the origin at the vertex with the x-axis bisecting the corner angle.
The no-slip condition is therefore
0
r
u u

= = on =
where 2 is the angle of the corner.
Applying them on the solution (7.18c) gives
u

: , ) , ) cos sin cos 2 sin 2 0 A B C D + + + =


, ) , ) cos sin cos 2 sin 2 0 A B C D + =
r
u :
, ) , ) , ) , ) sin cos 2 sin 2 2 cos 2 0 A B C D + + =
, ) , ) , ) , ) sin cos 2 sin 2 2 cos 2 0 A B C D + + + =
They are easily decoupled to give
, ) cos cos 2 0 A C + = (7.19a)
, ) , ) sin 2 sin 2 0 A C + = (7.19b)
, ) sin sin 2 0 B D + = (7.19c)
, ) , ) cos 2 cos 2 0 B D + = (7.19d)
Eqs(7.19a,b) are consistent only if
, )
, ) , )
cos cos 2
0
sin 2 sin 2

=

ie.,
, ) , ) , ) , ) 2 cos sin 2 cos 2 sin 0 = 1 1
] ]
which can be simplified as
, ) , ) tan 2 tan 2 = 1
]
, )
, )
, )
sin 2
sin
2
cos cos 2



1
]
=
1
]
, ) , ) , ) sin cos 2 2 sin 2 cos = 1 1
] ]
, ) , ) , ,
sin cos 2 sin 2 cos 1 1
] ]
, ) 2sin 2 cos = 1
]
(7.19e)
, ) sin2 sin 2 1 sin2 = + 1
]
, ) , ) 1 sin2 sin 2 1 = 1
]
where weve used the trigonometric identities
, ) sin cos cos sin sin A B A B A B + = +
, ) , ) 2sin cos sin sin A B A B A B = + +
Setting
, ) 2 1 x =
we have
sin2 sin
2
x
x

=
sin2 sin
2
x
x

= (7.19)
Now, eqs(19.c,d) can be obtained from eqs(19.a,b) by interchanging sine and cosine.
Hence, the counterpart of (7.19e) is
, ) , ) , ,
cos sin 2 cos 2 sin 1 1
] ]
, ) 2cos 2 sin = 1
]
which can be simplified to
, ) sin2 sin 2 1 sin2 = 1
]
, ) , ) 1 sin2 sin 2 1 = 1
]
sin2 sin
2
x
x

= (7.19)
Obviously, (7.19) & (7.19) cannot be satisfied simultaneously.
Thus, we have 2 independent solutions, namely,
1. , ) cos cos 2 r A C

= + 1
]
with
sin2 sin
2
x
x

= (7.19)
2. , ) sin sin 2 r B D

= + 1
]
with
sin2 sin
2
x
x

= (7.19)
7.3.4 Discussions
For a given corner subtending an angle 2, we can solve eq(7.19) or eq(7.19) to get x
and hence
1
2
x

= +
Both
r
u and u

, as calculated from (7.17), vary with r as


1 / 2 x
r r

= .
Thus, for x real and positive ( real and greater than 1),
0 = u at 0 r =
Furthermore, for a fixed , neither
r
u nor u

changes sign as r varies.


The flow is therefore of the simple form with no eddies, as shown in Fig.7.3a.
For complex, ie., of the form
p iq = + with p, q real
the measured value of any physical quantity is defined to be the real part of the
corresponding complex function.
For a fixed , both
r
u and u

are of the form


1 1 1 ln
Re Re Re
p iq p iq r
cr cr cr e
+
1 1 1 = =
] ] ]
where c is some complex constant (different ones for
r
u and u

, of course), and
weve used
, , , , exp ln exp ln
iq iq
r r iq r = =
Writing
i
c c e

= , where is real, we have


, ,
, ,
ln 1 1
Re cos ln
i q r p p
c r e c r q r

+
1
= +
]
which changes sign twice with r whenever
ln q r +
changes by 2.
Note that each sign change indicates the presence of a vortex (eddie).
Since
0
limln
r
r

, the sign changes occur with increasing rapidity as one


approaches the vertex. Thus, an infinite sequence of eddies is indicated.
Now, eq(7.19) is usually solved by graphical method as shown in fig.7.5.
Thus, the soloution
0
x is simply the ordinate of the intersection of the curve
sin x
y
x
= and the straight line
sin2
2
y

= .
As is clear from fig.7.5, no intersect is possible if
sin2
0.217
2

<
ie.,
sin2
0.217
2

>
Solving
sin2
0.217
2
C
C

=
gives
2 146.3
C

Eddies occur if
0
2 2 < .
7.4 Uniqueness & Reversibility of Slow Flows
Theorem
Consider viscous fluid in region V with closed surface S.
There is at most one solution to the slow flow eqs (7.3) which satisfies a given
boundary condition
, )
B
= u u x on S.
Proof
Let there be another solution
*
u that satisfies the same boundary condition.
Define the difference flow by
*
= v u u
and the difference pressure by
*
P p p =
The linearity of the slow flow eqs
2
p V = V u 0 V = u
implies
2
P V = V v 0 V = v
with the boundary conditions
0 = v on S.
In component form, these eqs become
2
i
i j j
P v
x x x

c c
=
c c c
0
i
i
v
x
c
=
c
Multiplying the 1
st
with
i
v , we have
2
i
i i
i j j
P v
v v
x x x

c c
=
c c c
Using
, )
i i
i i
i i i i
v P P v P
v P v
x x x x
c c c c
= + =
c c c c
we have
, )
2
i i
i
i j j
Pv v
v
x x x

c c
=
c c c
Integrating over V and applying the divergence theorem gives
2
i
i i
j j S V
v
Pv dS v dV
x x

c
=
c c
} }
The left hand side vanishes since 0 = v on S. Together with
2
i i i i
i i
j j j j j j
v v v v
v v
x x x x x x
| |
c c c c c
= +
|
|
c c c c c c
\ .
we get
0
i i i
i
j j j j V
v v v
v dV
x x x x

1 | |
c c c c
=
( |
|
c c c c
(
\ . ]
}
Applying the divergence theorem to the 1
st
term shows that it vanishes on account of
the boundary condition on S. Thus
2
,
0
i i i
i j j j j V V
v v v
dV dV
x x x
| |
c c c
= =
|
|
c c c
\ .
_
} }
Since the integrands are all non-negative, we must have
0
i
j
v
x
c
=
c
for all i, j at all V e x
which means
const = v for all V e x .
Since 0 = v on S, we have 0 = v everywhere so that
*
= u u . QED.
Reversibility
Let , )
1
u x , together with the attendant pressure field , )
1
p x , be the solution to the
slow flow eqs
2
p V = V u 0 V = u
with boundary conditions , ) , )
B
= u x u x on S.
Next, we change the boundary condition to , ) , )
B
= u x u x on S.
By inspection, we see that , )
1
u x , together with a pressure field , )
1
p c + x , c
being a constant, is a solution that satisfies the new boundary condition.
Furthermore, according to the uniqueness theorem, it is the only such solution.
Thus, we see that for flows governed by the slow flow eqs,
Reversed boundary conditions leads to reversed flow.
One example of this is the concentric cylinder experiment shown in Fig.2.6.
Note that for real fluids, reversal of flow is only partial since
1. The slow flow eqs themselves are only approximations.
2. Due to thermal and mechanical fluctuations, no boundary condition can be
exactly reversed in practice.
7.5 Swimming At Low R
Consider a battery powered mechanical fish consisting of a cylindrical body and a
plane tail which flaps as shown in Fig.7.6a.
It swims easily in water but makes no progress in syrup.
The problem it encounters in the latter case is caused by the reversibility of flow for
low Reynolds numbers as discussed in 7.4.
Swimming in low R situations thus requires motions that is not time-reversible.
This can be achieved by replacing the plane tail with a rotating helical coil.
Spermatozoa also used this mechanism, sending helical waves down their tails.
7.5.1 Swimming of a Thin Flexible Sheet
7.5.2 Basic Eqs
7.5.3 Dimensionless Form
7.5.4 Iterative Equations
7.5.5 1
st
order Solution
7.5.6 2
nd
order Solution
7.5.1 Swimming of a Thin Flexible Sheet
A simple model for ciliary propulsion is a thin extensible sheet which flexes
according to
s
x x = , ) sin
s
y a k x t = (7.21)
where , ) ,
s s
x y denote the coordinates of a particle on the sheet (see Fig.7.7).
Thus, a wave travels along the x direction with speed
c
k

=
while the particles on the sheet move only in the y direction with velocity
, ) cos
s
dy
a k x t
dt
=
This motion is not time-reversible since running the film backward entails the wave
moving in the opposite direction.
We shall show that, for / a small, 2 / k = being the wavelength, the flow
induced by this oscillatory flexing contains a steady component
2
2
2
a
U c

| |
=
|
\ .
(7.22)
in the x-direction. Thus, the sheet swims to the left in a fluid that is otherwise at rest.
7.5.2 Basic Eqs
As in 7.3, we work in the stream function representation with
u
y
c
=
c
v
x
c
=
c
(7.23)
so that
2
3
V = V u e
and the slow flow eq becomes
, ,
2 2
0 V V =
which, in Cartesian coordinates, is simply
2
2 2
2 2
0
x y

| | c c
+ =
|
c c
\ .
(7.24)
The boundary condition on the sheet is simply
s
= u u .
In terms of , we have
0
s
s
y y
u
y

=
c
= =
c
, , cos
s
s
y y
v a k x t
x


=
c
= =
c
(7.25)
on , , sin
s
y y a k x t = = .
7.5.3 Dimensionless Form
Without loss of generality, we shall solve eq(7.24) only at 0 t = .
Solutions at other times can be obtained by replacing kx with k x t .
To further facilitate manipulations, we introduce the dimensionless quantities:
' x kx = ' y ky = '
k
a

= (7.26)
On substituting into eqs(7.24-25) and then dropping the primes for clarity, we have,
(7.24):
2
2 2
2 2
0
x y

| | c c
+ =
|
c c
\ .
(7.27)
(7.25): 0
s
s
y y
u
y

=
c
= =
c
cos
s
s
y y
v x
x

=
c
= =
c
(7.28)
on sin
s
y y x = = .
7.5.4 Iterative Equations
For small, the boundary conditions (7.28) can be approximated by a Taylor series
expansion:
2
2
0 0
sin 0
s
y y y y
x
y y y

= = =
c c c
= + + =
c c c

2
0
0
sin cos
s
y y y
y
x x
x x y x

= =
=
c c c
= + + =
c c c c
(7.30)
Similarly, we write as a power series in ,
2
1 2 3
= + + + (7.31)
where the
n
s are independent of .
On substituting into
(7.27)
and using the fact that the coefficient of each power of must vanish, we have
2
2 2
2 2
0
n
x y

| | c c
+ =
|
c c
\ .
for all n.
By successive differentiations on (7.31), we have
2 1 2 3
0 0 0 0
n n n n
n n n n
y y y y
y y y y


= = = =
c c c c
= + + +
c c c c

2 1 2 3
1 1 1 1
0 0 0 0
n n n n
n n n n
y y y y
y x y x y x y x



= = = =
c c c c
= + + +
c c c c c c c c

On substituting into the 1
st
boundary conditions in (7.30), we have
2 1 2 3
0 0 0 y y y
y y y


= = =
c c c
+ + +
c c c

2 2 2
2 1 2 3
2 2 2
0 0 0
sin
y y y
x
y y y


= = =
(
c c c
+ + + + (
c c c
(

, )
2 1 2 3
0 0 0
1
sin
!
m m m
m
m m m
y y y
x
m y y y


= = =
(
c c c
+ + + + (
c c c
(

0 + =
Setting the coefficients of each power of to zero, we have
1
0
0
y
y

=
c
=
c
2
2 1
2
0 0
sin 0
y y
x
y y

= =
c c
+ =
c c

, )
1 1
1
1
0 0 0
sin sin
0
! 1 !
n n m m
m m n
n m
y y y
x x
y n y m y

+

+
= = =
c c c
+ + + + =
c c c

Doing the same to the 2


nd
boundary conditions in (7.30), we have
1
0
cos
y
x
x

=
c
=
c
2
2 1
0
0
sin 0
y
y
x
x y x

=
=
c c
+ =
c c c

, )
1 1
1
1
0
0 0
sin sin
0
! 1 !
n n m m
m m n
n m
y
y y
x x
x n y x m y x

+

=
= =
c c c
+ + + + =
c c c c c

To summarize, to get
m
, we must solve
2
2 2
2 2
0
m
x y

| | c c
+ =
|
c c
\ .
with the boundary conditions
, )
1 1
1
1
0 0 0
sin sin
0
! 1 !
n n m m
m m n
n m
y y y
x x
y n y m y

+

+
= = =
c c c
+ + + + =
c c c

, )
1 1
1
1
0
0 0
sin sin
0
! 1 !
n n m m
m m n
n m
y
y y
x x
x n y x m y x

+

=
= =
c c c
+ + + + =
c c c c c

which requires previous knowledge of { ;
1 1
, ,
m


.
7.5.4 Solution
7.5.5 1
st
order Solution
The 1
st
order solution
1
satisfies
2
2 2
1 2 2
0
x y

| | c c
+ =
|
c c
\ .
(7.32)
with boundary conditions
1
0
0
y
y

=
c
=
c
1
0
cos
y
x
x

=
c
=
c
A separable solution
, ) , )
1
X x Y y =
together with the boundary conditions suggest X to be sinusoidal so that we set
2
'' X X =
so that
1 2
cos sin X c x c x = +
In order to satisfy the 2
nd
boundary condition, we need
sin X x = , ) 0 1 Y =
so that
'' X X =
and
2
2
2
1 0
d
Y
dy
| |
+ =
|
\ .
Setting
y
Y e

=
gives
, )
2
2
1 0 + =
which has 2 pairs of double roots
1 =
The general solution is therefore
, ) , )
y y
Y A By e C Dy e

= + + +
If u is to be finite as y , we must have
0 C D = =
With , ) 0 1 Y = , we have
, ) 1
y
Y By e

= +
Finally,
1
0
0
y
y

=
c
=
c
means
, ) 1 sin 0
y
By B e x

+ + = 1
]
at 0 y =
ie., 1 B = so that
, )
1
1 sin
y
y e x

= + (7.34)
, )
1
1 1 sin sin
y y
y e x ye x
y


c
= + + = 1
]
c
(7.34a)
7.5.6 2
nd
order Solution
The 2
nd
order solution
2
satisfies
2
2 2
2 2 2
0
x y

| | c c
+ =
|
c c
\ .
(7.33)
with boundary conditions
2
2 1
2
0 0
sin 0
y y
x
y y

= =
c c
+ =
c c
2
2 1
0
0
sin 0
y
y
x
x y x

=
=
c c
+ =
c c c
Using the 1
st
order result
, )
1
1 sin
y
y e x

= + (7.34)
we have
, )
1
1 1 sin sin
y y
y e x ye x
y


c
= + + = (

c
, )
2
1
2
1 sin
y
y e x
y


c
=
c
2
1
2
0
sin
y
x
y

=
c
=
c
2
1
cos
y
ye x
x y


c
=
c c
2
1
0
0
y
x y

=
c
=
c c
so that the boundary conditions become
2 2
0
sin
y
x
y

=
c
=
c
2
0
0
y
x

=
c
=
c
(7.35)
A separable solution
, ) , )
2
X x Y y =
with
1 2
cos sin X c x c x = +
, )
y
Y A By e

= + (7.35a)
obviously cannot satisfy (7.35)
To proceed, we rewrite the 1
st
eq. in (7.35) as
, )
2
0
1
1 cos2
2
y
x
y

=
c
=
c
so that the cosine term is of the manageable type (7.35a).
On the other hand, the constant term requires
2
to be of the form
, ) , ) , )
2
X x Y y f y = + (7.35b)
so that the boundary conditions become
, ) , ) , ) , )
1
' 0 0 1 cos 2
2
X x Y f x + =
, ) 0 sin2 0 Y x =
which requires
, )
1
cos 2
2
X x x = , ) ' 0 1 Y = , )
1
' 0
2
f =
, ) 0 0 Y = (7.35c)
Plugging into (7.35a) gives
, ) 0 0 Y A = =
, ) ' 0 1 Y A B B = + = =
so that
y
Y ye

=
Furthermore, for (7.35b) to remain a solution, f itself must be a solution of (7.33), ie.,
(4)
0 f =
so that f is of the form
3 2
f Ay By Cy D = + + +
and
2
' 3 2 f Ay By C = + +
Now, u is proportional to f so that to keep the former finite at y , we need
0 A B = =
Plugging in the boundary condition gives
1
'
2
f C = =
so that, finally, we have
1
2
f y =
where D, which is of no physical significance, was dropped for convenience.
Putting everything together, we have
2
2
1 1
cos 2
2 2
y
ye x y

= + (7.36)
so that
, )
2 2
1 1
2 1 cos2
2 2
y
y e x
y


c
= + +
c
Together with
, )
1
1 1 sin sin
y y
y e x ye x
y


c
= + + = (

c
(7.34a)
we have
1 2
y y y

c c c
= + +
c c c

2
1 1
sin cos2
2 2
y y
ye x y e x

( | |
= + + +
|
(
\ .
(7.37)
Reverting to unprimed quantities, this becomes
2
1 1 1
sin cos2
2 2
ky ky
kye kx ky e kx
a y


c ( | |
= + + +
|
(
c
\ .

or
u
y
c
=
c
2
1 1
sin cos2
2 2
ky ky
akye kx a ky e kx

( | |
= + + +
|
(
\ .

2 2
1 1
sin cos2
2 2
ky ky
ye kx c ky e kx

( | |
= + + +
|
(
\ .

Generalizing to arbitrary time, we have
, ) , )
2 2
1 1
sin cos2
2 2
ky ky
u ye kx t c ky e kx t

( | |
= + + +
|
(
\ .
(7.38)
which has a steady component
2
1
2
U c = (7.22)
7.6 Flow in a Thin Film
7.6.1 Basic Conditions
7.6.2 Basic Equations
7.6.1 Basic Conditions
Consider a viscous fluid in steady flow between 2 rigid boundaries
0 z = and , ) , z h x y =
Let U and L be typical flow speed and length scale in the horizontal ( ) direction.
The condition for a thin film flow is
h L (7.39)
Now, the no-slip condition must be satisfied at 0 z = and z h = .
As z increases from 0 to h, u

1
st
increases to order U, then drops back to 0, thus
U
O
z h
c
| |
=
|
c
\ .
u

Meanwhile,
z
c
c
u

also experiences a sign change somewhere mid-stream so that


2
2 2
U
O
z h
c
| |
=
|
c
\ .
u

By comparison, the horizontal gradients V u



and
2
V u

, which are of
U
O
L
| |
|
\ .
and
2
U
O
L
| |
|
\ .
, respectively, are seen to be much weaker owing to (7.39).
Although the magnitude of
z
w

= u e is not specified, the no-slip condition still


applies, so that its horizontal gradients are also expected to be much weaker than the
vertical one.
Thus, the viscous term can be approximated as
2
2
2
z

c
V
c
u
u
From the incompressibility condition,
0
u v w w
x y z z
c c c c
+ + = V + =
c c c c
u

we see that
w U
O
z L
c
| |
=
|
c
\ .
and hence
, )
Uh
w O O U
L
| |
= =
|
\ .
u

Order of magnitude estimates also give


U Uh U
w
z L Lh L
c
V = V + +
c
u u


, , 1,1,
Uh h
U U U
L L
| | | |
=
| |
\ . \ .
u
so that
, )
2
1,1,
U h
L L
| |
V
|
\ .
u u
2
2 2
1,1,
U h
z h L

c
| |
|
c
\ .
u

Thus, , ) V u u may be neglected when compared to


2
2
z

c
c
u
if
2
2
U U
L h
ie.,
2
1
Uh
L

or
2
1
UL h
L
| |
|
\ .
(7.40)
which is the 2
nd
condition for a thin film flow. [the 1
st
is (7.39)]
7.6.2 Basic Equations
Under the conditions eqs(7.39,40) and in the absence of external forces, the steady
state Navier-Stokes eqs simplify to
2
2
1
0 p
z

c
= V +
c
u
0 V = u
i.e.,
2
2
p u
x z

c c
=
c c
2
2
p v
y z

c c
=
c c
2
2
p w
z z

c c
=
c c
0
u v w
x y z
c c c
+ + =
c c c
(7.41)
As shown in the last section, w and
2
2
w
z
c
c
are both smaller by a factor / h L than
their horizontal counterparts. According to (7.41), so is
p
z
c
c
.
Setting p to be independent of z, the 1
st
2 eqs in (7.41) can be integrated to give
1 u p
z A
z x
c c
= +
c c
1 v p
z C
z y
c c
= +
c c
2
1
2
p
u z Az B
x
c
= + +
c
2
1
2
p
v z Cz D
y
c
= + +
c
(7.42)
where A, B, C, D are functions of x, y only.
From
2
2 2
U
O
z h
c
| |
=
|
c
\ .
u

eq(7.41) implies
2
p p U
O
x y h

c c
| |
=
|
c c
\ .

so that
2
UL
p O
h

| |
=
|
\ .
(7.44)
Consider now the stress tensor
j
i
ij ij
j i
u
u
p
x x

| | c
c
= + +
|
|
c c
\ .
(7.43)
The largest term in the parenthesis is
2
U UL
O O p
z h h

c
| | | |
= =
| |
c
\ . \ .
u

Hence
ij ij
p (7.45)
which means the thin-film flow behaves like an inviscid flow in this aspect.
7.7 Flow in a Hele-Shaw Cell
AHele-Shaw cell flow is a thin film flow between 2 parallel plane boundaries.
Usually, fluid flow is driven by horizontal pressure gradient to past by cylindrical
objects in the gap as shown in Fig.7.8.
Applying the no-slip condition on 0 z = and z h = to (7.42) gives
0 B =
2
1
0
2
p
h Ah
x
c
= +
c
0 D =
2
1
0
2
p
h Ch
y
c
= +
c
so that
1
2
p
A h
x
c
=
c
1
2
p
C h
y
c
=
c
and
, )
1
2
p
u h z z
x
c
=
c
, )
1
2
p
v h z z
y
c
=
c
(7.46)
Note that
p
v y
p
u
x
c
c
=
c
c
and hence the direction of flow and the streamline patterns, are all independent of z.
Eliminating p in (7.46) by cross differentiation, we have
0
u v
y x
c c
=
c c
(7.47)
so that at a given z, the flow is similar to a 2-D irrotational flow.
However, if we calculate the circulation around any closed curve C, we find
, )
C
udx vdy I = +
}
, )
1
2
C
p p
z h z dx dy
x y
| | c c
= +
|
c c
\ .
}
, )
1
2
C
z h z p d

= V
}
x

, ) j
1
2
C
z h z p

=
0 = (7.48)
where weve used the fact that p is a physical quantity and hence must be
single-valued.
Now, (7.48) is valid irrregardless of the global topology of the fluid domain.
This behavior is markedly different from that of true 2-D flows for which can take
on finite values.
Thus, if we place a flat plate at an angle of attack to the oncoming stream, the
streamline patterns will be as shown in Fig.4.6a but never as in Fig.4.6b.
Another example is the flow into a rectangular opening as shown in Fig.7.9b, which is
to be contrasted with that of high R flow shown in Fig.7.9a.
7.8 Adhesive Problem
It is well known that a rather large force is required to pull a disc away from a rigid
plane if the two are separated by a thin film of viscous fluid.
Here we study the problem in terms of the steady state thin film flow eqs and attribute
all the time dependence of the flow to the changing boundary conditions. Justification
of this assumption will be left as an exercise.
Let the radius of the disk be a and the height of the fluid be , ) h t . [see Fig.7.10]
By symmetry of the setup, we expect the (unsteady) flow to be of the axisymmetric:
, ) , ) , , , ,
r r z z
u r z t u r z t = + u e e
In terms of cylindrical coordinates, the relevant thin film eqs are (cf.7.41)
2
2
r
p u
r z

c c
=
c c
2
2
0
z
u
z

c
=
c
, )
1
0
z
r
u
ru
r r z
c c
+ =
c c
Using the fact that p is independent of z, we can integrate the 1
st
eq. and get
1
r
u p
z A
z r
c c
= +
c c
2
1
2
r
p
u z Az B
r
c
= + +
c
where A, B are functions of r & t only.
Sticking in the no-slip boundary condition at 0 z = and , ) z h t = , we have
0 B =
2
1
0
2
p
h Ah
r
c
= +
c
so that
, )
1
2
r
p
u h z z
r
c
=
c
Thus
, )
2
2
1
2
r
u p
h z z
r r
c c
=
c c
and the incompressibility condition becomes
, ) , )
2
2
1 1
0
2 2
z
p p u
h z z h z z
r r r z
c c c
+ =
c c c
i.e.,
, )
2
2
1 1
2
z
u p p
h z z
z r r r
( c c c
= +
(
c c c

which integrates to
2
2 3
2
1 1 1 1
2 2 3
z
p p
u hz z C
r r r
( c c
| |
= + +
| (
c c
\ .

No-slip conditions at 0 z = then gives
2
2 3
2
1 1 1 1
2 2 3
z
p p
u hz z
r r r
( c c
| |
= +
| (
c c
\ .

The boundary condition at the moving disk is
z
dh
u
dt
= at , ) z h t =
so that
2
3
2
1 1
12
dh p p
h
dt r r r
( c c
= +
(
c c

which can be rearranged to give
2
2 3
1 1 12 p p p dh
r
r r r r r r h dt
c c c c
| |
+ = =
|
c c c c
\ .
(7.49)
Integrating, we get
2
3
6 p dh
r r D
r h dt
c
= +
c
2
3
3
ln
dh
p r D r E
h dt

= + +
where D, E are functions of t only.
To avoid a singularity at 0 r = , we set 0 D = .
With the boundary condition
0
p p = at r a = ,
0
p being the atmospheric pressure,
we have
, )
2 2
0 3
3 dh
p r a p
h dt

= +
The force exerted on the disk by the fluid is therefore
, )
2
0
0 0
a
F p p rdrd

=
} }
, )
2 2
3
0
6
a
dh
r a rdr
h dt

=
}
4 4
3
6 1 1
4 2
dh
a a
h dt

(
=
(

4
3
3
2
dh
a
h dt

= (7.50)
Thus, if one tries to pull up the disk, i.e., 0
dh
dt
> , then 0 F < , so that the pull by the
fluid is downward. Furthermore, the
3
h

dependence makes F rather large for small


h.
7.9 Thin Film Flow Down a Slope
Consider the 2-D problem of a layer of viscous fluid spreading down a slop under the
action of gravity.
[See Fig.7.11 for the definitions of the coordinate system and various quantities.]
As in 7.8, we neglect
t
c
c
u
and attribute all time dependencies to that of the
boundaries. In the presence of gravity, the thin film flow eqs become
2
2
1
0 sin
p u
g
x z

c c
= + +
c c
1
0 cos
p
g
z

c
=
c
(7.52)
The 2
nd
eq can be integrated to give
, ) cos , p gz f x t = + (7.52a)
Consider the free surface
, ) , z h x t =
Its unit tangent and normal vectors in the x-z plane are (see 3.4b.2)
2
1
1,
1
h
v
x
h
x
c
| |
=
|
c
\ .
c
| |
+
|
c
\ .
2
1
, 1
1
h
n
x
h
x
c
| |
=
|
c
\ .
c
| |
+
|
c
\ .
Under the thin film approximation
h L (L is horizontal length scale)
we have
1
h h
x L
c
c

so that
, )
1, 0 v = x , )
0, 1 n z =
Consider now the stress tensor
j
i
ij ij
j i
v
v
p
x x

| | c
c
= + +
|
|
c c
\ .
The stress vector at the surface is
n z = t
1 13
u w u
z x z

c c c
| |
= +
|
c c c
\ .
t
3 33
2
w
p p
z

c
= +
c
t [see (7.45)]
Hence. the continuity of the stress across an interface gives
0
u
z

c
=
c
(7.53)
0
p p = (7.53a)
on , ) , z h x t = .
Putting (7.53a) into (7.52a) gives
0
cos p gh f = +
so that
, )
0
cos p g h z p = +
cos
p h
g
x x

c c
=
c c
The 1
st
eq. in (7.52) thus becomes
2
2
cos sin
u h
g g
z x

c c
=
c c
(7.54)
Now,
1
h h
O
x L
c
| |
=
|
c
\ .

so that unless is very small, eq(7.54) can be approximated by
2
2
sin
u
g
z

c
=
c
Integrating, we get
sin
u g
z A
z

c
= +
c
2
1
sin
2
u z g Az B

= + + (7.54a)
where A, B are functions of , x t .
Applying (7.53) at z h = , we have
0 sin
g
h A

= +
Applying the no-slip condition at 0 z = , we have
0 B =
Hence, (7.54a) becomes
1
sin
2
z
u h z g

| |
=
|
\ .
(7.55)
sin
u z h
g
x x

c c
=
c c
so that the incompressibility condition gives
sin
w z h
g
z x

c c
=
c c
which can be integrated to give
2
sin
2
z h
w g A
x

c
= +
c
Putting in the no-slip condition at 0 z = , we have
0 A =
so that
2
sin
2
z h
w g
x

c
=
c
Putting this into the surface wave condition,
h h
w u
t x
c c
= +
c c
at z h = [see eq(3.18)]
we have
2
sin 0
2
h h h
u g
t x

| | c c
+ + =
|
c c
\ .
at z h = (7.55a)
Now, (7.55) gives
2
sin
2
h
u g

= at z h =
so that (7.55a) becomes
2
sin 0
h h h
g
t x

c c
+ =
c c
(7.56)
The characteristics of this obey (see 3.9a3)
2
1 sin 0
dt dx dh
h g

= =
so that
1
h c =
2
1
sin dx g
c
dt

=
2
1 2
sin g
x c t c

=
Hence, the general solution of (7.56) is
2
sin g
h f x h t

| |
=
|
\ .
(7.56a)
where f is an arbitrary function.
Now, (7.56a) is of the wave-steepening type discussed in 3.9.3.
Thus, taking its partial, we have
2 sin
1 '
h g h
ht f
x x

c c
| |
=
|
c c
\ .
'
2 sin
1 '
f
g
tff

=
+
which blows up at
C
t t = , where
j
2 sin
1 ' 0
C
t t
g
tff

=
+ =
In practice, surface tension effects set in to prevent instability of the type depicted in
Fig.3.16c. The wave profile at the front therefore remains in the form of a nose
shown in Fig.3.16b and Fig.7.11.
According to Huppert, beyond
C
t , h is described by the similarity solution
sin
x
h
g t

= (7.57)
[see H.E.Huppert, J.Fluid Mech 173, 557-94 (1986)]
The position
N
x of the front edge is obtained from the constant volume condition
0
N
x
hdx A =
}
(7.58)
where A is the area of the cross section of the fluid in the x-z plane.
Putting in (7.57), we have
3/ 2
0
2
sin sin 3
N
x
N
A xdx x
tg tg


= =
}
or
1/ 3
2
9 sin
4
N
A g
x t

| |
=
|
\ .
(7.59)
7.10 Lubrication Theory
When a solid body slides over another one, the frictional resistance it experiences is
usually comparable in magnitude to the normal force it exerts on the other.
However, if there is a thin film of fluid between them, the frictional resistance can be
reduced drastically. This is called lubrication.
The main reason for this is that, in thin film flows, the ratio between the magnitudes
of the tangential and normal stresses is of the order [see 7.6]
2
1
U
h
h
UL
L
h


7.10.1 Slider Bearing
7.10.2 Flow Between Eccentric Rotating Cylinders
7.10.1 Slider Bearing
Consider the 2-D system shown in Fig.7.12.
The rigid lower boundary at 0 z = moves with velocity U past a stationary block of
length L. The space between them is filled with viscous fluid, and the pressures at
both ends of the bearing are equal to the atmospheric value
0
p .
Working with the 2-D version of the thin film flow eqs (7.41), we can begin at the
solution
2
1
2
dp
u z Az B
dx
= + + (7.42)
where p is a function of x only.
The boundary conditions are
u U = at 0 z =
0 u = at z h =
which, when plugged into (7.42), gives
U B =
2
1
0
2
dp
h Ah U
dx
= + +
so that
, )
1
1
2
dp z
u h z z U
dx h
| |
= +
|
\ .
, )
1
2
dp U
z h z
dx h
(
= +
(

(7.60)
The integral form of the incompressibility condition is that the mass flux Q is the
same across all cross-sections of the film. In terms of (7.60), we have
0
h
Q udz =
}
, )
0
1
2
h
dp U
z h z dz
dx h
(
= +
(

}
3 2
1 1 1 1
1
2 2 3 2
dp U
h h
dx h
| | | |
= +
| |
\ . \ .
3
1
12 2
dp U
h h
dx
= + (7.61)
Rearranging terms gives
3
12
2
dp U
h Q
dx h

| |
=
|
\ .
which integrates to
0 3
0
12
2
x
U
p p h Q dx
h

| |
=
|
\ .
}
2 3
0 0
1 1
6 2
x x
U dx Q dx
h h

| |
=
|
\ .
} }
(7.61a)
where
0
p p = at 0 x = .
Now, we also have
0
p p = at x L = , so that
2 3
0 0
1 1
0 6 2
L L
U dx Q dx
h h

| |
=
|
\ .
} }
ie.,
2 2
0 0
3 3
0 0
6 1
2 12 1
L L
L L
Udx dx
h h U
Q
dx dx
h h

= =
} }
} }
(7.62)
In the special case of a plane slider bearing, h vary linearly from
1
h at 0 x = to
2
h
at x L = , ie.,
, ) , )
2 1 1
x
h x h h h
L
= +
Using the formulae
, )
, )
2
1 dx
b a bx
a bx
=
+
+
}
, ) , )
3 2
1
2
dx
a bx b a bx
=
+ +
}
we have
2
2 1 2 1
1
1 dx
h h h h h
h x
L L
=

| |
+
|
\ .
}
, ) , )
2
2 1 1 2 1
L
h h h L h h x
=
+ (

, )
2 1
L
h h h
=

2 3
2 1 2 1
1
1
2
dx
h
h h h h
h x
L L
=

| |
+
|
\ .
}
, ) , )
3
2
2 1 1 2 1
2
L
h h h L h h x
=
+ (

, )
2
2 1
2
L
h h h
=

so that
, )
2
2 1 2 1 2 1 0
1 1
L
dx L L
h h h h h h h
| |
= =
|

\ .
}
, )
, )
2 1 3 2 2 2 2
2 1 2 1 2 1 0
1 1
L
dx L L
h h
h h h h h h h
| |
= = +
|

\ .
}
whereupon eq(7.62) gives
, )
2 1
2 1
2
U h h
Q
h h
=
+
Also, Eq(7.61a) becomes
0 2 1
2 2
2 1 1 2 1 2 1 1
1 1 1 1
6
p p UL h h L
U
h h h h h h h h h h
| | | || |
= +
| | |
+
\ . \ .\ .
2 1
2 2
2 1 1 2 1 1
1 1 1 1 UL h h
h h h h h h h h
(
| |
=
( |
+
\ .

2 1
2 2 2 2
2 1 1 2 1 1
1 1 1 1 1 1 ULh h
h h h h h h h h
(
| || | | |
= +
( | | |

\ .\ . \ .

2 1
2 2 2
2 1 2 1 1 2
1 1 1 1 1 ULh h
h h h h h h h h
(
| |
= +
( |

\ .

, )
, )
2
2 1 1 2
2 2 2
2 1
UL
h h h h h h
h h h
( = +

, )
, ) , )
1 2
2 2 2
2 1
UL
h h h h
h h h
=

(7.63)
For
2 1
h h h < < , the last expression is positive so that
0
p p > and we have an
upward lift. Hence, lubrication occurs if h decreases in the direction of flow as
shown in Fig.7.12.
7.10.2 Flow Between Eccentric Rotating Cylinders
Consider the setup shown in Fig.7.13.
Viscous fluid fills the narrow gap between a fixed outer cylinder of radius
, ) 1 r a c = + , and an inner, offset, cylinder of radius a which rotates with peripheral
velocity U.
This serves as a simple model for an axle rotating in its housing.
Geometry
Let the centers of the outer and inner cylinders be at , ) 0, 0 = x and , ) , 0 q = x ,
respectively.
The quantity , ) h u then provides a convenient measure of the height of the fluid.
Simple trigonmetric maneuver gives
, ) 1 h a c o = +
where
2 2 2
2 cos a a o q q = +
, ) t u = +
sin sin
a
u
q
=
Now, and hence are small. Keeping only 1
st
order terms in them, we have
1
sin sin
sin
a a
u u
q q

| |
=
|
\ .

, ) cos cos t u = + 1
]
, ) , ) cos cos sin sin t u t u = +
cos cos sin sin u u = +
2
cos sin
a
q
u u +
1 cos 1 cos a a
a a
q q
o u
| | | |
+
| |
\ . \ .

and
, ) cos 1 cos h a a
a
q
c u c u
| |
=
|
\ .
(7.64)
where
offset of centers
difference of radii
eccentricity
a
q

c
= = =
The minimum of h is at 0 u = :
, ) , ) 0 1 0 h ac = >
Hence, 0 1 s s .
For 0 = , the cylinders are coaxial.
For 1 = , they are in contact at 0 u = .
Solution
For small gaps, curvature effects may be neglected so that the results for the plane
slider bearing are applicable here with straightforward adaptations such as x au ,
u u
u
, etc.
Thus, eq(7.60) becomes
, )
1
2
dp U
u z h z
a d h
u
u
1
= +
(
]
(7.65)
where z is the distance across the gap, measured along the radial direction.
Similarly, eq(7.61) becomes
3
1
12 2
dp U
Q h h
a d u
= + (7.66)
Using
, ) 1 cos h ac u = (7.64)
the counterparts to the integrals in (7.62) are
, )
2
2 2 2 2
0 0
1
1 cos
d d
I a
h a
t t
u u
c
u
2
= =

} }
, )
2 2
3 3 3 2 3
0 0
1
1 cos
d d
I a
h a
t t
u u
c
u
= =

} }
These can be evaluated by contour integral as follows.
1
st
, let
i
z e
u
=
so that
, )
1 1 1
cos
2 2
i i
e e z
z
u u
u
| |
= + = +
|
\ .
i
dz ie d izd
u
u u = =
whereupon
2 2 2 2 2 2
2
1 4
2 1
1 1
2
C C
dz zdz
I
a ia
z z iz z
z
c c

= =
1 | | | |
+ +
| |
(
\ . \ . ]
} }
2
3 3 3 2 3 2 3 3
2
1 8
2 1
1 1
2
C C
dz z dz
I
a ia
z z iz z
z
c c

= =
1 | | | |
+ +
| |
(
\ . \ . ]
} }
where C is the unit circle centered at the origin.
Now, the poles are the roots of
2
2
1 0 z z

+ =
i.e.,
, )
2
2
1 1 1
1 1 1 z

= =
Since 0 1 s s , they are both real.
Furthermore, 1 z z
+
= so that only one of them is inside C.
Obviously, z z
+
> , so that only the pole at z

that contributes to the integrals.


With the help of
2
z z

+
+ =
2
2
1 z z

+
=
we have
, ) , )
2 2 2 2 2
4
C
zdz
I
ia
z z z z
c
+
=

}
, )
, )
2 2 2
4
2
z z
d z
i
ia dz
z z
t
c

+
=
1
=
(

(
]
, ) , )
2 3 2 2
8 1 2z
a
z z z z
t
c

+ +
1
=
(

(
]
, )
, )
3
2 2
8 z z
a z z
t
c
+
+
+
=

, )
3/ 2
2 2
2
1 a
t
c
=

, ) , )
2
3 3 3 2 3 3
8
C
z dz
I
ia
z z z z
c
+
=

}
, )
, )
2 2
3 2 3 3 2
8 1
2
2
z z
d z
i
ia dz
z z
t
c

+
=
1
=
(

(
]
, ) , )
2
3 4 2 3 3
8 2 3
z z
d z z
a dz
z z z z
t
c

+ +
=
1 | |
= ( |
|

(
\ . ]
, )
, )
4 2 3 3
2 8
z z
z z z d
a dz
z z
t
c

+
+
=
1 | |
+
= ( |
|

(
\ . ]
, )
, )
, )
4 5 2 3 3
4 2 8 2 2 z z z z z
a
z z z z
t
c
+ +
+ +
1
+ +
=
(

(
]
, )
, )
2 2
5
2 3 3
16
2 2 z z z z z
a z z
t
c
+ +
+
1 = +
]

, )
, )
2 2
5
2 3 3
16
4 z z z z
a z z
t
c
+ +
+
= + +

, )
, )
2
5
2 3 3
16
2 z z z z
a z z
t
c
+ +
+
1
= + +
]

, )
2
5/ 2 2
2 3 2
4
2
2 1 a
t

c
| |
= +
|
\ .

, )
, )
2
5/ 2
2 3 2
2
1 a
t
c
+
=

Thus, (7.62) becomes


2
3
2
UI
Q
I
=
, )
, )
, )
5/ 2
2 3 2
3/ 2 2
2 2
1
2
2 2
1
a
U
a
c
t
t
c

=
+

2
2
1
2
Ua

c

| |
=
|
+
\ .
(7.67)
From (7.66), we see that
dp
du
is an even function of since h is.
Hence, p, and hence cos p u , are odd functions of .
The horizontal force on the inner cylinder
cos
x
S
F p dS u =
}
therefore vanishes.
The calculation of the vertical force is too involved to be included here.
We merely list the result:
, )
2 2 2
12
1 2
U
F
t
c
=
+
(7.68)
Of particular interest is the factor
, )
1/ 2
2
1

which makes
F as 1 .
In other words, by moving horizontally towards the outer cylinder, the inner one can
substain arbitrarily large vertical loads.
Let us consider the flow at u t = .
From (7.64), we see that
, ) 1 h ac = + at u t =
From (7.67), we have
2
1
2
Q
U
h

+
=
+
at u t =
We can therefore write (7.66) as
2 3
12
dp U Q
a
d h h

u
| |
=
|
\ .
2 2
1
12 1
2
U
a
h

+
| |
=
|
+
\ .
2
2 2
1
12
2
U
a
h

| | +
=
|
+
\ .
0 > at u t =
which implies an adverse pressure and hence reversed flow in that neighborhood.
Differentiating (7.65), we have
, )
1 1
2 2
u dp dp U
h z z
z a d a d h
u
u u
| | c
= +
|
c
\ .
, )
1
2
2
dp U
h z
a d h u
=
, )
3
6
2
2
Uh U
Q h z
h h
| |
=
|
\ .
, )
2
3 2
6 1
2
2 2
U h U
a h z
h h

| |
=
|
+
\ .
Hence
2
2 2
6 1
2 2
z h
u U h U
a
z h h
u

c

=
| | c
=
|
c +
\ .
2
2 2
6 1
3 2
U h
a
h

| |
=
|
+
\ .
, )
2
2 2
6 1 1
1 cos
3 2
U
a a
h

c u c

| |
=
|
+
\ .
2
2 2
2 3 3
1 cos
2
Ua
h
c
u

| |
=
|
+
\ .
2
2 2
2 1 4
cos
2
Ua
h
c
u

| | +
=
|
+
\ .
, )
2
2 2
2 1 4
cos
2
Ua
h
c
u

| |
+
| =
|
+
\ .
(7.69)
For this to be negative, we need
, )
2
2
1 4
cos 0
2

u

+
<
+
(7.69a)
Now, reversed flow, if occurs, is expected around u t = where cos 1 u > or
cos 1 u s . Rewriting (7.69a) to reflect this, we have
, )
2
2
1 4
1
2

>
+
0
z h
u
z
u
=
c
<
c
so that no reversed flow occurs.
Rewrite the criterion as
, ) , )
3 2 2
0 4 2 1 1 3 1 > + + = + + (7.69b)
The critical values of are the roots of the polynomial on the right side, i.e.,
1 =
, )
1
3 13
2

=
However, since 0 1 s s , the only valid value is
, )
1
3 13 0.30
2

+
= +
Writing
, ) , )
2
3 1
+
+ =
we see that near
+
so that o
+
= + , eq(7.69b) becomes
, )
, )
0 13 13 o o o o o
+
> + = +
so that to satisfy (7.69b), we must have 0 o < , i.e.,

+
<
Conversely, if

+
> (7.70)
there will exist a range of such that 0
z h
u
z
u
=
c
>
c
so that reversed flow is expected.
This feature, though of theoretical interest, is usually overwhelmed by other
complications in practice.
8. Boundary Layers
8.1 Prandtls Paper
8.2 Steady 2-D Boundary Layer Eq.
8.3 Boundary Layer On Flat Plate
8.4 High R Flow In Converging Channel
8.5 Rotating Flows Controlled By Boundary
Layers
8.6 Boundary Layers Separation
8.1 Prandtls Paper
8.2 Steady 2-D Boundary Layer Eq.
We now derive the eq. for a steady, incompressible, 2-D boundary layer adjacent to a
rigid wall at 0 y = .
The no-slip condition for a stationary boundary means
0 u v = = at 0 y = (8.2)
which differs from the inviscid flow boundary condition only by the additional
demand of 0 u = . Our main concern is therefore to see how u merges from 0 to the
inviscid flow value.
For a thin boundary layer, we expect
u u
y x
c c
c c
(8.4a)
To get an order of magnitude estimates, let
0
U be some typical value of u.
Let the distance for u to change by an amount
0
U be of order L in the x-direction,
and in the y-direction. Hence, (8.4a) means
0 0
U U
L

ie.,
L (8.4)
The Navier-Stokes eqs. are
2 2
2 2
1 u u p u u
u v
x y x x y

| | c c c c c
+ = + +
|
c c c c c
\ .
2 2
2 2
1 v v p v v
u v
x y y x y

| | c c c c c
+ = + +
|
c c c c c
\ .
(8.5)
0
u v
x y
c c
+ =
c c
The 3
rd
eq. in (8.5) means
0
v u U
y x L
c c
c c

so that v is of order
0
U
L

. Hence v u .
The other 2 eqs. in (8.5) then imply
p p
x y
c c
c c

which means we may write , ) p p x = in the boundary layer.


Similarly, from
2
0
2 2
u U
x L
c
c

2
0
2 2
u U
y
c
c

we deduce
2 2
2 2
u u
y x
c c
c c

These estimates can be used to simplify (8.5) into the boundary layer eqs.,
2
2
1 u u dp u
u v
x y dx y

c c c
+ = +
c c c
(8.1)
0
u v
x y
c c
+ =
c c
(8.2)
Eq(8.1) provides the means to estimate the magnitude of the layer thickness .
Now,
2 2
0 0 0 0
. . .
U U U U
L H S
L L L

+
Assuming viscosity effect to be significant within the layer, we have
2
0 0
2
U U
L

ie.,
0
1
L U L R

=
where R is the Reynolds number.
Hence, thin boundary layer needs large R.
By interpreting x as the direction tangent, y as that normal, to the boundary, (8.1-2)
are applicable to curved boundaries. Whats worth mentioning is that although
p p
x y
c c
c c
, we still have , ) p p x .
Let , ) U x be the slip velocity at 0 y = for the corresponding inviscid flow.
For a thin boundary layer, we expect , ) u U x at the edge y = , where,
according to the Bernoulli theorem,
2
1
2
p U const + = along a streamline. Hence,
dp dU
U
dx dx
= (8.8)
so that a rise in U is accompanied by a drop in p and vice versa.
To ensure the boundary layer merges smoothly into the mainstream inviscid flow, we
impose the boundary condition on (8.1-2)
, ) u U x as or
y y

(8.9)
8.2.1 Example
Consider the problem
'' ' 1 u u + = ; , ) 0 0 u = , , ) 1 2 u = (8.10)
where 0 > is a small constant.
Integrating, we have
, )
1
'
ln 1 '
1 '
du
y u C
u
= = +

}
or,
/
2
1 '
y
u C e

=
so that
/ /
2 3 4
y y
u y C e dy y C e C

= = + +
}
Putting in the boundary conditions, we have
3 4
0 C C = +
1/
3 4
2 1 C e C

= + +
which is easily solved to give
3 1/
1
1
C
e

=

and
4 1/
1
1
C
e

=

so that
/
1/
1
1
y
e
u y
e


= +

(8.11)
Now,
1/
1 e

.
Also
/
1
y
e

unless y s .
Hence, u as given by (8.11) may be splitted into 2 parts:
1. Mainstream part: 1
M
u y = + for y > .
2. Boundary layer part:
/
1
y
BL
u e

= for 0 y s < .
Note that
0
lim
M
u u

= with y fixed.
0
lim
BL
u u

= with / y fixed.
and
/ 0
lim lim
BL M
y y
u u

=
which is equivalent to (8.9).
Alternative Approach
Another approach which is more akin to that taken in fluid dynamic problems is to
take advantage of the smallness of at the earliest stage. Thus, (8.10) simplifies to
' 1
o
u
with solution
o
u y c = +
There is now only 1 arbitrary constant so that only 1 of the 2 boundary conditions can
be satisfied. On making , ) 1 2
o
u = , we obtain an outer solution
1
o
u y = +
What we did so far is similar to obtaining the mainstream inviscid flow:
The problem is simplified by invoking the smallness of a parameter to drop the
highest derivatives.
However, this lowering the order of the system reduces the number of arbitrary
constants in the solution so that not all boundary conditions can be satisfied.
Thus, an inner solution, or boundary layer, is needed to satisfy the other boundary
conditions.
To accentuate the relatively rapid change of u inside the boundary layer, we shift to
the variable
y
Y

=
so that (8.10) becomes
2
2
d u du
dY dY
+ =
On invoking 1 , we have the inner (boundary layer) eq.,
2
2
0
i i
d u du
dY dY
+
which, on integrating, gives
1
ln
i
du
Y C
dY
| |
= +
|
\ .
ie.,
2
Y i
du
C e
dY

=
so that
3 4
Y
i
u C e C

= +
On making this satisfy the no-slip boundary condition , ) 0 0 u = at 0 y = , we
get
3 4
0 C C = +
so that
, )
3
1
Y
i
u C e

=
Matching the inner & outer solutions with
0
lim lim
i o
Y y
u u

=
we get
3
1 C =
so that the full solution is
/
1
1
y
y
u
e

+

as 0 for fixed
/
y
y

as found before.
8.3 Boundary Layer On Flat Plate
If the fluid is inviscid, a uniform stream approaching a flat plate at zero incidence
angle suffers no deflection. Hence , ) U x const = . According to (8.8), 0
dp
dx
= so that
the boundary layer eqs (8.1-2) reduce to
2
2
u u u
u v
x y y

c c c
+ =
c c c
(8.12)
0
u v
x y
c c
+ =
c c
(8.13)
For a semi-infinite plate that extends from 0 x = to x = , it seems natural to seek a
similarity solution:
, )
u
h
U
= with
, )
y
g x
= (8.14)
and g is to be determined.
1
st
of all, (8.13) can be satisfied if we introduce the stream function so that
u
y
c
=
c
v
x
c
=
c
(8.15)
With the help of (8.14), we have
, ) , ) , ) , , x y u x y dy k x = +
}
, ) , )
dy
Uh d k x
d

= +
}
, ) , ) , ) Ug x h d k x = +
}
If the plate is a streamline, is constant on it. For convenience, we can set
, ) ( , 0) , 0 0 x y x = = = =
and get rid of , ) k x by writing
, ) , ) , ) , ) , )
0
, x Ug x h d Ug x f

= =
}
(8.16)
where
, ) , )
0
f h d

=
}
so that , ) 0 0 f =
Thus,
, )
, )
1 df df
u Ug x U
y g x d d


c c
= = =
c c
(8.17)
Using
2
y dg dg
x g dx g dx
c
= =
c
we have
dg dg df
v U f g
x x dx g dx d


| | | | c c c
= + =
| |
c c c
\ . \ .
df dg
U f
d dx

| |
=
|
\ .
(8.18)
Similarly,
2 2
2 2
u d f dg d f
U U
x x d g dx d


c c
= =
c c
2 2
2 2
u d f U d f
U
y y d g d


c c
= =
c c
(8.18a)
2 3 3
2 3 2 3
u U d f U d f
y g y d g d


c c
= =
c c
so that (8.12) becomes
2 2 3
2 2 2 3
df dg d f df dg U d f U d f
U U U f
d g dx d d dx g d g d



| |
=
|
\ .
or
2 3
2 3
dg d f d f
Ug f
dx d d


= (8.18b)
which is self-consistent only if
1
dg
g c
dx
=
3 2
1 3 2
0
d f d f
cUf
d d


+ =
where
1
c is a constant.
Integrating the 1
st
eq. gives
2
1 2
1
2
g c x c = + (8.18c)
where
2
c is another constant.
Now, eqs(8.18a) show that the partials of u are all proportional to
1
g

, which blows
up as 0 g . The only place such extremes can be tolerated is at the leading edge of
the plate where 0 x = . Hence, we set
2
0 c = in (8.18c).
To summarize, we now have
1
2 g c x =
1
2
dg c
dx x
=
1
2
y
c x
=
, )
1
2 U c x f = (8.19)
df
u U
d
=
1
2
df c
v U f
d x

| |
=
|
\ .
(8.17,18)
3 2
1
3 2
0
d f cU d f
f
d d
+ = (8.20)
Note that the value of at the origin 0 x y = = depends on the path one chooses to
take the limit. Thus, = if we approach along the y-axis but 0 = if we
approach along any other straight line y mx = on which
1
2
x
m
c
= .
Since (8.20) is a 3
rd
order ODE, therell be 3 arbitrary integration constants (besides
1
c ) in a general solution. Their determination require 3 boundary conditions.
From (8.16), we already have
, ) 0 0 f = .
Next, as the flow approaches the leading edge of the plate along the x-axis ( 0 ),
we expect 0 u = at 0 = so that (8.17) gives
, )
0
' 0 0
df
f
d

=
=
Finally, far away from the plate ( y or ) the flow should be , ) , 0 U = u .
Hence, (8.17) gives
, ) ' 1 f =
while (8.18) gives
, ) lim ' 0 f f

=
Finally, inspection of the eqs shows that
1
c serves only as a scale for x so that its role
is more or less redundant for a similarity solution. For convenience, we can set it to
1
c
U

=
so that (8.20) becomes
3 2
3 2
0
d f d f
f
d d
+ = (8.20)
which is free of any parameters of the problem.
Solution to this problem must be obtained numerically.
The results are shown in fig.8.8.
Of interest is that 0.97
u
U
= at 3 = .
The boundary layer thickness can be estimated from
2
U
y
x

= to give
x
O
U

| |
=
|
\ .
(8.22)
as shown in Fig.2.14.
The horizontal stress on the plate is
0 0
xy
y y
u v u
y x y

= =
| | | | c c c
= + =
| |
c c c
\ . \ .
where
0
0
y
v
x
=
c
| |
=
|
c
\ .
since , ) , 0 0 v x = for all 0 x > .
Uisng (8.18a) and = , we have
, )
2
2
0
'' 0
xy
U d f U
f
g d g

=
= =
Since
1
2
2
x
g c x
U

= =
we have
, )
2
2
0
'' 0
2 2
xy
U d f U
U U f
x d x



=
= = (8.23)
which decreases as x increases.
Plate of Length L
Assuming (8.23) to hold for a plate of finite length L, the drag on it is
, ) , )
0 0
2 2 '' 0 2 2 '' 0
2
L L
xy
U dx
D dx U f U UL f
x

= = =
} }
where the factor 2 accounts for the equal contributions from both the upper & lower
side of the plate. In terms of the Reynolds number
UL
R

= , we have
, )
2 1/ 2
2 2 '' 0 D U Lf R

= (8.24)
Note that D is proportional to L . Hence
0
lim 0 D

=
Numerically, , ) '' 0 0.4696 f = .
Both (8.24) & the velocity profile agrees with experiment unless R is very high, at
which case, the boundary layer becomes unstable and turbulence ensues. The critical
value of R for this onset is about
5 6
10 ~10 .
8.4 High R Flow In Converging Channel
We consider the high R, 2-D flow between 2 plane walls
0 y = and tan y x =
A narrow slit at the origin provides the means for the necessary intake or discharge to
maintain a steady flow.
For the mainstream flow, the simplest ansatz is purely radial:
, ) ,
r r
u r = u e
The incompressibility condition
, ) 0
r
u
ru
r r r

c c
V = + =
c c
u
then becomes
, ) 0
r
ru
r
c
=
c
Therefore
r
Q
u
r
= (8.25a)
where Q is a constant and the sign is inserted so that 0 Q > represents inflow.
Next, the 2-D Euler eqs in polar coordinates are [cf. eq(2.22), 2.4]
, )
2
1
r
r
u u p
u
t r r

c c
+ V =
c c
u
, )
1
r
u u u p
u
t r r


c c
+ V + =
c c
u
which, for our steady flow ansatz, become,
1
r
r
u p
u
r r
c c
=
c c
, ) p p r =
Thus,
r
u is also a function of r only and, upon integration,
2
1
1 1
2
r
u p c

= +
Using (8.25a), we have
2
2
1 1 2
1
2
r
Q
p u c c
r

= + = +
Boundary Layer on 0 y =
The mainstream flow, according to (8.25a), is
, )
Q
U x
x
= (8.25)
The boundary eqs
dp dU
U
dx dx
= (8.8)
2
2
1 u u dp u
u v
x y dx y

c c c
+ = +
c c c
(8.1)
0
u v
x y
c c
+ =
c c
(8.2)
thus become
2
2 3
dp Q Q
U
dx x x
= =
2 2
3 2
u u Q u
u v
x y x y

c c c
+ = +
c c c
(8.26)
Following the approach of 8.3, we try a similarity solution
, ) , ) , )
Q
u U x h h
x
= = with
, )
y
g x
=
Introducing the streamline function to satisfy (8.2), we have
u
y
c
=
c
v
x
c
=
c
and
, ) , ) , ) , , x y u x y dy k x = +
}
, ) , )
Q dy
h d k x
x d

= +
}
, ) , ) , )
Q
g x h d k x
x
= +
}
Setting the wall 0 y = to be a streamline, is constant on it. For convenience, we
can set
, ) ( , 0) , 0 0 x y x = = = =
and get rid of , ) k x by writing
, ) , ) , ) , ) , )
0
,
Q Q
x g x h d g x f
x x

= =
}
where
, ) , )
0
f h d

=
}
so that , ) 0 0 f =
Thus
, ) ' '
Q Q
u g f f
y x y x


c c
= = =
c c
and, with
2
'
'
y g
g
x g g


c
= =
c
we have
2
' '
'
g g g g
v Q f f f
x x x x g


( | | c
= = + +
( |
c
\ .
2
' '
1 '
Qg g g
x f x f
x g g

( | |
= +
( |
\ .
, )
2
'
'
Qg g
f x f f
x g

(
=
(

Similarly,
2 2
1 1 ' '
' '' ' ''
u g Q g
Q f f f x f
x x x g x g

( | | | | c
= + = +
( | |
c
\ . \ .
''
u Q
f
y xg
c
=
c
2
2 2
'''
u Q
f
y xg
c
=
c
so that (8.26) becomes
, )
2 2
' '
' ' '' ' ''
Q Q g Qg g Q
f f x f f x f f f
x x g x g xg

( | | ( | |
+
( | | (
\ . \ .
2
3 2
'''
Q Q
f
x xg

| |
= +
|
\ .
which simplifies to
2
2
2
'
' 1 '' 1 '''
g x
f x ff f
g g Q
| |
= +
|
\ .
(8.26a)
This can be consistent by setting
1
g c x =
so that (8.26a) becomes
2
2
1
' 1 ''' f f
c Q

= +
To make this parameter free, we set
2
1
c
Q

= , where Q is used to ensure the reality


of
1
c . Hence
2
' 1 ''' f f = (8.27a)
where the +/- sign corresponds to Q positive/negative, ie., to in/out flow.
One of the boundary conditions associated with (8.27a) was already found to be
, ) 0 0 f =
Another one should be provided by the no-slip condition
0 u v = = for all 0 y =
Since
0
0
y
v
=
= is automatically satisfied by (8.27), we get only one new condition
, ) ' 0 0 f =
The 3
rd
boundary condition is obtained by merging the solution with the mainstream
flow:
y
Q
u
x

=
ie.,
, ) ' 1 f =
Setting
, ) , ) ' F f =
eq.(8.27a) can be transformed into a 2
nd
order eq.,
2
1 '' F F = (8.28)
with boundary conditions
, ) 0 0 F = , ) 1 F = (8.29)
Summary
Taking stock of progress so far, we have
g x
Q

=
Q y
x

=
Q f = (8.27)
'
Q
u f
x
=
2
' '
Q
Qg
v f f
x x

= =
, )
2
' ''
u Q
f f
x x

c
= +
c
2
''
Q u Q
f
y x
c
=
c
' F f =
where the +/- sign corresponds to in/out flow.
The ODE to solve is
, )
2
'' 1 0 F F =
with boundary conditions
, ) 0 0 F = ; , ) 1 F =
Solutions
Using
2
1
'' '
2
d
F F
dF
=
the ODE becomes
, )
2 2
1
' 1 0
2
d
F F
dF
=
which is integrated to
2 3
2
1 1
'
2 3
F F F c
| |
=
|
\ .
(8.29a)
From the mainstream flow (8.25a), we have
2
r
du Q
dr r
=
so that
2
0
r
y
du Q
dr x
=
=
The boundary layer quantity that should merge to this is
, ) , )
2 2
'
u Q Q
F F
x x x

c
= + = (

c
Since , ) 1 F = , this means
, ) ' 0 F =
Substituting back to (8.29a) gives
2
2
3
c =
so that (8.29a) becomes
2 3
1 1 2
' 0
2 3 3
F F F
| |
=
|
\ .
which can be rearranged to give
, ) , ) , )
2
2 3
3
' 2 3 2 1
2
F F F F F = + = +
For the outflow case, we have
, ) , )
3
' 2 1
2
F F F = +
Since , ) 0 0 F = , the right hand side is imaginary at 0 = . Hence, there is no
solution that can satisfy the required boundary conditions.
For the inflow case, we need to solve
, )
3
' 2 1
2
F F F = + (8.29b)
To get rid of the square root, we set
2
2 G F = +
so that
2 ' ' GG F =
and (8.29b) becomes
, )
2
3
2 ' 3
2
GG G G =
or
, )
2
1
' 3
6
G G =
Integrating gives
1
3 2
1
tanh
3 6 3 3
dG G
c
G


= = +

}
or
4
3 tanh
2
G c
| |
= +
|
\ .
2
4
3tanh 2
2
F c
| |
= +
|
\ .
(8.30)
where the
i
c s are constants.
Now, (8.30) satisfies , ) 1 F = automatically. To satisfy , ) 0 0 F = , we need
1
4
2
tanh 1.14
3
c

=
That
4
c can have 2 possible values is an example of the non-uniqueness of flow at
high R. [see 9.7]
The flow pattern for
4
1.14 c = is shown in fig.8.9 and is typically observed in
experiment.
The flow pattern for
4
1.14 c = involves reversed flow close to the wall.
Using the fact that , ) 1 1 F , the layer thickness can be estimated from
1
Q
x

= =
ie.,
x
Q

(8.31)
which decreases as one approaches the corner.
8.5 Rotating Flows Controlled By Boundary
Layers
8.5.1 Almost Uniform Rotation: Basic Eqs.
8.5.2 Steady, Inviscid Flow
8.5.3 Ekman Boundary Layers
8.5.4 The Interior Flow
8.5.5 Unsteady Flow: Spin-Down
8.5.1 Almost Uniform Rotation: Basic Eqs.
The time rate of change of a vector Q as seen from a fixed and a rotating frame are
related by
F R
d d
dt dt
| | | |
= +
| |
\ \
Q Q
Q
where is the angular velocity vector of the rotation.
Setting = Q x , we have
F R
= + u u
x
Taking the time derivative of the whole eq. gives
F R
F F F
d d d
dt dt dt
| | | | | |
= +
| | |
\ \ \
u u x

, )
R
R
R R
d d
dt dt
| | | |
= + + +
| |
\ \
u x
u x
, ) 2
R
R
R
d
dt
| |
= + +
|
\
u
u x
For fluid motion,
d
dt
is replaced with
D
Dt
. Thus, we have
, ) 2
F R
F F R R R
F R
t t
c c
| | | |
+ V = + V + +
| |
c c
\ \
u u
u u u u
u x
where the 3
rd
and 4th terms on the right are the Coriolis and centrifugal acceleration,
respectively.
[For a more rigorous derivation, see R.E.Meyer, Introduction to Mathematical Fluid
Mechanics, Chap 5, Wiley (71)]
At a given instance of time, quantities in the 2 frames are related by simple coordinate
transformations. Hence, provided the 2 frames are kept in the same orientation, we
have
, ) , )
F R
V = V Q Q
Dropping the subscript R for clarity, the Navier-Stokes eqs in the rotating frame are
simply
, ) , )
2
1
2 p
t

c
+ V + + = V + V
c
u
u u
u x u
(8.34)
0 V = u
Using
, ) , ) , ) , ) , ) , ) = a b c d a c b d a d b c
we have
, ) , ) , )
2
2 2
= x x x x
Using
, ) , ) , ) , ) , ) V = V + V + V + V a b a b b a a b b a
we have
, ) , )
2
1
2
V = V + V a a a a a [cf.
2
1
2
f f f V = V ]
, ) , ) , ) , ) V = V + V = V = x x x x
, ) , ) , ) , )
2
2 2 V = V = x x x x
, ) , ) , )
2
2 2 2 2 V = V + V = V = x x x x x x x x
where we have used
, ) 0
k
ijk ijk kj ijj
i
j
x
x

c
V = = = =
c
x
, )
i
j j ij i
i
j
x
x

c
V = O = O = O (

c
x
, )
i
j j ij i
i
j
x
x x x
x

c
V = = = (

c
x x
Thus
, ) , )
2 2
2 2
V = V V x x x
, ) , )
2
2 2 2 = = x x x
or
, ) , )
2 1
2
= V x x (8.35)
Thus, for const = , (8.34) becomes
, )
2
*
1
2 p
t

c
+ V + = V + V
c
u
u u
u u
(8.34a)
where
, )
2
*
1
2
p p = x (8.36)
which contains the centrifugal forces, is called the reduced pressure.
With the understanding that only reduced pressures will be used in the following, we
can drop the subscript * for the sake of clarity.
Let
U be typical value of u ,
L be typical length scale of flow.
We are interested in cases where the flow u is small compared to the rotation of the
system, ie.,
1
U
L
=
O
(8.36a)
Now,
, )
2
U
O
L
| |
V =
|
\
u u
, ) O U = O u
Eq.(8.36a) therefore implies
, )
1
U
O
L
V
| |
=
|
O
\
u u
u

Eq(8.34a) thus simplifies to
2
1
2 p
t

c
+ = V + V
c
u
u u (8.37)
0 V = u (8.38)
The relevant Reynolds number is
2
L
R

O
= [not
UL

]
and the flow will be considered inviscid if 1 R .
8.5.2 Steady, Inviscid Flow
Let the coordinate system attached to the rotating frame be , , x y z with the z-axis
parallel to the axis of rotation, so that 0, 0, .
With , , u v w u , we have
0 0 , , 0 v u v u
u v w

i j k
u i j
For a steady, inviscid flow, Eq(8.37) becomes
1
2
p
v
x

(8.39)
1
2
p
u
y

(8.40)
1
0
p
z

(8.41)
while (8.38) becomes
0
u v w
x y z



(8.42)
From (8.41), we see that p is independent of z.
So are u and v by way of (8.39-40).
Eq(8.42) then put w also independent of z.
Thus, u is independent of z, which is known as the Taylor-Proudman theorem.
8.5.3 Ekman Boundary Layers
Consider the steady flow between 2 boundaries that rotate with slightly different
angular velocities.
In the rotating frame attached to 1 of the boundaries, the fluid flow is obviously small
so that the almost uniform rotation eqs. (8.37-38) apply.
We assume the flow consists of 2 components.
One is the inviscid interior (mainstream, high R) flow that obeys eqs(8.39-42).
The other is the so called Ekman boundary layer flow for which the viscous term
2
V u can be approximated by
2


V u.
Let the boundaries be planes perpendicular to the z-axis.
The angular velocities of the boundaries at 0 z = and z L = are and , ) 1 O + ,
respectively. As stated earlier, 1 .
Obviously,
2
2
2
z

c
V =
c
u
u (8.43a)
Consider 1
st
the boundary layer on 0 z = .
Borrowing the results from 8.5.2, and the help of (8.43a), eqs. (8.37-38) become
2
2
1
2
p u
v
x z

c c
O = +
c c
(8.43)
2
2
1
2
p v
u
y z

c c
O = +
c c
(8.44)
2
2
1
0
p w
z z

c c
= +
c c
(8.45)
0
u v w
x y z
c c c
+ + =
c c c
(8.46)
The arguments presented in 8.2 can be applied here by the transformations
, ) , x x y and y z
Thus, (8.4a) becomes
,
u u u
z x y
c c c
c c c
,
v v v
z x y
c c c
c c c
(8.4a)
To get an order of magnitude estimates, let
0
U be some typical value of u and v.
Let the distance for u or v to change by an amount
0
U be of order L in the x-y plane,
and in the z-direction. Hence, (8.4a) means
0 0
U U
L

ie.,
L (8.4)
Eq. in (8.46) means
0
w u v U
z x y L
c c c
+
c c c

so that w is of order
0
U
L

. Hence , w u v .
Eq(8.43-5) then imply
or
p p p
z x y
c c c
c c c

so that we may write , ) , p p x y = in the boundary layer.


Hence,
p
x
c
c
and
p
y
c
c
remain close to their inviscid interior values, which obey
1
2
I
I
p
v
x
c
O =
c
(8.39)
1
2
I
I
p
u
y
c
O =
c
(8.40)
The boundary layer eqs(8.43-4) can therefore be written as
, )
2
2
2
I
u
v v
z

c
O =
c
(8.47)
, )
2
2
2
I
v
u u
z

c
O =
c
(8.48)
, ) , ) 8.47 8.48 i + gives
, )
, )
2
2
2
I I
u iv
v iu v iu
z

c +
O + =
c
, ) 2
I I
i iv u iv u = O + + 1
]
Setting
, )
I I
f u iv u iv = + +
we have
2
2
2
f
i f
z

c
= O
c
(8.48a)
Consider the ansatz
a z
f e =
where a is a constant. Eq(8.48a) is satisfied if
2
2 a i = O
or
/ 4
2 2
i
a i e


O O
= = [
/ 2 i
i e

= ]
, )
2
cos sin 1
4 4
i i


O O
| |
= + = +
|
\ .
Setting
*
z z

O
=
the general solution to (8.48a) is
, ) , )
* *
1 1 i z i z
f Ae Be
+ +
= +
where A and B are arbitrary functions of x and y.
Hence
I I
u iv u iv f + = + +
, ) , )
* *
1 1 i z i z
I I
u iv Ae Be
+ +
= + + +
To match the interior flow, we need
0 f as z
which requires 0 B = , so that
, )
*
1 i z
I I
u iv u iv Ae
+
+ = + +
The no-slip condition 0 u v = = at the wall 0 z = gives
0
I I
u iv A = + +
so that
, )
, )
, )
*
1
1
i z
I I
u iv u iv e
+
+ = +
, ) , )
*
* *
1 cos sin
z
I I
u iv e z i z

1 = +
]
Equating real and imaginary parts, we have
, )
* *
* *
1 cos sin
z z
I I
u u e z v e z

=
, )
* *
* *
1 cos sin
z z
I I
v v e z u e z

= +
or
, )
*
* *
cos sin
z
I I I
u u e u z v z

= + (8.49)
, )
*
* *
cos sin
z
I I I
v v e v z u z

= (8.50)
Taking the partials
*
* *
cos sin
z
I I I
u u u v
e z z
x x x x

c c c c
| |
= +
|
c c c c
\ .
*
* *
cos sin
z
I I I
v v v u
e z z
y y y y

| | c c c c
=
|
c c c c
\ .
and putting them into the incompressibility condition (8.46), we have
w u v
z x y
c c c
= +
c c c
, )
* *
* *
1 cos sin
z z
I I I I
u v u v
e z e z
x y y x

| | | | c c c c
= + +
| |
c c c c
\ . \ .
From (8.39-40), we have
2
1
2
I I
v p
y x y
c c
=
c O c c
2
1
2
I I
u p
x x y
c c
=
c O c c
so that
0
I I
u v
x y
c c
+ =
c c
and
*
*
sin
z
I I
w u v
e z
z y x

| | c c c
=
|
c c c
\ .
or
*
*
*
sin
z
I I
w u v
e z
z y x


| | c c c
=
|
c O c c
\ .
With
*
0
0
z
w
=
= , we have
*
*
* *
0
sin
z
z
I I
u v
w e z dz
y x


| | c c
=
|
O c c
\ .
}
At the edge of the Ekman layer where
*
z , we have
*
* *
0
sin
z
I I
E
u v
w e z dz
y x

| | c c
=
|
O c c
\ .
}
Now,
*
* *
0
sin
z
I e z dz

=
}
, )
, )
1
0
0
1 1 1
Im Im Im
1 1 2
i z z iz
e dz e
i i


+

1
= = = =
(

]
}
so that
1
2
I I
E
u v
w
y x
| | c c
=
|
O c c
\ .
(8.51)
1
2
I

=
O
where
I
is the z-component of the vorticity of the interior flow.
If the boundary is rotating with angular velocity
B
O relative to the rotating frame,
(8.51) is generalized to
1
2
E I B
w


| |
= O
|
O
\ .
(8.52)
Similarly, if the upper boundary at z L = rotates with angular velocity
T
O relative
to the rotating frame, we have
1
2
E T I
w


| |
= O
|
O
\ .
(8.53)
8.5.4 The Interior Flow
According to 8.5.2,
I
u ,
I
v , and
I
w are all independent of z.
This means
E
w at the top and bottom of the interior flow must match.
Eqs(8.52-3) thus imply
1 1
2 2
I B T I
(8.54)
or
I T B

In the case of Fig.8.10,
0
B
,
T

so that
I I
I
v u
x y




or, in terms of cylindrical coordinates , , r z ,
1
0 0
r z
I
I
r
r r z
ru


e e e


1
z I z
d
ru
r dr

e e
where weve used the fact 0
I
u
z

.
Integrating, we have
2
1
1
2
I
ru r c


or
1
1
2
I
c
u r
r


If
I
u

is to be regular at 0 r , we must set the constant


1
0 c , so that
1
2
I
u r


Thus, the angular velocity of the fluid is the average of those of the boundaries.
In other words, the motion of the fluid is entirely controlled by the boundary layers.
We now obtain the rest of the solution. Using (8.51), we have
1 1
2 2
zI I
u

The incompressibility condition in cylindrical coordinates



1 1
0
I zI
rI
u u
ru
r r r z




then reduces to

1
0
rI
ru
r r

which gives
2 rI
ru c
However, if
rI
u is to be regular at 0 r , we must set the constant
2
0 c , so that
0
rI
u
The secondary flow is therefore purely in the z direction. (see fig.8.10).
8.5.5 Unsteady Flow: Spin-Down
Consider the situation depicted in fig.8.11.
Initially, the fluid, together with its boundaries at z L = , rotate about the z-axis with
angular velocity , ) 1 O + .
At 0 t = , the angular velocities of both boundaries are reduced to .
Obviously, the fluid will eventually spin-down to the same angular velocity.
What interests us is the relevant time-scale.
Physically, we expect the spin down to begin with the formation of Ekman layers on
both boundaries. At this point, the interior inviscid flow still rotates essentially with
angular velocity , ) 1 O + . As is the case in 8.5.4,
I
u and
I
v are independent of z
but not
I
w . In fact, according to eqs(8.52-3), 0
I
w > ( 0
I
w < ) near the lower
(upper) Ekman layer.
Thus, as time goes by, the Ekman layers extend towards the interior flow, which is the
essence of the spin-down process.
The almost uniform rotation eqs(8.37-8) for the inviscid interior flow are
1
2
I I
I
u p
v
t x
c c
O =
c c
(8.56)
1
2
I I
I
v p
u
t y
c c
+ O =
c c
(8.57)
1
I I
w p
t z
c c
=
c c
(8.57a)
0
I I I
u v w
x y z
c c c
+ + =
c c c
(8.46)
Eliminating p from (8.56-7) gives
2 0
I I I I
u v v u
t y x y x
| | | | c c c c c
O + =
| |
c c c c c
\ . \ .
which, with the help of (8.46), becomes
2 0
I I I
u v w
t y x z
| | c c c c
+ O =
|
c c c c
\ .
(8.58)
As discussed earlier, 0
I
w
z
c
<
c
so that the vorticity in the interior decreases with time:
0
I I I
u v
t t y x
| | c c c c
= <
|
c c c c
\ .
in agreement of the Helmholtz vortex theorem (5.3).
Now,
I
u and
I
v are independent of z so that (8.58) can be integrated over the
vertical span of the interior fluid to give
j 2 0
I
I
L w
t
c
+ O =
c
(8.58a)
where L is the height of the interior fluid and j
I
w is the difference of
I
w at its top
& bottom.
By eq(8.51),
1
2
I I
w

=
O
on
top
bottom
| |
|
\ .
of the
lower
upper
| |
|
\ .
Ekman layer (8.59)
Hence, (8.58a) becomes
2 0
I
I
L
t


c
+ O =
c
(8.60)
with solution
/
exp 2
t T
I
A t Ae
L


| |
O
= =
|
\ .
where A is an arbitrary function of , ) , x y and
2
L
T

=
O
(8.61)
is called the spin down time.
Applying the initial condition
2
I
= O at 0 t =
we have
2 A O =
and
/
2
t T
I
e

= O
In terms of cylindrical coordinates , ) , , r z and assuming axisymmetry, we have
1
0 0
r z
I
I
r
r r z
ru

c c c
=
c c c
e e e

, )
/
1
2
t T
z I z
d
ru e
r dr



= = O e e
which is easily integrated to give
2 /
1
t T
I
ru r e c



= O +
Regularity at 0 r = then sets
1
0 c = so that
/ t T
I
u r e



= O (8.62)
which decribes the main dissipation process of the excess rotation.
The other components of
I
u are obtained from (8.58) and (8.55) as
/ t L
rI
r
u e
L


= O
/
2
t L
zI
z
u e
L


= O (8.63)
[cf. the last part of 8.5.4]
The streamlines of this secondary flow are described by
2
zI
rI
dz
dz u z
dt
dr
dr u r
dt
= = =
which integrates to
ln 2ln z r const = +
or
2
zr const =
(see Fig.8.11).
8.6 Boundary Layers Separation
9. Instability
9.1 The Reynolds Experiment
9.2a Interface Waves
9.2 Kelvin-Helmholtz Instability
9.3 Thermal Convection
9.4 Centrifugal Instability
9.5 Instability of Parallel Shear Flow
9.6 General Theorem on Stability of Viscous
Flow
9.7 Uniqueness & Non-Uniqueness Of Steady
Viscous Flow
9.8 Instability, Chaos, And Turbulence
9.9 Instability At Very Low Reynolds Number
9.1 The Reynolds Experiment
9.2 Kelvin-Helmholtz Instability
Consider a deep layer of inviscid fluid of density
2
moving with uniform speed U
over another stationary deep layer of density
1 2
> . [see Fig.9.4]
Let the interface be , ) , y x t = .
Asmall traveling wave disturbance there can be described by
, )
, )
,
i k x t
x t Ae


= (9.2)
with the understanding that only the real part of any complex quantity has physical
meaning.
The dispersion relation is (detailed derivation is given in 9.2a),
, ) , ) , )
2 2 2
1 2 2 1 2 1 2 1 2
kU k U k g k ( + = + + +

(9.3)
where is the coefficient of surface tension.
This can be written as
R I
i

=
where
2
1 2
R
kU


=
+
, ) , )
2 2 2
1 2 1 2 1 2
1 2
1
I
k U k g k

( = + +

+
Thus, is complex if
I
is real, ie.,
, ) , )
2 2 2
1 2 1 2 1 2
0 k U k g k ( + + >

or
, )
2
1 2
1 2
1 2
U g
k
k



> +
+
(9.4)
in which case, the dispersion
+
indicates an exponential growth with time of any
disturbance. This is known as the Kelvin-Helmholtz instability.
The minimum flow
C
U that can induce such an instability is given by
, )
2
1 2
1 2
1 2
min
C
k
U g
k
k



(
= +
(
+

Setting
, ) , )
1 2
g
f k k
k
= + with 0 k >
we have
, )
1 2 2
df g
dk k
= +
The extremum of f are therefore at
, )
1 2
C
g
k

=
with
, ) , )
1 2
2
C
f k g =
which is obviously a minimum.
Hence
, )
2
1 2
1 2
1 2
2
C
U
g



=
+
(9.6)
which says that both gravity & surface tension serves to stabilize the system by
raising the value of
C
U .
Kelvin-Helmholtz instability can also occur in a continuously stratified fluid with
0
0
d
dy

<
The buoyancy frequency [see 3.8]
2 0
0
g d
N
dy

= (9.7)
then serves as a measure of the stabilizing effects of the density distribution.
We leave it as an exercise (Ex.9.2) to show that the system becomes unstable only if
the Richardson number
, )
2
2
/
N
J
dU dy
= (9.8)
is less than
1
4
somewhere in the flow.
Such instabilities are observed in the atmosphere, sometimes in the form of clear air
turbulence and sometimes marked by distinctive cloud patterns.
9.2a Interface Waves
We now generalize the surface waves results in 3.2-4 to waves at the interface of 2
fluids.
All quantities related to the upper and lower fluids are labeled 2 and 1, respectively.
Mechanical equilibrium in the unperturbed state then requires
1 2
> .
The interface is at , ) , y x t = .
For ease of reference, we list below the basic eqs governing surface waves.
0 u
t x y
c c c
+ =
c c c
on , ) , y x t = (3.18)
, )
2
1
2
p
gy G t
t

c
+ + + =
c
u (3.19)
2 2
2 2
0
x y
c c
+ =
c c
(3.24)
Eqs(3.18,19) are linearized as:
t y
c c
=
c c
on 0 y = (3.21)
, )
p
gy G t
t

c
+ + =
c
(3.19a)
with the boundary condition
0
p p = on y =
and setting
, )
0
p
G t

=
Eq(3.19a) becomes
0 g
t


c
+ =
c
on 0 y = (3.22)
For the interface waves, eqs(3.21,19a,24) become
1
t y
c c
=
c c
2
t y
c c
=
c c
on 0 y = (3.21a)
, )
1 1
1
1
p
gy G t
t

c
+ + =
c
(3.19b)
, )
2 2
2
2
p
gy G t
t

c
+ + =
c
(3.19c)
2 2
1 1
2 2
0
x y
c c
+ =
c c
(3.24a)
2 2
2 2
2 2
0
x y
c c
+ =
c c
(3.24b)
The boundary conditions are
1
0
y
y

c
=
c
2
0
y
y

c
=
c
(A1)
1 2
p p p = = , on y = (A2)
and from (3.21a),
1 2
y y
c c
=
c c
on 0 y = (A3)
Consider the ansatz
, ) i k x t
Ae


= (3.23)
, )
, )
1 1
i k x t
f y e


= , )
, )
2 2
i k x t
f y e


= (B)
To satisfy (3.24a,b) and (A1), we need
2
1 1
'' 0 f k f =
2
2 2
'' 0 f k f =
with the boundary conditions
, )
1
0 f = , )
2
0 f =
Thus
1 1
k y
f c e =
2 2
k y
f c e

=
where
1 2
, c c are constants.
The interface condition (A3) then demands
2 2
c c C = =
so that (B) becomes
, )
1
k y i k x t
Ce

+
=
, )
2
k y i k x t
Ce

+
=
Eq(3.21) gives
kC i A = (3.21b)
, )
, )
1
1
i k x t
p
i C gA e G

+ = , )
, )
2
2
i k x t
p
i C gA e G

+ = (C)
Since
1 2
, G G are functions of t only, they must vanish identically.
Consistency then demands
, ) i k x t
p Pe

=
where P is a constant, so that (C) become
1
P
i C gA

+ =
2
P
i C gA

+ =
Eliminating P gives
, ) , )
1 2 1 2
0 i C gA + + = (3.21c)
Eq(3.21b,c) are compatible only if
, ) , )
1 2 1 2
0
k i
i g


=
+
ie.,
, ) , )
2
1 2 1 2
0 kg + =
or
2 1 2
1 2
kg


| |
=
|
+
\ .
1 2
1 2
kg


| |
=
|
+
\ .
The phase velocity is therefore
1 2
1 2
p
g
v
k k


| |
= =
|
+
\ .
and the group velocity:
1 2
1 2
1 1
2 2
g p
d g
v v
dk k


| |
= = =
|
+
\ .
9.2a.1 Surface Tension
The effect of the surface tension at the interface is to modify the interface condition
(A2) to
2
2 1 2
p p
x

c
=
c
at , ) , y x t =
where is the coefficient of surface tension. Thus, eq(C) is modified into
, )
, )
1
1
1
i k x t
p
i C gA e G

+ =
and
, )
, ) , )
, )
2
2 1
2
1
i k x t i k x t
i C gA e G p k Ae


+ =
As before,
1 2
0 G G = = and
, )
1
i k x t
p Pe

=
so that
1
P
i C gA

+ =
and
2
2 2
k P
i C g A


| |
+ =
|
\ .
Eliminating P gives
, ) , )
2
1 2 1 2
0 i C A g k 1 + + + =
]
which is compatible with (3.21b) only if
, ) , )
2
1 2 1 2
0
k i
i g k


=
+ +
ie.,
, ) , )
3 2
1 2 1 2
0 k kg + + =
or
3
2 1 2
1 2 1 2
k
kg


| |
= +
|
+ +
\ .
9.2a.2 Uniform Flow
We now consider the case where the flow of the upper fluid tends to a uniform speed
U sufficiently far away from the interface.
To emphasize this fact, we write
, )
1 1
1 1 1
, , u v
x y
| | c c
= =
|
c c
\ .
u , )
2 2
2 2 2
, , U u v U
x y
| | c c
= + = +
|
c c
\ .
u
where
, ) , ) , ) , )
1 1 2 2
0 u y v y u y v y = = = =
The linearized version of eqs(3.18) is
1
t y
c c
=
c c
2
U
t x y
c c c
+ =
c c c
on 0 y = (3.21)
which combines to give an interface condition
1 2
U
y y x
c c c
=
c c c
on 0 y =
to supplement
2
2 1 2
p p
x

c
=
c
at , ) , y x t =
The linearized version of eqs(19,24) are
, )
1 1
1
1
p
gy G t
t

c
+ + =
c
(3.19)
, )
2 2 2 2
2
2
1
2
p
U U gy G t
t x

c c
+ + + + =
c c
(3.19)
2 2
1 1
2 2
0
x y
c c
+ =
c c
(3.24a)
2 2
2 2
2 2
0
x y
c c
+ =
c c
(3.24b)
Consider the ansatz
, ) i k x t
Ae


= (3.23)
, )
, )
1 1
i k x t
f y e


= , )
, )
2 2
i k x t
f y e


= (B)
The solutions to (3.24a,b) are still
, )
1 1
k y i k x t
C e

+
=
, )
2 2
k y i k x t
C e

+
=
Eq(3.21) gives
1
0 i A kC + =
, )
2
0 i kU A kC = (3.21b)
Eq(3.19,) become
, )
, )
1
1 1
1
i k x t
p
i C gA e G

+ =
, )
, ) , )
, )
2 2
2 2 2 1
2
1 1
2
i k x t i k x t
i C ikUC gA e G U p k Ae


+ + = (C)
Since
1 2
, G G are functions of t only, we must have
1
0 G = and
2
2
1
2
G U =
Consistency also demands
, )
1
i k x t
p Pe

=
where P is a constant, so that (C) become
1
1
P
i C gA

+ =
, )
2
2
2 2
k P
i kU C g A


| |
+ =
|
\ .
Eliminating P gives
, ) , )
2
1 1 2 2 1 2
0 i C i kU C g k A 1 + + + =
]
(3.21c)
which becomes, with the help of the 1
st
eq in (3.21b),
, ) , )
2
2 1
2 2 1 2
0 i kU C g k A
k


1
+ + =
(
]
Eq(3.21b,c) are compatible only if
, )
, ) , )
2
2 1
2 1 2
0
k i kU
i kU g k
k




=
+
ie.,
, ) , )
2
3 2
1 2 1 2
0 kg k kU + + =
or
, ) , )
2 2 2 3
1 2 2 2 1 2
2 0 kU k U kg k + + = [cf. Ex.3.6]
so that
, ) , )
2 2 2 2 2 3
2 2 1 2 2 1 2
1 2
1
kU k U k U kg k

1
1 = +
]
(
]
+
, ) , )
2 2 2
2 1 2 1 2 1 2
1 2
1
kU k U k g k

1
1 = + + +
]
(
]
+
or
, ) , ) , )
2 2 2
1 2 2 1 2 1 2 1 2
kU k U k g k 1 + = + + +
]
(9.3)
9.3 Thermal Convection
Consider a viscous fluid resting between 2 horizontal rigid boundaries at 0 z and
z d . A temperature difference T is maintained between the boundaries, with
the lower one being hotter. If the density of the fluid decreases with rising
temperature, the fluid becomes top-heavy. Nonetheless, if T is increased from 0
slowly by small steps, the fluid can remain stable up to a critical value whereupon an
organized cellular motion sets in. (see fig.9.6)
9.3.1 Linear Stability Theory
9.3.2 Stability to Finite-Amplitude Disturbances
9.3.3 Experimental Results
9.3.1 Linear Stability Theory
For small temperature changes, the fluid density at temperature T can be
approximated by
, ,
1 T T 1 =
]
(9.9)
where is the volume coefficient of thermal expansion, and the density at
temperature T .
Since the change in is usually slight, the fluid is still approximately incompressible:
0
D
Dt

= and 0 V = u (9.10)
Assuming the viscosity to be temperature independent, the conservation of
momentum still takes the form of the Navier-Stokes eqs.
2
D
p
Dt
= V + V +
u
u g (9.11)
Thermal conduction in a fluid is a rather complicated process. For an introductory
discussion, see Landau & Lifshitz, Fluid Mechanics, chapter V.
Here, we simply generalize the elementary heat diffusion eq.
2
T
T
t

c
= V
c
to include convective effects, ie.,
2
DT
T
Dt
= V (9.12)
Here, is the thermal diffusivity of the fluid. It is related to the thermal conductivity
by
p
c

=
where
p
c the specific heat at constant pressure.
What (9.12) describes is a situation where the change of the amount of heat inside a
fluid element is due solely to the exchange of heat by conduction with the surrounding
fluid. Neglected are all other energy sources (eg., work done on fluid element by
stresses exerted by the surrounding fluid) and sinks (eg., energy dissipation due to
viscosity).
In the unperturbed state of rest, the temperature , ,
0
T z must satisfy the steady state,
no flow version of (9.12), namely
2
0
2
0
d T
dz
= (9.12a)
of which solution is
, ,
0 1 2
T z c z c = +
Using the boundary conditions (see Fig.9.6)
, ,
0
0
l
T T = and , ,
0 l
T d T T = A
we have
2 l
T c = and
1 2 l
T T c d c A = +
which is easily solved to give
, ,
0 l
z
T z T T
d
= A + (9.13)
The corresponding density distribution eq(9.9) thus becomes
, , , , , ,
0 0
1 z T z T = 1
]
(9.14)
with the accompanying hydrostatic pressure is given by the static, stationary version
of (9.11):
, ,
0
0
0
dp
z g
dz
= + (9.15)
Consider a slight perturbation to the system such that
, ,
0 1
T T z T = + , ,
0 1
z = +
, ,
0 1
p p z p = + , ,
1 1 1 1
, , u v w = = u u (9.16)
where the perturbations, distinguished by the subscript 1, are assumed small and
functions of , , , , , x y z t .
Eqs(9.9) thus become
, ,
0 1 0 1
1 T T T 1 + = +
]
which, with the help of (9.14) becomes
1 1
T = (9.17)
Eqs(9.10) is now
1
1 1
0
t


c
+ V =
c
u
1
0 V = u (9.18)
Eq(9.11) is
, , , , , ,
2 1
0 1 0 1 0 1 1
D
p p
Dt
+ = V V + + V +
u
u g
which, with the help of (9.15) becomes
, , , ,
2 1
0 1 1 0 1 1 1
D
p
Dt
+ = V + + V +
u
u g
which in term can be linearized to
2 1
0 1 0 1 1
p
t

c
= V + V +
c
u
u g (9.19)
Finally, eq(9.12) becomes
, , , ,
2
0 1 0 1
D
T T T T
Dt
+ = V +
which, with the help of eqs(9.12a,13), becomes
2 0 1
1 1
dT DT
w T
dz Dt
+ = V
which in term can be linearized to
2 1
1 0 1
'
T
wT T
t

c
+ = V
c
(9.20)
where, according to (9.13)
0
0
'
dT T
T const
dz d
A
= = =
For small density changes, , ,
0
z in eq(9.19) may be approximated by the constant
. Eqs(9.17-20) then become a set of PDEs with constant coefficients.
We now aim to obtain an equation involing
1
w alone.
To begin, we eliminate
1
p in (9.19) by taking the curl of the whole equation:
, , , ,
2
1 1 1
t

c
| |
V V = V = V
|
c
\ .
u g g
Using (9.17) to replace
1
, we have
, ,
2
1 1
T
t

c
| |
V V = V
|
c
\ .
u g
Taking the curl again and using the vector identities
, , , ,
2
V V = V V V a a a
, , , , , , , , , , V = V V + V V a b b a a b a b b a
we have
, , , , , ,
2 2 2
1 1 1
T T
t

c
| |
1
V V = V V V
|
]
c
\ .
u g g (9.20a)
where weve also used (9.18).
Since , , 0, 0, g = g , the z-component of (9.20a) is
, ,
2
2 2 2 1
1 1 2
T
w g g T
t z

1 c c
| |
V V = + V
| (
c c
\ .
]
2 2
1 1
2 2
T T
g
x y

| | c c
= +
|
c c
\ .
(9.21)
Rewriting (9.20) as
2
1 1 0
' T wT
t

c
| |
V =
|
c
\ .
1
T can be eliminated from eq(9.21) by operating on the whole equation with
2
t

c
| |
V
|
c
\ .
, so that
2 2
2 2 2 2
1 1 2 2
w g T
t t x y t

| | c c c c c
| || | | |
V V V = + V
| | | |
c c c c c
\ .\ . \ .
\ .
2 2
0 1 2 2
' gT w
x y

| | c c
= +
|
c c
\ .
(9.22)
Since derivatives of x and y are always in the combination
2 2
2
2 2
x y

c c
V = +
c c
, we see
that
1
w is isotropic in the x-y plane. The independent variables thus fall into 3
groups: t, , , , x y , and z. Aseparable ansatz thus takes the form
, , , , , ,
1
, w W z f x y g t = (9.23)
Now, a successful separation means that we can write
F
F
c
F
=

(9.23a)
where F is any one of the separated functions,
F
c a constant, and an ordinary
differential operator that involves only functions and derivatives of the independent
variables of F.
The complicated form of (9.22) means that the most general form of separation is not
easily achieved. However, the situation will be greatly simplified if the operators
for 2 of the separated functions take the simplest possible forms. Since the
primary changes are in the z direction, we should simplify f and g so that
2 2
1
f a
f

V = (9.24)
1 dg
s
g dt
= (9.24a)
where the peculiar form of the proportionality constant for f is chosen to conform with
Achesons notation, as well as for future convenience.
Note that (9.24a) can be solved immediately to give
, ,
1
s t
g t c e =
One useful property of eq(9.23a) is
1
1 1 F
w
w F c w
F
= = (9.23b)
Let
d
D
dz
= ..
With the help of eqs(9.24, 24a, 23b), we have,
, ,
2 2 2
1 1
w D a w V =
so that eq(9.22) becomes
, ,
2 2 2
1
w
t t

c c
| || |
V V V
| |
c c
\ .\ .
, , , , , ,
2 2 2 2 2 2
1
s D a s D a D a w
1 1
=
] ]
2
0 1
' gT a w =
Cancelling fg from both sides gives
, , , , , ,
2 2 2 2 2 2 2
0
' s D a s D a D a W gT a W
1 1
=
] ]
(9.25)
Eq(9.25) is a 6
th
order ODE. Its general solution thus contains 6 arbitrary constants.
However, since the equation is homogeneous, only 5 of them are independent. (The
6
th
, taken as the overall constant multiplication factor, is immaterial).
Since the boundary conditions always appear in pairs (1 each for the upper & lower
boundary), the integration constants and boundary conditions can never be matched
entirely by themselves. As will be shown later, there are 6 boundary conditions. We
therefore have an eigenvalue problem whereby a solution satisfying all the boundary
conditions exists only for some special values of the parameters in (9.25).
The no-slip condition gives us
1 1 1
0 u v w = = = (BC1)
1 1 1
0
u v w
t t t
c c c
= = =
c c c
(BC2)
1 1 1
0
n n n
n n n
u v w
x x x
c c c
= = =
c c c
(BC3)
1 1 1
0
n n n
n n n
u v w
y y y
c c c
= = =
c c c
(BC4)
for all n, at 0, z d = .
Since the boundaries are kept at fixed temperatures, we have
1 1 1
1
0
n n
n n
T T T
T
t x y
c c c
= = = =
c c c
(BC5)
for all n, at 0, z d = .
We must now translate these conditions into ones on Wso that they can be applied to
(9.25).
(BC1) together with (9.23) gives us
0 W = at 0, z d = . (BCa)
(BC3,4), with 1 n = , and the incompressibility condition (9.18) gives us
0 DW = at 0, z d = . (BCb)
(BC5), with 2 n = , gives
2 2
1 2 2
0 T
x y
| | c c
+ =
|
c c
\ .
on 0, z d =
so that (9.21) becomes
2 2
1
0 w
t

c
| |
V V =
|
c
\ .
on 0, z d =
ie.,
, , , ,
2 2 2 2
0 s D a D a W
1
=
]
at 0, z d =
On expansion, we get
, ,
4 2 2 2
2 0 D a s D a W
1
+ + =
]
at 0, z d =
which, on applying (BCa), becomes
, ,
4 2 2
2 0 D a s D W
1
+ =
]
at 0, z d = (BCc)
Eqs(BCa-c) thus provide 6 boundary conditions to (9.25).
Now, (9.25) is an ODE with constant coefficients.
Its solution is therefore of the form
z
e

which, on substituting into (9.25), gives


, , , , , ,
2 2 2 2 2 2 2
0
' s a s a a gT a
1 1
=
] ]
(9.25a)
For given values of a and s, this is a 3
rd
degree polynomial of
2
.
Let the roots be
2
i
, 1, 2, 3 i = .
The general solution to (9.25) is
, ,
3
1
i i
z z
i i
i
W Ae Be

=
= +
_
As mentioned before, only 5 of the 6 constants ,
i i
A B are independent.
The determination of , , , ,
i i
A B s so that the boundary conditions as well as (9.25a)
are satisfied is a straightforward but tedious task which is not particularly
illuminating.
For illustrative purposes, it may be more fruitful to attack another related problem
where the boundary conditions are simplified to
2 4
0 W D W D W = = = at 0, z d = (BC)
which is easily seen to be satisfied by the ansatz
sin
N z
W
d

= 1, 2, 3, N =
The secular equation for the eigenvalue s is simply (9.25a) with
2
2
N
d

| |
=
|
\ .
.
Setting
2
2 2 2 2
*
N
a a a
d

| |
= = +
|
\ .
(9.25a) becomes
2 2 2 2
* * * 0
' s a s a a gT a 1 1 + + =
] ]
or
, ,
2
2 2 4
* * 0 2
*
' 0
a
s a s a gT
a
+ + + + =
with solution
, , , ,
2
2
2 4 4
* * * 0 2
*
1
4 '
2
a
s a a a gT
a

1
| |
= + + + (
|
( \ .
]
Using
0
'
T
T
d
A
=
we have
, , , ,
2
2
2 4
* * 2
*
1
4
2
T a
s a a g
d a

1
A
= + +
(
(
]
Hence, for 0 T A > , the factor inside the square root is always positive so that s is
always real.
Furthermore, if
, , , ,
2
2 2
4 4
* * 2
*
4
T a
a g a
d a

A
+ > + (9.26a)
we have 0 s
+
> , so that the perturbation grows exponentially with time and the
system is unstable.
Eq(9.26a) can be simplified as
2
4
* 2
*
T a
g a
d a

A
>
3
6 2 2
2 *
2 2 2
1 g T a N
a
d a a d

| | A
> = +
|
\ .
(9.27b)
Now, a is not a parameter related to the system. It is simply some unknown constant
related to the horizontal length scale via eq(9.24). Hence, (9.27b) is satisfied as soon
as the left side is greater than the minimum of the right side with respect to both a and
N. The minimum with respect to N can be found by inspection to be at 1 N = . That for
a satisfies
, ,
3 2
2 2
2 2
3 2 2 2
2 3
2 0 a a a
a d a d
| | | |
+ + + =
| |
\ . \ .
or
2
2 2
2
3 0 a a
d
| |
+ + =
|
\ .
2
2
2
2
a
d

=
so that the instability criterion is
3
2 2 4
2 2 4
2 3 27
2 4
g T d
d d d


| | A
> =
|
\ .
which, in terms of the Rayleigh number
3
g Td

A
= (9.27)
is simply
4
27
4

>
For the boundary conditions (BCa-c), the instability criterion is
1708 >
which corresponds to 3.1/ a d .
9.3.2 Stability to Finite-Amplitude Disturbances
We have just shown that a critical value of T exists, above which disturbances
grow exponentially with time so that even infinitesimal ones may become finite as
t .
A related question is whether a critical value of T also exists, below which any
disturbances die out eventually.
In our idealized system of thermal convection, the answer is yes and the 2 critical
values are equal. Thus, for 1708 , all disturbances subside eventually and the
system will return to the unperturbed state of rest.
9.3.3 Experimental Results
Criteria for the onset of instability as calculated from the linear theory are usually in
good agreement with experiment. However, after the system become unstable, the
predicted exponential growth of the disturbances cannot be substained indefinitely
since non-linear effects will eventually become substantial enough to halt the growth.
The system then reaches a steady state which cannot be described by the linear theory.
With respect to the thermal convection discussed earlier, this means only the critical
value
c
of the Rayleigh number agrees with observation.
Other interesting observations that are beyond the reach of the linear theory include:
1. Convection patterns such as the 2-D rolls and hexagonal cells in the x-y plane
(see fig.9.7).
2. Shifts of the unperturbed steady state convection to other, often time- dependent,
patterns as one continues to increase to values well beyond
c
.
3. Effects due to variation of surface tension in case of a free upper surface.
(Benard instabilities).
9.4 Centrifugal Instability
Consider 2 concentric, rotating, circular cylinders with the gap between them filled
with a viscous fluid.
Let the inner and outer cylinders, together with their associated quantities, be labeled
1 and 2, respectively.
As discussed in 2.4.2, the steady flow
B
u Ar
r

(2.31) (9.28)
with
2 2
1 1 2 2
2 2
1 2
r r
A
r r


2 2
1 2
1 2 2 2
2 1
r r
B
r r

(2.32) (9.29)
is an exact solution to the Navier-Stokes eqs that satisfies the no-slip boundary
condition.
If the cylinders rotates in the same sense, instability occurs if
1 c
, where
c

is some critical value that depends on


2 1 2
, , r r , and . [see eq(9.41)]. The ensuing
flow consists of counter-rotating Taylor vortices superimposed on the main rotary
flow. [see Fig.9.8]
9.4.1 Linear Stability Theory
9.4.2 Inviscid Theory: the Rayleigh Criterion
9.4.3 Experiments
9.4.1 Linear Stability Theory
Since the flow is axisymmetric, all quantities related to the unperturbed, steady
floware functions of r only.
Small quantities describing the perturbation will be distinguished by a prime. They
are assumed to be functions of r, z and t.
For example, we write
j ', ', '
r z
u U u u

= + u (9.30)
0
' p p p = +
The Navier-Stokes eqs. in cylindrical coordinates were derived in 2.4:
, )
2
2
2 2
1 2
r r
r r
u u p u u
u u
t r r r r


c c c
| |
+ V = + V
|
c c c
\ .
u
, )
2
2 2
1 1
2
r r
u u u p u
u u u
t r r r r


c c c
| |
+ V + = + V +
|
c c c
\ .
u (2.22)
, )
2
1
z
z z
u p
u u
t z

c c
+ V = + V
c c
u
where
r z
f u u u f
r r z

c c c
| |
V = + +
|
c c c
\ .
u
2 2
2
2 2 2
f r f
r r r r z
( c c c c
| |
V = + +
| (
c c c c
\ .

The incompressibility condition is:
, ) 0
z
r
u u
ru
r r r z

c c c
V = + + =
c c c
u
The unperturbed steady flow satisfies
, )
0
U r

= u e , )
0 0
p p r =
2
0
1 U dp
r dr

=
2
1
0
d dU
r U
rdr dr r

| |
=
|
\ .
(2.30)
Note that
0
U
r

c
V =
c
u
so that
0 0 0
0
r z
u u u

V = V = V = u u u
Using (2.30) to cancel out the unperturbed parts, eq(2.22) become
, ) , )
2 2
2
' 1 1 ' '
' ' 2 ' ' '
r r
r r
u p u
u U u u u
t r r r

c c
| |
+ V + = + V
|
c c
\ .
u

, )
, )
2
2
' ' ' 1
' ' ' ' '
r
r
u U u u dU
u u u u
t dr r r

+ c
| |
+ + V + = V
|
c
\ .
u

, )
2
' 1 '
' ' '
z
z z
u p
u u
t z

c c
+ V = + V
c c
u

and
, )
'
' 0
z
r
u
ru
r r z
c c
V = + =
c c
u
where
2 2
2
2 2
r z
c c
V = +
c c

The linearized version is


2
2
' 2 ' 1 ' '
'
r r
r
u U u p u
u
t r r r

c c
| |
= + V
|
c c
\ .

2
2
' 1
' ' '
r
u dU U
u u u
t dr r r

c
| | | |
+ + = V
| |
c
\ . \ .

2
' 1 '
'
z
z
u p
u
t z

c c
= + V
c c

(9.31)
and
, )
'
' 0
z
r
u
ru
r r z
c c
V = + =
c c
u
To simplify the mathematics, we assume the gap between the cylinder to be small, ie.,
2 1 1
d r r r =
Now
2
2
'
'
r
r
u
u O
d
| |
V =
|
\ .
2 2
1
' '
r r
u u
O
r r
| |
=
|
\ .
so that
2
2
'
'
r
r
u
u
r
V
and similarly for ' u

.
Eq(9.31) thus simplify to
2
2 ' 1 '
'
r
U u p
u
t r r

c c
| |
V =
|
c c
\ .

(9.33a)
2
' ' 0
r
dU U
u u
t dr r

c
| | | |
V + + =
| |
c
\ . \ .

(9.33b)
2
1 '
'
z
p
u
t z

c c
| |
V =
|
c c
\ .

(9.33c)
Using
1
' ' ' '
r r r r
u u u u
O O
r d r r
| | c
| |
= =
| |
c
\ .
\ .

the incompressibility condition becomes


' '
0
r z
u u
r z
c c
+ =
c c
(9.33d)
Finally, from (9.28), we see that
2 2
2
dU U B B
A A A
dr r r r

+ = + + =
so that (9.33b) is simplified to
2
' 2 ' 0
r
u Au
t

c
| |
V + =
|
c
\ .

(9.33b)
Eliminating ' p between (9.33a,c) gives
2
' ' 2 '
0
r z
u u U u
t z r r z

c c c c
| || |
V =
| |
c c c c
\ .\ .

Further eliminating '


z
u using (9.33d), we have
2 2 2
2
2 2 2
' ' 2 '
0
r r
u u U u
t z r r z

| | c c c c
| |
V + =
| |
c c c c
\ .
\ .

or
2
2 2
2
2 '
' 0
r
U u
u
t r z

c c
| |
V V =
|
c c
\ .

(9.34)
Still further eliminating ' u

using (9.33b), we have


2
2
2 2
2
4 '
' 0
r
r
AU u
u
t r z

c c
| |
V V + =
|
c c
\ .

(9.38a)
which is a 6
th
order linear PDE for '
r
u alone.
Now, the coefficients of (9.38a) are all constants except for the term
U
r

.
Thus, for
1 2
O O , we can write
, )
1 2
1
2
U
r

O = O + O
so that all the coefficients in (9.38a) are constants. In which case, we have
2
2
2 2
2
'
' 4 0
r
r
u
u A
t z

c c
| |
V V + O =
|
c c
\ .

(9.38b)
We must now establish and then satisfy the relevant boundary conditions.
No-slip condition obviously requires
' ' ' 0
r z
u u u

= = = (BC1)
' '
0
n n
r r
n n
u u
z
c c
= =
c c
(BC2)
' '
0
n n
n n
u u
z

c c
= =
c c
' '
0
n n
z z
n n
u u
z
c c
= =
c c
for all n on the surfaces
1 2
, r r r = .
That is, only partials with respect to r can be finite.
The incompressibility condition then implies
'
0
r
u
r
c
=
c
on
1 2
, r r r = (BC3)
From (9.34), we have
2 2
' 0
r
u
t

c
| |
V V =
|
c
\ .

on
1 2
, r r r = (BC4)
Since
2 2
2
2 2
r z
c c
V = +
c c

we have
2 2 2 2 4 4 4
2 2
2 2 2 2 4 2 2 4
2
r z r z r z r z
| || | c c c c c c c
V V = + + = + +
| |
c c c c c c c c
\ .\ .

so that (BC4) becomes
2 4 4
2 4 2 2
2 ' 0
r
u
t r r r z

(
| | c c c c
+ =
( |
c c c c c
\ .

on
1 2
, r r r = (BC4a)
Our task is to find a solution to (9.38b) that satisfy (BC1,2,3,4a).
Eq(9.38b) is obviously separable.
Since the partials with respect to z are all of even order, a periodic solution can be
supported. Thus, we consider the ansatz
, ) ' cos
s t
r r
u u r e nz =

(9.35)
so that
'
'
r
r
u
su
t
c
=
c
2
2
2
'
'
r
r
u
n u
z
c
=
c
2
2
2
'
cos
s t r
r
u
e nz D u
r
c
=
c

where
d
D
dr
= .
, )
2 2 2
' cos
s t
r r
u e nz D n u V =

Thus, (9.38b) becomes


, ) , )
2
2 2 2 2 2
4 0
r r
s D n D n u A n u
(
O =


(9.38)
while the boundary conditions at
1 2
, r r r = simplify to
(BC1,2): 0
r
u =

(BC5)
(BC3): 0
r
Du =

(BC6)
(BC4a):
, )
2 4 2 2
2 0
r
sD D n D u
(
=


ie.,
4 2 2
2 0
r
s
D n D u

(
| |
+ =
|
(
\ .


(BC7)
Since (9.38) is homogeneous and of even order, there is an odd number (=5) of
independent integration constants. The number of boundary conditions are even (=6).
Therefore, it is an eigenvalue problem for which a parameter, eg. s, can take on only
special values.
In general, s can be complex.
Let
R
s s i = + , where both
R
s and are real.
The time dependence of '
r
u is then
R
s t i t
e e

.
Instability occurs if 0
R
s > .
The threshold to instability is therefore at 0
R
s = .
For non-oscillatory evolution, 0 = , so that the threshold is at 0 s = .
The equations decribing this marginal state are
(9.38):
, )
3
2 2 2 2
4 0
r r
D n u A n u O =

(9.38a)
subject to boundary conditions at
1 2
, r r r = :
(BC5): 0
r
u =

(BC6): 0
r
Du =

(BC7):
, )
4 2 2
2 0
r
D n D u =

(BC7a)
Shifting to a more natural coordinate defined by
1 1
2 1
r r r r
x
r r d

= =

we have
1 dx d d
D
dr dx d dx
= =
Eq(9.38a) thus becomes
3
2
2 2 2
2 2
1
4 0
r r
d
n u A n u
d dx

| |
O =
|
\ .

which can be rearranged to give
, )
3
2 2 6
2
2 2
4
0
r r
d A n d
nd u u
dx
| | O
=
|
\ .

On setting
a nd =
4
2
4A d
T

O
=
it becomes
3
2
2 2
2
0
r r
d
a u Ta u
dx
| |
=
|
\ .

(9.38b)
subject to boundary conditions at 0,1 x = :
(BC5): 0
r
u =

(BC6): 0
r
du
dx
=

(BC7a):
4 2
2
4 2
2 0
r
d d
a u
dx dx
| |
=
|
\ .

(9.40)
Using,
2 2
1 1 2 2
2 2
1 2
r r
A
r r
O O
=

(2.32)
the Taylor number becomes
3 2 2 3 2 2
1 1 2 2 1 1 2 2
2 2
1 2 1
4 2 d r r d r r
T
r r r
O O O O O O
=
+
(9.41)
where the last equality made use of the narrow gap appoximation
1 2
r r .
For a given a, eq(9.38b) and the boundary conditions (9.40) constitute an eigen
problem with eigenvalues , ) T a . Let
, ) , ) min
l
T a T a = for given a.
The threshold of instability corresponds to the minimum of
l
T ranging over all a.
Remarkably, eq(9.38b) & (9.40) take the same form as the thermal instability eqs
(9.25-6) with 0 s = . Hence, the threshold for centrifugal instability is also
1708 T > (9.42)
Likewise, the critical value of n is approximately
3.1
d
.
The streamlines of the secondary flow are shown in Fig.9.8.
Since the period of the flow in the z direction is
2
n

, the height of each cell is


3.1
d
H
n

=
Although the final steps of our derivation invoke the assumption
1 2
O O , eqs(9.41-2)
turns out to quite adequate whenever
1 2
, O O are of the same sign.
Note that for
1 2
, 0 O O > , eq(9.41-2) implies the necessary condition for instability is
that
2
r O decreases sufficiently quickly with r.
9.4.2 Inviscid Theory: the Rayleigh Criterion
9.4.3 Experiments
9.5 Instability of Parallel Shear Flow
9.5.1 The Inviscid Theory
9.5.2 The Viscous Theory
9.5.3 Experimental Results
9.5.1 The Inviscid Theory
Consider the 2-D, inviscid, incompressible, flow between 2 flat plates at y L = .
(see Fig.9.10).
The basic eqs are the Euler eqs
1 u u u p
u v
t x y x
c c c c
+ + =
c c c c
(E1)
1 v v v p
u v
t x y y
c c c c
+ + =
c c c c
(E2)
and the incompressibility condition
0
u v
x y
c c
+ =
c c
(C1)
Consider the parallel shear flow
, )
0
, 0 U y = 1
]
u (9.46)
where U is an arbitrary function of y.
Substituting (9.46) into the basic eqs, we have
(E1):
1
0
p
x
c
=
c
(E2):
1
0
p
y
c
=
c
(C1): 0 0 =
Hence,
0
u is a solution to the basic eqs with
0
p p const = = .
Consider a general and presumably small 2-D disturbance.
All quantities related to the disturbance will be distinguished by the subscript 1.
They are in general functions of , ) , , x y t .
For example,
, )
0 1 1 1
' , U y u v = + = + 1
]
u u u
0 1
p p p = +
The basic eqs becomes
, )
1 1 1 1
1 1
1 u u dU u p
U u v
t x dy y x
| | c c c c
+ + + + =
|
c c c c
\ .
(E3)
, )
1 1 1 1
1 1
1 v v v p
U u v
t x y y
c c c c
+ + + =
c c c c
(E4)
1 1
0
u v
x y
c c
+ =
c c
(C2)
of which the 1
st
two can be linearized to
1 1 1
1
1
'
u u p
U U v
t x x
c c c
+ + =
c c c
(E5)
1 1 1
1 v v p
U
t x y
c c c
+ =
c c c
(E6)
where a prime denotes derivative with respect to y.
Since all coefficients in these eqs are independent of t and x, the natural ansatz is
, ) , )
, )
1
, ,
i k x t
f x y t f y e

=

(9.47)
where f can be u, v, or p. Thus
(E5): , )
1
' i kU u U v ikp

+ =

(E7)
(E6): , )
1
' i kU v p

=

(E8)
(C2): ' 0 iku v + =

(C3)
Eliminating p

from (E7,8) gives


, ) , ) ' ' '' ' ' ikU u i kU u U v U v k kU v + + =

Further eliminating of u

through (C3) gives


, ) , )
1
' ' '' '' ' ' U v kU v U v U v k kU v
k
+ + + =

which can be simplified to
, ) , )
1
'' '' 0 kU v U k kU v
k
+ = 1
]

or
2
''
'' 0
kU
v k v
kU
| |
+ =
|

\ .

(9.48)
Boundary conditions to (9.48) are the impenetrable wall condition
0 v =

at y L = (9.49)
Eq(9.48) is therefore an eigenvalue problem for .
The following trick is due to Lord Rayleigh (1880).
Let the complex conjugate of v

be denoted by
*
v

.
, )
*
9.48
L
L
dyv

}

gives
2
* 2
''
'' 0
L L
L L
kU
v v dy v k dy
kU

| |
+ =
|

\ .
} }

(9.50)
Now, integration by part gives, with the help of (9.49),
2
* * *
'' ' ' ' '
L L L
L
L
L L L
v v dy v v v v dy v dy


1 = =
] } } }

so that (9.50) becomes
2 2
2
''
' 0
L L
L L
kU
v dy v k dy
kU

| |
+ =
|

\ .
} }

(9.50a)
Writing
R I
i = +
the imaginary part of (9.50a) is
2
2
''
0
L
I
L
U
k v dy
kU

}

(9.51)
For 0
I
= , this means
2
2
''
0
L
L
U
v dy
kU

}

which is possible only if '' U changes sign somewhere in the interval.
Since instability results from exponential growth with time, which in turn requires
0
I
> , this gives us the Rayleighs inflection point theorem, which says:
A necessary condition for the linear instability of an inviscid shear flow , ) U y is
that , ) '' U y changes sign somewhere in the flow.
9.5.2 The Viscous Theory
Consider now the case where the fluid is viscous.
The discussion in 9.5.1 is modified as follows.
The basic eqs are now the Navier-Stokes eqs instead of the Euler eqs.
2 2
2 2
1 u u u p u u
u v
t x y x x y

| | c c c c c c
+ + = + +
|
c c c c c c
\ .
(NS1)
2 2
2 2
1 v v v p v v
u v
t x y y x y

| | c c c c c c
+ + = + +
|
c c c c c c
\ .
(NS2)
and the incompressibility condition
0
u v
x y
c c
+ =
c c
(C1)
With the same ansatz as (9.47), the counterpart of eqs(E7,8) are
, ) , )
2
1
' '' i kU u U v ikp k u u

+ = + +

(NS7)
, ) , )
2
1
' '' i kU v p k v v

= + +

(NS8)
while (C3) remains unchanged:
' 0 iku v + =

(C3)
, ) 7 ' NS gives
, ) , )
2
1
' ' '' ' ' ' ' ''' ikU u i kU u U v U v ikp k u u

+ + = + +

(NS7a)
, ) 8 ik NS gives
, ) , )
2
1
' '' k kU v ikp ik k v v

= + +

(NS8a)
, ) , ) 7 8 NS a NS a gives
, ) , ) ' ' '' ' ' ikU u i kU u U v U v k kU v + +

, )
2 2
' ''' '' k u u ik k v v
(
= + +


Eliminating u

through (C3) gives


, ) , )
1
' ' '' '' ' ' U v kU v U v U v k kU v
k
+ + +

, )
2
'' '''' ''
i
ikv v ik k v v
k

(
= + +
(


which can be simplified to
, ) , )
1
'' '' kU v U k kU v
k
+ (


2 4
'''' 2 ''
i
v k v k v
k
( = +


(NS9)
Introduce the stream function
, )
, ) i k x t
y e



=

so that
1
u
y
c
=
c
1
v
x
c
=
c
and
' u =

v ik =

Eq(NS9) becomes
, ) , ) '' '' i kU ik U k kU (


2 4
'''' 2 '' k k ( = +


which can be rearranged to give
, ) , )
2 4
'''' 2 '' '' '' i k k kU k U k kU ( + = + (


, ) , )
2
'' '' kU k kU = +

(9.53)
In terms of , the no-slip boundary conditions are
' 0 = =

at y L = (9.54)
We leave it as an exercise to show that for a plane Poiseuille flow defined as
, )
2
max 2
1
y
U y U
L
| |
=
|
\ .
(9.55) (see Ex.2.3)
solutions of (9.53-4) leads to a curve of marginal stability shown in fig.9.11.
Of particular interest is the fact that instability occurs for a band of wavenumbers k
(slashed region in fig.9.11) if
max
5772
U L
R

= > (9.56)
In contrast, the flow (9.55) is stable for all wavenumbers k in an inviscid flow.
Thus, viscosity plays a dual role.
According to (9.56), it is stabilizing since it raises the value of threshold value of
max
U to reach instability.
On the other hand, it is de-stabilizing because the flow would have been stable if
viscosity is absent altogether.
9.5.3 Experimental Results
Criterion (9.56) for the Poiseuille flow has been confirmed experimentally.
However, it applies only to systems with extremely low level of background
turbulence. Furthermore, non-linear effects are significant so that the criterion applies
only when the amplitudes of the disturbances are sufficiently low.
9.6 General Theorem on Stability of Viscous
Flow
Theorem:
Consider an incompressible viscous fluid occupying region , ) V t that is enclosed
within a sphere of diameter L.
Let , ) , t u x be a bounded solution to the Navier-Stokes eqs in , ) V t satisfying the
boundary condition , ) ,
B
t = u u x on the boundary , ) S t of , ) V t . Thus, there exists
M
u such that
, ) ,
M
u t > u x for all , ) V t e x at all t.
Let , )
*
, t u x be another solution satisfying the same boundary condition but different
initial conditions at 0 t = .
The difference flow
*
= v u u (9.59)
thus satisfies
0 = v on , ) S t (9.60)
The kinetic energy E of v defined as
, )
2
1
2
V t
E dV =
}
v (9.61)
satisfies
2 2
2
0 2
3
exp
M
t
E E u
L

1 | |
s
( |
\ . ]
(9.57)
where
0
E is its initial value.
Thus if
3
M
u L
R

= < (9.58)
then
0 E as t
and the flow is unstable.
Before the actual proof of the theorem, we 1
st
establish the following relation:
, )
2
1
i i
j i
j j V t
dE v v
v u dV
dt x x

1
| |
c c
(
=
|
|
c c
(
\ .
]
}
(9.62)
9.6.1 Proof of Eq(9.62)
9.6.2 Proof of the Theorem
9.6.1 Proof of Eq(9.62)
Both u and
*
u are solutions of the Navier-Stokes eqs., therefore
, )
2
1
p
t

c
+ V = V + V
c
u
u u u
, )
2 *
* * * *
1
p
t

c
+ V = V + V
c
u
u u u
Subtracting, we have
, ) , ) , )
2
* * *
1
p p
t

c
+ V V = V + V
c
v
u u u u v
Now,
, ) , ) , ) , ) , ) , )
* * *
V = + V + = V + V + V 1
]
u u u v u v u u v u u v
so that by setting
, )
*
1
P p p

=
we have
, ) , )
2
*
P
t

c
+ V + V = V + V
c
v
v u u v v
or, in index notations,
2
*
i i i i
j j
j j i j j
v u v P v
v u
t x x x x x

c c c c c
+ + = +
c c c c c c
Multiplying by
i
v and sum over i gives
2
2 2
*
1 1
2 2
i i
i j j i i
j j i j j
u P v
v v u v v
t x x x x x

c c c c c
| | | |
+ + = +
| |
c c c c c c
\ . \ .
v v (A)
We now try to put as many as possible the spatial derivatives in divergence form.
Thus
, )
j j
i
i j i i j i j i i
j j j j
u v
v
v v u v v u v u v
x x x x
c c
c c
= + +
c c c c
j
i
i j i j
j j
u
v
v v u v
x x
c
c
= +
c c
where 0
j
j
v
x
c
= V =
c
v
so that
, )
j
i
i j i j i i j
j j j
u
v
v v v v u u v
x x x
c
c c
=
c c c
(B)
Similarly
, )
2 2
* 2 2
* * *
j
j j j
j j j j
u
u u u
x x x x
c
c c c
= + =
c c c c
v v
v v (C)
where
*
*
0
j
j
u
x
c
= V =
c
v
, )
i
i i i
i i i i
P v P
v P v P v
x x x x
c c c c
= + =
c c c c
(D)
i i i i
i i
j j j j j j
v v v v
v v
x x x x x x
| |
c c c c c
= +
|
|
c c c c c c
\ .
so that
i i i i
i i
j j j j j j
v v v v
v v
x x x x x x
| |
c c c c c
=
|
|
c c c c c c
\ .
(E)
Puuting (C-E) into (A), we have
2 2
*
1 1
2 2
i i i i
i j i j i i i j
j j j j j
v v v v
v v u u v P v u v
t x x x x x

| |
c c c c c c
| |
= + + +
|
|
|
c c c c c c
\ .
\ .
v v
Upon integrating with
, ) V t
dV
}
the divergence term vanishes since
, )
2
*
1
2
i
i j i j i i
j j V t
v
v v u u v P v dV
x x

| |
c c
+ +
|
|
c c
\ .
}
v
, )
2
*
1
2
i
i j i j i i i
j S t
v
v v u u v P v n dS
x

| |
c
= + +
|
|
c
\ .
}
v

, )
*
1
2
i
j i i j i j
j S t
v
v u v u P v n dS
x

| |
c
= + +
|
|
c
\ .
}
0 = since 0 = v on , ) S t
Hence
, ) , )
2
1
2
i i i
i j
j j j V t V t
v v v
dV u v dV
t x x x

| |
c c c c
| |
=
|
|
|
c c c c
\ .
\ .
} }
v
The convective (Reynolds) theorem gives
, ) , ) , )
2 2 2
1 1 1
2 2 2
V t V t S t
d
dV dV dS
dt t
c
| |
= +
|
c
\ .
} } }
v v v u n

, )
2
1
2
V t
dV
t
c
| |
=
|
c
\ .
}
v since 0 = v on , ) S t
Thus
, ) , )
2
1
2
i i i
i j
j j j V t V t
d v v v
dV u v dV
dt x x x

| |
c c c
=
|
|
c c c
\ .
} }
v
which is (9.62).
9.6.2 Proof of the Theorem
From
2
0
i
ij
j
v
A
x
| |
c
+ >
|
|
c
\ .
for any
ij
A
we have
2
2
2 0
i i
ij ij
j j
v v
A A
x x
| |
c c
+ + >
|
|
c c
\ .
(9.63)
which holds even when, as will be assumed hereafter, summation over repeated
indices is implied.
Choosing
i j
ij
u v
A

= , we have
2
2
2 0
i j i j
i i
j j
u v u v
v v
x x
| |
| | c c
+ >
|
|
|
c c
\ .
\ .
or
, ,
2
2
1
2 2
i i
i j i j
j j
v v
u v u v
x x

| |
c c
> +
|
|
c c
\ .
On substituting into (9.62), we have
, ,
, ,
2
2
2
2
i
i j
j V t
dE v
u v dV
dt x

(
| |
c
(
s
|
|
c
(
\ .

}
Using
, ,
, , , ,
2
2 2 2
2 2
i j M j M
V t V t
u v dV u v dV u E

s =
} }
we have
, ,
2
2 2
1
2
i
M
j V t
dE v
u E dV
dt x

(
| |
c
(
s
|
|
c
(
\ .

}
(9.64)
On the other hand, setting
ij j i
A h v = in (9.63) gives
, ,
2
2
2
i i
j i j i
j j
v v
h v h v
x x
| |
c c
>
|
|
c c
\ .
which, upon using
, ,
2 2
2
j
i
j i j i i
j j j
h
v
h v h v v
x x x
c
c c
= +
c c c
becomes
, , , ,
2
2
2 2 j
i
j i i j i
j j j
h
v
h v v h v
x x x
| | c
c c
> +
|
|
c c c
\ .
On integrating over , , V t , the divergence term vanishes since 0 = v on its boundary
, , S t . Hence
, ,
, , , ,
2
2
2 j
i
i j i
j j V t V t
h
v
dV v h v dV
x x
| | ( c
c
>
| (
|
c c
(
\ .
} }
, ,
, ,
2 2
V t
dV = V
}
h h v (9.64a)
Now, (9.64a) holds for arbitrary h.
If we manage to find an h such that
2
0 C V > > h h (9.65)
for all x in , , V t for all t, where C is some real positive constant, (9.64a) becomes
, , , ,
2
2
2
i
j V t V t
v C
dV C dV E
x
| |
c
> =
|
|
c
\ .
} }
v
which, when substituted into (9.64), gives
, ,
2 2
1
M
dE
u C E
dt

s (9.66)
Integrating, we have
, ,
2 2
0
exp
M
t
E E u C

(
s
(

which is our theorem eq(9.57) with
2
2
3
C
L

= (9.66a)
Our task is therefore 2-fold.
1
st
, we must show that such an h do exist and find the maximum C it affords.
2
nd
, we need to find the optimal h that can maximize C among all hs.
At this point, the dimension of the system enters into consideration.
As stated in the theorem, we shall assume , , V t always lies inside a sphere, center at
the origin, of diameter L. Hence, all points of the system satisfy L s x .
The easiest construction is perhaps of the type
, ,
r
h r = h e
so that
, ,
2 2 2
2
1 d
r h h
r dr
V = h h
Setting , , h r ar = , where a is a constant, we have
2 2 2
3 f a a r = V = h h where
2
L
r s .
Since
2
2 0
f
a r
r
c
= s
c
for all r,
f is a monotonically decreasing function of r.
Its minimum is therefore at
2
L
r = where
2 2
3
4
a L
f a =
The value of a that maximize this is given by
2
1
3 0
2
aL =
ie.,
2
6
a
L
=
so that
, ,
2
6
h r r
L
=
2 2
2 2
6 6
3 f r
L L
| |
= V =
|
\ .
h h
and
2 2
2
6 6 9
3
4
L
r
C f
L L
=
| |
= = =
|
\ .
which is only slightly less than the value (9.66a) given in the theorem.
Eq(9.66a) is achieved with
, , tan
r
h r
L L

=
so that
2
2
sec 0
dh r
dr L L

| |
= >
|
\ .
for
2
L
r s
Thus, h is monotonically increasing with minimum at 0 r = .
Furthermore
2
2
dh h
f h
dr r
= +
2 2
2 2
2
sec tan tan
r r r
L L r L L L L

| | | |
= +
| |
\ . \ .
2
2
tan
r
L r L L

| |
= +
|
\ .
2
2
1 tan
L r
L r L

| | | |
= +
| |
\ . \ .
2
2
1 tan x
L x

| | | |
= +
| |
\ . \ .
where
r
x
L

=
The minimum of f is given by
2
2
2 2
tan sec 0 x x
x x
+ =
which implies
sin cos 0 x x x + =
ie.,
2 sin2 x x =
whose solution is
0 x = or 0 r =
At which point, (9.66a) is achieved and the theorem is proved.
9.7 Uniqueness & Non-Uniqueness Of Steady
Viscous Flow
Theorem
Consider a fixed region V of fluid that is enclosed in a sphere of diameter L.
Let u and
*
u be 2 steady solutions of the Navier-Stokes eqs in V with the same value
, )
B
u x on the boundary of V.
Let
M
u be an upper bound to u in V.
If
3
M
u L
R

= < (9.68)
the 2 flows are identical, ie.,
*
= u u
In other words, bounded steady viscous flow with 5 R < is unique.
Proof
The proposition of this theorem is just the steady state version of the theorem in 9.6.
Thus, the t limit of (9.57) is satisfied, ie.,
2 2
2
0 2
3
lim exp 0
M
t
t
E E u
L

1 | |
s =
( |
\ . ]
where (9.68) ensured the coefficient of t is negative. QED.
9.7.1 An Example of Non-Uniqueness of Steady
Flow
9.7.2 Hysteresis
9.7.1 An Example of Non-Uniqueness of Steady
Flow
We now consider some variants to the Taylor experiment of 9.4 involving viscous
fluid occupies the gap
1 2
r r r between 2 coaxial cylinders.
The differences we introduce are:
1.
2
0 .
2. Fluids are also bounded by stationary plane walls at 0, z L with L adjustable.
(see fig.9.13).
The system is characterized by 3 dimensionless numbers:
1. Radius ratio
1
2
r
r
.
2. Reynolds number
1 1
rd
R

where
2 1
d r r .
3. Aspect ratio
L
d
.
Benjamin & Mullin (1982) had performed measurements on such an apparatus with
1
2
0.6
r
r
12.61 359 R (9.71)
The salient points were:
1. Some 20 different stable steady flows were observed.
2. On theoretical grounds, they inferred the existence of 19 other steady flows that
were unstable and hence unobservable.
3. All observed flows were of an axisymmetric cellular nature as shown in fig.9.8.
4. Different flows are distinguished by different number of cells and/or different
sense of rotation within each cell.
5. Appearance of a particular flow pattern depends on the way the boundary
conditions (9.71) were achieved from an initial state of rest.
6. If 359 R were achieved by small steps from 0, the same flow consisting 12
cells was always observed.
For more details, see
T.B.Benjamin, T.Mullin, J.Fluid Mech 121, 219-30 (1982).
9.7.2 Hysteresis
Measurements on very short cylinders, with 4 , indicated that transitions between
2-cell & 4-cell modes were decribed by a state diagram shown in fig.9.14.
Such diagrams can be studied using catastrophic theory, a good reference of which is
J.M.T.Thompson, Instabilities & Catastrophes in Science & Engineering,
Wiley (1982).
Folds in the surface implies non-uniqueness of solutions in parts of the R plane;
hence hysteresis.
The middle sheet of the fold corresponds to unstable solutions which are not
observable.
9.8 Instability, Chaos, And Turbulence
Consider again the Taylor vortex apparatus with
1 2
and a temperature
difference
2 1
0 T T T maintained between the cylinders.
This may serve as a crude model for the atmosphere as a rotating fluid under
differential heating.
Ref: R.Hide, Q.J.R.Met.Soc. 103,1-28 (1977).
If is sufficiently small, a weak differential rotation is observed (see Fig.9.16a).
As is increased in small steps past a critical value
C
that depends on T ,
baroclinic instabilities set in which amplify non-axisymmetric waves.
Further increasing increases the amplitude of these waves leading to a meandering
jet structure reminiscent of atmospheric jet streams (see Fig.9.16b).
Amplitude, shape, and wavenumber of this jet can be either steady or oscillatory.
At still higher values of , complicated, aperiodic fluctuations set in and the system
become chaotic (see Fig.16c).
9.9 Instability At Very Low Reynolds Number
The theorem of 9.6 guarantees stability for flows of sufficiently low R.
However, it applies only if u is prescribed on some, possibly varying, boundary.
For flows with free boundaries, instabilities are possible for arbitrarily small R.
Viscous Fingering in Hele-Shaw Cell.
A Hele-Shaw cell (see 7.7) is formed by pressing 2 flat sheets of transparent plastics
together with syrup filling the small gap (~2mm) in between.
A hole is drilled on the top sheet for the insertion of a syringe.
Air is then injected by the syringe.
In principle, one might expect the air to displace the syrup in a symmetrical manner,
so that the air-syrup interface is circular.
However, such an interface is found to be unstable. Ripples soon form and develop
into fingers as shown in Fig.9.20.
Such behavior is observed whenever a more viscous fluid is displaced by a less
viscous one. It is called the Saffman- Taylor instability.
Buckling of Viscous Jets
A falling jet of viscous fluid is approximately symmetrical about a vertical axis if the
height H is below a critical value
C
H . (see Fig.9.21a).
If
C
H H , the jet becomes unstable and buckles (see Fig.9.21b).
What makes it interesting is that, in contrast with the other instabilities discussed so
far, it occurs only if R is less than some critical value.

Das könnte Ihnen auch gefallen