Version date: March 15, 2010 1

2 Flows without inertia

These notes are heavily based on lectures delivered by Keith Moffatt in DAMTP. Much
of the material should be found in any sufficiently mathematical textbook on the subject.
The books by Batchelor and Acheson are good references for the material in this and later
sections of the course, and some precise references are given where appropriate.

A small number of proofs are included here, but they are non-examinable. All non-
examinable material is indicated by being surrounded by brackets: [** · · · **].

2.1 Introduction
Recall, from previous courses, the momentum equation

Dui ∂
ρ = ρFi + σij (1)
Dt ∂xj

due to Cauchy, where ui is the ith component of the fluid velocity, and σij is the stress
tensor. For a Newtonian fluid we propose that the stress tensor is linearly related to the
velocity gradient tensor ∂ui /∂xj . Assuming that the fluid is isotropic, and proposing the
most general relation between σij and ∂ui /∂xj we are lead to the constitutive relation for
an incompressible Newtonian fluid:

σij = −pδij + 2µeij (2)

where µ is the shear viscosity of the fluid and p is the pressure. Inserting (2) into (1)
leads to the Navier–Stokes equation:
 
∂u
ρ + u · ∇u = ρF − ∇p + µ∇2 u . (3)
∂t

When µ = 0 this reverts to the Euler equation already seen, but the limit µ → 0 is not
a straightforward one: immediately we can see that this involves ignoring the highest
derivatives of u and therefore it is a singular perturbation.

2.2 Orders of magnitude of terms

Let L be a typical length scale for a given flow situation. This is not uniquely defined, but
usually there are sensible candidates: consider for example flow past a cylinder of cross-
sectional radius L, or a sphere of radius L, or flow in a channel of width L. Similarly,
let U be a typical velocity for the flow, usually derived from that imposed at a boundary
surface. Let T be a typical timescale (often it happens that T is of the same magnitude
as L/U , but this is not always the case).

Example: Oscillating sphere

Consider a rigid sphere of radius a whose centre moves at a velocity U0 ez cos ωt, sur-
rounded by fluid at rest at infinity. Appropriate choices are then L = a, U = U0 and
2 Version date: March 15, 2010

T = 2π/ω. Then, considering the orders of magnitude of the terms in (3) we have
 
 ∂u 
  ∼ U
 ∂t  T
U2
|u · ∇u| ∼
L
 2 
ν∇ u ∼ νU
L2
Now we can investigate the relative importance of these terms. Firstly,

|u · ∇u| UL
∼ (4)
|ν∇2 u| ν

defines the Reynolds number Re ≡ U L/ν which measures the relative importance of the
nonlinear and viscous terms. Secondly,
 ∂u 
 
∂t L2
∼ (5)
|ν∇2 u| νT

defines the Stokes number S ≡ L2 /(νT ) which is the ratio of the unsteady to the viscous
terms. Note that if T ∼ L/U then S ∼ U L/ν and so S and Re are equivalent.
If Re  1 and S  1 then we should be able to neglect the material derivative term
on the LHS of (3) which would yield the Stokes equations (in the absence of body forces):

1
0 = − ∇p + ν∇2 u (6)
ρ
0 = ∇·u (7)

Solutions of the Stokes equations will be focus of this chapter.

2.3 Poiseuille Flow

[see Batchelor pages 181-183 for a discussion of several cases including this one.]

There is a special case in which the inertia terms are not approximately small, but
identically zero. This arises for unidirectional flow where the velocity is a function only
of a second coordinate.
Consider a steady pressure-driven flow between two planes y = ±b. Assume that the
velocity field is u = (u(y), 0, 0).

Then a simple calculation shows that both parts of Du/Dt are identically zero. Therefore
such a flow must satisfy

0 = −∇p + µ∇2 u

i.e.
∂p d2 u ∂p ∂p
= µ , = =0
∂x dy 2 ∂y ∂z
Version date: March 15, 2010 3

So p = p(x) but ∂p/∂x is a function of y which is a contradiction unless ∂p/∂x =

constant = −G, say. Then
d2 u
= −G/µ
dy 2
and, requiring no slip on y = ±b (i.e. u = 0 on y = ±b) implies
G 2
u(y) = − (y − b2 )
2µ
which is a parabolic velocity profile. The total volume flux (per unit length in the z-
direction is given by
Z b
2Gb3
Q = u(y)dy = .
−b 3µ
In the case that the streamlines are curved rather than straight (as here) we can see that
u · ∇u ∼ κu2 where κ is the curvature of the streamlines. If the curvature is small in this
case then we may still be able to use the Stokes equations as an approximation to the
flow.

2.4 Properties of the Stokes Equations

[Batchelor, section 4.8, pages 227-228]

Consider a volume V containing fluid, bounded by a surface S on which the velocity

field is prescribed: u = U(x) on S.

Remark: Linearity.

The Stokes equations (6) - (7) are linear: if {u1 , p1 } and {u2 , p2 } are any two solutions,
then {λ1 u1 + λ2 u2 , λ1 p1 + λ2 p2 } is also a solution (and satisfies the linearly combined
boundary conditions on all boundaries).

Remark: Reversibility.

The Stokes equations are reversible. This is the case λ1 = −1, λ2 = 0 of the remark above,
in fact. Reversing the boundary conditions implies that the flow is reversed everywhere
and is still a solution.

Theorem 1 (Solutions to the Stokes Equations are unique) . Suppose that {u(1) , p(1) }
and {u(2) , p(2) } are two solutions satisfying the same boundary condition, i.e. u(1) = U(x)
on S and u(2) = U(x) on S. Then u(1) = u(2) and p(1) − p(2) = const everywhere in V .

[** Proof: Consider the velocity field ũ = u(2) − u(1) , so that ũ = 0 on S. Similarly, let
(2) (1) (2) (1)
p̃ = p(2) − p(1) , ẽij = eij − eij and σ̃ij = σij − σij . Then from the consitutive law we
have

σ̃ij = p̃δij + 2µẽij

and the Stokes equations imply ∂/∂xj (σ̃ij ) = 0. Now, consider

Z Z
2µ ẽij ẽij dV = ẽij σ̃ij dV since ẽkk = 0
V
ZV
∂
= (ũi )σ̃ij dV using σ̃ij = σ̃ji
V ∂x j
Z
∂
= − ũi (σ̃ij ) dV using ũ = 0 on S
V ∂x j
= 0.
4 Version date: March 15, 2010

(1) (2)
Hence ẽij ≡ − in V , and so eij = eij .
Now, we can also compute, using ũ = 0 on S, that
Z Z
|ω̃|2 dV = 2 ẽij ẽij dV
V V

and therefore this quantity is also zero, implying that ω (1) ≡ ω (2) everywhere in V .
(1) (2)
Putting these two results together implies that ∂/∂xj (ui − ui ) = 0 everywhere in V ,
and, since the velocity fields coincide on S they must therefore be equal everywhere in V .
From this it follows also that the pressure fields can differ by at most a constant. 2 **]

Theorem 2 (The Minimum Dissipation Theorem) . Let u(1) be the unique Stokes
flow satisfying u(1) = U(x) on S. Let u(2) be any kinematically possible flow, i.e. one
that merely satisfies ∇ · u(2) = 0 in V and u(2) = U(x) on S. Then
Z Z
(1) (1) (2) (2)
eij eij dV ≤ eij eij dV
V V

with equality only if u(2) ≡ u(1) . In other words, the Stokes flow has a smaller rate
of viscous energy dissipation Φ than any other flow in V (recall the definition of Φ =
R
2µ V eij eij dV ).

[** Proof: Consider the velocity field ũ = u1 − u2 as before. Then we can compute that
Z Z Z Z
(1) (1) ∂ ũj (1) ∂ (1)
2µ ẽij eij dV = ẽij σij dV = σij dV = − ũj σij dV = 0
V V V ∂x i V ∂xi

Hence
Z Z  
(2) (2) (1) (1) (2) (1)
eij eij − eij eij dV = ẽij eij + eij dV
V
ZV Z
(1)
= ẽij ẽij dV + 2 ẽij eij dV ≥ 0,
V V

with equality if and only if ẽij = 0, i.e. if and only if u(1) = u(2) . 2 **]

Theorem 3 (Unsteady Stokes flow relaxes monotonically to steady Stokes flow)

. Consider a time-dependent solution u(x, t) to the unsteady Stokes equation

∂u
ρ = −∇p + µ∇2 u
∂t
subject to the steady boundary condition u(x, t) = U(x) on S. Then the kinetic energy
of the flow decreases monotonically to its minimum value which is attained when the flow
becomes steady Stokes flow.

[** Proof: Consider the rate of change in time of the viscous energy dissipation:
Z Z
d ∂
eij eij dV = 2 eij eij dV
dt V ∂t
ZV
1 ∂
= σij eij dV since eii = 0 ∀ t
µ V ∂t
Z  
1 ∂ ∂ui
= σij dV note that σij = σji
µ V ∂t ∂xj
Z
1 ∂2
= σij σik dV (8)
µρ V ∂xj ∂xk
Version date: March 15, 2010 5

substituting in the Stokes equation again. Now suppose that we maintain the steady
boundary condition u(x, t) = U(x) on S for all t, so that ∂u/∂t = 0 on S, i.e.
∂/∂xk σik = 0 on S, then, integrating (8) by parts we obtain
Z Z   
d 1 ∂ ∂
eij eij dV = − σij σik dV ≤ 0.
dt V µρ V ∂xj ∂xk
so that Φ decreases monotonically to its minimum value, obtained when ∂σij /∂xj ≡ 0 in
V. 2 **]

2.5 2D Stokes flow

Recall that in 2D the incompressibility condition ∇ · u = 0 may always be satisfied by
introducing a streamfunction ψ(x, y) such that

u = ∇ × (ez ψ) = (∂ψ/∂y, −∂ψ/∂x, 0).

Recall also that

ω = (0, 0, −∇2 ψ).

Therefore, taking the curl of (6) we obtain

∇2 ω ≡ ∇2 ∇2 ψ = 0

which is the biharmonic equation, often written ∇4 ψ = 0.

2.6 The paint-scraper problem

(a) (b)
(a) Paint scraper in the frame where the wall is stationary and the scraper moves at velocity U .
(b) In the frame where the scraper in stationary and the wall moves at a velocity −U .

Choosing a frame of reference in which the scraper (the line θ = α) is steady and the
wall (at θ = 0) moves with velocity −U in the x-direction we have to solve ∇4 ψ = 0 in
the wedge 0 ≤ θ ≤ α subject to the boundary conditions ur = −U , uθ = 0 on θ = 0 and
ur = uθ = 0 on the line θ = α.
From dimensional analysis we see that ψ has dimensions of length × velocity, and
given that there is only one variable with each of these units available, i.e. r and U
respectively, we propose a solution in the (variable-separable) form ψ(r, θ) = rU f (θ)
since θ is dimensionless. Now we reduce the biharmonic equation to an ODE for f (θ) and
apply the boundary conditions.
First, it is convenient to introduce the notation F (θ) = f + f 00 since
U U
∇2 ψ = (f + f 00 ) = F (θ)
r r
So then
U
∇4 ψ = (F + F 00 ) = 0
r3
6 Version date: March 15, 2010

which implies

F (θ) = A0 cos θ + B 0 sin θ

for some constants A0 and B 0 . Then we find the general solution for f :

f (θ) = A cos θ + B sin θ + Cθ cos θ + Dθ sin θ.

The radial velocity boundary condition ur = (1/r)∂ψ/∂θ ≡ U f 0 (θ) on θ = 0, α implies

f 0 (0) = −1, f 0 (α) = 0.

Two further boundary conditions are given by the fact that we require the lines θ = 0, α
to be streamlines on which ψ = 0, since ψ = 0 at r = 0. Therefore

f (0) = 0, f (α) = 0.

Applying these four boundary conditions determines the four constants of integration
A, B, C, D:

α(α − θ) sin θ − θ sin(α − θ) sin α

f (θ) = .
sin2 α − α2
It is instructive also to find the pressure distribution in the fluid. From integrating the
Stokes equation (6) this is found to be
 
2µU α sin θ + sin α sin(α − θ)
p = p0 +
r α2 − sin2 α

from which we see that p → ∞ as r → 0, so an infinite force is needed to keep the scraper
in contact with the wall.

2.7 Flow in a corner generated by remote forcing

[Acheson section 7.3, pages 229-232]

In this section we will investigate the generation of steady flow in a corner, between
rigid boundaries fixed to be at θ = ±α. We suppose that there is some remote device,
e.g. a rotating cylinder, that drives the flow, see sketch.

We ask for the structure of the flow near the corner, i.e. at small r.
Similar to the previous problem, we guess that there is a separable solution for the
streamfunction, and take ψ(r, θ) = rλ f (θ) - taking the leading-order term in the r-
dependence. Then, defining F (θ) = λ2 f + f 00 in a manner also similar to the previous
Version date: March 15, 2010 7

subsection, we have
∇2 ψ = (λ2 f + f 00 )rλ−2 ≡ rλ−2 F (θ)
⇒ ∇4 ψ = (F 00 + (λ − 2)2 F )rλ−4 = 0
⇒ F (θ) = A0 cos(λ − 2)θ + B 0 sin(λ − 2)θ
⇒ f (θ) = A cos λθ + B sin λθ + C cos(λ − 2)θ + D sin(λ − 2)θ
which has four linearly independent functions as long as λ 6= 1. Now we need to satisfy
the boundary conditions f (±α) = 0 and f 0 (±α) = 0. Symmetry considerations imply
that ur (r, θ) is odd in θ and uθ (r, θ) is even in θ. Hence ψ(r, θ) must be even in θ, and so
B = D = 0. Now we are left with the pair of equations
f (α) = A cos λα + C cos(λ − 2)α = 0
0
f (α) = −λA sin λα − C(λ − 2) sin(λ − 2)α = 0
So, for a nonzero solution for A and C we see that we require
sin 2µα = −µ sin 2α (9)
where µ = λ − 1. There is no obvious solution to this except λ = 1 which is degenerate
(we ruled this case out earlier in the computation in fact). Graphically it is clear that
real solutions to (9) exist as long as α > αc where the critical value αc ≈ 73◦ , see figure.

When α < αc we can continue the search by looking for complex solutions for λ = p+iq.
What does this mean physically? To begin with, linearity means that we are allowed to


consider ψ = 12 rλ f (θ) + c.c. = Re(rλ f (θ)).
We have that
∂ψ

uθ = − = −Re λrλ−1 f (θ)
∂r
λ p+iq
Let λ − 1 = p + iq, so that r = r = rp (cos(q log r) + i sin(q log r)). Then on θ = 0 we
have that
uθ |θ=0 = Crp cos(q log r + δ)
for some constants C and δ. Sketch (a) and (b) below indicate how this function oscillates
infinitely often as r → 0.

(a) (b) (c)

(a) Plot of cos(q log r) against r. (b) Plot of r cos(q log r) against r. (c) Sketch of sequence of
p

eddies in the flow, rotating in alternate directions and (rapidly) decreasing in strength as r → 0.

The sketch in (c) illustrates that the flow describes an infinite sequence of eddies
as r → 0. For example, in the case 2α = π/6 we can compute that 2αp ≈ 4.72 and
2αq ≈ 2.20. In this case the ratio of intensities of successive eddies as r → 0 is ≈ 1/412.
8 Version date: March 15, 2010

2.8 A sphere moving at constant speed

[**

[Acheson, section 7.2, pages 223- 226; Batchelor, section 4.9, pages 230 - 233]

This and the next section summarise a classic problem in Stokes Flow. The material
is included because the result is widely quoted and used, but we won’t rely on it later in
the course. We take a fixed frame of reference F so that the fluid at infinity is at rest,
and the sphere, of radius a is moving at a constant speed U (wlog in the z direction) and
is instantaneously located at the origin r = 0:

We naturally make use of spherical polar coordinates (r, θ, φ) and introduce a ‘Stokes
streamfunction’ ψ(r, θ). This is a like the streamfunction for 2D flow in Cartesian coor-
dinates, but for axisymmetric flows in 3D. It is defined as the function ψ such that

1 ∂ψ 1 ∂ψ
ur = , and uθ = − (10)
r2 sin θ ∂θ r sin θ ∂r
(and uφ = 0). It can be seen straightforwardly (Acheson section 5.5, page 173) that
Dψ/Dt ≡ u · ∇ψ is zero, and hence ψ is constant on streamlines. The velocity field is
computed equivalently from writing
 
ψ
u = ∇ × 0, 0, . (11)
r sin θ

Now we need to specify the problem mathematically: we first discuss appropriate bound-
ary conditions and then the governing equation to solve.

Boundary conditions:

u →)0 as r→∞
ur = U cos θ
on r=a
uθ = −U sin θ

i.e. on r = a we require ψ = 21 U a2 sin2 θ and ∂ψ/∂r = U a sin2 θ. From (10a) we see that
we also require ψ = o(r2 ) as r → ∞, i.e. ψ/r2 → 0 as r → ∞ so that we guarantee that
|u| → 0 as r → ∞.
From (11) we find that

1
ω = ∇ × u = (0, 0, − D2 ψ
r sin θ
Version date: March 15, 2010 9

where
∂2 sin θ ∂ 1 ∂
D2 ≡ 2
+ 2
∂r r ∂θ sin θ ∂θ
is known as the ‘Stokes operator’. Then ∇2 ω = 0 becomes
 
2 1 4
∇ ω = 0, 0, − D ψ
r sin θ

where D4 ψ ≡ D2 (D2 ψ). This is therefore the version of the biharmonic equation that we
are required to solve in the axisymmetric case.
We look for solutions in the separable form

ψ(r, θ) = f (r) sin2 θ

and find that

 
2f
2
D ψ ≡ f − 2 sin2 θ
00
r
= F (r) sin2 θ,

which is convenient, because then

 
2F
D ψ = F − 2 sin2 θ = 0.
4 00
r

We see that the equation F 00 − 2F/r2 = 0 is homogeneous so we expect solutions to be

in the form F (r) = rλ . Substituting this and solving we obtains λ = −1 and λ = 2 as
linearly independent solutions. Therefore

F (r) = Ãr2 + C̃/r

which implies
2f
f 00 − = Ãr2 + C̃/r
r2
and therefore

f (r) = Ar4 + Br2 + Cr + D/r.

The boundary conditions then imply that A = B = 0 and we have f (a) = U a2 /2,
f 0 (a) = U a which yields two linear equations for the coefficients C and D. These have
the unique solution
3 1
C= U a, D = − U a3 .
4 4
And so, finally we have the solution
   
D 2 1 2 3r a
ψ(r, θ) = Cr + sin θ = U a − sin2 θ. (12)
r 4 a r
The term ∝ r dominates for large r and is known as the ‘Stokeslet’ term. The term ∝ 1/r
does not contribute to the vorticity of the flow, and is a ‘dipole’ contribution. The ‘dipole’
term decays much more rapidly at large r.
It is now also easy to compute the pressure field around the sphere. Since the vorticity
ω = (0, 0, ω(r, θ)) is given by

D2 vψ 2C
ω=− = 2 sin θ
r sin θ r
10 Version date: March 15, 2010

we can use the Stokes equation in the form

∂p 4Cµ
− = µ(∇ × ω)r = 3 cos θ
∂r r
and integrating this we obtain
2Cµ
p = p∞ + cos θ (13)
r2

2.9 Computation of the drag force on a sphere

Recall that the force F on the spherical surface is given by
Z
Fi = σij nj dS
S

which, using the divergence theorem on a large but finite volume of fluid confined between
the sphere S and a large sphere S∞ ‘at infinity’, is equal to the expression
Z
Fi = σij nj dS
S∞

because ∂/∂xj (σij ) ≡ 0 in V. Then, as r → ∞, the contribution from the dipole term
vanishes and we need only integrate the ‘Stokelet’ terms:
C
u∼ (2 cos θ, − sin θ, 0)
r
So that the force (acting in the −z direction) is given by
Z
F = |F| = − (σrr cos θ − σrθ sin θ dS. (14)
S∞

We now need to substitute in for the components of the stress tensor, and then for the
pressure from (13):

∂ur 6Cµ
σrr = −p + 2µ = −p∞ − 2 cos θ
∂r
 r 
∂  uθ  1 ∂ur
σrθ = 2µerθ ≡µ r + = 0.
∂r r r ∂θ

We obtain
Z π
1 z}|{ 2
F = 6Cµ cos2 θ 2π r| sin
{zθ dθ}
0 r2
z}|{
where the is from the φ integral and the |{z} is the usual element of surface area in
spherical polar coordinates. So

F = 8πµC = 6πµaU (15)

i.e. the drag force on the sphere is F = −6πaU. This is sometimes known as Stokes Law.
In fact we could have deduced all of it from dimensional analysis apart from the constant
6π. So the integration work above is all in aid of computing that coefficient of 6π.

**]
Version date: March 15, 2010 11

2.10 Flow in a Thin Layer

As previously in this chapter, we continue to assume that inertia is negligible. We consider
fluid in a 2D thin layer 0 < y < h(x, t):

Let h0 be a typical thickness (for example, an average value of h(x, t)). Let L be the scale
of variation of h(x, t) in the x-direction. For example, define
 
 h 
L ∼ min  .
∂h/∂x 

We now make the additional fundamental assumption that h0  L. This assumption

defines what we mean by ‘a think layer’. Therefore we have
   
 ∂u 
    ∂u 
 
 ∂y   ∂x 

since the LHS is O(U/h0 ) and the RHS is O(U/L). Since we know that
 
∂ψ ∂ψ
u= ,− , 0 = (u, v, 0)
∂y ∂x

we must have
   
 ∂ψ 
   ∂ψ  .
 

 ∂y   ∂x 

As we would usually solve ∇4 ψ = 0 this approximation reduces the biharmonic equation

to
∂4ψ
=0
∂y 4

(there are small cross-terms here that we ignore). We can integrate this once to obtain

∂ 3ψ ∂2u
≡ = −G(x, t)/µ
∂y 3 ∂y 2

where G(x, t) is some function of x and t. Therefore we must identify G(x, t) with ∂p/∂x,
a function only of x and t but not of y. So we can integrate twice more with the boundary
conditions u = 0 on y = 0 (a rigid boundary) and u = U on y = h(x, t) - a prescribed
velocity. This yields the ‘locally parabolic’ profile
1 Uy
µu(x, y, t) = − y(y − h)G(x, t) + .
2 h
12 Version date: March 15, 2010

The volume flux of fluid (per unit length in the third dimension) is given by
Z h
G(x, t) h3 Uh
Q(x, t) = u(x, y, t) dy = +
0 2µ 6 2
so the relation between Q(x, t) and h(x, t) is roughly cubic. Incompressibility supplies the
further condition that
∂Q ∂h
=−
∂x ∂t
which can be seen immediately from considering the flow between two points x and x+δx:

This in turn implies that

 
∂h ∂Q ∂ G(x, t) 3 U h
=− = − h +
∂t ∂x ∂x 12µ 2
 
1 ∂ ∂p 1 ∂
= h3 − (U h).
12µ ∂x ∂x 2 ∂x
For a given h(x, t), this equation determines p(x, t) (we hope that that this equation comes
with boundary conditions for p at two locations x = x0 , x = x1 .

Notes: [

] (i) There is a 3D generalisation of this ‘thin film equation’ which is

∂h 1 1
= ∇ · (h3 ∇p) − ∇ · (uh)
∂t 12µ 2
where ∇ ≡ (∂/∂x, ∂/∂y, 0) and u = (u, v, 0), i.e. there is no velocity in the z-direction.

(ii) We can justify the neglect of inertial forces a posteori as follows. Let U0 be a
typical velocity, then
U2
|u · ∇u| ∼ 0
  L
 2   ∂2  U 0
ν∇ u ∼ ν
 ∂z 2 u ∼ ν h2

0
Version date: March 15, 2010 13

so the ratio of these is

|u · ∇u| U0 2
= h = Re ε
2
|ν∇ u| νL 0

where Re = U0 h0 /ν is a Reynolds number and ε = h0 /L  1 is the slow scale of variation

in the x-direction. Then we can safely neglect inertia if the product Re ε  1.

Example 1. A thrust bearing.

Consider a hammer driven by an external force Fe onto an axisymmetric drop of honey

on a rigid table at z = 0:

How can we determine the height h(t) of the fluid drop?

Set u = v = 0 and w = −dh/dt on z = h(t). Then, using axisymmetry, and therefore
plane polars (r, θ):
 
12µ dh 2 1 ∂ ∂p
=∇ p≡ r (16)
h3 dt r ∂r ∂r

While the fluid remains entirely within the gap (which we assume) we suppose it occupies
a region 0 ≤ r ≤ a(t), giving a (constant) fluid volume of V0 = πa2 h. Note that we really
require a  h to use ‘lubrication theory’. Note that the pressure p(r, t) = pa atmospheric
pressure at r = a. We are also neglecting surface tension forces at r = a. Integrating (16)
w.r.t. r twice, we obtain

∂p 12µ dh r2
r = + C(t)
∂r h3 dt 2
∂p 12µ dh r C(t)
⇒ = 3
+
∂r h dt 2 r
6µ dh r2
⇒p = + C(t) log r + D(t).
h3 dt 2
Since p is finite at r = 0 we must have C(t) ≡ 0. Using atmospheric pressure at r = a we
can determine D(t) and hence we have

3µ dh 2
p = pa + (r − a2 ). (17)
h dt
Having computed the pressure in the fluid we can now use this expression to compute the
excess force F exerted by the fluid on the hammer:
Z a
F = p(r, t) − pa 2πr dr
0
Z a
3µ dh
= 2π r3 − a2 r dr
h3 dt 0
3µπ dh 4
= − 3 a (18)
2h dt
where we note that dh/dt < 0.
14 Version date: March 15, 2010

Now, if we neglect the inertia of the hammer (i.e. it is just sitting on top of the fluid
drop) then F = Fe , constant. In this case we may integrate (18) again, after substituting
V0 2


for a4 using the fact that a4 = πh . This gives
 1/4
3µV02 1
h(t) =
8πFe (t − t0 )1/4

where t0 is a constant of integration. Thus the height of the drop decreases very very
slowly ∼ (t − t0 )−1/4 as t → ∞.

The hammer, of course (why?) never makes contact with the table.

Example 2. Gravitational spreading.

In this example we consider how a drop of viscous liquid spreads out axisymmetrically
on a horizontal surface under gravity. This is motivated, for example, by the spreading
of hot lava from a volcano.
The drop lands on the table as a sphere, say (so that the flow is axisymmetric):

...and then ‘very quickly’ becomes a ‘thin film’ occupying the region 0 ≤ r ≤ a(t):

In contrast to Example 1, here the pressure is not constant in the layer, it is hydrostatic:

p = pa + ρg(h − z) (19)
Version date: March 15, 2010 15

where the free surface Γ is given by z = h(r, t). We now have two expressions for ∂p/∂r:
one from differentiating (19) and one from the thin film approximation of the Stokes
equations:
∂p ∂h ∂p ∂2u
= ρg , and =µ 2 (20)
∂r ∂r ∂r ∂z
since u = (u(r, z, t), 0, 0). For this case, since the upper boundary is stress-free, the
appropriate boundary conditions for the velocity are

u=0 on z = 0,
∂u
=0 on z = h.
∂z
Combining the two equations in (20) and integrating twice w.r.t. z we obtain
ρg ∂h
u(r, z, t) = z(z − 2h)
2µ ∂r
which is a ‘semi-parabolic’ profile for the velocity:

The outward radial flux of material is straightforwardly computed to be

Z h
2πρgr 3 ∂h
Q = 2πr u dz = − h (21)
0 3µ ∂r
...and now we can use the final ingredient which is conservation of mass:
∂h ∂Q
2πr =−
∂t ∂r
and, combining this with the expression (21) we obtain the evolution equation for h(r, t):
 
∂h g 1 ∂ ∂h
= rh3 . (22)
∂t 3ν r ∂r ∂r
This equation looks rather like a diffusion equation for h, but with a nonlinearity due
to the factor of h3 . Just like the usual linear heat equation, it has a similarity form of
solution, that is, we look for a solution in the form

h(r, t) = t−α H(η), where η = r/tβ

where the exponents α and β need to be determined. We use the evolution equation (22)
and the (constant) volume condition:
Z a(t)
V = h(r, t)2πr dr = constant (23)
0

to determine α and β, as follows.

We see that the derivatives of the similarity variable η w.r.t. r and t are given by
∂η ∂η βη
= t−β , and =− .
∂r ∂t t
16 Version date: March 15, 2010

Substituting these into (22) we obtain

   
dH −α−1 g −4α−2β d dH
− αH + βη t = t ηH 3 , (24)
dη 3νη dη dη
which must hold for all t, implying

α+1 = 4α + 2β
⇒ 3α + 2β = 1 (25)

From the constant volume condition we have that

Z a/tβ
V = t−α H2πt2β η dη
0

must be independent of t, hence

2β − α = 0. (26)

Solving the two equations (25) and (26) we find the unique solution α = 1/4, β = 1/8.
Therefore a(t) ∝ t1/8 .

Computing the shape H(η) of the drop

We can now recast the evolution equation (24) and the constant-volume condition (23) in
terms of the similarity variable η to solve for the shape of the droplet as it spreads out.
From volume conservation we introduce A as the value of η at which we hit the boundary
of the drop, i.e. the height falls to zero: H(A) = 0. We have
Z A
V = 2π H(η)η dη (27)
0

and hence A = a/t1/8 . The evolution equation (24) becomes

   
1 1 dH g d dH
− H+ η = ηH 3
4 8 dη 3νη dη dη

which is a nonlinear ODE for H(η). It integrates once in the form given, multiplying
through by a factor of η, to yield
dH 3ν 2
ηH 3 + η H = constant
dη 8g
but in fact this constant of integration must be zero since H and dH/dη are both finite
(in fact for a smooth solution we would expect dh/dη = 0) at η = 0. So we can cancel a
factor of ηH and obtain
dH 3ν
H2 + η=0
dη 8g
which can be integrated w.r.t. η to yield
1 3 3ν 1 2 3ν 2
H + η = constant = A
3 8g 2 16g
since A is defined to be the value of η at which H(η) = 0. Therefore the shape of the
drop is
9ν
1/3
H(η) = A2 − η 2 .
16g
Version date: March 15, 2010 17

Further, the drop size A can be directly related to the volume V since
Z A
9ν
1/3
V = 2π A2 − η 2 η dη
0 16g
 1/3
3π 9ν
⇒V = A8/3
4 16g
or equivalently, in dimensional units, we have
 1/8
1/3 t
a(t) = CV (28)
T

where T = ν/(gV 1/3 ) is the characteristic viscous timescale for the spreading under
gravity, and C is a pure number that we have computed via the thin film theory:
 10 1/8
2
C= = 0.77921...
35 π 3
The 1/8th power law relation (28) for the increase in the horizontal size of the drop over
time has been verified experimentally (Huppert, 1982).

2.11 The Hele-Shaw Cell

This is another classic thin-film case, with a surprising link to the potential flows studied
earlier in the course.
A Hele-Shaw cell consists of a pair of glass plates fixed at a small constant separation
(h ≈ 1mm) with fluid trapped between them:

we assume that there is a locally parabolic velocity profile between the rigid upper and
lower boundaries.
We then examine the thin film flow around objects placed in the cell. If we assume
that the scale of the obstacles L is much greater than the plate separation h then the thin
film equations are valid.
1
Recall that the volume flux Q = − 12µ h3 ∇p in the thin film. Therefore, since the film
height h is constant, the average velocity is given by
1 1 2
ū(x, y) = Q=− h ∇p = ∇φ(x, y)
h 12µ
where φ = −h2 /(12µ)p is a 2D potential! So the averaged flow is potential flow, at least
as long as L  h.

For example, flow at velocity U past a circular cylinder r = a in a Hele-Shaw cell

should look like
18 Version date: March 15, 2010

with φ ∼ U x = U r cos θ as r → ∞, and (impermeability) ū · n = 0 on r = a, hence

∂φ/∂r = 0 on r = a. Incompressibility ∇ · ū = 0 implies ∇2 φ = 0 and we have the general
solution
 
a2
φ(r, θ) = U r + cos θ,
r
which we saw earlier in the course, in chapter 1. However, it should be noted that we
cannot satisfy the no-slip boundary condition on r = a:
1 ∂φ
uθ = |r=a = −2U sin θ 6= 0
r ∂θ
so there must be a small region, of width O(h), around the cylinder in which the thin-film
approximations are not valid. This is an example of a boundary layer, and we will see
more of these later in the course.

A further limitation of the Hele-Shaw cell is that we cannot sustain flows with non-
zero circulation κ. Mathematically, such flows contained additional contributions to the
potential φ such as φ = κθ/(2π) which is not single-valued (but this did not matter in
the section on potential flow since the only quantities with physical meaning were the
derivatives of φ). In this context we have φ = −h2 /(12µ)p and so φ is proportional to
the pressure field and has a physical significance. So we cannot have solutions where φ
has a contribution ∼ θ. If we attempt to generate circulation by, for example, rotating
the cylinder, then the circulation is killed off by viscous effects within the boundary layer
around the cylinder, leaving the flow outside this with zero circulation.

Two-fluid flow in the Hele-Shaw cell

When a more viscous fluid advances, displacing a less viscous one, the flat interface
between them is stable to small disturbances. If, however, the more viscous fluid is
receding, then the flat interface is unstable and a ‘fingering instability’ occurs.
Version date: March 15, 2010 19

(a) (b)

In (a) the sketch shows viscous liquid with a pressure field pL (x, y) on the left and air,
within which the pressure pA < pL is constant, on the right. The lines show contours of
constant pL , with the rightmost one being the liquid-air interface at which pL = pA . Since
ū ∝ ∇p in the liquid, the middle section, where the contours are closer together, movs
faster to the right than the top and bottom sections and so the flat interface is stabilised
(the sinusoidal perturbation decays).
In (b) we suppose that pA > pL and the air is pushing the liquid back towards the
left. The fast motion in the liquid is still in the centre and so the sinusoidal perturbation
is amplified and the interface is no longer flat, but contains a (large) deformation in the
centre.