You are on page 1of 71

Slow Viscous Flows

John Lister
(Zachary Ulissi)
Michaelmas Term, 2009-2010
Contents
1 Summary of Fluid Mechanics 3
1.1 Mass Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Stress Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Stress Tensor Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 The Momentum Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Newtonian Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.1 Kinematic boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.2 Rigid boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.3 Dynamic boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7 The Reynolds Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7.1 Notes about the Reynolds Number . . . . . . . . . . . . . . . . . . . . . . 7
2 Stokes Equations 8
2.1 Simple Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Instantaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Linear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Reversibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.4 Forces and Torque Balance . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.5 Work Balances Dissipation (since there is no KE) . . . . . . . . . . . . . . 9
2.2 Three theorems based on dissipation integrals . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Example of the dissipation theorem . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Example with rigid particles . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.3 Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Representation by potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Complex Variables in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Papkovich-Neuber Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.3 Notes on the Papkovich-Neuber Solutions . . . . . . . . . . . . . . . . . . 12
2.4 Solutions for points, spheres, and cylinders . . . . . . . . . . . . . . . . . . . . . 13
2.4.1 Pseudo-tensors vs true tensors . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.2 Solution due to a point force . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.3 Fundamental Solutions of Stokes Flow . . . . . . . . . . . . . . . . . . . . 14
2.4.4 Point Source Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.5 Force Dipoles, Stresslets, Rotlets, Etc. . . . . . . . . . . . . . . . . . . . . 15
2.4.6 Example: Rigid Sphere in Translation . . . . . . . . . . . . . . . . . . . . 16
2.4.7 Simple Derivation of Force/Stress on a Translating Sphere . . . . . . . . . 17
2.4.8 Gravitational Settling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.9 2D singularities and Flow Past a Cylinder . . . . . . . . . . . . . . . . . . 17
2.4.10 Uniform Flow Past a Stationary Cylinder . . . . . . . . . . . . . . . . . . 18
2.5 The Motion of Rigid Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5.1 The Resistance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.2 The Faxen Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1
3 Applications 21
3.1 Marangoni Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.2 Thermophoresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.3 Surfactant Eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Deformation of Drops (Rallison 1984, Stow 1994, Ann. Rev. Fluid. Mech.) . . . 25
3.2.1 Small deformations (Ca 1) in uniform streams (U

=Ex . . . . . . . . 25
3.2.2 Larger Deformations in Pure Strain . . . . . . . . . . . . . . . . . . . . . 26
3.2.3 Larger Deformations for 1 (Taylor 1964) . . . . . . . . . . . . . . . . 26
3.3 The Rayleigh Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Neglect of inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4 Long Thin Flows and Immersed Bodies 29
4.1 Long thin ows I: Lubrication theory . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Lubrication Theory for Pigs and Slugs . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Long Thin Flows II: Extensional Flows . . . . . . . . . . . . . . . . . . . . . . . . 39
4.4 Long Thin Flows III: The Two Fluid Case . . . . . . . . . . . . . . . . . . . . . . 40
4.5 Slender Body Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.6 Elastohydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.6.1 The Beam equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6.2 Example: Thin Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.6.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.6.4 Example: Compression of a rod of length l, radiusr . . . . . . . . . . . . . 47
4.6.5 Winkler Elastic Foundation Model . . . . . . . . . . . . . . . . . . . . . . 49
4.6.6 Example: Falling Cylinder Onto a Winkler Substrate (Skotheim+Mahadevan
PRL, 2004) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5 Handouts 52
5.1 Summary of Fluid Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.2 Integral Representations of Stokes Flow . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 2D Singularities and Flow Past a Cylinder . . . . . . . . . . . . . . . . . . . . . . 62
5.4 The Greens Function for Stokes Flow . . . . . . . . . . . . . . . . . . . . . . . . 64
5.5 Two Solutions in Lubrication Theory . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.6 Extensional Flow Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.7 Small Deformations of a Viscous Drop by Uniform Pure Strain E . . . . . . . . . 69
5.8 Example: Swimming Flagella . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2
Chapter 1
Summary of Fluid Mechanics
The continuum assumption deals with innitesimally small uid particles that have a well
dened density (x, t), pressure p(x, t), and velocity u(x, t), where x(t) is the position of the
particle. From the Eulerian perspective (static frame of reference), a time derivative will be /t,
but in a frame of reference following a uid parcel the derivative will be D/Dt = /t +u .
1.1 Mass Conservation
In general, Eulers equation is:

t
+ (u) = 0
D
Dt
+ u = 0
For incompressible uids (everything in this course):
D
Dt
= 0 u = 0
1.2 Stress Tensor
Stress is a force per unit area across a surface. For example, a surface force balance on an
innitesimal tetrahedron shows is linearly related to surface normal n by:
= n
Where is the stress tensor and is the stress by the outside acting on the inside of a surface
with outward normal n. An angular momentum balance, for most materials (except, for example,
magnetic ones), yields:

ij
=
ji
1.2.1 Stress Tensor Sheet
In the continuum approximation, the forces on a small of uid at (x, t) can be divided into those
acting on and proportional to the volume, like gravity, and those acting on and proportional to
the surface area, like pressure or friction due to relative motion. (In reality, surface forces are
exerted over a thickness of a few intermolecular distances so, for example, molecules in relative
motion travel through a mean-free path before depositing their momentum in a collision). Since
the volume forces on the small blob and its inertia are both proportional to its volume V , whereas
the surface forces on the blob are of order V
2/3
, the surface forces must balance by themselves
in the limit V 0. This observation can be used as follows to show that the stress (x, t; n)
(force per unit area) acting on a surface with normal n is linearly related to n by a second-rank
tensor .
Consider a small tetrahedron of uid aligned with local, rectangular coordinate axes e
(1)
, e
(2)
, e
(3)
.
Denote the forces exerted by the exterior of the tetrahedron on the surfaces of the tetrahedron by
F
(k)
acting on the three faces with outward normals in the three negative coordinate directions
e
(k)
and F acting on the sloping face, which has outward normal n and area A.
3
Since the surface forces must balance by themselves in the limit as V 0, we obtain:
F =

k=1,3
F
(k)
=

k=1,3
F
(k)
(By Newtons third law)
A =

k=1,3
A
(k)

(k)
where A
(k)
is the area of the k
th
surface of the tetrahedron. From geometry, A
(k)
= A n e
(k)
.
Thus the stress can be written as:
=
_
_

k=1,3
e
(k)

(k)
)
_
_
n
= n
where
(x, t) =

k=1,3
e
(k)

(k)
)
is the stress tensor, which is independent of the direction of n. The components of the stress
tensor are given by:

ij
=

k=1,3
e
(k)
j

(k)
i
But e
(k)
j
=
jk
, and hence
ij
=
(j)
i
is the ith component of the force per unit area exerted
on the uid by a surface with normal in the jth coordinate direction.
The most important statement about the stress tensor is that the force per unit area (stress)
exerted on the uid by a surface with unit normal n pointing out of the uid (equal, of course,
to minus the stress exerted by the uid on the surface) is given by:
= n
1.3 The Momentum Equation
Consider an arbitrary xed control volume used for mass conservation. The rate of change of
the total momentum within the control volume is due to the outow of momentum through the
4
boundary, and to the forces acting on the uid, which comprise body forces (per unit volume)
and surface forces (per unit area). Thus:
d
dt
_
V
udV =
_
V
(u)u ndS Momentum ux
+
_
V
FdV Body forces
+
_
V
ndS Surface forces
Use of the divergence theorem gives
_
V

t
(u
i
)dV =
_
V

x
j
(u
i
u
j
)dV +
_
V
F
i
dV +
_
V

x
j
(
ij
)dV
Also, since this holds for arbitrary control volumes, the integrands must equate to give

Du
Dt
+u
_

t
+ (u)
_
= F +
(actually (
T
),but well soon show that =
T
). The second term is zero by conservation
of mass, so

Du
Dt
= F +
1.4 Energy
Taking the integral of the momentum dotted with velocity of the uid, the rate of internal viscous
dissipation is obtained:
D =
_
V
_

_
udV =
_
V
e
ij

ij
dV =
_
V
e : dV
where e is the strain rate tensor
e
ij
=
1
2
_
u
i
x
i
+
u
j
x
i
_
Notice that e is symmetric, that is e
ij
= e
ji
and e
ii
= 0. The rate of work by surface forces on
the uid is:
_
dV
u ndS
1.5 Newtonian Fluids
Stresses are produced by uid deformation. If relationship between stress/deformation rate
u
i
/x
j
is local, linear, instantaneous, and isotropic, the uid is Newtonian. If it is also incom-
pressible
= pI + 2e = pI +d
Where is the dynamic viscosity and d is teh deviation stress, which depends only on e =
1/2[(u+(u)
T
] and not on the vorticity = u. For incompressible Newtonian uids with
constant viscosity:

Du
Dt
= F p +
2
u u = 0
which together are known as the Navier-Stokes equations. The rate of viscous dissipation is then:
D = 2
_
V
e : edV
Often, F = and F can be incorporated into a modied pressure p + = P.
5
1.6 Boundary Conditions
1.6.1 Kinematic boundary conditions
At uid/uid interfaces, using mass conservation and no mixing:
[u n] = 0, Mass conservation
[u n] = 0, No stress
The brackets indicate terms across the interface (i.e. matching, but not necessarily 0, velocities).
1.6.2 Rigid boundary
A hard impermeable boundary yields instead:
u n = 0, No ux
u n = 0, No slip
1.6.3 Dynamic boundary conditions
In the absence of surface tension [ n] = 0. With surface tension:
[ n] = n
s

where =
s
n (specied since interfaces may be curved) and is the coecient of surface
tension.
1.7 The Reynolds Number
Once again, the Navier-Stokes equations are:

Du
Dt
= p +f +
2
u u = 0
Supposed L, U, and L/U are representative length, velocity and time scales of a given body:
Examining the scaling of each term

Du
Dt

U
2
L

2
u
U
L
2
The ratio of these two terms is dened as the Reynolds number Re:
Re
Inertial Forces
Viscous Stresses
=
UL

=
UL

where = /. If Re 1, we have the Stokes equations, which neglect the small inertia term

2
u = p F u = 0
The Stokes equations are very useful for:
1. Large - Magma, glass
2. Small L - microorganisms, microuidics
3. Thin long lms - lubrication theory
6
1.7.1 Notes about the Reynolds Number
1. There is an assumption that Re scaled with L, but u and
2
u may not have the same
length scale:
2. L
1
will be the lengthscale for the viscous term (
2
u), while L
2
will be the length scale for
the boundary term (u )
3. The length scales may very throughout the ow, such as in the case of a ow past a body
(the reynolds number in the far eld will be dierent)
4. The time scale may not be L/U if there is an externally imposed time scale (e.g. an
oscillating body, T
1
)
7
Chapter 2
Stokes Equations
The Stokes equations are:
=
2
u p = F u = 0 (2.1)
2.1 Simple Properties
2.1.1 Instantaneous
Since there is no time derivative term (no /t), there is no inertia or memory. The ow only
knows about the current boundary conditions and applied forces, and responds instantaneously.
If the boundary conditions or forces are changing w/ time, the ow is called quasi-steady.
2.1.2 Linear
The response of the ow is linearly proportional to the forcing (double the force, double the
velocity eld). Additionally, it is possible to superimpose solutions onto the same geometry.
2.1.3 Reversibility
If the sign of all forces is changed, the sign of the velocity eld will change (F u). If
the sign of the forces is changed and history of application is reversed, the ow retraces its own
history exactly.
This can be used with symmetry arguments to rule something out. For example:
Migration is impossible due to the contradiction. Three more phenomena to be described:
1. There will be no rotation when falling.
Reection across both axes, then a reversal of force/velocity will yield a contradiction.
8
2. There will be no migration in the boundary layer.
Reversing the ow, then reecting along the y axis yields a contradiction
3. Swimming is not possible.
2.1.4 Forces and Torque Balance
Becaues there is no inertia, all forces must balance
= F
_
V
ndS +
_
V
FdV = 0
This allows for a consistency check on any stress boundary conditions. Similarly,
u = 0
_
V
u ndS = 0
is a consistency check on velocity BC. Likewise, moments/torques balance.
2.1.5 Work Balances Dissipation (since there is no KE)
The dissipation (as dened previously):
D = 2
_
V
e
ij
e
ij
dV
=
_
(
ij
+
ij
)e
ij
dV
=
_ _

ij
u
i
x
j
+pe
ii
_
dV
=
_ _

x
j
(
ij
u
i
) u
i

x
j

ij
_
dV
=
_
V
u ndS +
_
u FdV
work balances dissipation
2.2 Three theorems based on dissipation integrals
Lemma 2.2.1. If u
I
is incompressible ow, u
I
= 0 and u
S
is a Stokes ow w/ body force
F
S
(
S
= F
S
) then:
2
_
V
e
I
: e
S
dV =
_
u
I

S
ndS +
_
V
u
I
F
S
dV
Proof. ll this in later, using the the same method as for the dissipation of work.
Theorem 2.2.1. Solutions of Stokes equation are unique
Proof. Suppose u
1
and u
2
are Stokes ows with the same boundary conditions and body force
(i.e. F
1
= F
2
in V and either u
1
= u
2
or
1
n =
2
n on V ), let:
u

= u
1
u
2
u

is also a Stokes ow with F

= 0 and either u

= 0 or n = 0 on V . From the previous


lemma:
2
_
e

: e

dV = 0 e

= 0 in V
Therefore, u

is a solid body motion (no strain) with (u + x), and usually u

= 0. Therefor:
u
1
= u
2
Solution is unique
9
Theorem 2.2.2. Reciprocal Theorem: If u
1
,u
2
are Stokes ows (in a certain domain), then:
_
V
u
1

2
ndS +
_
V
u
1
FdV =
_
u
2

1
ndS +
_
V
u
2
F
1
dV
Proof. Apply previous lemma to show that both sides are:
2
_
e
1
: e
2
dV
This is essentially Greens 2nd theorem.
Theorem 2.2.3. Maximum Dissipation Theorem: Among all ows in a certain domain, that
satisfy given velocity boundary conditions and u = 0, the dissipation is minimized by stokes
ow w/e F = 0.
Proof. First,
0 2
_
(e e
S
) : (e e
S
)dV = 2
_
(e : e e
S
: e
S
)dV + 4
_
e
S
: (e
S
e)dV
The second term is zero by the previous lemma, with:
u
I
= u
S
u
where u
I
= 0 on the boundary and F
S
= 0. Therefore, the remaining term is:
0 2
_
(e : e e
S
: e
S
)dV = D D
S
2.2.1 Example of the dissipation theorem
The dissipation theorem allows for bounds to placed on the drag of irregularly shaped objects.
For example, in the case of an arbitrary volume using inscribed and circumscribed circles:
If body A is inside body B then an inequality can be derived by letting u
S
be a Stokes ow
past A and u
I
be the Stokes ow past B plus the solid body motion in the gap between A and B.
This inequality can be derived for the inner and outer circles, allowing the appropriate bounds
to be calculated:
(6a
1
U)U F u (6a
2
U)U
2.2.2 Example with rigid particles
Adding rigid particles to a Stokes ow with given external velocity boundary conditions increases
the dissipation and, if the particles are force free and couple free, increases the apparent viscosity.
This does not say anything about more or less particles, just that the addition of some particles
will increase the viscosity. For example, does not say anything about shear thickening suspensions
of silica.
10
2.2.3 Inertia
Inertia will increase drag (for example, consider that Du/Dt is part of F).
2.3 Representation by potentials
Assuming no body force (F = 0), a conservative ow with a modied pressure:

2
u = p (1) u = 0 (2)
Taking the divergance of (1):
(1)
2
p = 0
(1)
2
= 0

2
(1)
4
u = 0
In 2D, if we use a velocity eld:
u = (0, 0, )
z
=
2

4
= 0
Using axisymmetric sphericals (r, , ) with:
u =
_
0, 0,

r sin
_
This results in:

0
=
E
2

r sin
E
4
= 0 where E
2
=

2
r
2
+
sin
r
2

1
sin

Many explicit solutions can be found in coordinate systems in which the operators
2
,
4
, E
2
,
and E
4
are separable (see Happel and Brenner).
2.3.1 Complex Variables in 2D
Writing z = x +iy and z = x iy allows:

2
= 4

2
zz
If f(x, y) is analytic, meaning:
f = f(z)
f
z
then Ref and Imf are harmonic. Similarly,
4
= 0 implies can be written as:
= Im(z(z) +(z))
where and are analytic. Some very clever solutions can be found by conformal mapping
techniques. An example of a very tricky solution is the combination of two bubbles:
Of course, the problem is that the complex methods only work in 2D.
11
2.3.2 Papkovich-Neuber Solution
This will be the main way to solve for Stokes ows. Let:
p =
2

can always be solved with an integral representation


(x) =
1
4
_
p(x

)
|x x

|
dV

Using an equivalence with the normal denition of p:

2
(u ) = 0
u = where
2
= 0
But:
u = 0
2
=
The idea is then to think of a function for that satises this relationship:
=
1
2
x +
2
= 0
This comes essentially from completing the squares with the :

2
(A) = (
2
A) B +A
2
B + 2A B
Hence any Stokes ow (where F = 0) can be written in terms of a harmonic vector and a
harmonic scalar :
2u = ( + x) 2 p = (2.2)
which is called the Papkovich-Neuber representation of Stokes ow.
2.3.3 Notes on the Papkovich-Neuber Solutions
1. Any potential ow of the form:
2u =
is also a Stokes ow but p = 0 and
ij
= u/2 + u
T
/2 =
i

j
dier from the
corresponding inviscid ow.
2. Sometimes it is possible to nd another harmonic scalar with:
= x 2
If so, one can dispense of by:

= +
For example, if has a spherical harmonic expansion then we can do all the terms except
for the uniform strain as
=
1
2
x

x
where

is the far eld strain. In this case we can write


2u =

x +(x ) 2
p =
2e =

+ () x
where
2
= 0 and

is a constant, symmetric, traceless matrix.


12
3. Conversely, if = , then u can be calculated using:
= x 2
which is generally easier than (2)
4. Lambs Solution is frequently referred to in literature and uses three harmonic scalars
expanded in spherical harmonics, which is essentially the same as the method just above.
2.4 Solutions for points, spheres, and cylinders
Points and spheres have no intrinsic orientation or direction, so spherical harmonics are
used. Let:
r = |x|
2
1
r
= 0 (when r = 0)
All other functions where 0 as r 0 can be obtained from:
1
r
,
1
r
,
1
r
,
1
r
, . . .
Harmonic functions which are bounded at r = 0 can be obtained from:
r
1
r
, r
3

1
r
= x, r
5

1
r
, . . . r
2n+1
()
n
1
r
This is a generalization of other schemes for generating harmonic functions, such as using Leg-
endre polynomials:
_
r
n
r
n1
_
P
m
n
()
_
cos n
sin n
_
Recall that:
x = I
_
or
x
i
x
j
=
j
x
i
=
ij
_
r =
x
r
_
from 2rr = r
2
= (x
j
x
j
)[same as /x
i
(x
j
x
j
)] = 2x
_
f(r) =
x
r
f

(r)
For example:

1
r
=
x
r
3

1
r
=
I
r
3
+
3xx
r
5

k
=
3(
ij
x
k
+
jk
x
i
+
ki
x
j
)
r
5

15x
i
x
j
x
k
r
7
These functions depend only on x and r and have no preferred orientation, which is useful for
problems with spherical symmetry. It is possible to form Papkovich-Neuber potentials , by
multiplying by constant tensors and taking an appropriate number of dot products. For example:
A
1
r
, B
1
r
, C
1
r
, (D )
1
r
,
1
r
are all vectors
2.4.1 Pseudo-tensors vs true tensors
Tensors have no right-handedness dened. We can distinguish between true and pseudo tensors
by checking whether they keep or change signs under reection. Examples of true tensors and
pseudo-tensors are:
True: U, F, x,
Psuedo: , U x, = u
13
All of the pseudo-tensors have a right-handedness obtained form some time of cross product.
The product of two true tensors is pseudo, and vice versa. An important distinguishing property
is behavior under transformations. Under rotation, both psuedo and true tensors stay the same,
but under reection they change:
T

ijk
= R
ij
R
jm
R
kn
, . . . T
lmn
where the sign depends on whether it is a pseudo or true tensor and R is the reection matrix.
We need u to be true, so , should also be true.
2.4.2 Solution due to a point force
Consider:
= F(x) u = 0, u() 0
Because of the linearity of Stokes ow, the answer must be linear in F. In addition, there is no
orientation of the ow, due to rotational symmetry. So, try:
=
F
r
Since F (1/r) is pseudo, using (F )(1/r) would yield = F (1/r) which is too singular
(blows up at r = 0). The velocity eld is then:
2u =
_

_
F x
r
_

2F
r
_
=
_
F I
r

(F x)x
r
3

2F
r
_
=
_
F
r

(F x)x
r
3
_
The stress can then be calculated
= 3
(F x)xx
r
5
On any sphere r = R, n = x/R then:
u n =
2
2
F n
R
n =
3(F n) n
R
2
Checking the mass ux:
_
r=R
u ndS =

R
F
__
r=R
ndS = 0
_
= 0
As expected, there is no net ow from the point force. Since we want to specify a force:
F =
_
r=R
ndS = 3F
__
n
i
n
j
dS
R
2
=
4
ij
3
_
Thus, is chosen to be:
=
1
4
The solution to this problem was achieved using only some thinking, and the rest came from the
Papkovich-Neuber solution. All that really needed to be done was identify the coecient.
2.4.3 Fundamental Solutions of Stokes Flow
The fundamental solutions of Stokes ow, upon which other solutions are linearly dependent,
are as follows:
u = F J(x) = F K(x) p =
F x
4r
3
J =
1
8
_
I
r
+
xx
r
3
_
K =
3
4
xxx
r
5
Where J is known as the Oseen tensor.
14
Example: Flow due to a point force
From u = 0 and = F(x):
J = 0 K = I(x)
Thus, the velocity prole will fall o proportionally to 1/r
2.4.4 Point Source Flow
From mass conservation and radial symmetry, the velocity eld must be:
u =
Qx
4r
3
This could also be derived from:

Q
r
or Q
1
r
2.4.5 Force Dipoles, Stresslets, Rotlets, Etc.
Further solutions can be obtained by taking gradients of the stokeslet and source solutions. For
example, consider:
The solution for small d, due to the linearity of the Stokes equations, will be:
u = F J(x d) F J(x)
Taking a Taylor expansion:
u
j
= F
i
(d
i
)J
ij
+. . .
As the displacement goes to zero, the force will go to innity while maintaining a constant Fd.
The term F
i
d
j
can be split into three parts:
1. An isotropic part with no ow ( J = 0)
1
3
F
k
d
k

ij
15
2. A symmetric traceless part, representing a stresslet of strength S:
S
ij
=
1
2
(F
i
d
j
+F
j
d
i
) +
1
3
F
k
d
k

ij
3. An antisymmetric part, representing a Rotlet of point couple G:

1
2

ijk
G
k
where G = d F
2.4.6 Example: Rigid Sphere in Translation
Consider:

2
u = p u = 0 u = U(on r = a) u(r = ) = 0
We need harmonic functions of U and x that are linear in U, decay at r = , and are true
vectors (not pseudo). First, some possibilities for U and :
1
2
=
U
r
, U
1
r
= U
1
r
The second possibility for yields the same behavior as , so it must be the rst possibility.
Higher order terms are not possible because any additional gradients will require more dot
products with U and x, which would break linearity. Thus:
u = A
_
u
r
+
(u x)x
r
3
_
+B
_

U
r
3
+
3(U x)x
r
5
_
u =
_
u
r
+
(u x)x
r
3
_
+U
_

1
r
_
With the boundary condition u = U on r = a. Solving for the coecients:

a


a
3
= 1,

a
+
3
a
3
= 0 = a
3
/4, = 3a/4
Thus, the velocity eld is:
u =
3
4
U
_
a
r
+
a
3
3r
3
_
+
3
4
(U x)x
_
a
r
3

a
3
r
5
_
=
_
1
r
_
Stokeslet +
_
1
r
3
_
quadropole
In order to calculate the pressure and vorticity, either the Stokes relations can be used, or
the form can be inferred. They must be of the form (due to dimensional and scaling/vector
arguments):
p a
U x
r
3
=
3
2
a
U x
r
3

U x
r
3
= a
3
2
U x
r
3
16
Force and Stress on a Translating Sphere
The next step is calculating the stress on the sphere from the Stokes equations and available
information:
= pI +[u + (u)
T
] x = I r =
x
r
r
n
= nr
n1
x
r
This leads to:
=
3
2
U x
r
3
I+
3
4
__
a
r
3

a
3
r
5
_
(xU +Ux)
+
_
Ux +xU + 2(x U)I
_
a
r
3

a
3
r
5
__
+ (U x)2xx
_
3a
r
5
+
5a
3
r
7
__
Evaluating at the surface of the sphere:
n|
r=a
=
_
x
r
_
r=a
=
3
2

a
(u n) n +
3
4a
[2((uU n) n +U) +O + 4(U n) n] =
3U
2a
This remarkably leads to a constant across the sphere. Hence:
_
ndS =
3U
2a
4a
2
= 6aU
2.4.7 Simple Derivation of Force/Stress on a Translating Sphere
The far-eld is dominated by the 1/r stokeslet term with coecient F = 6aU. Since the force
exerted at r = is dominated by this term and the force must balance the total exerted across
r = a, we can deduce without any further work that the drag on a sphere is:
Drag = 6aU
2.4.8 Gravitational Settling
A force balance for small particles using Stokesian drag and gravity is:
4
3
(
s
)a
3
g 6aU U =
2
9
(
s
)ga
2

This relation to the square of the particle radius allows small particles to stay suspended in
solution while larger ones may fall out.
2.4.9 2D singularities and Flow Past a Cylinder
The harmonic functions in 2D are:
ln r, (ln r), . . .
and when bounded at the origin:
1, r
2
ln r = x, = r
2n

n
(ln r)
Using a potential = F/2 yields:
u = F J
2D
p =
F x
2r
2
= F K
2D
where
J
2D
=
1
4
_
ln rI +
xx
r
2
_
K
2D
=
1

xxx
r
4
This is the 2D Stokeslet solution, which corresponds to a line force F per unit length. Note that
u does not converge to 0 at innite distance for a line force, although it will converge to 0 for
line dipoles, line quadropoles, etc.
17
2.4.10 Uniform Flow Past a Stationary Cylinder
Motivated by the previous solutions for uniform ow past a sphere, we try:

2
= U ln r +U ln r +U
to solve the Stokes equations with
u(r = a) = 0 u(r ) U
(The U is partly motivated by the change in reference frame.) However we can nd that, while
we can satisfy the boundary conditions on the cylinder with:
u =
_
U
_
ln
r
a
+
1
2
_
1
a
2
r
2
__
+
(U x)x
a
2
_
a
4
r
4

a
2
r
2
__
p7 = 2
U x
r
2
Drag = 4U
There is no choice of that allows u U as r . The problem is that inertia cannot be
p7?
neglected far from the cylinder since
(u )u

2
u

U
2
/r
U/r
2
=
Ur

1 as r
This scaling suggests that inertia needs to be brought back into the equations when r =
O(aRe
1
).
After scaling lengths with a and velocities with U, the steady Navier-Stokes equations give:
Re(u )u = p +
2
u u = 0, r > 1 u = 0, r = 1 u z, r
Where Re = Ua/ 1. A somewhat ad hoc way of obtaining an approximate solution is to
recognize that the (u )u term only has a leading-order eect where u z. This motivatees
the Oseen equations:
Re
z
u = p +
2
u u = 0
which are linear (with solutions by fourier transform), no less accurate than the Stokes equations
for small r, and correct at leading order for large r.
A rather better way of proceeding is to solve the original equations by matching asymptotic
expansions for Re 1 between r = O(1) and r = O(Re
1
). The zeroth-order solution for
r = O(1) is the above Stokes solution with determined by matching to be:
=
1
ln 1/Re
+O
1
[ln 1/Rey]
2
F =
4U
ln 1/Re

n
c
n
[ln 1/Re]
n
This depends only weakly on a and corrections are quite large. Intertia cannot be neglected
far from a sphere either, or from any other 3D body, but the solution for r = O(Re
1
) does not
aect the zeroth-order Stokes solution for r = O(1), and comes in at rst order instead.
2.5 The Motion of Rigid Particles
We can get some information for particles of various shapes using the dissipation theorems and
bounding the results by inscribing and circumscribing spheres. An alternate, more detailed
formulation is:
18
2.5.1 The Resistance Matrix
A rigid particle moving with velocity U +x through a uid, otherwise at rest, exerts a force
F and couple G on the uid. By linearity, both F and G must be linear to U, . The grand
resistance matrix is:
_
F
G
_
=
_
A B
C D
_ _
U

_
where the tensors A, . . . , D depend on the size, shape, orientation, etc of the body. The dissipa-
tion is then:
_
u ndS =
_
(U + x) ndS U F + G
Hence, from the reciprocal theorem (with F = 0):
[U
1
,
1
]
_
A B
C D
_ _
U
2

2
_
= [U
2
,
2
]
_
A B
C D
_ _
U
1

1
_
For all U
1
,
1
, U
2
,
2
. Thus A = A
T
, D = D
T
means the matrix is diagonalisable and also
that B
T
= C. Thus, forces due to pure rotation balance forces associated with translation. Since
the dissipation must be positive, A, B, C, D must all be positive, denite, and invertible.
A few examples:
1. If a particle has 3 independent planes of reectional symmetry, then B = C = 0. (i.e. if
three independent vectors dotted with B,C are zero, B,C must be zero.
2. A cube falls with the same speed and no rotation in all orientations because A I (from
symmetry).
3. A and D are known for ellipsoids (w/ B, C = 0) and hence also known for discs and rods
(which are just limiting cases).
The resistance matrix (with R
1
being the mobility matrix) is sometimes extended to:
_
_
F
G
S
_
_
=
_
_
_
_
_
_
U U

_
_
and is called the grand resistance matrix.
It is possible to describe the force, couple, and stresslet exerted by a particle placed in a
background linear ow with:
U

= U

x +E

x
A linear ow L gives leading order eects for small particles of characteristic length a placed in
some background ow U

(x) of larger lengthscale L:


U

(x) = U

(0) + (x )U

(0) +O
_
a
2
L
2
_
2.5.2 The Faxen Relations
These relations concern a particle placed in an arbitrary unbounded ow U

(x). The goal is to


calculate the particle velocity u
P
u
P
= U + x
Let u

be the perturbation ow u(x) U

(x), which satises u

() 0 and u
P
= U +
xu

(x) on the particle. let u be a test ow due to translation with arbitrary velocty

V . By
linearity, (x) =

V

for some

. The reciprocal theorem for u

, u gives
V
_
Part.

ndS =

V
_
(U + x U

(x))

ndS
19
But V is arbitrary and the contribution from U +x is given by the resistance matrix. Hence
F = A U +B
_
U


ndS
= Unknown + Unknown + Background Flow
which is Faxens rst relation. Similar formulas can be derived for the couple and stresslet using
other test ows.
2.5.3 Example
For a sphere

n =
3
2a
I
The matrices A, B are
A = 6aI B = 0 (Spherical Symmetry)
which, used with Faxens rst relation, gives
U =
F
6a
+
1
4a
2
_
r=a
U

(x)dS
where the second term is just the average velocity on the surface. The expansion of the far eld
is:
U

(x) = U

(0) + (x )U

(0) +
1
2
xx : U

(0) +. . .
which leads to
1
4a
2
_
U

dS = U

(0) + 0 +
1
8a
2
__
xxdS
_
: (U

(0)) +. . .
The odd terms in this series (e.g.
_
xxxdS) are zero by symmetry. The even terms are isotropic,
and give (
2
)
n
U

(0). But
4
U

(0) through the harmonic requirement of the Stokes equation


(same for 6,8,etc). Thus, the Taylor series naturally terminates after just a few terms.
u =
F
6a
+u

(0) +
a
2
6

2
U

(0)
One application of this is two falling bodies. Its possible to calculate the fall velocity to
O(a
3
/R
3
) with a 1/r
4
dependence on each other.
20
Chapter 3
Applications
3.1 Marangoni Flow
Marangoni ow is driven by gradients in surface tension () gradients between immiscible uids.
A few examples:
1. Temperature eects: As the temperature increases, the surface tension decreases. For a
water/air interface, (d/dt)/ = 1/50K. A classic example is
2. Surfactant gradients:
surfactant
diagram
3. Alcohol/water
wineglass
3.1.1 Boundary Conditions
Surface tension can be represented as a surface stress (units N/m) derived from a surface energy
J/m
2
, which acts across lines in an interface. The surface stress tensor is (I n n)
s
, which
essentially just kicks out the relevant tangential directions.
A force balance on a small bit of interface gives
[ n]
+

+
s

s
= 0
where

s
(I n n)
is the gradient operator in the surface. Hence, the jump in the uid stresses is
[ n]
+

=
S
((I n n)
= (I n n)
s
+ (
s
n) n + ( n
s
) n
=
s
+ n
which is the surface tension crossed with the curvature.
21
3.1.2 Thermophoresis
An immiscible drop in a temperature eld migrates from a cold region to a hot region. We
assume a background temperature gradient H = (T)

. From a scaling argument, T Ha


implies Ha

, and

= d/dT < 0. The driving force from a surface tension dierence is


proportional to F Ha
2

. The viscous stress U/a acts proportional to the area. Thus,


the drag is proportional to Ua F and U Ha

/.
This leads to a second thermal problem within the bubble and the eld. The heat equation
is
T
t
+u T =
2
T
The temperature gradient approaches the gradient H at innite distance. Since the temperature
must be continuous at the surface of the particle
[T] = [k n T] = 0 On |x| = a
where k is the thermal conductivity (q = kT) and is the thermal diusivity = k/C
p
.
We assume that the Peclet number is small
Pe =
Ua

=
Advection
Diusion
1
so that the heat equation becomes

2
T = 0
Thus, for an insulating drop (k
i
= 0)
T = T
0
+H x(1 +a
3
/2r
3
)
which is a harmonic scalar linear in H. On the surface
T = T
0
+
3
2
H x
We can expand around the temperature dierence, and take the rst term of the temperature
dierence is small
(T) (T
0
) + (T T
0
)

(T
0
)
The next step is to connect the surface tension gradients to the uid problem. The surface stress
is driven by an interfacial force density
F
s
= (
1
n
2
n) = [ n]
+

= (I n n) (3H

/2) (
s
n) n
The last term is also dependent on H through the surface tension. The divergence of the normal
for a sphere is just 2/a (as expected). The force acting on the droplet is then
F =
3
2
H

(I 3 n n)
0
2
a
n
For the simplest case of a drop with equal viscosity to the surrounding uid ( = 1), the velocity
prole is
u(y) =
_
r=a
J(y x) F
s
(x)dS
=
3
2

1
8
H
_
(I 3 n n)
_
I
R
+
RR
R
3
_
dS
=
3

a
16
H G(y)
22
where R = y x and G is a dimensionless function of y. By the symmetry of the integral
G(y) =
_
r
a
_
I +
_
r
a
_
y y
where y = y/r. Continuinc
u y =
3

a
16
( +)H y
Thus, a sphere will remain spherical and translate with a velocity
U =
3

a
16
( +)H =
1
5

aH

Its possible to nd the coecients , from


G
ii
= 3 +
y
i
G
ij
y
j
= +
3.1.3 Surfactant Eects
The concentration C of surfactant in a small material area of surface changes due to various
physical processes
1. Surface diusion with
Flux = D
s

s
C
Where the diusion coecient is on the order of 10
9
m
2
/s
2. Adsorption from bulk with rate
r = k(C C
0
)
Where C
0
is the equilibrium concentration and the rate constant is on the order of 1000/s.
3. Changing surface area
(AC) = 0 C =
A
A
C
to calculate the rate of change of surface area its necessary to consider a material volume
The volume of the uid element is then
V = L A n
dV
dt
= ( u)V
d
dt
L = (L )u
where L is an arbitrary length vector (notice in the next step it cancels on both sides).
d
dt
(A n) = [( u)I u] A n
We now take n d/dt( nA) and use n

n = 0 since n n = 1.
d
dt
(A) = A( u n n : u] = (
s
u)A
23
Expanding this last term
d
dt
(A) = A
s
(u
s
+ (u n) n)
= A(
s
u
s
+ (
s
n)(u n) + ( n
s
)(u n))
= Dilation of Flow in Surface + Dilation of Surface + 0
(since
s
is in the tangential direction only)
Combining all of these eects into a total time derivative on the concentration
DC
Dt
= C[
s
u
s
+ (u n)
s
n] +D
S

2
s
C k(C C
0
)
Example: Rigidication of Interfaces
Consider a rising bubble with surfactants present. Any ow leads to a concentration gradient
in surfactant that creates a gradient in surface tension thus driving a reverse ow. A steady
solution in the frame of the bubble has a steady concentration prole of surfactant concentration
and no mass ux into or out of the surface (u n = 0). Consider a linearized calculation with
C = C
0
+C

(x) with C

C
0
(to be checked later). Thus
D
s

2
s
C

kC

= C
0

s
u
s
After neglecting the nonlinear terms u C

and C

s
u
s
. Since the solution should be linear
in u and have spherical symmetry,
u
s
= A(I n n) u
for some A (O(1)). Then

s
n = (I n n) (x/a) =
I n n
a
which is the same as the divergence of the normal vector, (x/r), and the n n (x/a) term
vanishes. Thus

s
n =
2
a
Which isnt surprising as this is the curvature of a sphere. Plugging this in

s
u
s
= A
s
(I n n) U =
2
a
AU n
Simplifying the surface divergence

s
= (I n n)
s
( n n) = (
s
n) n + ( n
s
) n = (
s
n) n
Guessing that the surface concentration is proportion to the normal velocity
C

(u n)
Since

2
s
(u n) =
s

_
I n n
a
U
_
=
2
a
2
U n
The solution for C

is then
C

=
2C
0
A(u n)/a
k + 2D
s
/a
2
The assumption of small C

is thus ne as long as surface diusion is fast (Ua/D


s
1) or
surface absorption is fast (U/ka 1), or if the ow is very slow. The problem is completed
24
by the Papkovich-Neuber representation of ow inside and outside the bubble. The new set of
boundary conditions for this example are
u(r = ) = U u(r = 0) bounded
and on the surface
u = u
s
[ n n] = n C
d
dC
Solving the problem, the drag is
F = 4aU
_
3(

+)/2 + 1
(

+) + 1
_

=
aC
0
d
dC
3(a
2
k + 2D
s
)
So the main eect is to increase the apparent viscosity of the bubble. As the eect becomes
stronger, the solution approaches that of a rigid sphere.
3.2 Deformation of Drops (Rallison 1984, Stow 1994, Ann.
Rev. Fluid. Mech.)
To leading order, a small drop in an external ow with velocity U

is advected with velocity


U

(0) and is deformed by gradients of the ow x U

(0) against the restoring eect of surface


tension.
The strength of this behavior can be represented in a new dimensionless quantity
Viscous Stress
Surface Tension Pressure
=
E
/a
= Ca =
U

Which is just the capillary number or perhaps U/. Low capillary numbers lead strong strong
surface tensions.
3.2.1 Small deformations (Ca 1) in uniform streams (U

= E x
There is deformation due to a symmetric and traceless ow E (ellipsoidal symmetry). Thus, a
new shape might be expected
r = a
_
1 +
x D x
r
2
_
+O(D
2
)
where D is also a symmetric, traceless (to conserve volume) tensor and D = O(Ca). In a general
time dependent problem, D need not be proportional to E. Only in the equilibrium case is the
time evolution of D determined by E.
n = (r a())
=
x
r
2a
_
D x
r
2

(x D x)x
r
4
_
25
We note that | n| = 1 +O(D
2
) as required. Since | n| = 1,
n ( n ) n = 0

s
n = n =
2
r
+
6x D x
r
4
a
=
2
a
+
4(x D x)
a
3
+O(D
2
)
Solution using Papkovich-Neuber gives

D =
5
2 + 3
E
6
a
40( + 1)
(2 + 3(19 + 16)
D
3.2.2 Larger Deformations in Pure Strain
However, beyond a critical capillary number the bubble breaks
However, bubbles with small values of can handle a large amount of deformation.
3.2.3 Larger Deformations for 1 (Taylor 1964)
With parameters
L R
0
L
R
0
Ca
3
Ca
crit

1/6
A preliminary argument can be constructed from a scaling argument. Since 1, the shear
stress exerted by the bubble on the ow can be neglected. Thus
u
ext
Ez EL
We rst start by neglecting the pressure gradients within the bubble. Moving along the bubble
with a slice of uid, we see a hole of radius R(z), collapsing with a velocity of EzdR/dz, with
a strain rate proportional to vel/R, due to the normal stress dierence /R between the inside
and the outside. Thus the energy is related to
E

R
0

a
R
0
C
a
26
The volume of the bubble is approximately
R
2
0
L a
3

L
R

_
a
R
0
_
3
C
3
a
The ow inside the bubble is a Poiseuille ow, but with wall velocity Ez and no net ux.
The pressure drop along the length can be estimated from pipe ow
p
L

()EL
R
2
0
The pressure drop must be p /R
0
. The viscosity ratio must then be


ER
0
_
R
0
L
_
2
Ca
6
3.3 The Rayleigh Instability
Consider a small asymmetric perturbation to a cylinder of one uid immersed in another
If the radius is only changed slightly (|| a, |
z
| 1), the normal is
n = (r ()) (1, 0,
z
)
which is basically normalized since
z
is very small. The divergence of the normal is then
n =
1
r

zz

1
a


a
2

zz
The shape is stabilized by the axial curvature
zz
but destabilized by the azimuthal curvature
/a
2
. Consider disturbances of the form
= cos(kz)e
st

n
n =
1
a

1 a
2
k
2
a
2

The ow will be unstable if 1 > a
2
k
2
(that is, for suciently long wave length disturbance). The
growth rate s depends on the dynamics. In the general case, with inertia, and
0
,
i
,
0
,
i
is in
Tomotika (PRS1935). We will consider the simple case with no inertia and no viscosity in the
external ow. Linearized kinematic boundary conditions are used u|
r=a
=
t
= s. The dynamic
boundary conditions
[ n] = ( n) n = 0 (outside)
The tangential stress is then
u
z
+
w
r
= 0
27
The normal stress is
p
0
p + 2
w
r
=

a
+
_
1 a
2
k
2
a
2
_
u
s
To solve for the Stokes ow inside the cylinder, harmonics functions are needed of the form
cos(kz) (see Morse/Fesbach or derive by separation of variables). The Papkovich-Neuber rep-
resentation is then
= AI
0
(kr) cos(kz)
= (BI
1
(kr) cos(kz), 0, 0)
where I
0
, I
1
are the modied Bessel function (which are well behaved at r = 0. Using I

0
(x) =
I
1
(x), (xI
1
)

= xI
0
) and thus nd that
u = (AkI
1
+BkrI
0
2BI
1
) cos(kz)
w = (AkI
0
BkrI
1
) sin(kz)
p = 2BkI
0
cos(kz)
Substituting these equations into the boundary conditions,
s =

2a
(1 k
2
a
2
)
I
1
(ka)
2
k
2
a
2
[I
0
(ka)]
2
(1 +k
2
a
2
)[I
1
(ka)]
2
The rst part of the term could have been guessed from intuition, but the second requires the
full model (which Lord Reyleigh was able to do in 1892). Plotting this function as a function of
ka
Interestingly, the most unstable distrubance has innite wavelength, because this minimizes the
internal stretching w/z and external drag is ignored (while a small eect, at long enough length
scales it becomes important). A full model would take into account changes in the external ow.
A cylindrical bubble in a viscous uid is most unstable at innite wavelength because this
minimizes du/dz and ignores internal Poiseuille ow. For a more realistic model with 0 < <
3.3.1 Neglect of inertia
A scaling argument for the neglect of intertia (when ak = O(1)) shows
u
a
2

p
a


a
2
u as s

a
as we identied.
28
Chapter 4
Long Thin Flows and Immersed
Bodies
4.1 Long thin ows I: Lubrication theory
Consider a long thin ow with at least one eectively no-slip boundary condition.
Assuming that the velocity gradients satisfy /z 1/h /x 1/L then scaling the
divergence of u shows
u
U
L

w
h
u

x
w

z

u
L
We then consider the scalings in the horizontal components of the Navier Stokes Equations,
u u = p +
2
u
_

u
2
L
_
=
_
p
L
_
+
_
u
L
2
+
u
h
2
_
_
uh

h
L
_
=
_
p
uL
h
2
_
+
_
h
2
L
2
+ 1
_
Thus, we can neglect inertia if the modied Reynolds number is small
uh

h
L
1
We can also neglect the
2
x
in
2
and use just the
2
z
since L h. Finally, we can expect the
pressure to be on the order of uL/h
2
, since the pressure term must balance 1. In the vertical
components
_
uw
L
_
=
_
p
h
_
+
_
w
h
2
+
w
L
2
_
_
uh

h
L
h
2
L
2
_
= [1] +
_
h
2
L
2
+
h
4
L
4
_
Since all of the terms except 1 are small, at leading order p
z
p/h 0. Thus, the ow in the
thin layer is quasi-parallel to the boundary. If we let u = (u, v) and = (
x
,
y
) denote the
29
components of the velocity and gradient in the horizontal direction (parallel to the surface), then
the three governing equations of lubrication theory will be
Momentum Conservation

2
u
z
2
= p
Depth Integrated Volume Flux q =
_
h
0
udz
Depth Integrated Mass Conservation
h
t
+ q = 0
Thus, the global geometry also matters. These three equations can besolved with various bound-
ary conditions at z = 0, h
Squeeze Films
A squeeze lm is a thin layer bounded above and below by a moving surface. Thus, the boundary
conditions will be u(z = 0) = U
1
and u(z = h) = U
2
. The prole and volume ux will then look
like
u =
p
2
z(h z) +U
1
+ (U
2
U
1
)
z
h
q =
h
3
12
p +
1
2
h(U
1
+U
2
)
The last term is just the thickness times the average velocity from the two moving sides. The
height will then change like
h
t
=
1
12
(h
3
p)
1
2
[(U
1
+U
2
)h]
which is called Reynolds equation. We are usually given velocities U
1
, U
2
, h
t
) and then calculate
the pressure gradient and force.
Free-Surface Flows
Free-surface ows have a free upper surface, so that u(z = 0) = 0 and u/z = 0. The prole is
then
u =
p
2
z(2h z)
q =
h
3
3
p
h
t
=
1
3
(h
3
p)
Marangoni Flows
For a Marangoni ow, the top boundary condition (at z = h) will instead be

u
z
=
30
Example: Viscous Drop Approaching a Wall
We rst assume that
1. Surface tension is suciently strong to keep the drop spherical
2. 1 so that the drop is eectively rigid.
We can solve this problem in two ways, either with a scaling argument or by solving the equations
exactly. First with the scaling method. The relevant length scales will be
L (ah
0
)
1/2
By geometry
u (L/h
0
)

h
0
Radial velocity vertical velocity
u (a/h
0
)
1/2

h
0
through mass conservation
p
uL
h
2
0

a
h
2
0

h
0
From pipe-ow
F L
2

a
2
h
0

h
0
ga
3
Thus, the falling rate can be balanced through the graviational force and resistance to the
uid ow. Going about the full solution is a little more complicated. The gap thickness, as a
function of the distance r from the wall, is just
h(r, t) = h
0
(t) +
r
2
2a
+O
_
r
4
a
3
_
Additionally, the heigh is considered to be shrinking (

h
0
< 0). By mass conservation (integrating
h/t + q = 0),
r
2

h
0
r = 2rq(r)
For the rigid boundary boundary conditions, the velocity prole will be (with simple thin lm
ow)
u =
1
2
p
r
y(h y) r q =
h
3
12
p
r
r
Solving for the pressure gradient and integrating:
p(r, t) =
_
r
6

h
0
rdr
h
3
=
6

h
0
h
3
0
_

r
rdr
(1 +r
2
/2ah
0
)
3
= p

h
0
a
h
2
0
(1 +r
2
/2ah
0
)
2
The total force is just the integral of the pressure, so
F = 2
_

0
(p p

)rdr =
6a
2

h
0
h
0
We can see that the scaling argument retrieved the solution, except for the factor of 6. We can
now check the assumptions going in to the model
31
1. High pressure in the gap will deform the drop by h over lengthscale L. The deformation
can be estimated
p

h
L
2
h
h
0

p
/a

h
0

a
2
h
2
The deformation is small if the rst term is small compared to h
2
/a
2
, and is called the
capillary number

h
0

C
a

h
2
a
2
Equivalently, the Bond number must also be small
Bo
ga
2


h
a
Either way, as h continues to decrease (the sphere falls), the deformation assumption will
break down.
2. Shear stress against the bubble can drive circular ow over lengthscale L
The stress balance on the sphere will be
u
h
0

u
D
L

u
D
u

_
a
h
0
_
1/2
1

so that the rigid boundary condition u = 0 at z = h is ok if (a/h


0
)
1/2
, which will
eventually be invalid as well.
So, two extensions to the model are necessary to treat these two issues:
1. If (a/h
0
)
1/2
, we can neglect the shear stress exerted by the drop on the squeeze lm
(the drop is essentially inviscid compared to the lm). We then replace the boundary
condition at z = h by u/z = 0.

h
0
/F increases by a factor of (1/3)/(1/12) = 4 so that
the bubble falls four times as fast.
2. If Bo h/a, we get a dimple, which is a more complicated situation.
Example: Dimple Drainage between a Bubble and a Wall
32
We assume that h
0
/a Bo = ga
2
/ 1. We also assume, to be checked later, that
h
m
h
0
R a. The essential idea is that the small gap h
m
controls the leakage ux out of
the dimple, which is nearly stagnant with a uniform pressure p
d
:
p
d
= p
0
+
_
2
a

2
h
_
A vertical force balance on the bubble gives
R
2
(p
d
p
0
) =
4
3
a
3
g
Thus, R aBo
1/2
(Bo 1) and
2
h 2/a R. In the dimple, p
d
is constant, so that
1
r

r
_
r
h
r
_
= Constant
Solving this dierential equation directly
h = k
1
r
2
+k
2
+k
3
ln r = h
0
_
1
r
2
R
2
_
where k
3
was set to zero since the ln term is singular at the origin. The trapped volume in the
dimple will then be (see the example sheet ).
check this
V h
0
R
2
Free surface lubrication theory (neglecting gravity) gives the leakage ux as
q =
h
3
3
(h
xx
)
x
The drain rate is clearly just

V = 2Rq. We then need to match the two regimes (inside the
dimple, and the rest of the drop).
Similarity Solution (Quasi-steady state)
We rst look for similarity scalings
h
x
h
0
Only h
0
changing in time
q
h
4
x
3

h
0
t
Mass Conservation
h
x
2
1
So,
h
0
t
1/4
h t
1/2
x t
1/4
q t
5/4
We then dene the similarity variable and height prole
=
x
Bt
1/4
h(x, t) = At
1/2
H()
with suitable choices of A and B. We then obtain (from q = h
3
(h
xx
)
x
/2)
H
3
H

= 1 H

( ) = 1
We need to be careful about the number of boundary conditions needed for a general nonlinear
equation. In fact, the problem has a unique solution to within translations. We nd that
H
1
2
C
2
C = 1.2098
The unscaled condition h
xx
= 1/a xes all of the remaining constants (and allows the determi-
nation of q and thus the fall rate).
33
Example: Gravity Currents on an Inclined Plane
We modify the z balance in the thin lm equations to
p
z
= g cos p = g cos (h(x, y, z) z)
and the x balance to

2
u
z
2
=
p
x
g sin
We can calculate u and q through mass conservation
h
t
+
g sin
3
h
3
x
=
g cos
3
(h
3
h)
So that we now have an evolution equation for the prole. Many similarity solutions exist for
various geometries (see the example sheet and handouts).
ll this in
Similarity Solutions
Consider
At

f
_

x
Bt

_
We recognize that

x
(At

f()) =
At

Bt

()
_
dx(At

f()) = Bt

_
At

f()d

t
(At

f()) =
At

t
(f f

())
Thus, operations look like natural scaling laws.
A Spherical Heat Pulse in 3D
The governing equation for heat diusion is
T
t
=
1
R
2

r
_
r
2
T
r
_
4
_
Tr
2
dr = Q
Looking at the various scalings
T
t

T
r
2
Tr
3
Q r (t)
1/2
T
Q
(t)
3/2
We then the similarity solution
T =
Q
(t)
3/2
f
_

r
(t)
1/2
_
in the governing equations

3
2
f
1
2
f

=
1

2
(
2
f

4
_
f
2
d = 1
34
The rst equation can be simplied and integrated

3
2

2
f
1
2

3
f

=
1
2
(
3
f)

= (
2
f

1
2

3
f =
2
f

+c =
2
f

f = Ae

2
/4
Where c = 0 since f must be zero at innity. We can apply the integral condition to nd that
A is
A =
1
(2)
3/2
4.2 Lubrication Theory for Pigs and Slugs
Rigid particle in a tube
Consider a rigid tube with radius a, lled with liquid of viscosity . Within this tube place a
close-tting, axisymmetric, force-free particle (a pig in a tube).
We apply a pressure gradient G = dP/dx, and try to nd the steady state translation speed.
We assume a separation of length scales a a L. We use scaling methods to estimate p/L
pa
2
aL
u
a
Shear stress balance
p
L

u
a
2
p
L

u

2
a
2
From Stokes eqn and lubrication (p
x
u
yy
)
We thus have two estimates for p/L, but they dier by a factor of . In order to resolve
this paradox, we expand the pressure term in powers of
p(x) = p
0
(x) +p
1
(x) +. . .
We can suppose that p
0
u/
2
a
2
but insist that [p
0
]
L
0
= 0, and let
u = u
0
+u
1
+. . .
Near the pig: If we consider ows in the thin gap between the pig and the wall
At leading order in we have

2
u
0
y
2
=
dp
0
dx
u
0
(0) = U u
0
(h) = 0
u
0
=
1
2
dp
0
dx
y(y h) U(1 y/h)
Integrating to nd the leakage ux and solving for the pressure gradient
Q
0
= 2a
_
h
0
u
0
dy
Q
0
2a
=
(h)
3
12
dp
0
dx
U
h
2
dp
0
dx
=
6

3
h
3
_
Q
0
a
+Uh
_
35
We can solve for Q
0
through the constraint [p
0
]
L
0
= 0
0 = [p
0
]
L
0
=
_
L
0
dp
0
dx
dx Q
0
= au
_
L
0
h
2
dx
_
L
0
h
3
dx
Far from the pig: This region must be supplying ux Q
0
, so we just need to nd the pressure
gradient G = p
x
to give this ux. The ow here is governed by the Stokes equation
G =
1
r
d
dr
_
r
du
dr
_
u(r = a) = U
The solution and volume ux are then
u =
G
4
(a
2
r
2
) U
Q
0
= 2
_
a
0
rudr = a
2
U +
G
8
a
2
Equating the uxes allows the pressure gradient to be calculated
G =
8
a
2
_
Q
0
a
2
+U
_
=
8U
a
2
_
1

a
_
L
0
h
2
dx
_
L
0
h
3
dx
_
Transition region: The transition region is the area in front of the pig, on the order of a. The
transition region poses a signicant challege; next to the pig simple lubrication theory is needed,
far from the pig Poisseuille ow if sucient, but in the transition reigon the full Stokes equaion
is required. We rst investigate its importance through scaling arguments before attempting a
full solution. In the transition region
p p
s
p
s
a

U
a
2
p
s

U
a
The ratio p
s
to p
0
is then
p
s
p
0
=
U/a
UL/
2
a
2


2
a
L
1
So, the transition region doesnt really matter for calculating the pressure drop. Finally, to
calculate p we use a force balance (correct to O())
pa
2
= 2a
_
L
0

u
0
y

y=h
dx =
2
a
_
L
0
_
2
h
+
3Q
0
ah
2
_
dx
So, the solution for p looks like
36
Long Bubble in a Tube (Slug)
This problem was rst considered by (Bretherton, JFM 1961). Consider a tube of radius a
containing uid of viscosity and a large bubble of negligible viscosity. The bubble is big, in
that the volume is larger than 4/3a
3
.
Suppose that U is given and further that the capillary number is small (strong surface tension)
Ca
U

1
We wish to address two questions: the shape of the bubble and the pressure gradient. For the
pig problem, the pressure gradient was
G =
8
a
2
_
U +
Q
a
2
_
This result was independent of the details of the pig derived from the Stokes ow in the far eld.
We thus need to calculate Q and h(x). Because surface tension is strong (Ca 1), we take as
an ansatz that the bubble has two hemispherical caps joined together by a cylinder. There will
be a jump in the curvature where these two caps meet.
Where is constant the pressure in the lm is uniform, meaning that there is no pressure gradient
and thus no driven ow (there will only be plug ow driven by the wall). This is true everywhere
except in the nose and tail regions, where the curvature changes (from 1/ to 2/). We then
seek a scaling solution for the size of the nose and tail.
Width
Change in Thickness
Change in Curvature
1
a

2
Thus, will scale like (a)
1/2
. The change in which drives a ow in the gap.
/a

2
aCa
2/3
a
aCa
1/3
a
Shape of the nose:
37
The pressure in the lm will be
p = const h

p
x
= h

noting that the pressure is higher inside the bubble than out. The velocity prole and volume
ux will then be
u =
1
2
dp
dx
y(y 2h) U
Q
2a
=
_
h
0
udy = Uh
1
3

3
h
3

_
dp
dx
= h

_
= Uh +

3

3
h
3
h

In the cylindrical part of the bubble, the curvature is constant and h = 1 and thus the ux will
be
Q = 2aU
Substituting this into the ux through the nose region
h 1 =

2
3Ca
h
3
h

h(x ) 1 h(x )
1
2
x
2
a
where the outer solution is just a spherical cap. Choosing a similarity variable
= x
_
3Ca

3
_
1/3
reveals the Landau-Levich equation
h
3
h

= h 1 h( ) = 1 h( )
1
2

a
_
1
3Ca
_
2/3

2
For large negative , we can linearize by letting h = 1 +H and then
H = H

H = Ae

, e
/2
_
Acos

3
2
+Bsin

3
2

_
, x
for negative . The only sensible solution is
h = 1 +e

0
where
0
is an arbitrary shift of the origin. We nd that when 1,
h
1
2
(0.643)
2
= 1.337Ca
2/3
a
Tail region
We use the same rescaling and uid mechanics so the lm thickness is again governed by the
Landau-Levich equation. Only the boundary conditoins change so
h( ) 1
h( )

2a
_
1
3Ca
_
2/3

2
Again, we linearize by letting
h = 1 +H H

= H
Now we must choose the solution
h = 1 +Bcos

3
2
(
1
) e
(
1
)/2
38
where
1
is an arbitrary choice of the origin. B must be chosen to give the correct curvature as
. The necessary pressure gradient is then
G =
8
a
2
_
U +
Q
a
2
_
Q = 2aU
=
8U
a
2
_
1
2
a
_
=
8U
a
2
_
1 2.7Ca
2/3
_
4.3 Long Thin Flows II: Extensional Flows
Imagine a long thin viscous thread, falling under gravity (e.g. a thread of honey).
We assume it is long and thin, so that
z
1/L and L a. There should be no tangential
stress on the thread (inviscid air), so that

w
r
= 0
w = w(z, t)
Using local mass conservation,
[ u = 0]
1
r

r
(ru) =
w
z
The solution is then
u =
r
2
w
z
u
r
=
1
2
w
z
If we apply mass conservation to a slice
A
t
+

z
(Aw) = 0
DA
Dt
= A
w
z
which shows that there is a thinning due to a stretching of the ow. We next consider an axial
stress balance on a slice
where f is a generic body force. The axial force balance is then

z
(A
zz
) +p
ext
A
z
+fA = 0
We then need to determine
zz
, through a stress balance

zz
= p + 2
w
z
39
but we dont yet know the internal pressure. We can obtain this from a radial stress balance
p
ext
= p + 2
_
u
r
=
1
2
w
z
_

zz
= p
ext
+ 3
w
z
where 3 is often called the Trouton viscosity. Returning to the axial stress balance and substi-
tuting in this term for
zz
3
1
A

z
_
A
w
z
_

p
ext
z
+f = 0
There are severeal possible variants to this model
1. We could include surface tension by adding a /a term when considering the radial stress
balance.
2. We could consider a unidirectional extension of a thin sheet, similar to above but with

x
u =
z
w and h instead of A. The trouton viscosity would then be 4.
3. We could include inertia by adding Dw/Dt to the RHS of the axial stress balance
4. We could add a lateral force, but this changes the thread shape quickly since there is little
resistance to lateral motion.
4.4 Long Thin Flows III: The Two Fluid Case
Previous examples raise the question: when are boundary conditions eectively rigid or stress-
free?
Example
Consider a gravity driven curent along the interface between two uids.
After a long time, H/R 1 so that the current is thin and lubrication style arguments hold.
We also assume that the pressure within the drop is hydrostatic. In the static case, the pressure
within the drop is only a function of z
p = p
0

1
gh
1
+
0
g(h
1
z)
= p
0
+
2
gh
2

0
g(h
2
+z)
Thus
h
1
=

2

1
h
2
This is reminiscent of a oating iceberg. We can obtain the same result by considering a vertical
force balance. The dynamics are dominated by a radial pressure gradient that drives radial ow.
p
r
= (
2

0
)g
h
2
r
= (
2

0
)

1
g
h
r
= g
h
r
(
2

0
)

1
40
The rate of spreading is determined by a balance between
r
p and the viscous dissipation. We
can estimate the dissipation through scaling arguments. For the case when R H
Kinematics U R/t
Volume consdervation R
2
H V
0
a
3
Radial dynamics (spreading force)
p
r
a
3
g
a
3
H
R
Viscous Dissipation
s
U
L
s
A
s


s
RA
s
L
s
t
where
s
is the relevant viscousity for the dissipation mechanism, L
s
is the lengthscale asso-
ciated with the shear, and A
s
is the area over which the dominant shear stress acts. The radial
balance is then
rhog
a
3
H
R


s
RA
s
L
s
t
Suppose the viscosity is the same everywhere (
1
=
2
= ). We can introduce the dimensionless
number

ga

t =

s

R
2
/a
2
L
s
H
A
which is a radial balance and with volume conservation H a
3
/R
2
. These give equations for
H() and R(). There are only two dimensionless variables left , H/R. There are four regimes
of interest for various values of these parameters
Internal Strain ( R/H), Toee Stretch Regimes
We imagine the extenal uid is inviscid and the internal uid is in extensional ow. The relevant
scales are

s
= L
s
R A
s
RH

R
2
a
2
R
a

_

_
1/2
External Sheer (H/R R/H)
Here the internal uid is still approximately in plug ow, but is resisted by external shear. The
external uid sees a disc of radial stokeslets, similar to a series of sliding plates.

s
= L
s
= R A
s
= R
2
41

R
2
a
2
R
a
3
/R
2

_
R
a
_
5

1/5
=
R
a
The transition from the rst regime to the second can be found by setting the two terms for
equal to each other
R
2
a
2
R
H

R
2
a
2

R
H
Internal Shear (H/[Rlog R/H] < H/R)
In this case there is a Poiseulle component of the internal ow, but the external uid is no long
dragged along.

s
L
s
H A
s
R
2
R
a

_

_
1/8
Nose Push ( H/Rlog[R/H])
At very low viscosities the interior uid can more along the current without a large pressure
gradient, so the only resistance is having to push the external uid out of the way. The external
uid then sees an average force directed outwards

s
= A
s
HR L
s
H ln[R/H]

_
R
a
_
1
log R/a
4.5 Slender Body Theory
Consider the case of a solid, possibly exible body in a Stokes ow [eg a agellum or thread in
a viscous uid).
42
The body has thickness 2d(s), a centreline X(s), and a characteristic length 2l such that d(s)/l
1. We also assume that the radius of curvature of the centreline is much larger than d(s).
We want to know what the resistance will be to a prescribed motion of the centreline V =

X,
in some background ow u

(x). Recall that for an incompressible body we can write


u(y) = u

(y) +
_
s
J(y x) f(x)ds J =
1
8
_
I
r
+
xx
r
3
_
Where J is the Oseen tensor and the distribution of Stokeslets f(x) is as yet unknown.
Because the body is slender (long, thin), we can approximate the ow by placing the stokeslets
on the centreline rather than on the surface. Then
u(y) = u

(y) +
_
l
l
J[y x(s, t)] f(s)ds
We have the boundary conditions that
u(x

) = V
for x

S. We can use this to determine f(s)


Now J 1/r as v 0, so that the dominant contribution to the integral will come from near
the point s

.
v = u

(x

) +
1
8
_
l
l
_
I
r
+
xx
r
3
_
f(s

)ds
= u

(x

) +
f(s

)
8
_
l
l
_
I
r
3
+
(s s

)
2
p(s

)p(s

)
r
3
_
Using the fact that
_
l
l
dx

d
2
+x
2
= 2 log
_
l
d
__
1 +
d
2
l
2
+ 1
__
= 2 log
l
d
+O(1)
and integrating by parts on the second term
v = u

) +
f(s

)
8
_
I +p(s

)p(s

)
_
2 log
l
d
f(s

) = (v u

(x

))
4
log l/d
(I +ff)
1
= (v u

(x

))
4
log l/d
_
I
1
2
pp
_
1
Thus,
f(s

) =
4
log l/d
_
I
1
2
pp
_
(v u

(x

))
Example: Translation of a Rigid Rod
A rod in this framework is just a slender curve with constant tangent vector p = p
0
. The
background ow in this case is u

= 0. There are two cases of interest.


43
1. Broadside motion: In thiscase, v p
0
so that
f(s

) =
4
log l/d
V
The resistance to motion is
F =
_
l
l
f(s

)ds

=
8l
log l/d
v
which is suggestive of the fact that low Reynolds number motion is only weakly dependent
on the shape (think of a moving sphere).
2. Lengthwise Motion: In this case (v p
0
)p
0
= v so that
f(s

) =
4
log l/d
1
2
v =
2
log l/d
v
The resistance to motion is
F =
4l
log l/d
v
We note several things
1. For a given force F, the lengthwise motion is only twice as fast as the broadside motion.
2. The drag scales with l and not d (apart from log l/d), this is reminiscent of the result for
a sphere with length l.
3. There are errors of size 1/(log l/d)
2
in the calculation. In practice these errors are big, so
if we wanted 1% accuracy from the rst term in the expansion we would need l/d e
10
.
4. There are also O(1) errors in v(x

) at the ends of the rod. These are unimportant when


calculated F.
4.6 Elastohydrodynamics
Recall that the displacement u

in an elastic solid satises Naviers equation

2
u

t
2
= F + (

ij
=

( u

)
ij
+ 2

ij
e

ij
=
1
2
_
u

i
x
j
+
u

j
x
i
_
We note that u

is a displacement not a velocity eld and that elastic solids are usually com-
pressible ( u

= 0).
Here

and

are called the Lame coecients with dimensions of Pa (pressure).

is
commonly known as the shear modulus. Consider the form of
ij
when
u

=
_
_
y
0
0
_
_
F = 0 =

2
u

t
2
Some typical values of

and

are

(GPa)

(GPa)
Cartilage 3 10
5
9 10
5
Rubber 0.04 0.03
Steel 100 78
In this case, were interested in slow viscous ows over elastic boundaries, usually quasi-static,
so we can neglect inertial terms. Were left with

= F
Noting that in this form Naviers equation is very similar to Stokes equation. In particular,
we can use many of the same techniques (Papkovich-Neuber potentials, developed for elasticity
theory, complex potentials, and the Airy-stress functions is similar to the concept of stream
functions).
44
4.6.1 The Beam equation
We would like to describe the deformation of thin beams and rods (both beams and rods can
support shear forces N(x) and tensions T(x), versus a string that can only have a tension). For
convenience, we dene N to act normal to the x-axis.
We assume that the beam follows the curve y 0 before displacement and has the prole W(x)
after displacement. We start by assuming small displaceements W

(x) 1. Using a horizontal


force balance
dT
dx
= 0
For the vertical force balance
T
_
dW
dx

x+x

dw
dx

x
_
+
dN
dx
x +q(x)x = 0
T
d
2
w
dx
2
+
dN
dx
+q = 0
We then need to relate the shear forces N to the moment M. To do so, we take moments about
X.
0 = q
1
2
x
2
+N(x +x)x M(x +x) M(x) N =
dM
dx
Next, we relate M to W through Naviers equation. Doing so requires a fairly non-intuitive
calculation.
Relating M and W
Consider a cylindrical rod bending into an arc of a circle of radius R.
We assume that there is a neutral surface that is unstretched by the bending. The surface at z
from the centreline is stretched from l to l +l, so
l +l
R +z
=
1
R
l
l
=
z
r
We also assume that the shress through thickness is
= E
l
l
E =

+ 2

where E is Youngs modulus. Thus


M =
_
zdA =
E
R
_
z
2
dA
B
R
B = E
_
y
2
dA
45
where B is the Bending Stiness. For small amplitudes
1
R
=
d
2
w
dx
2
M = B
d
2
w
dx
2
Returning to the beam equation
B
d
4
w
dx
4
+T
d
2
w
dx
2
+q = 0
As an example we can show that B = Ea
4
/4 for a rod with radius a.
4.6.2 Example: Thin Plate
Consider a thin plate of thickness h. We make three simplifying assumptions
1. Stresses normal to hte midplace of the plate arising from loading are assumed small com-
pared to stresses in the plane of the surface.
2. Points which lie on the normal to the undened midplane lie on the same normal to the
deformed midplane (deformations are locally a rotation).
3. The midplane is a neutral plane (i.e. does not stretch under deformation).
Now, if we use the rst assumption and the assumption of thinness, we get

zz
=

xz
=

yz
= 0 Plane Stress State
For a full justication, see Section 6.3 of (Howell/Kozyre/Ockendon, CUP 2009). In particular,
the result that
zz
= 0 together with Naviers equation implies
u

=
2

+ 2

xx
+

yy
)
Here, we neglect deformations in the y direction (
yy
= 0), and therefor

xx
= 4

__

xx
=
u

x
_
The second assumption impliues that deformation is locally a rotation, so that
u

= sin =
dw
dx

xx
= 4

d
2
w
dx
2
M =
_
h/2
h/2
(
xx
)d =
h
3
12
4

_
d
2
w
dx
2
Finally,
M = B
d
2
w
dx
2
B =
h
3
3

where B is then the bending stiness per unit width, and is dierent from the rod case. We then
obtain the beam equation from before
B
d
4
w
dx
4
+T
d
2
w
dx
2
+q = 0
A few notes:
1. Accounting for deformations in the y direction and assuming the tension is isotropic we
can get
B
4
w +T
2
w +q = 0
2. Non-zero tension can only arise from boundary forces. Other methods of generating tension
generally involve stretching the midplane and thus do not t into the assumptions for this
model.
46
4.6.3 Boundary Conditions
There are three sets of boundary conditions are commonly used
1. Clamped Ends: w = 0, w
x
= 0 at a point (or boundary)
2. Hinged Ends: w = 0, w
xx
= 0 assuming the moment is zero at a point.
3. Free Ends: w
xx
= 0, w
xxx
= 0 (M, N = 0). This also implies no tension.
We can also get point forces. Image that q = F(x x

). Integrating the beam equation will


yield
F = B
_
d
3
w
dx
3
_
x
+

4.6.4 Example: Compression of a rod of length l, radiusr


What values of P = T allow for nontrivial solutions? Assuming there is no external forcing
(i.e. q = 0, the beam equation reduces to
B
d
4
w
dx
4
+P
d
2
w
dx
2
= 0
Hinged Ends
The boundary conditions are
w(0) = 0 = w
xx
(0) = w(l) = w
xx
(l) = 0
The general solution is
w = A
1
+A
2
x +A
3
cos
_
_
P
B
x
_
+A
4
sin
_
_
P
B
x
_
Application of the boundary conditions forces A
1
= A
3
= 0. If A
2
= 0,
_
P/Bl = n and thus
there are nontrivial solutions (P = Bn
2

2
/l
2
for n = 1, 2, . . . ). The threshold for the lowest
mode is then
P =

2
B
l
2
which is the critical required force.
Clamped Ends
As an excercise show that P = 4
2
B/l
2
.
take this
from biologi-
cal hw
Machins Problem (J. Exp. Bio. 1958, Machin, Wiggens/Golstein PRL 1998)
Consider a very simple model of how sperm agella could operate: consider a long slender rod
of length l, diameter d, with one free end and the other end subject to an oscillating gradient.
47
We assume a shape from the beam equation
B

4
w
x
4
+T

2
w
x
2
+q = 0
B

4
w
x
4

w
t
= 0
B

4
w
x
4

4
log(l/d)
w
t
= 0
Where from from slender body theory, and B = Er
4
/4 from elasticity theory. We look for
solutions of the form
w(x, t) = Re[e
it
g(x)]
and thus get the dierential equation
Bg

= ig
Moving to dimensionless variables
l

=
_
B

_
=
x
l

G =
g
l

L =
l
l

= iG G(0) = 0 G

(0) = f G

(L) = G

(L) = 0
The general solution to this is
G() =
3

j=0
A
j
exp(
j
)
j
= exp
_
i
8
+
ij
2
_
We could determine the coecients A
j
for any length l, but here we only consider the simpler
limites of very short and very long laments.
1. L 1 (l l

): Expand the exponentials in powers of for small and nd that G() = f


at leading order
2. L 1 (l l

): We need
j
with the negative real part so
G() = A(exp[1 c +is)] exp[(s +ic)]) c = cos(pi/8) s = sin(/8)
Since G

(0) = f,
A =
1
2
f[s c i(s +c)]
To prevent lateral motion, we must apply a normal force at x = 0 of magnitude F = N =
Bw
xxx
(0). The component of this force in the x direction is
Fw
x
(0) = Bw
xxx
(0)w
x
(0)
Now,
G

(0) =
_
0 L 1
fe
i/4
L 1
w
xxx
(0, t) =
_
0 L 1
f
l
2
(cos t cos /4 sin t cos /4) L 1
Since w
x
(0, t) = f cos t, we get
< F
x
(0) >=
_
B
f
l
2
1

2
f
1
2
=
B
l
2
f
2
2
3/2
L 1
0 L 1
This ties in with reversibility, as L 1 the rod behaves like a straight lament and the motion
is not reversible, while when L 1 there is a nonreversible traveling wave.
Unfortunately, this does not give the agellum shape that is observed in experiments (a
continuous wave), so Machin concluded that the agellum must be active, rather than passive
elastic laments (i.e. not driven by the attachment to the cell). Therefore, the agella must
generate moments along its length (and this has been conrmed experimentally.
48
4.6.5 Winkler Elastic Foundation Model
Consider a thin, compressible layer of a linear elastic material bonded to a rigid substrate.
This is applicable to a thin layer of cartilage attached to a bone in a joint. Imagine applying
some pressure distribution p(x) to the surface of the elastic layer (z = 0), e.g. some king of
viscous ow over the surface layer.
We assume that p(x) has a typical value p and acts over some horizontal lengthscale l
p
, so
we want to know the deformation within the layer in response to the pressure. We assume that
1. h
l
l
p
(the elastic layer is thin)
2. w

h
l
where w

is a typical vertical displacement (small deformations).


We recall that Naviers equation for the displacement eld u

= (u

, w

), in a linearly elastic
isotropic solid.
= 0
ij
=

( u)
ij
+ 2

1
2
_
u

i
x
j
+
u

j
x
i
_
We take the vertical component of Naviers equation with the assumptions h?
l
l
p
and w

h
l
0 = (

+ 2

2
w

z
2
+ (

2
u

xz
+

2
w
z
2
= O
_
w

h
2
l
_
+O
_
u

h
l
l
p
_
+O
_
w

l
2
p
_
Usiung these order of magnitude estimates, the thinness assumption and compressibility we see
that the leading order term in the small parameter h
l
/l
p
is given by the term containing
2
w

/z
2
so

2
w
z
2
= 0 w

= A(x)z +B(x)
Since the base is rigid (w

(z = h
l
) = 0)
w

(x, z) = A(x)[z +h
l
]
So, by continuity of the vertical traction at the surface (z = 0)
p(x) = n ( n)
z=0
For small deformations the normal is approximately vertical so
p(x) = n ( n)
z=0
= (

+ 2

)
w

Solving for w
w(x, z) = p(x)
h
l

+ 2

_
z
h
l
+ 1
_
The real quantity of interest is the displacement of the free surface, so setting z = 0 we nd that
w(x, 0) =
h
l

+ 2

p(x) =
w
p(x)
which is basically an elastic spring (and
w
isthe Winkler stiness compliance). Since w(x, 0)
p(x), the layer behaves like a matress of linearly elastic springs.
49
4.6.6 Example: Falling Cylinder Onto a Winkler Substrate (Skotheim+Mahadevan
PRL, 2004)
Consider a rigid cylinder translating parallel and close to a Winkler substrate in a viscous ow
We still assume a rigid body below the elastic layer and no slip conditions on the surface. When
there is no elastic layer, we expect no vertical/normal force from the reversibility argument for
Stokes ow. However, the elasticity breaks reversibility in this case, as well see. In steady state
dQ
dx
= 0
Lubrication theory in a uid (Reynolds equation) tells us
0 =
d
dx
_
h
3
12
dp
dx
+
1
2
Uh
_
Moving to dimensionless variables
H
h
h
0
X
x
(2h
0
R)
1/2
P
p
U(2R)
1/2
/h
3/2
0
Reynolds equation becomes
d
dx
_
H
3
P
dx
+ 6H
_
= 0
The gap thickness is given by
H = 1 +x
2
+P
U(2R)
1/2
h
5/2
0

w
We can now eliminate H and the dimensionless Reynolds equation becomes an equation just for
P. We need two BCs for P, names p() = 0. Were interested in the limit of a small amount
of elasticity (small deformations) so that 1 and
w
is small. A power expansion in is then
appropriate for P.
P(X) = P
0
(X) +P
1
(X) +. . .
At O(
0
dP
0
dX
=
A
(1 +X
2
)
3

6
(1 +X
2
)
2
P
0
= A
_
3
8
arctan X +
X
8
(5 + 3X
2
(1 +X
2
)
2
_
6
_
1
2
arctan X +
X
2(1 +X)
2
_
+A

50
Through the BC, A

= 0 and A = 8, so the arctan terms cancel. Finally


P
0
=
2X
(1 +X
2
)
2
As expected, it integrates to zero so there are no net normal forces (consistent with viscoelasticity)
and is antisymmetric. At the rst order in ,
dP
1
dX
=
B
(1 +X
2
)
3
+
24X(X
2
1)
(1 +X
2
)
6
P
1
(X) =
3
5
(3 5X
2
)
(1 +X
2
)
5
Since B = B

= 0 from the BC. To calculate the magnitude of the normal force, we integrate
over the pressure.
F
_

(P
0
+P
1
+. . . )dx =
3
8
+O(
2
)
Where the P
0
term cancels from antisymmetry. Thus, the presence of the elastic layer gives rise
to a normal force, which would not exist without the elastic layer present. Using numerics we
can calculate P in terms of without the assumption of small . The result is
51
52
Chapter 5
Handouts
53
5.1 Summary of Fluid Mechanics
54
55
56
57
58
59
5.2 Integral Representations of Stokes Flow
60
61
5.3 2D Singularities and Flow Past a Cylinder
62
63
5.4 The Greens Function for Stokes Flow
64
5.5 Two Solutions in Lubrication Theory
65
5.6 Extensional Flow Example
66
67
68
5.7 Small Deformations of a Viscous Drop by Uniform
Pure Strain E
69
5.8 Example: Swimming Flagella
70

You might also like