Chapter 3

Slow viscous flow

Contents
3.1 Stokes flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Translating sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Lubrication flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Hele-Shaw flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.1 Stokes flow

For flows where the Reynolds number:
ρU D
Re =  1,
µ
we can neglect the term ρDu/Dt in the Navier–Stokes equation so that the equations governing
the fluid flow become the Stokes equation and the incompressibility condition:
µ∇2 u = ∇p, (3.1)
∇ · u = 0. (3.2)
In the case of gravity driven flows, equation (3.1) is replaced by:
µ∇2 u = ∇P − ρg. (3.3)
Unlike the Navier–Stokes equation, the Stokes equations are linear, which means that we can
construct solutions using the principle of linear superposition. Furthermore, since there are
no time derivatives the solutions are time reversible under the reflection (u, t) → −(u, t) (see
demonstrations by G.I. Taylor http://www.youtube.com/watch?v=51-6QCJTAjU (t=13:13)).

31
32 3.2 Translating sphere

This is bad news for small water-living creatures as it leads to the scallop theorem: if the
swimming motion of a micro-organism is time reversible, it produces no net forward motion.
Consequently, micro-organisms use wave motions along their flagella for propulsion, rather
than the side-to-side motion employed by fish.
Videos: https://www.youtube.com/watch?v=NBH3UvlZo90, https://www.youtube.com/watch?v=LShQiec4t

3.2 Translating sphere

Let us consider the Stokes flow generated by a sphere of radius a moving at speed U . Using
spherical polar coordinates centred on the sphere, the velocity of the sphere is given by
U ex = (U cos θ, −U sin θ, 0), where θ represents the angle between the first vector of the
coordinate frame and the direction of motion.

The potential flow solution for a sphere of radius a moving at speed U is given in spherical
polar coordinates by:  3
a3

a
u = ∇φ = U 3 cos θ, −U 3 sin θ, 0 .
r 2r
However, although this satisfies the boundary condition for the r component of velocity, u, it
does not satisfy the boundary condition for v. This solution also had the unrealistic property
that the drag force on the sphere was identically zero: the D’Alembert paradox.
The flow is two-dimensional, so we may assume that the velocity is of the form:

u = (u(r, θ), v(r, θ), 0) ,

so that ∇ · u = 0 becomes:
1 ∂ 1 ∂
r2 u +


2
(v sin θ) = 0, (3.4)
r ∂r r sin θ ∂θ
while equation (3.1) reduces (or expands) to:
     
µ ∂ 2 ∂u 1 ∂ ∂u ∂v ∂p
2
r + sin θ − 2u − 2 − 2v cot θ − = 0, (3.5)
r ∂r ∂r sin θ ∂θ ∂θ ∂θ ∂r
     
µ ∂ 2 ∂v 1 ∂ ∂v ∂u v 1 ∂p
r + sin θ + 2 − − = 0. (3.6)
r2 ∂r ∂r sin θ ∂θ ∂θ ∂θ sin2 θ r ∂θ

Given the boundary conditions, it is natural to seek a solution for the velocity of the form:

u = f (r) cos θ, v = g(r) sin θ,

where f (r) and g(r) satisfy f (a) = U , g(a) = −U and f and g both tend to zero as r → ∞.
Substituting into equation (3.4) gives:

r df
f+ + g = 0. (3.7)
2 dr
Chapter 3 – Slow viscous flow 33

Careful inspection of equation (3.5) shows that all the terms coming from the fluid velocity
are proportional to cos θ, which suggests that the pressure is of the form:

p(r, θ) = p0 + h(r) cos θ,

where p0 is a constant. Substituting into equations (3.5) and (3.6), we obtain:

   
µ d 2 df dh
r − 4f − 4g − = 0, (3.8)
r2 dr dr dr
   
µ d 2 dg
r − 2g − 2f + h = 0. (3.9)
r dr dr
Thus, we have three coupled linear ODE for f , g and h. Eliminating h by differentiating
equation (3.9), adding equation (3.8) and substituting for g from equation (3.7), we obtain:

d4 f 8 d3 f 8 d2 f 8 df
4
+ 3
+ 2 2
− 3 = 0. (3.10)
dr r dr r dr r dr
This is a fourth order Cauchy equation and has solutions of the form rm where m are the
roots of:
m4 + 2m3 − 5m2 − 6m = 0,
that is, m = −3, −1, 0 and 2. The general solution for f is of the form:
C D
f = Ar2 + B + + 3. (3.11)
r r
The boundary conditions as r → ∞ implies that we must have A = B = 0, hence f is of the
form:
C D
f= + 3.
r r
Substituting into equation (3.7):
r df C D
g = −f − = − + 3.
2 dr 2r 2r
Hence, applying the wall boundary conditions (f = U , g = −U at r = a), we obtain:

3aU U a3
C= , D=− ,
2 2
that is:
a3
 
3a
u= − U cos θ, (3.12)
2r 2r3
a3
 
3a
v=− + U sin θ. (3.13)
4r 4r3
(3.14)

Finally, from equation (3.9), we obtain h = 3µU a/2r2 , giving:

3µU a
p = p0 + cos θ. (3.15)
2r2
From this solution, we can calculate the drag force acting on the sphere:
Z Z
F = − f dS = − n · τ dS.
S S
34 3.2 Translating sphere

Note that the sign is negative because this is the force applied by the fluid onto the boundary.
This force is directed along the axis, so we need only consider the component in the direction
(cos θ, − sin θ, 0). The normal to the surface is −er , so:
Z
FD = (τrr cos θ − τrθ sin θ)dS.
S

To find τrr and τrθ , we need the components of the strain-rate tensor Err and Erθ , which, in
spherical polar coordinates, are given by:
 
∂u 1 ∂  v  1 ∂u
Err = , Erθ = r + .
∂r 2 ∂r r r ∂θ

Remember:
τrr = −p + µErr , τrθ = 2µErθ .
Hence, for r = a:
3U
Err = 0, Erθ = sin θ,
4a
and therefore:
3µU 3µU 3µU
τrr cos θ − τrθ sin θ = −p0 cos θ − cos2 θ − sin2 θ = −p0 cos θ − .
2a 2a 2a
The term p0 cos θ integrates to zero over the surface of the sphere, leaving:

3µU
FD = − S,
2a

where S = 4πa2 is the surface area of the sphere. We obtain:

FD = −6πµaU, (3.16)

which is the Stokes drag force on a sphere, a classical result in fluid dynamics.
If the sphere is falling due to gravity, then the associated force is:

4πa3
FW = − ∆ρg,
3
where ∆ρ is the difference in density between the material of the sphere and the surrounding
fluid, giving a fall speed for the sphere equal to:

2∆ρga2
U= .
9µ

It is tempting to try to perform the same calculation in two dimensions for a moving cylinder,
however, the solution fails because, the equivalent general solution to equation (3.11) is:

D
f = Ar2 + B + C log(r) + .
r2
and it is not possible to apply the boundary conditions at r = a and r → ∞ simultaneously.
This result is known as Stokes paradox and was explained by Oseen who mentioned that the
result was derived for small Reynolds numbers and that, in the case of the cylinder, large
distances and therefore large Reynolds numbers were involved.
Chapter 3 – Slow viscous flow 35

3.3 Lubrication flows

In many practical applications, the fluid flows in thin films where the boundaries are nearly
parallel. These flows are “nearly” unidirectional and we can exploit this property to find
approximate solutions.

3.3.1 The slider bearing flow

Consider a slider bearing where the fluid flows in the gap between the surface y = 0 moving
at velocity (U, 0, 0) and a fixed block at y = h(x) of length L, where the two ends are at equal
pressure p0 . We consider the case where the gap between the block and the moving surface
varies slowly with x, so that |dh/dx|  1. Specifically, we wish to consider the case:
(d2 − d1 )
h(x) = d1 + x,
L
where |h0 | = |d2 − d1 |/L  1. Since this is a two dimensional problem, we shall neglect z so
that the velocity:
u = (u(x, y), v(x, y)),
satisfies the following equations:
∂u ∂v
+ = 0, (3.17)
∂x ∂y
 2
∂ u ∂2u

∂p
µ + 2 = , (3.18)
∂x2 ∂y ∂x
 2
∂2v

∂ v ∂p
µ 2
+ 2
= , (3.19)
∂x ∂y ∂y
together with the boundary conditions: u(x, 0) = U , u(x, h(x)) = 0 and v(x, 0) = v(x, h(x)) =
0.

Constant gap
For a constant gap, h0 = 0, we would set the x derivatives of the velocity to zero, so that:
dv
= 0,
dy
d2 u ∂p
µ 2 = ,
dy ∂x
d2 v ∂p
µ 2 = .
dy ∂y
From the first of these equations, we deduce that v(y) = 0. The third equation gives ∂p/∂y = 0
which implies that p is only a function of x:
d2 u dp
µ = = −G.
dy 2 dx
36 3.3 Lubrication flows

Hence, the pressure is given by

p = p0 − Gx,
but, p(L) = p0 implies G = 0 and hence:

d2 u
= 0.
dy 2

The boundary conditions yields: u = U (h − y)/h.

Variable gap
Let us now consider h0  1 and, for simplicity, rescale the equations (3.17) to (3.19) by
choosing appropriate scales for variation in x and y. The natural choice for y is the average
gap, h̄ = 21 (d1 + d2 ), so we shall define:

y = h̄y ∗ .

However, x derivatives only arise from changes in h so the appropriate scale is h̄/h0 :

h̄ ∗
x= x .
h0
We can additionally write the velocity and pressure in the form:

µU ∗ ∗ ∗
u(x, y) = U u∗ (x∗ , y ∗ ), v(x, y) = U v ∗ (x∗ , y ∗ ), p(x, y) = p (x , y ).
h̄
Substituting these scalings into equations (3.17)–(3.19), we obtain:

∂u∗ ∂v ∗
h0 + = 0, (3.20)
∂x∗ ∂y ∗
∂ 2 u∗ ∂ 2 u∗ ∂p∗
h02 ∗2 + ∗2 = h0 ∗ , (3.21)
∂x ∂y ∂x
2
∂ v ∗ 2
∂ v ∗ ∂p∗
h02 ∗2 + ∗2 = . (3.22)
∂x ∂y ∂y ∗

The boundary conditions on u∗ and v ∗ become u∗ (x∗ , 0) = 1, u∗ (x∗ , h∗ (x∗ )) = 0 and v ∗ (x∗ , 0) =
v ∗ (x∗ , h∗ (x∗ )) = 0, where h(x) = h̄h∗ (x∗ ). Noting that |h0 |  1, we seek solutions in the form:

u∗ = u∗0 + h0 u∗1 + . . . .

Substituting in equation (3.20), we find:

∂v ∗ ∗
0 ∂u0
= −h + ....
∂y ∗ ∂x∗

We thus write:
v ∗ = h0 v1∗ + h02 v2∗ + . . . . (3.23)
To find the size of the leading order pressure term, we need to look at equations (3.21) and
(3.22). From equation (3.21), we have:

∂p∗ ∂ 2 u∗0
h0 = + ...,
∂x∗ ∂y ∗2
Chapter 3 – Slow viscous flow 37

and from equation (3.22):

∂p∗ 2 ∗
0 ∂ v1
= h + ....
∂y ∗ ∂y ∗2
The pressure p∗ can be as large as 1/h0 and thus takes the form:

1 ∗
p∗ = p + .... (3.24)
h0 −1
Let us now consider the leading order terms in h0 in each of the equations:

∂u∗0 ∂v1∗
+ = 0, (3.25)
∂x∗ ∂y ∗
∂ 2 u∗0 ∂p∗−1
= , (3.26)
∂y ∗2 ∂x∗
∂p∗−1
= 0. (3.27)
∂y ∗

The last two equations are analogous to those we found for h0 = 0 the only change is that u∗0
is function of both x∗ and y ∗ . Since the leading order gives ∂p∗−1 /∂y ∗ = 0, we can integrate
equation (3.26) to obtain:

dp∗−1 y ∗2
u∗0 = + A(x∗ )y ∗ + B(x∗ ),
dx∗ 2
and applying the boundary conditions, we obtain:

1 dp∗−1 ∗ ∗ y∗
u∗0 = y (y − h∗
) + 1 − .
2 dx∗ h∗
To find v1∗ , we integrate equation (3.25):
y∗
∂u∗0 ∗
Z
v1∗ =− dy
0 ∂x∗
1 d2 p∗−1 ∗2 1 dh∗ 1 dp∗−1
 
∗ ∗ 1
= y (3h − 2y ) + − ∗2 y ∗2 .
12 dx∗2 2 dx∗ 2 dx ∗ h

Hence, applying the boundary condition (v ∗ = 0 at y ∗ = h∗ ), we obtain:

1 d2 p∗−1 ∗3 1 dp∗−1 ∗2 dh∗ 1 dh∗

h + h = ,
12 dx∗2 4 dx∗ dx∗ 2 dx∗
which can be rewritten as: ∗ 
dh∗

d ∗3 dp−1
h = 6 .
dx∗ dx∗ dx∗
This equation is called the Reynolds equation. Upon integrating, we obtain:

dp∗−1 A∗
 
1
=6 + ,
dx∗ h∗2 h∗3

for some constant A∗ .

Now, let us go back to the original unstarred variables. Recall that:
 
µU 1 ∗
p= p + ... ,
h̄ h0 −1
38 3.3 Lubrication flows

giving:
1 dp∗−1 dx∗
 
dp µU
= + ··· ,
dx h̄ h0 dx∗ dx
so the leading order reads:
dp 6µU
≈ 3 (h + A) , (3.28)
dx h
where A = h̄A∗ .
For the specific case of the slider bearing h(x) = d1 + (d2 − d1 )/Lx:

A L
 
3µU L 2
p(L) = p0 − +
d2 − d1 h h2 0
3µU L
= p0 + 2 2 [2d1 d2 + A(d1 + d2 )] ,
d2 d1

and hence, to get p0 = p(L), it follows that A = −2d1 d2 /(d1 + d2 ), and:

6µU (d2 − d1 ) x
 
dp d1
= − .
dx h3 L d1 + d2

To leading order, the fluid velocity reads:

3U (d2 − d1 ) x
 
d1  y
u(x, y) = − y(y − h) + U 1 − .
h3 L d1 + d2 h

Note that if d1 > d2 then p(x) > p0 within the bearing and so there is a positive upward force
on the bearing, whereas if d2 > d1 the force is negative.

3.3.2 Alternative method for the slider bearing flow

In the above, we used a formal expansion to find the dominant terms in the governing equa-
tions by scaling and non-dimensionalising the equations. However, with sufficient experience
it is possible to recognise the dominant terms in each equation without needing to rescale
everything. Let us return to the original dimensional equations:

∂u ∂v
+ = 0,
∂x ∂y
 2
∂ u ∂2u

∂p
µ 2
+ 2 = ,
∂x ∂y ∂x
 2
∂2v

∂ v ∂p
µ + = ,
∂x2 ∂y 2 ∂y

and seek out the largest terms in each equation by estimating the size of each term. Let us
denote U the size of u, V the size of v and P the size of p. We also need estimates for the
size of derivatives. Since variations in y occur over the gap h, let us denote ∂ /∂y as being
of size 1/h, and since h0 = dh/dx, we can estimate x derivatives as being of size h0 /h. Let us
now look at the terms in equation (3.17):

∂u ∂v
+ = 0,
∂x ∂y
U h0 V
h h
Chapter 3 – Slow viscous flow 39

Since these two terms have to balance to yield a two-dimensional flow, we can deduce:
V ∼ U h0 . (3.29)
Notice that this is equivalent to the scaling we deduced in equation (3.23). Equation (3.18)
has three terms,
 2
∂2u

∂ u ∂p
µ + = .
∂x2 ∂y 2 ∂x
U h02 U P h0
µ 2 µ 2
h h h
Since |h0 |  1, the first term is small compared with the second term, so we can deduce the
scale for pressure:
U
P ∼ µ 0, (3.30)
hh
the same scaling found in equation (3.24). Finally, equation (3.19) also has three terms:
 2
∂2v

∂ v ∂p
µ 2
+ 2
= .
∂x ∂y ∂y
V h02 V P
µ 2 µ 2
h h h
We find that the scale of ∂p/∂y is larger than the terms on the left-hand side by a factor
1/h02 . Note that equation (3.19) yields a smaller scale for P .
By neglecting all but the dominant terms in equations (3.17)–(3.19), we obtain:
∂u ∂v
+ = 0, (3.31)
∂x ∂y
∂2u ∂p
µ 2 ' , (3.32)
∂y ∂x
∂p
' 0. (3.33)
∂y
Equation (3.33) shows that the pressure is approximately independent of y and so, integrating
equation (3.32) and applying the boundary conditions at y = 0 and y = h(x), we get:
1 dp Uy
u= y(y − h) + U − ,
2µ dx h
so that:
1 d2 p
 
∂v ∂u 1 dp U dh
=− =− y(y − h) + − 2 y.
∂y ∂x 2µ dx2 2µ dx h dx
Upon integration and application of the boundary condition (v = 0 at y = 0), we get:
1 d2 p 2
 
1 dh 1 dp U
v= y (3h − 2y) + − y2.
12µ dx2 2 dx 2µ dx h2
The boundary condition (v = 0 on y = h) yields the Reynolds equation:
h3 d2 p h2 dh dp U dh
2
+ = .
12µ dx 4µ dx dx 2 dx
Integrating once returns:
dp 6µU
= 3 (h + A) ,
dx h
as found in equation (3.28) above.
40 3.3 Lubrication flows

3.3.3 Cylinder approaching a wall

Consider the motion of a cylinder towards a wall in the limit where the minimum gap d is
small compared to the radius of the cylinder a.

We use Cartesian coordinates with the origin located at the position on the wall nearest to
the cylinder with the y axis directed normal to the wall towards the cylinder and x directed
in the plane of the wall perpendicular to the cylinder axis. The velocity takes the form:

u = (u(x, y), v(x, y)).

In these coordinates, the position of the cylinder surface is given by:

p
h(x) = d + a − a2 − x2 .

However, we are interested in the flow in the region |x|  a, so:

x2
h(x) ≈ d + .
2a
Note that dh/dx = x/a is small provided that |x|  a. Let us choose the minimum gap d as
the scale for y. The scale for x is less obvious. Let us write:

x2
 
h(x) = d 1 + .
2ad
√
We can see that variations in h are felt for x ∼ ad, so we shall rescale x and y as follows:
√ d ∗
x= adx∗ = x , y = dy ∗ ,


d/a  1. With this scaling, h(x) = dh∗ (x∗ ), where:

p
where  =

x∗2
h∗ (x∗ ) = 1 + . (3.34)
2
The governing equations are:

∂u ∂v
+ = 0, (3.35)
∂x ∂y
 2
∂ u ∂2u

∂p
µ 2
+ 2 = , (3.36)
∂x ∂y ∂x
 2
∂2v

∂ v ∂p
µ 2
+ 2 = , (3.37)
∂x ∂y ∂y

with boundary conditions: u(x, 0) = v(x, 0) = 0, u(x, h(x)) = 0 and v(x, h(x)) = −V .
Chapter 3 – Slow viscous flow 41

The velocity v being of size V , we can expand:

v(x, y) = V (v0 (x∗ , y ∗ ) + v1 (x∗ , y ∗ ) + . . . ) ,

and therefore from conservation of mass:

∂u V
∼ ,
∂x d
so that u must be of size −1 V :

u(x, y) = V −1 u−1 (x∗ , y ∗ ) + . . . .

Thus, the flow mainly goes in the x direction. Turning to equation (3.36), we have:

∂p ∂2u ∂2u V ∂ 2 u∗−1

=µ 2 +µ 2 =µ 2 + ...,
∂x ∂y ∂x d  ∂y ∗2

and hence:
µV
.p∼
d2
Therefore, the pressure can be expanded in the following fashion:

µV −2 ∗
 p−2 (x∗ , y ∗ ) + . . . .


p(x, y) =
d
Substituting into equations (3.35) to (3.37) and keeping only the leading order terms in , we
have:
∂u∗−1 ∂v0∗
+ ∗ = 0, (3.38)
∂x∗ ∂y
∂ u∗−1
2 ∂p∗−2
= , (3.39)
∂y ∗2 ∂x∗
∂p∗−2
= 0, (3.40)
∂y ∗

with boundary conditions u∗−1 (x∗ , 0) = u∗−1 (x∗ , h∗ ) = 0, v0∗ (x∗ , 0) = 0 and v0∗ (x∗ , h∗ ) = −1.
As before, ∂p∗−2 /∂y ∗ = 0, so that the solution of equation (3.39), satisfying the boundary
conditions, is given by:
1 dp∗−2 ∗ ∗
u∗−1 = y (y − h∗ ). (3.41)
2 dx∗
From equation (3.38), we get:

1 d2 p∗−2 ∗2 1 dp∗−2 dh∗ ∗2

v0∗ = y (3h∗
− 2y ∗
) + y .
12 dx∗2 4 dx∗ dx∗
Applying the boundary condition at y ∗ = h∗ , we obtain:

h∗3 d2 p∗−2 h∗2 dp∗−2 dh∗

−1 = + .
12 dx∗2 4 dx∗ dx∗
The associated Reynolds equation is:
 ∗ 
d ∗3 dp−2
h = −12. (3.42)
dx∗ dx∗
42 3.3 Lubrication flows

Integrating once, we find:

dp∗−2 x∗ A∗
= −12 + .
dx∗ h∗3 h∗3
Since this flow is symmetric about x = 0, dp∗−2 /dx∗ = 0 at x∗ = 0 and so A∗ = 0. It follows:
6x∗ ∗ ∗
u∗−1 = − y (y − h∗ ).
h∗3
Recall from equation (3.34) that dh∗ /dx∗ = x∗ which yields:
6
p∗−2 (x∗ ) = p∗∞ + ,
h∗2
where p∗∞ is the dimensionless pressure at infinity.
Having found the pressure, we can now calculate the force that the cylinder exerts on the
fluid. This force is oriented in the y direction and its amplitude given by:
Z a
Fy = L fy dx,
−a

where L is the length of the cylinder and f = n · τ . Here:

1 
0

n= √ − h , 1 ,
1 + h0 2
where h0 = ∂h/∂x. For |x|  a, n ≈ (−h0 , 1) and hence:
 
dh du dv dv
fy = −p − µ + + 2µ .
dx dy dx dy
Since the pressure is of size −2 µV /d, it constitutes the dominant contribution to fy :
Z a
Fy = −L (p(x) − p∞ )dx.
−a

Changing variables to x∗ = d−1 x = −1 a−1 x, we get:

Z −1
−2 µV L 
p∗−2 (x∗ ) − p∗∞ d−1 dx∗


Fy = −
d −−1
Z −1
1
= −−3 6µV L x∗2 2
dx∗ .
−−1 (1 + 2 )
√
To perform this integral, we substitute x∗ = 2 tan u and obtain:
√ Z π
2
Fy = −−3 6 2µV L cos4 udu
2
−π
√
9 2  a 3/2
=− πµV L .
4 d
If the cylinder is falling towards the wall under a constant force F = −M g:
4M gd3/2 d
V = √ = − d,
9 2πµLa3/2 dt
and therefore:
1 1 2M g
p =√ + √ t,
d(t) d0 9 2πµLa3/2
where d0 is the gap at t = 0. The cylinder will not contact the wall in finite time.
Chapter 3 – Slow viscous flow 43

3.3.4 Free surface flows

Let us now consider the gravitational spreading of a blob of viscous fluid on a surface.

For simplicity, we shall consider a two-dimensional blob whose height is given by y = h(x, t).
In order to use lubrication theory, we shall assume that |dh/dx  1| so that the normal to
the surface is approximately (−h0 , 1). At leading order, the boundary conditions at y = h are:
 
∂v ∂u ∂v
−P + 2µ = −Patm , µ + = 0,
∂y ∂y ∂x

together with u = v = 0 at y = 0. The kinematic boundary condition at the free surface

requires that:
Dh ∂h ∂h
v= = +u . (3.43)
Dt ∂t ∂x
The Stokes equations read:

∇P = ρg + µ∇2 u,
∇ · u = 0,

and reduce to:

 2
∂ u ∂2u

∂P
=µ + 2 , (3.44)
∂x ∂x2 ∂y
 2
∂2v

∂P ∂ v
= −ρg + µ + , (3.45)
∂y ∂x2 ∂y 2
∂u ∂v
0= + . (3.46)
∂x ∂y

Rather than scale these equations, let us look for the dominant terms. We shall use U and V
to denote the size of the velocity components u and v respectively, and take 1/h as the size
of ∂ /∂y, and since h0 = dh/dx we can estimate x derivatives as being of size h0 /h.
Since this flow is driven by gravity, P and ρg must balance so that P ∼ ρgh. Equation (3.44)
gives:

∂2u ∂2u
 
∂P
=µ + 2 .
∂x ∂x2 ∂y
µU 02 µU
h0 ρg h
h2 h2
Hence, we can estimate that:
ρgh2 0
U∼ h,
µ
and neglect the x derivatives of u so that:

∂2u 1 ∂P
≈ .
∂y 2 µ ∂x
44 3.3 Lubrication flows

Furthermore, from equation (3.46):

ρgh2 02
V ∼ h0 U ∼ h .
µ

Hence, the sizes of the terms in equation (3.45) are:

∂2v ∂2v
 
∂P
= −ρg + µ + .
∂y ∂x2 ∂y 2
ρg ρg h04 ρg h02 ρg

Equation (3.45) reduces to:

∂P
= −ρg,
∂y
so that
P = −ρgy + A(x).

Furthermore, since |dv/dy| ∼ h02 (ρgh)/µ, we can also neglect this term in the boundary
condition. As P = Patm at y = h:

P = Patm + ρg(h − y), (3.47)

and therefore ∂P/∂x = ρg∂h/∂x. It follows:

∂2u ρg ∂h
2
≈ .
∂y µ ∂x

Again since |dv/dx| ∼ h03 (ρgh)/µ, we can neglect its contribution to the boundary conditions
and the boundary conditions on u simplify to u = 0 on y = 0 and du/dy ≈ 0 on y = h. The
leading order solution for |h0 |  1 is:

ρg ∂h
u=− y(2h − y). (3.48)
2µ ∂x

From equation (3.46):

"  2 #
ρg ∂ 2 h 2 ∂h
v= 2
y (3h − y) + 3 y2 ,
6µ ∂x ∂x

and so, at y = h:
"  2 #
ρg ∂2h 3 ∂h
v= 2 h +3 h2 .
6µ ∂x2 ∂x

Substituting into the kinematic boundary condition gives:

"  2 #
ρg ∂ 2 h 3
 
∂h ∂h 2 ρg ∂ 3 ∂h
= h + 3 h = h . (3.49)
∂t 3µ ∂x2 ∂x 3µ ∂x ∂x

This is a non-linear diffusion equation for h(x, t).

Chapter 3 – Slow viscous flow 45

3.4 Hele-Shaw flow

Let us consider now the case of the general flow in the gap between two parallel plates
separated by a distance h.

At the boundaries z = 0 and z = h, we impose no slip boundary conditions: u = 0. We

assume that the variations of u in the (x, y) plane are slow so that ∇2 u ≈ ∂ 2 u/∂z 2 .
The Stokes equations reduce to:
∂p ∂2u
≈ µ 2,
∂x ∂z
∂p ∂2v
≈ µ 2,
∂y ∂z
∂p
≈ 0,
∂z
and the mass conservation equation reads:
∂u ∂v ∂w
+ + = 0.
∂x ∂y ∂z
The velocity in the (x, y) plane is given by:
1 ∂p
u=− z(h − z),
2µ ∂x
1 ∂p
v=− z(h − z).
2µ ∂y
Integrating over z, we obtain the average velocity in the (x, y) plane:

1 ∂p 1 h h2 ∂p
Z
ū = − z(h − z)dz = − ,
2µ ∂x h 0 12µ ∂x
1 ∂p 1 h h2 ∂p
Z
v̄ = − z(h − z)dz = − ,
2µ ∂y h 0 12µ ∂y

which we can write as:

h2
ū = − ∇H p, (3.50)
12µ
where ∇H is the restriction of the gradient operator to the (x, y) plane.
Hence, the two-dimensional flow in the (x, y) plane is a potential flow with the velocity
potential Φ = −h2 /12µp. Furthermore, integrating the continuity equation between 0 and h,
we obtain:
∂ ū ∂v̄
+ = 0,
∂x ∂y
and so this flow is incompressible and hence:

∇H · ū = 0. (3.51)
46 3.4 Hele-Shaw flow