1

Chapter 6 Fluid Mechanics and Convection

1. Introduction
Any material that flows in response to an applied stress deform continuously.
In solids, stresses are related to rates of strains. In fluids, stresses are related to rates
of strains. Strain rates in fluids are a result of gradients in the velocities or rates of
displacement of fluid elements. Velocity gradients are equivalent to strain rate.
Newtonian or linear fluid: the rate of strain or velocity gradient is directly
proportional to the applied stress; the constant proportionality is viscosity.
The physics that governs fluid motions is the principles of mass, momentum, and
energy conservation together with the rheological or constitutive law for the fluid.
In geodynamics, mantle convection is the manifestation of the fluid behavior of
the mantle and is responsible for plate tectonics and continental drift; it plays a
dominant role in determining the thermal structure of the earth.
Thermal convection: when a fluid is heated from within or from below and cooled
from above, thermal convection can occur. The hotter fluid at depth is
gravitationally unstable with respect to the cooler fluid near the upper surface.
Buoyancy forces drive the convective flow. On many scales folding of crustal rocks
can be attributed to the fluid behavior of these rocks. A fluid instability can also
explain the formation of saltdomes due to diapiric upwelling of a buried layer of salt.

2. 1-D channel flows

The movement of the plates over the surface of the surface of the earth represents
a flow of mantle rock from accretionary plate boundaries to subduction zones. To
maintain mass balance, a complementary flow of mantle rock from subduction to
accretionary plate boundaries must occur at depth.
A model for this counter flow is asthenospheric flow confined in a horizontal layer
immediately below the lithosphere. Postglacial rebound data suggest the pressure of
a low-viscosity (of the order of 100 km thick) region beneath the oceanic lithosphere.
Seismic studies show that a region beneath the lithosphere has low seismic velocity,
particularly that shear waves are attenuated. All there observations favor a
low-viscosity region beneath the lithosphere.
p 0 p1
( A : the horizontal length of a
A
section of the channel) or the prescribed motion of one of the walls u 0 .

The flow occurs as a result of pressure gradient

Shear stress (force per unit area), or gradient in velocity, is exerted on the
horizontal planes in the fluid and at the walls. For a Newtonain fluid with constant
viscosity :
du
=
-----<1>
dy
u : velocity of fluid ( m ),
s

: shear stress, tangential stress on the surface normal to y direction (

N
),
m2

dynamic : viscosity SI unit is Pa s or CGS unit is poise,

kinematic viscosity = =

unit : m
(SI unit),
s

cm 2

(CGS unit).

The kinematic viscosity is a diffusivity similar to the thermal diffusivity . hile

describes how heat diffuses by molecular collisions, describes how momentum
diffuses.
Prandtl number : Pr (dimensionless quantity)

A fluid with small Prandtl number diffuses heat more rapidly than it does momentum;
the reverse is true for large value of Pr.

Consider the force balance applied on the element of a fluid as shown in figure 6-1.
(1). The net pressure force on the element in x direction is ( p1 p 0 )y (force per
unit length the channel in the direction normal to the page).

(2). Shear force resulting from velocity gradient is a function of y.

On the upper boundary at y:
- shear force in the x direction: - ( y )A .
On the lower boundary at y + y :

( y + y )A = ( y ) +

d
y A
dy

p1 : pressure at entrance
p 0 : pressure at exit

The net forces from (1) + (2) applied on the element must be zero if the fluid motion
is not accelerated, so

( p1 p0 )y + ( y ) + d y A ( y )A = 0

( p p0 )
d
= 1
,
dy
A

dy

----<2>

p p1 dp
dp
: horizontal pressure gradient if A 0 , where lim 0
=
= horizontal
A

0
A
dx
dx
d dp
pressure gradient
=
.
--- momentum equation
dy dx

When p1 > p 0 , a pressure difference tends to move the fluid in the +x direction, the
pressure gradient is negative.

The pressure drop in a channel is often expressed in

( p p0 )
terms of a hydraulic head H H 1
.
g
The hydraulic head is the height of fluid required to hydrostatically provide the

applied pressure difference p1 p0 .

Plug eq <1> =

du
(constitutive law) into eq <2>,
dy

d 2 u dp
=
dy 2 dx

------<3>

Integrating <3> gives

u=

1 dp 2
y + c1 y + c 2
2 dx

------<4>

To solve constants c1 , c 2 , we must satisfy the boundary conditions that u = 0 at

y = h (lower plate is fixed equivalent to no-slip boundary conditions), and u = u 0

at y = 0 . Therefore, eq <4> becomes

u=

u y
1 dp 2
y hy 0 + u 0
2 dx
h

------<5>

y
dp

= 0 , u = u 0 1 , the
dx
h
solution reduces to a linear velocity profile, known as Couette flow.

In Fig. 6-2 (a), if p1 = p 0 , no pressure gradient,

In Fig 6-2 (b), if u 0 = 0 , (upper plate is no-slip), u =

1 dp 2
y hy .
2 dx

If this is rewritten in terms of y , distance measure from the center of the channel;
i.e., y =

h
h
1 dp 2 h 2
y .
y = y , then u =
2
2
2 dx
4

The velocity profile is parabola that is symmetric about y =

h
.
2

3. Asthenospheric counter-flow
One model for the flow in the mantle associated with the movement of the surface
plates is a counter-flow beneath the lithosphere as shown in fig. 6-4.

The lithosphere is assumed to be a rigid plate of thickness hL , moving with velocity

u 0 . Beneath is the asthenosphere of thickness h and uniform viscosity . At the
base of the asthenosphere, we assume the mantle is stationary; that is, u = 0 at
y = h . At the top of asthenosphere or the base of lithosphere, u = u 0 at y = 0 .
Conservation of mass requires that the flow of material in the +x direction in the
lithosphere must be balanced by a counter flow in the asthenosphere; that is,
h

u 0 hL + udy = 0
0

Flux of material
in the lithosphere

-----<6>

Flux of material
(per unit distance normal to the page)
in the asthenosphere

By substituting eq <5> of u into eq <6>

u 0 hL

h 3 dp u 0 h
+
= 0,
12 dx
2

-----<7>

dp 12u 0 hL 1
=
+
.
dx
h 2 h 2

-----<8>

eq<5>:
u=

u y
1 dp 2
(
y hy ) 0 + u 0 ,
2 dx
h

1 dp y 2 hy 2 h u 0 y 2

0
2 dx 3
2
2h
Substitute eq <8> into eq <5>,
h

udy =

h
0

+u 0 h =

h 3 dp u 0 h
+
.
12 dx
2

y
1 y 2 y
h
u = u 0 1 + 6 L + 2
h
h 2 h
h

shown in fig. 6-4.

-----<9>

Shear stress LA on the base of the lithosphere at y = 0 due to the counter-flow in

the asthenosphere is
2u 0
h
-----<10>
2 + 3 L
h
h
The sign in eq <10> indicates that the asthenosphere exerts a drag force on the

LA =

du
dy

y =0

base of the lithosphere tending to oppose its motion.

e.g. = 4 1019 Pa s, hL = 100km, h = 200km, u 0 = 50 mm

yr

, we get

LA = 2.2MPa(22bars ) .
The asthenospheric counter-flow requires that the pressure gradient ( dp

dx

> 0 ),

increase with x; i.e., p must increase in the +x direction i.e. the direction of seafloor
spreading. This increase in pressure with distance from a ridge could be provided by
a hydrostatic head associated with topography. The ocean floor would have to rise
with distance from the ridge.

The pressure in the asthenosphere at a distance b beneath the ridge is given by

hydrostatic formula:
p = w gw + g (wr w + b )
-----<11>
w : sea water density,
w : the depth of ocean floor at a distance x from the ridge,
: density of mantle,
wr : ocean floor depth at the ridge.

Differentiate eq <11> with respect to x,

dp
dw
= ( w )g
.
dx
dx

Positive

dp
dw
requires a negative
or an ocean depth that decreases with x.
dx
dx

The slope of the seafloor

dw
is
dx

12u 0 hL 1
dw
dp
1
=
=
+ .
( w )g dx ( w )gh 2 h 2
dx
e.g. w = 1000 kg

, = 3300 kg

m3

, g = 10 m

s2

, dw

dx

-----<12>
= 7.2 10 4 .

Across the Pacific Ocean, x = 10,000km , this would give a decrease in depth or
7.2km .

No systematic decrease in ocean depth is observed. The gravity anomaly

predicted by asthenspheric counterflow has also not been observed in the Pacific.
We conclude that the shallow counterflow model for mantle convection is not correct
and that significant converctive flow occurs beneath the asthensphere.

4. Pipe flow applications to flows in aquifers and volcanic conduits.

Consider viscous flow through a circular pipe with a radius R and a length A , the
flow is driven by the pressure difference ( p1 p0 ) applied between the sections at a
distance A apart. Assume that the velocity of the fluid along the pipe u depends
only on distance from the center of the pipe r. Find u (r ) by writing a force balance
on a cylindrical control volume of radius r and length A .
(1). The net pressure force on the ends of the cylindrical control volume is

( p1 p0 )r 2

(a force along the cylinder axis in the direction of flow).

(2). Shear force acting on the cylindrical surface of control volume, results from the
shear stress on the cylindrical surface (r ) exerting a net frictional force 2rA (r )
on the control volume.
If the flow is steady, not accelerated, the forces from (1) + (2) must be equal to zero;
that is:

r 2 ( p1 p 0 ) = 2rA ,

p1 p0 dp
=
(pressure gradient along the pipe),
A
dx

r dp
.
2 dx

-----<13>

In cylindrical geometry, the shear stress is directly proportional to the radial

gradient of the velocity u :

du
.
dr

Substitutes <14> into <13>

du
r dp
=
.
dr 2 dx

-----<14>

-----<15>

Integrate <15> and apply the boundary condition u = 0 at r = R to give:

1 dp 2
(
u=
R r2 )
-----<16>
4 dx
The velocity profile in the pipe is a parabaloid of revolution, is known as Poiseuille
flow. The maximum velocity in the pipe umax occurs at r = 0 .
U max =

R 2 dp
4 dx

-----<17>

p 0 p1 dp
=
is negative when p1 > p 0 , and umax will be positive.
A
dx

The volumetric flow rate Q through the pipe is the total volume of fluid passing a
cross-section per unit time. The flow through an annulus of thickness dr and
radius r occurs at the rate 2rdru (r ) ; Q is the integral of this over a cross section,
R

Q = 2rudr .
0

-----<18>

Substitute <16> into <18> and carry out the integration,

Q=

R 4 dp
.
8 dx

-----<19>

If we divide Q by the cross-sectional area of the pipe R 2 , we obtain the mean

velocity u :
u=

R 2 dp
.
8 dx

-----<20>

Compare Umax in <19> and u in <20>:

1
u = u max .
2
The mean and maximum velocities in the pipe are directly proportional to the pressure
gradient and inversely proportional to the viscosity . The result is valid as long as
the flow is laminar.
In fluid mechanics, we often work in terms of dimensionless variables.
Introduce two quantities: a dimensionless pressure gradient or frictional factor f, and
Reynolds number Re to rederive the relation <20>.

the frictional factor:

4 R dp
,
u dx

-----<21>

uD
,
-----<22>

where D = 2R is the pipes diameter.

Use <21> & <22>, we can rewrite eq <20> as
64
f =
-----<23>
Re
The inverse dependence of f on Reynolds # Re:
At sufficiently higher Re, observed pressure drops become considerably higher than
those given by laminar theory. The flow in the pipe becomes unsteady with random
eddies. This is known as turbulent flow.

Reynolds number:

Re

The advantage of the dimensionless formulation of the problem is that the transition
to turbulent flow occurs at Re 2200 independent of the pipe radius, flow velocity,
or type of fluid considered ( and ). The mean velocity u corresponding to Re

10

2200 is 22 mm

for water with viscosity = 10 3 Pa s flowing in a pipe of 0.1

m diameter. This illustrates that most flows of liquids and gases are in the turbulent
regime.
No theoretical equivalent to the Newtonian relation between shear stress and rate of
strain as given in eq <1>. For turbulent flow, it is found by empirically

f = 0.3164Re 4 .

-----<24>

5. Artesian aquifer flows

Natural springs are usually due to the flow of groundwater from a high elevation
to a low elevation. The flow takes place through an aquifer or permeable formation.

In an idealized model for an aquifer in the shape of a semicircle of R . The

entrance of the aquifer lies a distance b above the exit and its cross section is assumed
to be circular with radius R. The hydrostatic pressure head available to drive flow
through the aquifer is gb , where is the density of water. Since the overall
length of the aquifer is R ( R >> b ), the driving pressure gradient is

dp gb
,
=
ds
R

-----<25>

where s is the distance along the aquifer. The volumetric flow rate produced by this
pressure gradient can be calculated from eq <19> if the flow is laminar.
Substitute <25> into <19> for
Q=

dp
dp
replaced by
,
dx
ds

gbR 4
.
8R

-----<26>

If the flow is turbulent, we can determine Q by using the empirical relation <24>
between f and Re.
Recast eq <24> into dimensional from:

11

4
4 R dp

=
0
.
3164
2
u2R .
dx
u

-----<27>

Re 1

The result of rearranging eq <27> so as to determine u is

4

1
4
1

7
4 2 4 1 dp 7 75 7
R .
u=

0.3164 dx

-----<28>

Since Q = R 2 u , we obtain the volumetric flow rate through the aquifer for turbulent
1 dp 1 dp gb
flow by multiplying <28> by R 2 and substituting for = =
dx ds R
in eq <25>,
1

gb 7 7
Q = 7.686 R 7 .
R
19

-----<29>

6. Flow through volcanic pipes.

Another natural example of pipe flow is the flow of magma through volcanic
conduits of nearly circular cross section. The upward flows of magma is driven by
buoyancy of the lighter magma relative to the denser surrounding rock. At the same
depth h, the lithostatic pressure in the surrounding rock is s gh ( s : density of
rock). The hydrostatic pressure in a stationary column of magma is A gh ( A :
magma density), assuming that the lithostatic and hydrostatic pressures are equal in
the pipe; that is, the walls of the pipe are free to deform as the magma is driven
upward, then the pressure gradient available to drive the magma up to the surface
( s A )g . The volumetric flow Q driven by the above pressure gradient through
a volcanic pipe of radius R is from eq <19> in page 8.
Q=

R 4 dp ( s A )gR 4
=
8 dx 8

dp
= ( s A )g ),
dx

-----<30>

if flow is laminar. From eq <28> and Q = R 2 u , the volumetric flow for turbulent

is:

Q = 14.8

19
7

[( s A )g ]7

3
7
A

1
7

-----<31>

12

7. Conservation of fluid in 2-D

Consider a viscous fluid flowing to 2-D plane x, y the corresponding velocity
components in x & y direction are u and v.

Consider a rectangular element with dimensions x and y as seen in fig. 6-10.

The flow rate per unit area at x + x is
u ( x + x ) = u ( x ) +

u
x .
x

The net flow rate out of the region between x and x + x per unit area normal to
the x direction is:
(1). u +

u
u
x u = x .
x
x

Similarly, the net flow rate per unit area normal to y direction out of the region
between y and y + y is:
(2). v +

v
v
y v = y
y
y

Therefore, the net rate at which fluids flow out of the rectangular region xy
= (1) area across which flow occur( = y ) + (2) area across which flow occur( = x )
=

u
v
xy + yx .
y
y

-----<32>

The total net outward flow rate per unit area of the rectangular is
u
v
xy + yx
u v
y
y
=
+ .
xy
x y

If the flow is steady (time-independent), and there are no density variations to

consider, there can be no net flow into or out of the rectangle. The consideration of
fluid or continuity equation is
u v
+
= 0.
-----<33>
x y

13

G
u v w G G
+ +
= u = 0 , where u = (u , v, w) , = , ,
x y z
x y z
G
G
This is appropriate for an incompressible fluid. (For compressible fluid, u = 0 .)

In 3-D,

( )

8. Force balance in 2-D

Force acting on the control volume; i.e. the rectangular element xy must be in
balance. Include in force balance are the pressure forces, viscous forces and gravity
force. At the moment, we neglect the inertial force associated with the acceleration of
a fluid element. This is appropriate for the slow motion of very viscous or high
Prandtl number fluids. The earths mantle behaves as a highly viscous fluid on
geological time scales. The viscosity of the mantle is about 1021 Pas; its density
and thermal diffusivity and about 4000 kg

2
and 1 mm

The Prandtl number

of the mantle is about 1023. Therefore, we can ignore the inertial force.

The balance of pressure, viscous, and gravity forces and the neglect of inertial force
are based on the Newtons second law of motion with the neglect of its acceleration.
Its equivalent to the conservation of momentum.
(1). Pressure Force:
The net pressure force acting on the element xy in the +x-direction per unit
area of the fluid element
p( x )y p( x + x )y
p( x + x ) p( x )
p
=
=
= .
xy
x
x
using Taylors series: p( x + x ) = p( x ) +

p
x
x

G
p

14

The net pressure force on the element in +y-direction per unit area of the fluid
element
p( y )x p( y + y )x
p( y + y ) p( y )
p
=
=
= .
xy
y
y
(2). Gravitational Force:
The gravitational body force on a fluid element is its mass times the acceleration
of gravity. The mass for the element xy = xy . g is the force of
gravity per unit area of the element and the gravity acts in the positive y direction.
Thus the net gravitation force per unit area of the element in the +y direction is
g .
(3). Viscous Forces:
The viscous forces acting on the element both parallel and to the surface of
the element. xx and yy are viscous normal stresses, the viscous forces per unit
area that act to the elements surfaces.

The stresses are positive in the

directions shown in fig. 6-12.

If there is no net torque about the center of the element,

xy = yx .
The net viscous force in the x-direction per unit cross-sectional area of the element is
xx ( x + x )y xx ( x )y yx ( y + y )x yx ( y )y
+
.
xy
xy

15

xx yx
+
.
y
x
yy xy
+
Similarly, the net viscous force in the y-direction per unit area is
.
y
x

Using Taylors series, the above equation simplifies to

For an idealized Newtons fluid, the viscous stresses are linearly proportional to the
velocity gradients.
Constitutive Law:

xx = 2

u
,
x

yy = 2

v
,
y
u

yx = xy = + .
y x

-----<34>

The total normal stress is the sum of the pressure and the viscous stress; that is:

xx = p xx = p 2

u
,
x

-----<35>

yy = p yy = p 2

v
.
y

-----<36>

The minus sign in front of xx and yy are the result of the opposite sign
conventions adopted for (+ for compression) and (+ for extension).
The viscous stress is the only contribution to the shear stress. Use eq<34> to rewrite
the viscous forces acting on the element of the rectangular,
2u 2 v
xx yx
2u

+
= 2 2 + 2 +
x
y
xy
x
y
yy
y

xy
x

= 2

2 v 2u
2v

+ 2 +
y
x

y 2
x

Applying the continuity equation:

-----<37>

-----<38>

u v
u v u v
+
= 0 + = + = 0
x y
x x y y x y

2v
2u
= 2
xy
x

-----<39>

2u
2v
= 2
xy
y

-----<40>

Use eqs <39> and <40> to replace the mixed derivative terms in eqs <37> & <38>,
we get:

16

2u 2u
x-direction: 2 + 2 ,
y
x
2v 2v
y-direction: 2 + 2 .
y
x

the net viscous forces per unit cross-sectional area in x and y direction, respectively.

In 3-D:

2u 2u 2u
+ 2 + 2
2
y
z
x

x-direction

2v 2v 2v
2 + 2 + 2
y
z
x

y-direction

2w 2w 2w
+ 2 + 2
2
y
z
x

2 u

z-direction

The force balance equations for an incompressible fluid with very large viscosity
undergoing steady flow in 2-D are obtained by adding the pressure, gravity and
viscous forces together and equating their sum to zero.
For x-direction,

y-direction,

2u 2u
p
+ 2 + 2 ,
x
y
x

-----<41>

2v 2v
p
0 = + g + 2 + 2 ,
y
y
x

-----<42>

0=

and gravity acts only in y-direction.

In order to eliminate the hydrostatic pressure variation in equation <42>, we
introduce
P = p gy ,
-----<43>
where the pressure P is the pressure generated by fluid flow.
Substituting <43> into <42> yields:
2u 2u
P
+ 2 + 2 ,
x
y
x

-----<44>

2v 2v
P
0=
+ 2 + 2 .
y
y
x

-----<45>

0=

17

9. The stream function

The continuity equation in 2-D for an incompressible fluid can be satisfied if we
introduce a stream function defined such that.

u=

,
y

v=

.
x

-----<46>

u v
+
= 0 (eq <33>) yields
x y

Substitution of <46> into

2 2
+
=0.
xy yx

-----<47>

Substitution of <46> into <44> and <45>,

0=

3
P
3
+ 2 + 3
x
x y y

0=

3
P
3
+ 3 + 2 .
y
y x
x

-----<48>

-----<49>

Eliminate the pressure from these equations and obtain a single differential equation
for if we take the partial devirative of eq <48> w.r.t. y and the partial devirative of
<49> w.r.t. x and then add them together,

4
4
4
< 48 > + < 49 >= 0 = 4 + 2 2 + 4 .
y
x
x
x y
y

-----<50>

This is a biharmonic equation. In terms of Laplacian operator 2 ,

2
2
2 = 2 + 2 ,
x
y
We can write eq <50> for the stream function in the form,
4 = 0 .

-----<51>

The stream function can be given a physical interpretation in terms of volumetric flow
rate between any two points in an incompressible steady 2-D flow. Consider two
points A and B, separated by an infinitesimal distance s as shown in fig 6-13.

18

The flows across AB can be calculated from the flows across AP and PB. Since the
mass consideration requires that no net flow is into or out of the triangle PAB. The
volumetric flow rate across AP into the triangle per unit distance normal to the figure
is uy ; similarly, the flow rate across PB out of the triangle is vdx . The net flow out
of PBA is thus uy + vx ; this must be equal to the volumetric rate of flow (per unit
distance in the 3rd dimension) into PAB across AB. In terms of the stream function
, uy + vx can be written:
uy + vx =

y +
x = d .
y
x

Thus, the small difference d is the volumetric flow rate between any two points
separated by the infinitesimal distance s . If the points are separated by an arbitrary
distance, the integrate of d between the points,
B

d =
A

A ,

-----<52>

gives the volumetric flow rate between the points, i.e. the difference between the
volumes of the stream function at any two points is the volumetric rate of flow across
any line drawn between the points.

10. Postglacial Rebound

Important information in the fluid behavior of the mantle comes from studies of
the dynamic response of the mantle to loading and unloading at the surface.
Mountains depress the underlying crust-mantle boundary. The mountain building
process is so slow that dynamic effects can be neglected; i.e. the mantle beneath a
mountain is essentially in hydrostatic equilibrium throughout the cycle of the
mountain. However, the growth and melting of ice sheets occur sufficiently fast so
that dynamic effects are important in the adjustment of the mantle to the changing
surface load. The thick ice sheet that covers GreenLand has depressed the surface
level several kilometers below the sea level. The load of ice has forced mantle rock
to flow laterally, allowing the earths surface beneath the ice to subside. When the
ice melted about 10000 years ago, the surface rebounded. The rate of rebound can
be determined by dating elevated beaches.
To determine the response of the mantle to the removal of an ice load, we consider
the flow in a semi-infinite, viscous fluid half-space ( y > 0 ) subjected to an initial
periodic surface displacement given by

wm = wmo cos

2x

where is the wave length and wm << .

19

The displacement of the surface w loads to a horizontal pressure gradient due to

hydrostatic load. When the surface is displaced upward (negative w), (w<0).
The pressure is positive (p>0). The fluid is in driven away from the region as the
displacement w decreases. When the surface is displaced downward (w>0), the
pressure is negative (p<0). The flow is driven into the region as the displacement w
decreases when the loading ice removed.

20

The return of the surface to an undeformed (w = 0) state is governed by the viscous

flow in half-space. The flow can be determined by solving the biharmonic equation
for the stream function. Since the initial surface displacement is of the form
cos

2x

, must also vary periodically with x. we can assume the solution of is

an arbitrary combination of sin

2x

and apply the method of separation of variable

2x

Y ( y),

and take

= sin

-----<53>

where Y ( y ) is to be determined. By substituting into the biharmonic equation

<50>, we obtain
d 4Y
2
2
4
dy

2
d Y 2
2 +
Y = 0.
dy

4

-----<54>

Solution of the constant coefficient differential equation <54> for Y are of the form
Y exp(my ) .
-----<55>
Substituting <55> into <54>, leads to the solution of m.
2

2
4
2 2 2
2 2 2
m 2
m +
= m
= 0

or m =

-----<56>

The two values of m provide two possible solution for Y:

2y
2y
exp
-----<57>
and exp
.


The differential equation <54> is 4th order, the solutions of y should be four; therefore,
two additional solutions are required. It can be verified by direct substitution that
2y
2y
y exp
-----<58>
and y exp
.


The general solution for Y is the sum of these four solutions from <57> and <58>,

= sin

2y
2y
2y
2y

2x

Ae
Bye
Ce
Dye
+
+
+
,

-----<59>
where A, B, C and D are determined by the appropriate boundary conditions. First, we
require the solution to be finite as y so that C = D = 0. The solution for
simplifies to

21

= sin

2x

2y

( A + By ) .

----<60>

The velocity components u and v can be obtained by differentiating according

to <46>, we find
2x

u=
e
= sin

2y

2
( A + By ) B ,

-----<61>

2y

2
2x
( A + By ) .
cos
-----<62>
=
e
x

Since the part of mantle that behaves like a fluid is overlain with a rigid lithosphere,
we force the horizontal component of the velocity u to be zero at y = w ; i.e., the

v=

no-slip boundary condition is applied at the upper boundary of the fluid half-space.
Because the vertical displacement w of this boundary is small, w << , its
appropriate to apply this condition at y = 0 . By setting u = 0 at y = 0 (no-slip
condition) in eq <61>, we find:

B=

2A

and

= A sin

2x

2
u = A
ye

2

2y

2y
1 +
,

2y

cos

2x

sin

2x

-----<63>

-----<64>

2y

2y
-----<65>
1 +
.

To evaluate A, its necessary to equate the hydrostatic pressure head associated with
the topography w to the normal stress at the upper boundary of the fluid half-space.
v
The former quantity is gw , and the latter from eq <36> is p 2 . Since w
y
is small, we can equate these stresses at y = 0 .
v=A

gw = p 2

v
y

hydrostatic pressure due to

topography w

-----<66>

viscous normal stress due

to velocity gradient

To apply the condition <66>, we first calculate the pressure p and displacement w
at y = 0 . The pressure can be found by inserting <64> for u into the horizontal
force balance <41>, we obtain:

22

p
2x
2
= 2A
sin
x

on y = 0 .

-----<67>

on y = 0 .

-----<68>

Integrating <67> w.r.t. x to give:

2x
2
p = 2A
cos

v
on y = 0 for eq <66> can be found by differentiating eq <65> w.r.t. y and
y
then evaluate the result on y = 0 .
We get
v
= 0 .
-----<69>
y y =0
Condition <66> thus simplifies to
2 A 2
2x
.
-----<70>
w y =0 =
cos

The surface displacement w is related to the velocity fluid by the fact that the time
derivative of w is just the vertical component of the surface velocity; i.e.,
2

w
y=w = v
t
Since w << , we can write:
w
t
From eq <65>,

y =0

y =0

=v

=A

-----<71>

-----<72>

y =0

y=w

cos

2x

-----<73>

so that

2
2x
w
cos
.
y =0 = A

t
Combining <70> & <74>, we find that w at y = 0
w
g
g
.
= w
= w
t
4
4

-----<74>

-----<75>

This can be integrated with the initial condition w = wm at t = 0 , to give

gt
gt
= wm exp
w = wm exp
-----<76>
.
4
4
The surface displacement decreases exponentially with time as fluid flows from
regions of elevated topography to regions of depressed topography, eq <76> can be
rewritten as
t

w = wm e ,

-----<77>

where r , the characteristic time for the exponential relaxation of the initial

23

displacement, is given by

r =

4 4
=
.
g
g

-----<78>

The viscosity of the mantle can be established from eq <78> once the relaxation time
for postglacial rebound has been determined.
Quantitative information on the rate of postglacial rebound can be obtained from
elevated beach terraces. Wave action over a period of time erodes a beach to sea
level. If sea level drops or if the land surface is elevated, a fossil beach can be
obtained by radioactive dating using C14 in shells and driftwoods. The uplift of the
beach terraces is compared with the exponential time dependence given in eq<77>.
Assume that uplift began 10000 years ago. And wm 0 = 300m with 30m of uplift
to occur in the future; i.e. w = 30m at t = 10 4 year later (the present).

Previously we have only considered the response to a spatially periodic surface

displacement. Since the problem is linear, solutions can be superimposed to
consider other distribution of surface displacement. More complete studies of
postglacial rebound include the flexural rigidity of the elastic lithosphere and a depth
dependent viscosity. If the ice sheets continue to melt during rebound, the sea level
increase must be taken into account. Table 6-2 summarizes the ocean mantle
viscosity from rebound studies.

11. Angle of subduction

As discussed in Chapter 3 (section 17), the oceanic lithosphere bends as its
subducted at a trench. The gravitational body force on the descending lithosphere is

24

directed vertically downward. It might be expected that under this force the
lithosphere would bend through 90 0 and descend vertically into mantle. However,
its observed that oceanic lithosphere straightens out after subduction and descends at
a finite angle of dip .

One explanation for why the lithosphere descends at an angle other than 90 0
(Table 6-3) is that pressure forces due to the induced flows in the mantle balance the
gravitational body forces. The pressure forces are due to the mantle flow induced by
the motion of the descending lithosphere, they are flow pressures relative to the
hydrostatic pressure. The dip of the lithosphere is thus a consequence of the balance
between the gravitational torque and the lifting pressure torque.
The pressure forces acting on the lithosphere can be calculated using 2-D viscous
corner flow model in fig 6-18. The trench is at x = 0. Assume the surface
y = 0, x < 0 moves with constant speed U toward the trench; the surface
y = 0, x > 0 is stationary. The descending lithosphere is the line extending from
x = 0 down to the positive x-axis at angle of dip. the velocity parallel to this line is
U . Distance measured along this line is r .

The line divides the viscous mantle into

two corners, the arc and the ocean corner. The motion of this line divides a flow in
the arc corner. The velocities of the dipping line and the surface induce a flow in he
oceanic corner.

25

To solve for the motions in both corners and the flow pressures on the dipping line,
we use the stream functions for the corner flows that satisfy the biharmonic equation
( 4 = 0 ). For the corner flow geometry, we write in the form:
y
-----<79>
x
where A, B, C, and D are constants determined by boundary conditions. There are
two stream functions with distinct values of these constants.
In the arc and oceanic corner which have different angles and conditions on their
bounding lines. The velocity corresponding to the stream functions can be obtained
from eq <46>
x

x
= B D tan 1 + ( Cx + Dy ) 2
u=
,
-----<80>
2
y
y
x +y

= ( Ax + By ) + (Cx + Dy ) tan 1 .

v=

x
= A + C tan 1 + ( Cx + Dy ) 2
.
2
x
y
x +y

-----<81>

Recall that:

1 y
=
tan
x
y

1 1
x
1 y
1 y
y
= 2
,
tan
=
= 2
.

2
2
2
2
y x x + y2
x
x
y x x +y
1+ 2
1+ 2
x
x

The pressure can be found by substituting <80> into <44> and integrating the
resulting expression for

2u 2u
P
P

(from
2 + 2 = 0 ). Alternatively, eq
x
y
x
x

<45> and <81> can be used to get

P=

2v 2v
P
P
(from
+ 2 + 2 = 0 ). Therefore,
y
y
x y

2 (Cx + Dy )
.
x2 + y2

-----<82>

The pressure in <82> is the pressure relative to the hydrostatic pressure; i.e., the

26

pressure associated with flow.

General expression for the constants of integration are complicated, here we
evaluate them for a particular value of the dip angel. For a dip of

, representative

Ryukyu arc, the boundary conditions for the arc corner are:
u = v = 0 on y = 0, x > 0 , or tan 1

y
= 0,
x

----<83>

U 2
y
on y = x , or tan 1 = .
2
x 4
Applications of these conditions lead to the expressions for C and D in the arc corner.
u=v=

U 2 2
U 2
2

C=
and D =
.
2
2

2 2
4
4

Thus, the pressure in the arc corner is:

Parc corner =

U 2 [x + (4 ) y ]
.

2 2
2

2
x +y
4

-----<84>

-----<85>

r 2
, the flow pressure on the top of the descending slab is:
2
4 U
8.558U
P=
=
.
-----<86>
2
r

2
r
4

The negative value of the flow pressure on the top of the descending slab gives the
effect of a suction force tending to lift the slab against the force of the gravity.
On x = y =

The pressure force varies as

1
along the upper surface of the slab and therefore has
r

a singularity in the idealized model as r 0 . However, the lifting torque on the

slab is the integral of the product rP over the upper surface of the slab. The lifting
torque per unit distance along the top of the slab is a constant, the torque on the slab is
thus proportional to its length.
The boundary conditions for the oceanic corner are:
u = U , v = 0 on y = 0, x < 0 or tan 1

U 2
y
on y = x or tan 1 = .
2
x 4
By substituting <80>, <81> into <87>, we find:
u=v=

y
= ,
x

27

U
2
2
C=
2

3
9
1 +

2
2

D=

3 9
,

+
4
2

3 3
2 2 + 2 21 + 2

9

4
U
2

-----<88>

The flow pressure in the oceanic corner is from substituting <88> into <82>,

U 3 2 4 0.462U
P=
=

r 2
r
2
9
4

-----<89>

where the flow pressure P is evaluated on the bottom of the descending slab. The
positive value of P means the induced pressure on the bottom of the slab also exerts a
lifting torque on the slab. The torque per unit distance along the slab is a constant.
The net lifting torque on the slab is the sum of the torques exerted by pressures on the
top and bottom of the slab. Comparison of <86> and <89> shows that the torque
exerted by the suction pressure in the arc corner far outweighs the lifting effect of
pressure on the bottom of the slab.

12. Diapirism
Diapirism: the buoyant upwelling of relatively light rocks that rise into the heavier
overlying rock; e.g., salt dome.
Initially a layer of salt is deposited at the surface by evaporation of seawater.
Subsequent sedimentation buries this layer under other heavier sedimentary rocks
such as shales and sandstones. At shallow depth, the strength of the salt layer is
sufficient to prevent gravitational instability from the inducing flow. As the depth of
the salt layer increases, the temperature of the salt increases because of the thermal
gradient. Thermally activated creep process allow the salt to flow upward to be
replaced by the heavier overlying sedimentary rocks. Eventually the upward flow of
the salt creates salt domes.
The deformation of the rocks above salt domes results in the formation of
impermeable traps for the upward migrating oil and gas. Other examples of
diapirism are in the mountain belt where the heat flow is high and volcanism heats
lower crustal rocks to sufficiently high temperature so that they can freely flow by
solid-state creep. If the heated rocks at depth are lighter than the overlying rocks,
the deeper rocks will flow upward to form diapirs.

28

We apply the same type of analysis used in postglacial rebound study to investigate
diapirism. Consider the geometry in fig 6-21. A fluid layer with thickness b and
density 1 overlies a second fluid layer of thickness b but with density 2 . Both
fluid layers have viscosity . The upper boundary of the top layer and the lower
boundary of the bottom layer are rigid surfaces. Take 1 > 2 , the gravitational
instability of heavy fluid overlying light fluid is known as the Rayleigh-Taylor
instability.

Take the undisturbed interface between the superposed fluid layers to be

y = 0, y = b and y = b are the upper and lower rigid boundaries.

As a result of

29

gravitational instability, the displacement of the disturbed fluid interface is denoted by

w.

Assume that w is given by w = wm 0 cos

2x

the flow in the upper layer has the form of eq <59>.

hyperbolic functions instead of exponentials,

. The stream function 1 for

We rewrite eq <59> using

2x
2y
2y
2y
2y
+ B1 sinh
+ C1 y cosh
+ D1 y sinh
A1 cosh

1 = sin

-----<90>

e x ex
e x + ex
cosh x =
( sinh x =
,
).
2
2
The stream function 2 for the lower layer is:
2x
2y
2y
2y
2y
+ B2 sinh
+ C 2 y cosh
+ D2 y sinh
A2 cosh
. ----<91>

The velocity in the layer is found by differentiating 1 and 2 based on eqs <46>,

2 = sin

u1 =

v1 =

u2 =

v2 =

cos

sin

2x
( A1 + C1 y ) cosh 2y + (B1 + D1 y )sinh 2y ,

sin

cos

D1
C1
2x
2y
2y
+ B1 + D1 y +
, -<92>
A1 + C1 y +
sinh
cosh

2
2
-----<93>

D2
C 2
2x
2y
2y
+ B2 + D 2 y +
,<94>
A2 + C 2 y +
sinh
cosh

2
2

2x
( A2 + C 2 y ) cosh 2y + (B2 + D2 y )sinh 2y .

-----<95>

We evaluate constants of integration by applying the boundary conditions of no-slip

on y = b ; that is,
u1 = v1 = 0 or y = b ,
u 2 = v 2 = 0 or y = b ,

and the continuity of u and v across the interface. For small displacements of the
interface w << , we require continuity u and v at the undisturbed location of the
interface, y = 0 . Thus, we require u1 = u 2 and v1 = v 2 on y = 0 .
By applying these conditions to eqs <92> to <95>, we obtain:
B1 +

C1
C 2
= B2 +
,
2
2

A1 = A2 ,

-----<96>
-----<97>

D1
C
2b

= B1 bD1 + 1 ,
A1 bC1 +
tanh
2

(B1 bD1 ) tanh 2b = A1 bC1 ,

-----<98>
-----<99>

30

D2
C 2
2b

= B2 bD2 +
,
tanh
A2 bC 2 +
2

(B2 bD2 ) tanh 2b = A2 bC 2 .

-----<101>

Shear stresses must also be continuous across the interface between the fluid layers.
For w << , and for equal viscosities across the interface, this condition can be
written:
v
u
u v
xy = yx 1 + 1 = 2 + 2
on y = 0 .
x
y
x
y
v
Since v is continuous at y = 0 , so is
, and eq simplifies to:
x
u1 u 2
=
on y = 0 .
-----<102>
y
y
Eq <102> requires that:
D1 D1
D2 D2

+ D1 = A2 +
+ D2
A1 +
+
+
2 2
2 2

with A1 = A2 , D1 = D 2 .
-----<103>
By substituting eq <99> & <101> and combining the result with the difference
between eqs <98> & <100>
2b
2b
2b
coth
0 = (C1 + C 2 )1 +
tanh

-----<104>

2b
only if
Eq <104> can be satisfied for arbitrary

-----<105>
C1 = C 2 .
If we add eqs <99> & <101> and use eq <105>, we deduce that
B1 = B 2 .

-----<106>

By using eqs <105> & <106> to simplify eq <96>, we get

C
B1 = 1 .
-----<107>
2
All the constant of integration can now be determined in terms of A1 by solving eq
<96>, <97> and <107>. We find that 1 in the upper layer is

1 = A1 sin

2x

cosh

2y

+ A1 sin

2x y
2b
2y
sinh
tanh

b 2b

2y
2y
1
y

sin
+
+ cosh

2b
2b
2b
2b
b
sinh

cosh

1
2b

tanh

2b 2b
2b

sinh cosh

-----<108>

31

The expression for 2 is obtained by replacing y with y .

The time rate of change of the interface displacement

w
must be equal to the
t

vertical of the fluid at the interface. Since x << , this condition can be written:
w
=v
t

y =0

The vertical velocity v can be evaluated by differentiating eq <108> w.r.t. x and

set y = 0 , we can rewrite eq <109>
w 2A1
2x
=
.
-----<110>
cos

t
To eliminate A1 from the equation of motion of the interface, we need to incorporate

the buoyancy force brought by the displacement of the interface into the analysis.
Compare two columns of fluids, one with the interface in the undisturbed location,
and the other with the interface displaced downward. Because fluid of density 1
replaces fluid of density 2 between y = 0 and y = w , the additional weight of
this fluid (1 2 )gw is felt as a normal stress or pressure on the disturbed interface.
It must be balanced by the net normal stress on the interface due to flow pressure and
normal viscous stress; that is, stress at y = 0 . On y = 0 , the normal viscous stress
v

yy = 2

y =0

By differentiating eq <108> w.r.t. x and y and evaluate the result on y = 0 , it can

be seen that this quantity is zero. Thus, the buoyancy force per unit area due to the
displacement of the interface is balanced solely by the net flow pressure exerted on
the interface. This condition can be written

(1 2 )gw = (P2 P1 ) y =0 .

-----<111>

Eq <111> provides a second relation between w and the flow field that allows us to
relate w to A1 and therefore convert eq <110> into an equation for w .
The flow pressure on y = 0 in the upper layer can be found by substituting <108>

32

into <48>

(P1 ) y =0

2A1 2
1

=
+

2b
2b
b 2b
sinh
cosh

1
2b
2x

.

tanh
cos

sinh 2b cosh 2b 2b

-----<112>

By carrying through the same procedure using 2 , we find:

<111> becomes

(P2 ) y =0 = (P1 ) y =0 .

-----<113>

(1 2 )gw = 2(P1 ) y =0 .

-----<114>

Eq <114> shows that with the heavy fluid above light fluid (1 > 2 ) , a downward
displacement of the interface w > 0 causes a negative pressure in the upper fluid
layer. This tends to produce a further downward displacement of the interface
leading to instability of the configuration. Upon substituting eq <112> into <114>,
we get

(1 2 )gw = 4A1 2 cos 2x +

2b
b
2b
2b
sinh

cosh

1
2b

. -----<115>

tanh

2b
2b 2b

sinh cosh

By solving <115> for A1 and substituting the resulting expression into <110>, we
get :

2
2b
1

tanh

2b
2b

2b
sinh
cosh
w (1 2 )gb

w .
=
4
t

2b

+
2b
2b

sinh
cosh

The solution of this equation is

------<116>

33

w = w0 e .

------<117>

1
+

2b sinh 2b cosh 2b

4
.
-----<118>
(1 2 )gb

2
2b
1

tanh
2b
2b

2b
sinh
cosh

The quantity a is the growth time ( for 1 > 2 ) of a disturbance. Fig. 6-23 is a plot

with a =

of the dimensionless growth time (1 2 )gb

dimensionless disturbance wave number

2b

as a function of the

If 1 > 2 (heavy fluid overlies on top), the interface is always unstable; i.e,
a > 0.
If 1 < 2 (light fluid overlies on top), the interface is stable because a is negative
for all .
For large , <118> reduces to
24

, a
.
( 1 2 ) gb 2 b
2

For every small

-----<119>

34

0, a

4
2 b
.
( 1 2 ) gb

-----<120>

When the heavy fluid lies on the top and the configuration is unstable, the disturbance
with the shortest time constant grows and dominates the instability. The wavelength
that gives the smallest value for a is:
. = 2.568b

-----<121>

The rate of growth of this dominant disturbance is obtained by substituting <121> into
<118> with the result.
13.04
a =
-----<122>
(1 2 )gb
The instability takes longer to grow, the more viscous the fluid and the smaller the
density difference. Though we only consider the stability problem for small
displacements, its expected that the wavelength of the most rapidly growing small
disturbance closely corresponds to the spacing between fully developed diapers. A
example in N. Germany (fig 6-24). The depth to the salt large is about 5 Km, and
the spacing of the salt dome is about 10-15 km, in agreement with eq <121>.

35

13. Folding
Folds are found in both sedimentary and metamorphic rock on all scales.
Folding occurs under a wide variety of conditions, but is often associated with
compressional tectonics. At relatvely low temperature, sedimentary rocks flow to
produce folds rather than fracture, pressure solution creep is thought to be a major
role. Sedimentary rocks are often saturated with water. When the differential
stresses are applied to the rock, the minerals dissolve in regions of high stress and are
deposited in regions of low stress. The result is the deformation of the rock. Pressure
solution Creep of sedimentary rocks can result in a linear relationship between stress
and strain rate, therefore, a Newtonian fluid behavior. Folds usually form in a
pre-existing layered structure with considerable variation in the material properties of
adjacent layers. If a uniform medium is subjected to compression, it will be
uniformly squeezed as fig 6-25(a). If the medium is composed of a series of week and
strong layers, folding will occur as fig 6-25 (b). The strong layer is referred to as
being competent; the week layer is referred to being incompetent.

One approach to the quantative study of folding is to consider an elastic (competent)

layer of thickness h embedded between two semi-infinite Newtonian viscous fluids
(incompetent). An end load P on the elastic layer causes it to buckle.

36

The deformation of a thin elastic plate under end loading is described by the
differential equation (3-74). The vertical component of the normal stress due to flow
on the fluids above and below the plate can be used to determined the force per unit
area q ( x ) on the plate. The fluids occupy semi-infinite half space. Assume the
deformation of the plate is given by
t

2x a
w = wm cos
e .

-----<123>

Since the plate forms the boundaries of the fluid half-spaces, the situation is identical
with the post glacial rebound. By symmetry, the solutions above and below the plate
are identical. Consider the solution below the plate and measure g positive
downward from the base of plate (fig 6-26). The appropriate solution of the
biharmonic equation is <59>,

= sin

2y
2y
2y
2y

2x

+
+
+
Ae
Bye
Ce
Dye
.

-----<59>

The velocity should be finite as y and requires C = D = 0. The rigidity of the

plate requires that u = 0 on the plate. Assume w << again, this boundary
condition can be applied at y = 0 . Therefore, eq <63> is applicable in the fluid below
the plate.
2y

2y
-----<63>
1 +
.

From eq <68>, the pressure Pb on the base of the plate at y = 0 is given by

= A sin

2x

2x
2
Pb = 2 A
.
cos

-----(124)

This can be rewritten in terms of w by using eq<74>

w
t

y =0

=A

cos

2x

-----<74>

37

2 w
Pb = 2
.
-----<125>

t
The pressure PT acting downward on the top of the plate is relate to the pressure Pb
-----<126>
acting upward on the base of the plate by: PT ( x ) = Pb (x ) .

<124> becomes:

There is no normal viscous stress on the plate since

v
according to eq <69>
y
q = PT Pb = 2 Pb .

y =0

v
vanishes on y = 0
y

= 0 . Thus, the net normal stress on the plate is

-----<127>

Substitute <125> into <127>, we obtain

2 w( x, t )
q( x, t ) = 4
.
-----<128>

t
With the force per unit area acting on the elastic plate, the equation for the deflection
of the plate is:
4w
2w
2 w
+
= 4
= q( x) .
P

4
2
x
x
t
Substitute eq <123> into <129>, we find:
D

a =

4
2

2 2
P
D

-----<129>

-----<130>

The wavelength corresponding to the smallest value of a is obtained by setting the

derivative of a w.r.t. equal to zero. The result is:
1

3D 2
= 2
.
P

-----<131>

This is the wavelength of the most rapidly growing disturbance. Write

P =h,

-----<132>
where is the stress in the elastic layer associated with the end load P , we get:
1

Eh 3
D
12 1 2

2
E
= h
2
(1 )

-----<133>

Its expected that when folds develop in an elastic layer of rock surrounded by rock
exhibiting fluid behavior, the initial wavelength of the folds has the dependence on
the thickness of the elastic layer and the applied stress <133>.
The observed dependence of fold wavelength on the thickness of a fold is given
in Fig 6-27, in good agreement with

(1 2 )
E

= 10 2 .

38

For E = 50GPa , = 0.25 , = 530 MPa , This is a high stress, but likely to be
about the compressional strength of sedimentary rocks at a depth of 2-5 km.
As the amplitude of a fold increase, its wavelength decreases somewhat and the
bending stress in the elastic layer exceeds the yield strength of the rock. The elastic
layer either fractures or plastically yield at points of maximum bending moment that
1
are at x = n , n = 0,1,2,....... . If a plastic bending occurs, an angular or chevron
2
fold would be expected (fig. 6-28(a)). Folds with nearly straight limbs of this type are
often observed. For a rounded told (fig 6-28 (b)), its likely that the dominant
member has also deformed in a fluid-like behavior. Assume the competent layer is a
Newtonian fluid with a viscous 1 embedded between two semi-infinite fluids with
viscous 0 (1 >> 0 ) . This approach is often referred to as the Biot theory of
folding.
To derive the bending of a free or isolated plate of viscosity , we follow the theory
of bending of an elastic layer given in section 3-9. The bending moment is given by
(eq 3-61);
h

M = 2h xx ydy .

The longitudinal stress xx in a viscous plate is given by eq <35>,

-----<134>

39

xx = p 2

u
.
x

For a free plate, yy must vanish its surface, and if the plate is thin, we can take

yy = 0 throughout the plate. From eq <36> and yy = 0 , we obtain:

yy = 0 = p 2

v
v
p = 2 .
y
y

The incompressible continuity eq <33> gives

-----<135>

v
u
= , and we can write eq <135>
x
y

as:
p = 2

u
.
x

-----<136>

By substituting <136> into <35>, we get:

xx = 4

u
.
x

-----<137>

Eq <134> for the bending moment in viscous plate becomes::

u
ydy
x
2

M = 4 2h xx ydy = 4 2h

(xx = 2

u
)
x

-----<138>

u
d 2w
By direct analogy with eq <3-70> xx = y 2 , the rate of strain rate
is

x
dx

given by:
u
3w
=y 2 .
x
x t

-----<139>

u
and
x
have opposite signs. Substitute <139> into <138> and carry out

The sign of this equation is opposite to eq <3-70> because the rate of strain
the strain rate xx

the integration, we get:

M =

h 3 3 w

.
-----<140>
3 x 2 t
Upon substituting the second derivative w.r.t. x of eq <140> into eq <3-60>, we obtain
the general equation for the bending of a thin viscous plate,

h3 5 w

2w
. Dw ''''+ Pw '' = q and M = Dw ''
=qP 2
3 x 4 t
x

----<141>

d 2M
d 2w
= q + P 2 .
<3-60>
dx 2
dx
Solution of this equation give the vertical displacement w of a viscous plate as a

40

function of time.
Consider a specific example of a viscous plate of length L embedded at one end
with a concentrated load Va applied at its other end. Since P = q = 0 , eq <141>
reduces to

h 3 5 w
3 x 4 t

=0

-----<142>

Integrate twice w.r.t. x yields:

h 3 3 w

= M = f1 (t )x + f 2 (t )
-----<143>
3 x 2 t
where f1 (t ) and f 2 (t ) are constants of integration that can depend on time. Since

the overall torque balance given in eq <3-78>

M = Va ( X L ) ,
must also be applicable to the viscous plate, we can identify f 1 and f 2 as
f1 = Va , f 2 = VaL .
Eq <143> becomes

h 3 3 w

= Va x + VaL .
-----<144>
3 x 2 t
Integrate eq <144> twice more w.r.t. x and satisfy the b.c.s for an embedded plate,

h 3 w Vax 2
x
w
=
= 0 at x = 0 , to get
-----<145>
L .
3 t
2
3
t
A final integration w.r.t time (t ) and application of the initial condition w = 0 at
w=

t = 0 , give:

3 Va x 2
x
w=
L t .
3
2 h
3
Compared eq <146> with eq <3-83> w =

-----<146>
Va x 2
x
L , both show the same spatial
2D
3

41

dependence of the deflection, but the deflection of an elastic plate is time-independent

and the deflection of a viscous plate increases linearly with time.
Now return to the viscous folding problem by considering the buckling of a viscous
plate contained between two semi-infinite viscous fluids. If the approximation

yy = 0 , we made in the bending moment of a viscous plate, the plate viscosity 1

must be much larger than the viscosity 0 of the surrounding half-space. Eq <141>s
t

2x a
solution can take to be the form <123> w = wm cos
e .

The responses of the

semi-infinite fluids to the deformation of the viscous plate are identical to their
responses to the bending of an elastic plate. The force q per unit area on the
viscous plate is given by eq <128>. Substitute <128> into <141>, we obtain:
8 0 w
2w
=
P 2 .
t
3 x 4 t
x
With w given by eq <123>, we have:

1 h 3 5 w

a =

1 2
4 2

0 + 2 1 h 3 .
3
P

-----<147>

-----<148>

The wavelength corresponding to the smallest value of a w.r.t. equal to zero;

the result is:
1

1 1 3
.
= 2h
6

-----<149>

This is the wavelength of the most rapidly growing mode.

14. Stokes Flow

A solid body will rise or fall through a fluid if its density is different from the density
of the fluid. If the body is less dense, the buoyancy force will cause it to rise; if its
more dense, it will fall. If the fluid is very viscous, the Reynolds number Re based
on the size of the body, the velocity at which the body moves through the fluid, and
the viscosity of the fluid will be small. In the limit Re << 1 , inertia forces can be
neglected. Eqs <33>, <44>, <45> are applicable.
If the body has a spherical shape, a relatively simple solution can be obtained in the
limit of a very viscous fluid. The resulting flow is known as Stokes Flow. The
application include the magma visiting through the lithosphere (plume). Stokes
solution can be used to estimate the size of a mantle heterogeneity (xenoliths)
entrained in mantle flow. One model for the ascent of magma in the mantle is that

42

magma bubbles rise under the buoyancy force. Stokes solution can be used to
estimate the rate of magma ascent as a function of the size of magma bubbles.
We derive an expression for the velocity of ascent or decent U of a spherical body
in a constant-viscosity fluid with different density. We first calculate the net force or
drag exerted by the fluid on a sphere, and then equate this force to the buoyancy force
responsible for the spheres solution. For calculating the drag on the sphere due to
its steady motion through the fluid, we can consider the sphere to be fixed and have
the fluid move past the sphere. We dont discuss the transient period during which
the sphere accelerates to its final steady or terminal velocity.

Consider a sphere radius a centered at the origin of a spherical coordinate system

( r , , ) as fig 6-31. The fluid approaches the sphere at z = with velocity
U in the z-direction. The flow is clearly symmetric about the z-axis. Thus,
neither the velocity nor the pressure p of the fluid depends on the azimuthal angle

. In addition, there is no azimuthal component of fluid motion; that is, the only
non-zero component of fluid velocity are the radial velocity U r and the meridional
velocity u . The continuity equation and the equations of motion for the slow,
steady, axisymmetric flow of a viscous incompressible fluid one, in spherical polar
coordinates with u = 0 ,
u = 0 0 =

1 2
1
(sin u ) ,
r ur +
2
r sin
r r

-----<150>

43

p + V 2 u = 0
0=

1 2 u r

u 2u
p

2
1
(u sin )
+
r
+ 2
sin r 2r 2
r
r r sin
r
r sin
r r
,

-----<151>

1 2 u

u
u 2 u r
1 p
1

2 2
0=
+ 2
+ 2
sin
+ 2
r
r
r r sin
r r sin
r r
.
-----<152>
We need to find a solution subject to the condition that the fluid velocity approaches
the uniform velocity U is the z-direction as r + . The radial and meridional
components of the uniform velocity are U cos and U sin , respectively.
Therefore, we can write:
u r U cos and

u U sin as r .

We must also satisfy the no-slip velocity boundary condition or r = a ,

u r = u = 0 on r = a .

-----<153>
-----<154>

The nature of the boundary conditions suggests that we try a solution of the form:
u r = f (r ) cos and u = g (r )sin .
-----<155>
Substitute eq <155> into eqs <150> ~ <152>, we obtain:
g=

1 d 2
(r f ),
2r dr

-----<157>

p cos
+
r
r2

d 2 df

-----<158>
4( f + g ) ,
r
dr dr

p sin d 2 dg
0=
-----<159>
+
2( f + g ) .
r

r dr dr

We can eliminate the pressure by differentiating eq <158> w.r.t. and substracting

0=

the derivative of eq <159> w.r.t. r to obtain:

1 d 2 df 4( f + g ) d 1 d 2 dg 2( f + g )
0= 2
+
r

. -----<160>
dr r dr dr
r
r dr dr
r2

The solution of eq <157> and <160> for the function f and g can be found as
simple powers of r . Thus we get:
f = cr n ,
------<161>
where c is a constant. Eq <157> gives:
g=

c(n + 2) n
r .
2

-----<162>

By substituting eqs <161> and <162> into eq <160>, we find that n must satisfy:
n(n + 3)(n 2 )(n + 1) = 0, n = 0, 3 ,2 ,1.

44

The function f and g are thus linear combinations of r 0 , r 3 , r 2 , and r 1 .

c 2 c3
+ + c4 r 2 ,
3
r
r
c
c
g = c1 + 23 3 2c 4 r 2 ,
2r
2r
where c1, c 2 , c3 , and c 4 are constants. The velocity components u r and u are
f = c1 +

given by
c
c

u r = c1 + 23 + 3 + c 4 r 2 cos .
r
r

c
c

u = c1 + 23 3 2c 4 r 2 sin .
2r
2r

Since u r and u must satisfy conditions <153> as r ,

c1 = U , and c 4 = 0 .

-----<156>
-----<157>

-----<158>

The no-slip condition on r = a (eq<154>) requires that:

3aU
a 3U
and c3 =
.
2
2
The final expressions for u r and u are:
c2 =

a 3 3a
u r = U 1 3 + cos ,
2r
2r

a 3 3a
u = U 1 3 sin .
4r
4r

-----<159>

-----<160>
-----<161>

The pressure associated with the flow can be found by substituting eqs <160> and
<161> into eq <152> and integrating with respect to ,
p=

3aU
cos .
2r 2

-----<162>

Both pressure forces and viscous forces act on the surface of the sphere. By
symmetry, the net force on the sphere must be in the negative z-direction. This net
force is the drag D on the sphere. We first calculate the contribution of the pressure
forces to the drag. The pressure force on the sphere acts in the negative radial
direction. The component of this force is, per unit area of the surface,
p cos =

3U
cos 2 .
2a

-----<163>

The pressure drag D p is obtained by integrating the product of the force per unit
area with the surface area element 2a 2 sin d over the entire surface of the sphere,

D p = 3aU sin cos 2 d = 2aU .

0

-----<164>

The viscous stresses acting on an area element of the spheres surface are the radial

45

viscous stress rr .

( rr )r =a

u
= 2 r ,
r r = a

-----<165>

u 1 u r
= r +
.
r r r r = a

-----<166>

and the tangential stress r

( r )r =a

By substituting eqs <160> & <161> into <165> & <166>, we find:
( rr )r =a = 0 ,

-----<167>

3U sin
.
-----<168>
2a
The nonzero tangential stress r is a force per unit area in the direction. The

( r )r =a

component of this force per unit area in the negative z-direction is:
3U sin 2
( r )sin =
.
-----<169>
2a
The viscous drag Dv is found by integrating the product of this quantity <169> with
the surface area element 2a 2 sin d over the entire surface of the sphere,

Dv = 3aU sin 3 d = 4aU .

0

-----<170>

The total drag on the sphere is the sum of the pressure drag and the viscous drag:
D = D p + Dv = 6aU .

-----<171>

This is the well-known Stokes formula for the drag on a sphere moving with a small
constant velocity through a viscous incompressible fluid. Stokes resistance law is
often written in dimensionless form by normalizing the drag with the product of the
pressure

1
f U 2 ( f is the density of the fluid) and the cross-sectional area of the
2

sphere a 2 . The dimensionless drag coefficient C D is thus:

D
12
24
CD
=
=
,
1
Re
2
2
Ua

(
)
f
fU a
2

where the Reynold number is given by

f U ( 2a )
Re =
.

-----<172>

-----<173>

The Stokes drag formula can be used to determine the velocity of a sphere rising
buoyantly through a fluid by equating the drag to the gravitational driving force. If
the density of the sphere s is less than the density of the fluid f . The net

46

upward buoyang force according to Archimedes principal is

4

F = ( f s )g a 3
-----<174>
3

Set this equal to the drag on the sphere 6aU and solve for the upward velocity
U to obtain

U=

2( f s )ga 2
9

This result is only valid for the Reynold number is of order 1 or smaller.
For larger Re #, the flow of a fluid about a sphere becomes quite complex.
Vortices are generated and the flow becomes unsteady. The dependence of the drag
coefficient C D for a sphere on Re # is given in fig. 6-23. The Stokes flow from eq
<172> is a valid approximation for Re < 1. The sharp drop in C D at Re = 3 10 5
is associated with the transition to turbulent flow. The dependence of C D on Re
for a sphere is similar to the dependence of f on Re for pipe flow given in fig 6-7.

In terms of the drag coefficient, the upward velocity of a sphere from eqs <172>,
<173> is given by:
1

8 ag ( f s ) 2
.
U =
------<176>
CD f
3

The drag coefficient C D can be obtained from Re # and fig 6-23. We can

estimate the velocity of magma ascent through the lithosphere. Refractory peridotite
xenoliths with a maximum dimension of about 0.3 m have been found in the basaltic
lava erupted in 1801 at Hawaii. An upper limit on the size of xenoliths that can be
entrained is obtained by setting the relative velocity U equal to the flow velocity of
the magma.
The viscosity of the basaltic magma is estimated to be 10 Pas. Assume

47

s m = 600 kg

m3

and a = 0.15m , we find from eq <175> that

U = 3 m ( 10.8 km ). The corresponding value of Re # from eq <173> with

s
hr

f = 2700 kg

m3

is 243. Therefore, the Stokes formula is only approximately valid.

Using eq <176> and the empirical correlation given in Fig. 6-32, we find
U = 0.87 m
10 mm

yr

and Re = 70 . On the other hand, we take a typical mantle velocity

, = 100 kg

m3

, = 10 21 Pas , and g = 10 m

s2

. Form eq <175>, we

find that the spherical bodies with radii less 38 km will be entrained in mantle flows.
The conclusion is that sizable inhomogeneities can be carried with the mantle rocks
during mantle convection.
One model for magma migration is that magma bodies move through the mantle
because of the differential buoyancy of the liquid. The velocity of a bubble of
low-viscosity fluid moving through a high-viscosity fluid because of buoyancy is
given by
U=

a 2 g ( f b )

3 f

-----<177>

where b is the density of the bubble and f is the density of the surrounding
fluid, and f is the viscosity of the ambient fluid. Eq <177> differs from eq
<175> in that the boundary conditions for <177> is free-slip; that is, u r = 0, r = 0
at r = a . Taking a = 0.5km, f b = 600 kg
U = 0.016 mm

yr

m3

, = 10 21 Pas , we find

. Even for a relatively large magma body, the migration velocity is

about 13 orders of magnitude smaller than that deduced from xenoliths.

The calculated velocity of 0.016 mm

hr

is unreasonably slow, because it would take

about 10Gyr to migrate 100 km. If the magma doesnt penetrate the lithosphere by
diaspirism, an alternative mechanism is hydrofracturing. Liquid under pressure can
fracture rock, it has been suggested that the pressure caused by the differential
buoyancy of magma can result in the propagation of a fracture through the lithosphere
along with the magma migrates.

48

15. Plume Heads and Tails

A simple steady-state model for the ascent of a plume head through the mantle is
given in fig 6-33. The plume head is modeled as a spherical diapir whose velocity is
given by the stokes flow solution. Using the solution of prob. 6-23, we can write the
terminal velocity U of the ascending plume head from eq <177>,
U=

a 2 g ( m p )
3 m

------<178>

where a is the radius of the plume head, p is the density of the hot plume rock, m
is the density of the surrounding mantle rock. We take T p to be the mean
temperature of the plume rock and T1 to be the temperature of the surrounding rock.

From eq <4-172>, we get

p m = m v (T p T1 ) .

------<179>

Substitute eq <179> into <178> to get the ascent velocity of the plume head,
U=

a 2 g m v (T p T1 )
3 m

------<180>

The plume tail is modeled as a cylindrical pipe and the buoyancy driven volume flux
Q p of the plume rock is given by eq <30>,
4
( m p )gR
,
Qp =
8
p

------<181>

49

where R is the radius of the plume tail and p is the viscosity of the plume rock. A
measure of the strength of a plume is the buoyancy flux b, which is defined by
B = Q p ( m p ) .

-----<182>

A combination of eqs <179>, <180>, <181> gives

2
4 2
2
gR m (T p T1 ) v
.
B=
8
p

-----<183>

The total heat flux in a plume Q H is related to the volume flux by

QH = m c p (T p T1 )Q p ,

-----<184>

where c p is the specific heat at constant pressure. A combination of eqs <179>,

<182>, <184> gives
QH =

cpB

-----<185>

In the steady-state model, the plume head neither gains nor loses fluid; this requires
that the mean flow velocity in the plume tail equals the ascent velocity of the plume
head U , therefore,
Q p = R 2U .

-----<186>

Once the plume flux B has been specified along with the other parameters, the radius
of the plume tail R can be determined by eq <183>, the heat flux in the plume QH
by <184>, the ascent velocity of the plume head U by <186>, and the radius of the
plume head a by <180>.
As pointed out in section 1-6, the ascent of mantle plumes forms the topographic
swells, referred as hotspots. The buoyancy flux associated with a mantle plume can be
determined from the rate of hotspot swell formation. We hypothesize that the excess
mass associated with the swell is compensated by the mass deficit of the hot (light)
plume rock impinging on the base of the lithosphere. Thus the buoyancy flux B is
given by
B = ( m w )AS u P ,
-----<187>
where m is the mantle density, w is the water density, AS is the cross-sectional
area of the swell in a vertical cross section perpendicular to the plume track, and u P
is the plate speed relative to a fixed hotspot reference frame.

50

Ex. The Hawaii hotspot.

Take u P = 90 mm

yr

, AS = 1.33km 2 (taken from fig. 1-19), w w = 2300 kg

and we find B = 7.4 10 3 kg

. Taking c p = 1.25 kJ

kg oK

m3

and v = 3 10 5 o K 1 ,

the plume heat flux from eq <185> is QH = 3 1011W , slightly less than 1% total
surface heat flux. The radius of the Hawaii plume R from <183> is 84km . (assume
B = 7.4 10 3 kg

m = 3300 kg

m3

, p = 1019 Pas , T p T1 = 200 o K , v = 3 10 5 o K 1 ,

, and g = 9.8 m

s2

). The radius R = 84 km is relatively small and

explains why plumes are very different to observe seismically. From eqs <179> and
<182>, the volume flux in Hawaii plume Q p = 12 km

yr

. However, the volume

flux of basalt Qv required to create the Hawaii Islands and seamount chains is
estimated to be 0.1 km

yr

. Thus, it only needs to melt ~1% of the plume flux to

generate hotspot volcanics at Hawaii.

The buoyancy fluxes for 43 plumes are given in Table 6-4. The total buoyancy flux
for three plumes B = 58.5 10 3 kg

. Taking c p = 1.25 kJ

kg oK

and

v = 3 10 5 o K 1 , the total plume heat flux from <185> QH = 0.244 1013 W

(~5.5% of the total globe heat flow Q = 4.43 1013 W ). In section 4-23, the basal
heating of the oceanic and continental lithosphere Qm is 1.58 1013 W . Therefore,
the derived plume heat flux is only 15% of the total heat flux associated with the basal
heating of the lithosphere. The difference can be attributed either to plumes that

51

impinge on the base of the lithosphere but are too small to have a surface expression
or to secondary mantle convection involving in the lower part of the lithosphere.
Ex. 2. Runion hotspot and the flood basalt province of the Deccan Traps in fig.
1-22.

From table 6-4, the present buoyancy flux of the Runion plume B = 1.4 10 3 kg

From <181>, the radius of the plume conduit R = 55km ; from <179> and <182>, the
volume flux Q p = 2.2 km
U = 0.23 m

yr

yr

; from <186>, the mean ascent velocity of the plume

. Assume the strength of the Runion plume has remained constant

for the last 60Myr , taking m = 10 21 Pas , we find from <180> that the radius of the
plume head a = 336 km . The corresponding volume of the plume head
V PH = 1.2 10 8 km 3 . The volume of the basalts in the Deccan Traps
V B 1.5 10 6 km 3 . Thus, it needs to melt ~1% of the plume head to form the flood
basalts of the Deccan Traps. Assuming the volume flux of the Runion plume
QP = 2.2 km

yr

and has remained constant over the 60 Myr life time of the plume,

the total volume flux through the plume tail has been 1.3 10 8 km 3 . For the ascent
velocity of the plume head U equal to 0.23 m

yr

, it could take about 12Myr for

the plume head to ascend from the CMB to the surface.

52

16.

Pipe flow with heat addition

We now consider the problems involving both fluid and heat transfer. Consider
an example of the heating of water in an aquifer as the heat balance on a thin,
cylindrical shell of fluid in a pipe. The thickness of the shell is r , and its length is
x (fig. 6-33). The heat conducted out of the cylindrical surface at r + r per unit
time is:

2 (r + r )xq r (r + r )

where q r (r + r ) is the radial heat flux at r + r . The heat conducted into the shell
across the inner cylindrical surface is:
2rxq r (r ) .
Since r is small, we can expand:
q r
r + ........
r
By neglecting higher power of r , we can write the net rate at which heat is
q r (r + r ) = q r (r ) +

conducted into the cylindrical shell through. Its inner and outer surface as:
Its inner and outer surface as:

dq
2x[rq r (r ) (r + r )q r (r + r )] = 2x r r + q r r . -----<188>

dr
In cylindrical coordinates, the radial heat flux q r is related to the radial temperature
gradient

T
by Fouriers law of heat conduction; i.e.,
r
T
,
q r = k
r

-----<189>

where k is the thermal conductivity of the fluid. Eq <188> for the net effect of
radial heat conduction can thus be rewritten as:
2T T
.
2xrk r 2 +
r
r
The amount of heat converted out of the shell at x + x by the velocity u (r ) per

unit time is given by:

2rrucT ( x + x ) .

53

The amount of heat converted into the shell at x per unit time is given by
2rrucT ( x ) .
By expanding T ( x + x ) = T ( x ) +

T
x + ........ , we find the net rate at which fluids
x

carry heat out of the shell is

2rruc[T ( x + x ) T ( x )] = 2rruc

T
x .
x

-----<190>

If the flow is steady so that the temperature of the fluid doesnt change with time and
if axial heat conduction is unimportant compared with advection of heat by flow, the
net effects of radial heat conduction and axial heat advection must balance. Therefore,
we can equate <190> with <188>,

uc

2T 1 T
T
.
= k 2 +
x
r

-----<191>

By equating axial heat advection to radial heat conduction, we also assumed that
viscous dissipation or frictional heating in the fluid is negligible.
We can determine the temperature distribution in the pipe using eq <191> for the
laminar flow in section 6-4. The velocity as a function of radius can be expressed in
terms of the mean velocity u by combining eqs <16> & <20> to give
r 2
u = 2u 1 .
R

-----<192>

If the wall temperature of the pipe is Tw , changing linearly along its length; i.e.,
Tw = c1 x + c 2 ,
-----<193>
where c1 and c 2 are constants. Accordingly, we assume that the temperature of
the fluid is given by

T = c1 x + c 2 + (r ) = Tw + (r ) .

-----<194>

(In this situation, the net contribution of axial heat conduction to the heat balance of a
small cylindrical shell vanishes). Thus is the difference between the fluid
temperature and the wall temperature. Substitute <192> and <194> into <191>,
r 2
d 2 1 d
.
2 cu 1 c1 = k 2 +
r dr
R
dr

-----<195>

The boundary conditions are

T = Tw

at r = R ,

-----<196>

qr = 0

at r = 0 .

-----<197>

The latter is required because there is no line source or sink of heat along the axis of
the pipe. Condition <196> is satisfied if

54

r=R = 0 ,

-----<198>

and condition <197> with the Fouriers law yields

d

= 0.
dr r =0

-----<199>

The solution of eq <195> that satisfies these boundary conditions is

cuc1 R 2
8k

r2 r4
3 4 2 + 4 .
R
R

-----<200>

The heat flux to the wall q w can be found by substituting eq <200> into Fourier law
<189> and evaluate the result at r = R ,
1
-----<201>
q w = cuRc1 . (c: specific heat)
2
The heat flux is a constant, independent of x . If c1 > 0 , Tw increases in the
direction of flow, the heat flows through the wall of the pipe into the fluid. If c1 < 0 ,
Tw decreases in the direction of flow, and heat flows out of the fluid into the wall of
pipe. The heat flux to the wall can be expressed in a way by introducing heat
transfer coefficient h between the wall heat flux and the excess fluid temperature
according to

q w = h T Tw = h ,

-----<202>

where the overbar represents an average over the cross section of the pipe. The
flow-weighted average excess fluid temperature is
R

2 urdr
0

R 2 u

11cuc1 R 2
.
48k

-----<203>

By combining eqs <201> & <203>, we get the heat transfer coefficient for laminar
flow in a circular pipe,
h=

48k
,
11D

-----<204>

where D = 2 R is the diameter of the pipe. Eq <204> is only valid for Reynolds #
< 2200. At greater Re #, the flow is turbulent.

In fluid mechanics, its often introducing a dimensionless measure for the heat
transfer coefficient, known as the Nusselt number Nu, which for pipe flow with heat
addition is defined as
Nu

hD 48
=
= 4.36 .
k
11

The Nu # measures the efficiency of the heat transfer process. If temperature

difference T Tw across a stationary layer of fluid of thickness D and thermal

55

conductivity k , the conductive heat flux qc would be:

qc =

k T Tw
q k
= w .
D
Dh

-----<205>

Thus the Nu # is:

Nu =

qw
qc

-----<206>

and heat transfer with fluid flow through the pipe is 4.36 times more efficient than
conductive heat transport through an equivalent stationary fluid layer across which the
same temperature difference is applied.

17.

Aquifer model for hot springs

The result derived from the precious section can be used to study the heating
water flowing through an aquifer surrounded by hot rocks. Consider the same
semicircular aquifer with circular cross section in fig. 6-9, the heat convected along
the aquifer is balanced by the heat lost or gained by conduction to the walls. We can
write:

R 2 cu

dT
= 2Rh Tw T ,
ds

-----<207>

where s is the distance measured along the aquifer from the entrance, u is the
mean velocity in the aquifer, T is the flow-averaged temperature of the aquifer fluid
and Tw is the temperature of the aquifer wall rock. Assume a laminar flow so that
the heat transfer coefficient h in eq <204> can be used. The coordinate s can be
related to the angle by
s = R .
-----<208>
Assume the wall temperature Tw is related to local thermal gradient by
Tw = R sin + T0 ,
-----<209>
where T0 is the surface temperature. Substitution of eqs <204>, <208>, <209> into
eq <207>, yields

R 2 cu d T 48
=
k R sin + T0 T .
R d 11

------<210>

The equation can be simplified through the introduction of the Pclet number Pe
defined by
Pe

cuR

.
-----<211>
k
The Pe number is a dimensionless measure of the mean velocity of the flow through
the aquifer. Its related to Re and Pr already introduces. Since (thermal
k
diffusivity ) =
, Pe can be written as
c

56

Pe =
Using the definitions of Re

uR

-----<212>

uD
( D = 2R ) and Pr , we can rewrite <212>

1 u 2R 1
= Re Pr .
-----<213>
2 2
The simplification of eq <210> is done by the introduction of a dimensionless
temperature defined by:
Pe =

T T0
.
R
With eqs <211> & <214>, we can put eq <210> into the form
11 R
d
+ = sin .
Pe
48 R
d

-----<214>

-----<215>

This is a linear 1st-order differential equation that can be integrated using an

integration factor. With the boundary condition that the water entering the aquifer is
at the surface temperature, T = T0 or = 0 at = 0 , the solution can be written
1

2
48 R
48 R 48 R 48 R
sin cos + exp
=

1 +
. -----<216>
11 RPe 11 RPe 11RPe
11RPe

The dimensionless temperature e at the exit of the aquifer, = , is

48 R 48 R
exp 11 RPe + 1 11 RPe

.
e =
2
48R
1+

11RPe

e (the exit temperature) is plotted as a function of

-----<217>

RPe
in fig 6-35. e is
R

57

RPe
= 5 . Thus, for given values of all parameters other than u ,
R
there is a particular flow rate through the aquifer that maximizes e . The maximum
e is about one-half of the maximum wall temperature at the base of the aquifer,

maximum when

since e =

1
1
, corresponds to Te = T0 + R , and Tw at =
is
2
2
2

T0 + R (T0 << R ) .

To understand why there is a maximum exit temperature, we show the mean

temperature of the water in the aquifer as a function of position in fig. 6-36 for three
flow rates. The dimensionless wall or rock temperature w ,

w =

Tw T0
.
R

RPe
= 1 , the water temperature follows the wall temperature
R
because of large heat transfer, and the exit temperature e is low. For very

For a low flow rate,

slow rate,

RPe
0 , = w = sin , and the exit temperature = the entrance
R

temperature and no hot spring. For high flow rate

transfer, and the water doesnt heat up.
the entrance temperature, no hot springs.

When

RPe
= 15 , there is very little heat
R

RPe
, the water temperature =
R

58

18.

Thermal Convection

Plate tectonics is a consequence of thermal convection in the mantle driven

largely by radiogenic heat sources and the cooling of the earth. When a fluid is heated
from below or from within and cooled from above, it will has dense cool fluid near
the upper boundary and hot light fluid at depth. The situation is gravitationally
unstable, and the cool fluids tend to sink and the hot fluid tends to rise.

Density variation caused by thermal expansion lead to buoyancy forces that drive
thermal convection. Thus its essential to account for density variations in the
gravitational body force term of the conservation of momentum equation. However,
the density variations are sufficiently small so that they can be neglected in the
continuity equation. This is known as the Boussinesq Approximation. It allows us to
use the incompressible conservation of mass equation <33>. The force balance
equations <41> & <42> are also applicable by adding the vertical buoyancy forces
due to small density variations, let
= 0 + ,
-----<219>
where 0 is a reference density and << 0 . Equation <42> can be rewritten:
0=

2v 2v
p
+ 0 g + g + 2 + 2
y
y
x

-----<220>

Use P = p 0 gy to eliminate the hydrostatic pressure corresponding to the

reference density 0 , and eqs <41> & <220> become
0=

2u 2u
P

+ 2 + 2 ,
x
y
x

<221>

0=

2v 2v
P
+ g + 2 + 2 .
y
y
x

<222>

59

Density variations due to temperature changes are given by:

= 0 v (T T0 ) ,
------<223>
where v is the volumetric coefficient of thermal expansion and T0 is the reference
temperature corresponding to 0 . Substitution of <223> into <222> gives
0=

2v 2v
P
+ 2 + 2 g 0 v (T T0 )
y
y
x

-----<224>

The last term in <224>, the buoyancy force per unit volume depends on the
temperature. Thus the velocity field cant be determined without simultaneously
solving for the temperature field. Therefore we require the heat equation. The
energy balance must account for heat transport by both conduction and convection.

Consider a small 2-D element shown in fig. 6-37. Since the thermal energy content
of the fluid is cT per unit volume, an amount of heat cTuy is transported
across the right side of the element by velocity u in x-direction, this is the energy
flow per unit time and per unit length in the direction to the figure. Then the
energy flow rate per unit area at x + x is cTu +

(cTu )x . The net energy

x

advected out of the element per unit time and per unit length due to flow in the
x-direction is thus

-----<225>
cTu + (cTu )x cTu y = (cTu )xy .
x
x

The same analysis in the y-direction gives

cTv + (cTv )y cTv x = (cTv )xy .

y
y

convection
-----<226>

The net rate at which the heat is conducted out of the element unit length has been
derived in <4-49>,

60

2T 2T
k 2 + 2
y
x

-----<227>

Conservation of energy requires that the combined transport of energy out of the
element by conduction and convection must be balanced by the change in the energy
content of the element. The thermal energy of the fluid is cT per unit volume.
Thus, this quantity changes at the rate

(cT )xy per unit length of the fluid. By

t

combining the effects of conduction, convection and thermal inertia, we obtain

2
2

(cT ) k T2 + T2 + (cuT ) + (cvT ) = 0 .

y
t
y x
x

-----<228>

By treating and c as constants and noting that

(uT ) + (vT ) = u T + v T + T u + v = u T + v T ,
y
x
x
y
x
y
x y
= 0 (continuity eq.)

and =

k
, we get the heat equation for 2-D flows
c
2T 2T
T
T
T
+u
+v
= 2 + 2 ,
t
x
y
y
x

-----<229>

T K
+ u T = 2T .
t

In <229>, we have neglected frictional heating in the fluid associated with the
resistance to flow and compressional heating associated with the work done by
pressure force in moving the fluid.
Note: Material derivative
In fluid mechanics, ones concern is normally with the fluid velocity as a function of
space and time, rather than trying to follow any particle displacements or paths in
solid mechanics as called the Lagrangian description. The description of the flow at
every fixed point as a function of time is called the Eulerian formulation. For any
quantity Q( x, y, z , t ) that represents a given property of the fluid, the total

differential change of the quantity Q o associated with the changes dx, dy, dz, dt is
given by
Q
Q
Q
Q
dQ =
dx +
dy +
dz +
dt.
x
y
z
t
The spatial increments by following a paricular particle are dx = udt , dy = vdy , dz = wdz.
Therefore, the time derivative of Q of a particular particle is,

61

dQ DQ Q
Q
Q
Q Q G G
G
=
=
+u
+v
+w
=
+ u Q, where u = (u, v, w).
dt
Dt
t
x
y
z
t

The quantity DQ / Dt is termed the substantial derivative, particle derivative or

material derivative and Q / t is called the local derivative.

19.

Linear stability analysis for the onset of thermal convection in

a layer of fluid heated from below.

b
At y = , cold boundary T = T0
2

At y =

b
, hot boundary T = T1
2

(T1 > T0 )

Assume no heat source in the fluid, buoyancy forces tends to drive convection when
fluids near the heated lower boundary becomes hotter and lighter than the overlying
fluid and tend to rise. Upper boundary is denser than the fluid below and tends to
sink. However, the motion doesnt take place for small temperature differences across
the layer because the viscous resistance of the medium to flow must be overcome.
We now determined the minimum (T1 T0 ) required for convection to occur.
For sufficiently small (T1 T0 ) , the fluid is stationary ( u = v = 0 ), and we can

= 0 exits. The energy

assume that a steady = 0 conductive state with
x
t

equation <229> then simplifies to

d 2Tc
=0,
-----<230>
dy 2

where subscript c indicates that this is a conduction solution. The eq <230> satisfies
the b.cs T = T0 at y =

b
b
and T = T1 at y = + , so the temperature profile is
2
2

T1 + T0 T1 T0
+
y.
-----<231>
b
2
The stationary conductive state will persist until T1 T0 reaches a critical value at
Tc =

which even the slightest further increase in temperature difference will cause the layer
to become unstable and convection to occur. Thus, at the onset of convection, the

62

fluid temperature is nearly conduction temperature profile and the temperature

difference T ,
T1 + T0 (T1 T0 )

y,
-----<232>
b
2
is arbitrary small. The convective velocities u , v are also infinitesimal when
T T Tc = T

motion first takes place. The energy eq that pertains to the onset of convection can
be written in terms of T by solving <232> for T and substituting into <229>. We
get
2T 2T
T
T
T v(T1 T0 )
+ u
+ v
+
= 2 + 2
t
x
y
b
y
x
T , u , v are small quantities, the nonlinear terms

-----<233>

u T
v T
and
on the left
x
y

side of <233> are much smaller than the remaining linear terms.
neglected and <233> can be written as:

Thus, they can be

2T 2T
T v
.
+ (T1 T0 ) = 2 +
2
t
b

x
y

-----<234>

The neglect of the nonlinear terms, the terms involving products of the small
quantities u , v , and T , is a standard mathematical approach to problems of
stability. Its known as a linearized stability analysis. Its valid for the study of the
onset of convection when the motion and the thermal disturbance are infinitesimal.
To summaries, the equations for the small perturbations of temperature T , velocities
u , v and pressure p when the fluid layer becomes unstable are

u v
= 0,
+
x y

-----<235>

0=

2u 2u
P
+ 2 + 2 ,
x
y
x

-----<236>

0=

2 v 2 v
P
0 v gT + 2 + 2 ,
y
y
x

-----<237>

2T 2T
T v
+ (T1 T0 ) = 2 + 2 .
t b
y
x

-----<238>

We have takes the buoyancy force at any point in the layer to depend only on the

63

departure of the fluid temperature from the basic conduction temperature at the point
( 0 v gT ) .
Eqs <235>-<238> are solved subject to the following b.c.s: the surface y =

b
are
2

isothermal and no flow occurs across them; i.e.,

b
T = v = 0 on y = .
2

-----<239>

If the boundaries of the layer are solid surface,

u = 0 on y =

b
.
2

(no-slip conduction)

-----<240>

b
b
are free surfaces; i.e., if there is nothing at y = to exert
2
2
a shear stress on the fluid, u needs not vanish on the boundaries. Instead, the shear

If the surfaces y =

stress yx must be zero on y = ,

2

yx = 0 on y =
b
2

v u
b
+
= 0 on y =
x y
2

-----<241>

b
v
b
Because v = 0 on y = , and
0 on y = , the free surface bcs are
2
2
x
u
b
= 0 on y = .
-----<242>
y
2

A simple analytic solution can be found if the free surface conditions are adopted.
Introduce the stream function defined in eqs <46> that automatically satisfies the
continuity equation <235>. Eqs <236> - <238> can be written:
0=

3 3
P
2 +
,
3
x
x y y

-----<243>

0=

3 3
P
0 g v T 3 + 2 ,
y
y x
x

-----<244>

2T 2 T
T 1

.
+ (T1 T0 )
= 2 +
t b
x
y 2
x

-----<245>

Eliminate the pressure from eqs <243> & <244> yields:

4
4
4
T
0 g v
0 = 4 + 2 2 2 +
.
4
x
x y
y
x

-----<246>

Eqs <237> and <238> can be solved for and T by the method of separation of

64

variables. The boundary conditions <239> & <242> are automatically satisfied by
the solutions of the form:
y 2x t
sin
e ,
b
y 2x t

T = T0 cos cos
e .
b

= 0 cos

-----<247>
-----<248>

For > 0 , The disturbance will grow with time, the heated layer is convectively
unstable. For < 0 , the disturbance will decay in time, and the layer is stable
against convection. We can determine by substituting <247> & <248> into
<245> & <246>, we find:

2 4 2
+ 2 +
b
2

(T T0 )2
T0 = 1
0 ,
b

-----<249>

4 2 2
2
2 + 2 0 =
0 g vT0 .

-----<250>

The disturbance amplitudes 0 and T0 can be eliminated by division, yielding an

equation to solve for . The growth rate is found to be

4 2 b 2

2 2
2
0 g v b 3 (T1 T0 )

2 + 4 b
= 2

2 2

2
b
4 b + 2
2

The dimensionless growth rate

.-----<251>

b 2
2b
(a
depends on only two quantities,

dimensionless wave number), and a dimensionless parameter known as Rayleigh

number Ra.
0 g v (T1 T0 ) b3
Ra =
.
-----<252>

In terms of Ra number, we can write <251> as:

4 2b 2

2 4 2b 2
+
Ra
2
2
b 2

=
.
2

2 4 b 2
+ 2

The growth rate is positive and there is instability if

-----<253>

65

2 4 2b 2
+
2
Ra >
.
4 2b 2

-----<254>

2
The growth rate is negative and there is stability if Ra is less than right side of eq
<254>. Convection just sets in when = 0 , which occurs when
2 4 2b 2
+
2
Ra Ra cr =
4 2b 2

-----<255>

2
The critical value of Rayleigh number Ra cr makes the onset of convection. If
Ra < Ra cr , disturbance decays with time; if Ra > Ra cr , perturbation will grow
exponentially with time.
Eq <255> shows that Ra cr is a function of the wavelength of the disturbance (see fig.
6-39). If Ra number and disturbance wavelength are such that the point lies above
the curve, the perturbation of wavelength is unstable; if the point lies below the
curve, convection cant occur with disturbance of wavelength .

Eg. If Ra= 2000, disturbance with 0.8

For

2b

0.8 and

2b

2 b

5.4 are convectively unstable.

5.4, convection cant occur. Fig 6-39 shows that there is

a minimum value of Ra cr . If Ra lies below the minimum, all disturbances decay,

66

the layer is stable, and convection cant occur.

The value of

2b

at which Ra is a minimum can be obtained by setting the

derivative of the right side of <255> w.r.t.

2b

equal to zero,

2
3
2
4 2b 2
Ra cr
4 2b 2 2 b 2 4 2b 2 2 b 4 2b 2
2
3 +
=
2
2
=0
+

2
2
2

2 b 2

or

2b

-----<256>

The value of the wavelength corresponding to the smallest value of the critical
Rayleigh number is

= 2 2b .
-----<257>
Substitution of this value back to eq <255> gives the minimum critical Rayleigh
number
27 4
min ( Ra cr ) =
= 657.5 .
-----<258>
4
When Ra exceeds Ra cr for convection to occur, one can think of the temperature
difference T1 T0 across the layer as having to exceed a certain minimum value or
the viscosity of the fluid having to lie below a critical value before convection sets in.
For examples, by increasing T1 T0 with other quantities fixed, when Ra reaches
657.5 (for heating from below with stress-free boundaries), convection sets in and the
aspect ratio of each convection cell is 2 . The minimum Ra cr and the
corresponding disturbance wavelength for no-slip velocity boundary conditions can
be solved numerically. For that case, min Ra cr = 1707.8 and = 2.016b .
The linear stability analysis for the onset of convection can be carried out for a
fluid layer heated uniformly from within and cooled from above. The lower
boundary is assumed to be insulating; i.e. no heat flows across the boundary. The
fluid near the top is cooled and denser; therefore buoyancy forces can drive fluid
motion if they are strong enough to overcome the viscous resistance. Thin type of
instability is directly applicable to the earths mantle. Since the earths interior is
heated by decay of the radioactive elements and the near-surface rocks are cooled by
heat conduction. The appropriate Ra for a fluid layer heated from within is
Ra H =

v 0 2 gHb5
,
k

-----<259>

67

where H is the rate of internal heat generation per unit mass. For no-slip bcs, the
minimum Ra cr is 2722 and the associated value of
min Ra cr = 867.8 and the associated value of

2b

2b

is 2.63; for free-slip bcs,

is 1.79.

We can estimate the value of Ra for the mantle. Based on the postglacial
rebound studies, we take = 10 21 Pas . The rock properties take k = 4W

m oK

= 1 mm s , and v = 3 10 5 o K 1 , g = 10 m 2 and an average density

s

0 = 4000 kg

m3

H = 9 10 12 W

. Based on Chapter 4s discussion, the heat source

kg

. If convection is restricted to the upper mantle, take b = 700km ,

we find Ra = 2 106 . For the entire mantle convection, take b = 2880km , we find

Ra = 2 109 . In either case, Ra is much greater than the minimum Ra cr . It was

essentially this calculation that led Arthur Holmes to propose in 1931 that thermal
convection in the mantle was responsible for driving continental drift.

20. A transient boundary layer theory for finite-amplitude thermal

convection.
The linear stability analysis determines whether or not thermal convection occurs, but
its not useful to determine the structure of convection when Ra exceeds Ra cr .
Because its linear, the linear stability analysis cant predict the magnitude of
finite-amplitude convection flows. To do so, the full non-linear equations must be
solved, but in general, can be only done numerically. Here we present approximate
solutions that are valid when the Rayleigh number is large and convection is vigorous.
For large value of Ra, a convecting fluid layer of thickness b heated from below is
largely isothermal. By symmetry, the isothermal temperature Tc of the bulk of the
fluid is given by
T1 T0
,
----<260>
2
where the upper boundary is maintained at temperature T0 and the lower boundary at
Tc = T0 +

T1 .

The thermal gradient between the cold upper boundary at temperature T0 and
the core at temperature Tc occurs across a thin thermal boundary layer adjacent to
the upper boundary. The thermal gradient between the core at temperature Tc and
the hot lower boundary e at temperature T1 occurs across a thin thermal boundary

68

layer adjacent to the lower boundary.

Consider a boundary-layer stability approach to the thickening thermal boundary
layers. We assume that the isothermal core fluid is in contact with the boundaries of
the fluid layer. The hot and cold thermal boundary layers form as the fluids adjacent
to the hot lower and cold upper boundaries are heated and cooled, respectively. The
boundary layers thicken until they become gravitational unstable and separate from
the boundaries. The hot lower boundary layer ascends into the isothermal core and
the cold upper boundary descends into the core. The end of the boundary-layer
growth is determined by a stability analysis of a fluid layer carried out in the previous
section.
The transient growth of the two thermal boundary layers can be described by 1-D
heat conduction analysis of Section 4-15. Consider the growth of a cold thermal
boundary layer adjacent to the upper boundary. Initially the core fluid with
temperature Tc is in contact with the boundary. Subsequently, conductive heat
losses to the surface result in the development of a thin cold thermal boundary layer.
The temperature distribution as a function of time from eq<4-113> is
Tc T
y
------<261>
= erfc
.
Tc T0
2 t
A similar expression can be written for the thickening hot boundary layer on the lower
boundary. The thickness of the thermal boundary layer from eq<4-115> is
yT = 2.32(t )

1/ 2

------<262>

The basic assumption in this approach is that a linear stability analysis can be applied
to the boundary layers. We assume that the boundary layers thicken until the
stability condition is satisfied, at which time they break away from the boundary
surfaces to be replaced by isothermal core fluid and the process repeats. The
breakaway condition is assumed to be given by the stability analysis for a fluid layer
developed in section 6-19. The applicable Ra number is given by an expression
similar to eq<252> where the boundary layer thickness b is replaced by yT from
eq<262>. The relevant temperature difference is
Tc T0 =

1
(T1 T0 )
2

-----<263>

from eq<260>. The critical value of the Rayleigh number Ra yT ,cr gives a critical
value for the boundary thickness yT ,cr
Ra yT ,cr

0 v g (T1 T0 ) yT3 ,cr 0 v g (Tc T0 ) yT3 ,cr

=
=
.

-----<264>

The critical value of this Rayleigh number for free surface boundary conditions is

69

Ra yt,cr = 657.5 . From eqs <262> and <264>, the time at which boundary layer
breakaway occurs t c is given by
2

1
tc =
5.38

2 Ra yT ,cr 3

.
0 v g (T1 T0 )

-----<265>

The mean heat flow q across the fluid layer during the time t c from equation
<4-117> is
q=

2k (Tc T0 )

(t c ) 2
1

k (T1 T0 )

(t c ) 2
1

-----<266>

The combination of eqs <265> and <266> gives

1

g (T1 T0 ) 3
q = 1.31k (T1 T0 ) 0 v
.
2 Ra yT ,cr

-----<267>

The Nusselt number Nu is defined in eq <6-267> as the ratio of the convective heat
flow across the layer q to the heat flow q c that conduction would transport,
Nu

q
,
qc

-----<268>

qc =

k (T1 T0 )
.
b

-----<269>

and

Upon substituting Equations <267> and <269> into Equation <6-333>, we obtain
Ra
Nu = 1.04
Ra yT ,cr

3
,

-----<270>

where Ra is the Rayleigh number based on the full layer thickness b and the overall
temperature difference (T1 T0 ) as defined in eq <252>. We find that the Nu is
proportional to the Ra to the one-third power. With Ra yT ,cr = 657.5 , we find

Nu = 0.12Ra1 3 .
-----<271>
Although this is an approximate solution, the dependence of the Nu on the Ra is
generally valid for vigorous thermal convection in a fluid layer heated from below
with free-surface boundary conditions.
We can apply this boundary-layer stability analysis to thermal convection in the
upper mantle. The depth of deep earthquakes associated with the descending
lithosphere at ocean trenches (about 660km) provides a minimum thickness for the

70

convecting part of the mantle. We assume that mantle convection is restricted to the
upper 700km and evaluate the Rayleigh number in eq<252>, with b = 700km ,

0 = 3700kg m 3 , g = 10m s 2 , v = 3 10 5 K 1 , T1 T0 = 1500 K , = 1mm 2 s 1 ,

and = 10 21 Pa s ; we obtain Ra = 5.7 105 . The mean surface heat flux q is
given by
q=

k (T1 T0 )
b

Nu =

0.120k (T1 T0 ) Ra

-----<272>

With the same parameter values and k = 4W m 1 K 1 , we find q = 85m W m 2 . This

is remarkably close to the Earths mean surface heat flow of 87m W m 2 given in
Section 4-4. However, such excellent agreement must be considered fortuitous.
Eq <265> for the time at which boundary layer breakaway occurs can be written
using Ra yT ,cr = 657.5 as
tc =

22.3b 2

Ra

-----<273>

Substituting the values given above for upper mantle convection we find
tc = 50.5 Myr . This is about one-half the mean age of subduction given in Figure
4-26.
The boundary-layer stability approach can also be applied to fluid layer that is
heated from within and cooled from above. In this case there is only a single
thermal boundary layer on the upper boundary of the fluid layer. The mean heat
flow out of the upper boundary q is related to the heat generation per unit mass in
the layer H by
q = 0 Hb .

----<274>

Applying eqs <265> and <266> to the upper boundary layer only, we have
1

g (T1 T0 ) 3
q = 2.62k (T1 T0 ) 0 v
,
Ra yT ,cr

-----<275>

where T1 is now the temperature of both the lower boundary and the isothermal core.
By combining eqs <274> and <275>, we can solve for the temperature of the

71

isothermal core with the result

1

Hb 4 Ra y ,cr 4
T1 T0 = 0
,

2.62k 0 v g
T

-----<276>

where T1 T0 is also the temperature rise across the fluid layer.

The efficiency with which convection cools the fluid layer can be accessed by
comparing the temperature rise across the internally heated layer given by eq <276>
with that which would be obtained if all the internally generated heat were removed

only by conduction T1c T0 . The dimensionless temperature ratio,

T1 T0
,
T1c T0

-----<277>

is thus a measure of convective efficiency for the internally heated fluid layer. The
smaller is, the more efficient convection is in removing the heat produced in the
fluid. Without convection, the temperature rise across the layer would be (see
section 4-6)
T1c T0 =

0 Hb 2

.
2k
By substituting <276> and <278> into <277> we find that

-----<278>

Ra
4
= 0.97 yT ,cr ,
Ra H

-----<279>

where Ra H is the Rayleigh number defined for a fluid layer heated from within in
eq <259>. The nondimensional temperature difference between the isothermal core
and the upper boundary decreases as convection becomes more vigorous with
increasing Rayleigh number. Taking Ra yT ,cr = 657.5 we find

= 4.91Ra1H4 .
Again, the dependence of the dimensionless temperature on Ra is generally valid for a
vigorous convecting fluid layer heated from within and cooled from above with free
surface boundary conditions, although the constant of proportionality is model
dependence. The dimensional temperature of the isothermal core is given by

Hb 2

= 2.45

Hb 2

Ra H1 4 .
-----<280>
k
2k
The boundary layer stability analysis for thermal convection in a uniformly
heated fluid layer cooled from above can be applied to thermal convection that occurs
T1 T0 =

72

throughout the whole mantle. For this case we take b = 2800km . Based on the
discussion in Chapter 4, we assume H = 9 10 12 W kg 1 . We also take

v = 3 10 5 K 1 , 0 = 4700kg m 3 , g = 10m s 1 , k = 4W m 1 K 1 , = 1mm 2 s 1 ,

and = 10 21 Pa s and obtain Ra H = 3 109 from eq<259>. From eq<280> and
these parameter values we have T1 T0 = 918K . This is about a factor of 2 too low.
The boundary-layer stability results just discussed give episodic bursts of convection.
This is clearly quite different than the steady-state subduction that occurs on the Earth.
However, as discussed in Section 1-20, episodic subduction has been proposed
explain the globe resurfacing that occurred on Venus about 500 Ma ago.

21. A steady-state boundary layer theory for finite-amplitude

thermal convection.
Here we will develop a thermal boundary layer analysis of vigorous steady convection
in a fluid layer heated from below and limit to the case with very large Prandtl
numbers so that inertia terms in the momentum equation can be neglected (Fig. 6-40).
The flow is divided into 2-D rolls of width

; alternative rolls rotate in opposite

directions. On the cold upper boundary, a thin thermal boundary layer (TBL) forms.
When two cold TBL from adjacent cells meet, they separate from the boundary and
from a cold descending thermal plume. Similarly, a hot thermal boundary layer
forms on the bottom of the cell. When two hot TBLs meet adjacent cells they form a
hot ascending plume. The buoyancy forces in the ascending and descending plume
drive the flow. The core of each cell is nearly isothermal viscous rotational flow.
Symmetry requires that the temperature Tc in the nearly isothermal core must be the
mean of the two boundary temperatures as shown in eq<260>.

73

An exact solution for the boundary layer model requires numerical method, but we
can get an analytic solution by making a number of approximations. We first obtain
the structure of the cold thermal boundary layer adjacent to the top boundary of the
fluid layer. Let y = 0 to be the upper boundary and y is measured positive
downward. Let x be the horizontal coordinate and x = 0 be at the center of the
ascending plume. We assume the horizontal fluid velocity on the top is a constant u 0
(actually the horizontal fluid velocity is zero at x = 0 and
maximum value near x =

and inverses to a

, the constant velocity u 0 is an average of the

horizontal velocity on the upper boundary). The thermal structure of thus boundary
layer was solved already in section 4-16. From eq <4-124>, the temperature in the
cold TBL is:
1

2
Tc T
u
y

= erfc
2 x
Tc T0

-----<281>

The cold TBL can be directly associated with the thickening oceanic lithosphere. By
integrating the surface flux across the width of the cell (eq (4-127)) from x = 0 to
x=

, we obtain a total rate heat flow Q out of the top of the cell per unit distance

along the axis of the roll is:

1

u 2
Q = 2k (Tc T0 ) 0 .
2

-----<282>

At the boundary between two cells, the cold TBLs from two adjacent cells turn
through 90o to form a cold, symmetrical descending plume, analogous to the
subduction lithosphere at a trench. Since very little heat conduction can occur during
thus transition from a TBL to a thermal plume, the distribution of the temperature in
the newly formed plume is the same as in the boundary layer. We assume the
convected heat in the plume just after its formation must equal the convected heat
before its formation and the thickness of the plume relative to the boundary layer must
u
be in the ratio 0 . Therefore, the temperature in the plume just as it is formed is
v0
given by:

Tc T
v
= erfc 0
2u 0
Tc T0

2u 0 2
.
x
2

As the plume descends along the boundary between two adjacent cells, its

-----<283>

74

temperature can be obtained by using the temperature in <283> as the initial

temperature distribution in Laplaces solution of heat conduction equation (4-157) by
y
identifying t as
. This is analogous to the time-dependent solution of the heat
v0
conduction equation for the structure of the cold TBL in section 4-16 which identified
x
t as
.
u0
The temperature distribution in the descending plume can be used to calculate the
total downward gravitational body force on the plume due to the negative buoyancy
relative to the isothermal core. The downward buoyancy force per unit volume on an
element of the plume is 0 g v (Tc T ) .
Thus

f b = 0 g v (Tc T )dx

-----<284>

is the downward buoyancy force per unit depth and per unit distance along the roll
axis on one-half of the cold plume. It is appropriate to replace the integral, since
T Tc at the edge of the plume. The total downward buoyancy body forces Fb on
the descending plume is obtained by integrating f b along the vertical extent of the
plume from y = 0 to y = b ,
b

Fb = f b dy ,

-----<285>

where Fb is a force per unit length of the plume along the roll axis.
f b in eq <284> is proportional to the heat content of a slice of the plume of thickness
dy . Since no heat is added to the descending plume along its length, this heat content
is a constant. Therefore, the buoyancy body force on the plume per unit depth f b is

independent of y and
Fb = f b b .

Substitution of <283> into <284> with x =

f b = 0 g v (Tc T0 )

-----<286>
x yields

2
v
x
u

0
0
erfc

dx
2u 0

u
= 2 0 g v (Tc T0 ) 0
v0

2u 0

u
= 2 0 g v (Tc T0 ) 0
v0

2u 0

2
erfczdz
0
1

2
.

-----<287>

75

Thus the total downward gravitational body force Fb on one-half of the symmetric
plume is
u
Fb = f b b = 2 0 g v b(Tc T0 ) 0
v0

2u 0

2
.

-----<288>

Though here we only consider the cold TBL and the plume, the problem is
symmetrical and can be directly applied to the hot TBL and ascending plume in which
Tc T0 is replaced by Tc T1 . The total upward body force on the ascending hot
plume is equal to the downward body force on the cold descending plume as given in
eq <288>.
Determination of the viscous flow in the isothermal core requires a solution of the
biharmonic equation. However, the analytic solution cant be obtained for the bcs of
this problem. We instead approximate the core flow with the linear velocity profile
shown in fig. 6-41, i.e.,
y

u = u 0 1 2 ,
b

v = v 0 1 4 .

-----<289>
-----<290>

To conserve the fluid,

v0
= u0b
2

------<291>

The assumed velocity profiles dont satisfy the required bcs; e.g., the condition
u = 0 at x = 0 ,

is not satisfied. However, the assumed profile are reasonable

approximations to the actual flow near the center of the cell.

The shear stress on the vertical boundaries of the core flow is given by eq <34>,

cv =

4v
v
= 0 ,
x

-----<292>

76

and the shear stress on the horizontal boundaries is given by:

2u
u
ch =
= 0.
b
( y )

-----<293>

The rate at which work is done on each vertical boundary by the shear stress is
b cv v 0 per unit distance // the roll axis. The rate at which work is done on each

horizontal boundary is ch u 0 . The rate at which the buoyancy force does work on
2
each of the plumes is Fb v0 . The rate at which work is done on the plumes by the
gravitational body forces must equal the rate at which work is done on the boundaries
by the viscous forces; this gives
2 Fb v0 = 2b cv v 0 + ch u 0
-----<294>
Substitution of eqs <288> & <292> & <293> into <294> yields
1

2 2v0 2 u0 2
.
0 g vu0 (Tc T0 )
+
=

2b 2
2 u0

-----<295>

T T0

After eliminating the core temperature Tc using eq <260> Tc = T0 + 1

and
2

the vertical velocity v 0 using eq <291> v0 = u 0 b . We solve for the horizontal

2

velocity u 0
7

3

2b

Ra 3
u0 =
,
2
b

3
1
+

4
16b

-----<296>

where Ra is the Rayleigh number to a fluid layer heated from belows that has been
defined in eq <252>.
From eq <282> and the obtained mean velocity along the upper boundary of the cell,
we can find the total rate of heat flow through the cell Q,
5

Q=

3

2b

k (T1 T0 )
1
3

2
3

4
1
+

4
16b

Ra 3 .

-----<297>

The Nu is defined as the ratio of the convection Q to the heat flow rate by conduction
Qc in the absence of convection,
Nu =

Q
,
Qc

-----<298>

77

where Qc =

k (T1 T0 )
.
b
2

-----<299>

Upon substitution of <297> and <299> into <298>, we obtain

2

Nu =

3

2b

1
1
3

2
3

2
1 +
4
16b
4

1
3

1
3

Ra .

-----<300>

The aspect ratio of the cells, i.e., the ratio of the horizontal width to the vertical
thickness,

2b

, remains unspecified.

According to linear stability theory, the aspect ratio of the most rapidly growing
disturbance is

(2b )

= 2 (see eq <247>). However, for finite-amplitude convection,

we determine the aspect ratio for which the Nu is a maximum. This is the aspect
ratio of the cells that is most effective in transporting heat across the fluid layer at a
fixed Ra. We therefore require
Nu
= 0,


2b

-----<301>

and find that

2b

= 1.

-----<302>

For this value of the aspect ratio, the horizontal velocity is

u0 = 0.271 Ra 3 ,
b

-----<303>

and the Nusset # is

1
3

Nu = 0.294Ra .

-----<304>

Although the solution is approximate, the derived relation of Nu and Ra is exact.

Numerical calculations show that the constant 0.294 in Nu-Ra relation should be
reduced to 0.225.
If we assume that the mantle convection is restricted to the upper mantle, say
b = 700km , the boundary layer analysis for a fluid layer heated from below can be
applied to estimate Ra in the upper mantle. We take
b = 700km, 0 = 3700 kg
2

m3

, g = 10 m

s2

, v = 3 10 5 0 K 1 , T1 T0 = 1000 0 K ,

= 1 mm s and = 10 21 Pas , and obtain Ra = 5.7 105 . The mean surface heat

78

flux q is
q=

2Q

2Qc

Nu =

k (T1 T0 )
b

Nu =

With the above parameter values and k = 4W

k (T1 T0 )
b
m 0K

( 0.294 ) Ra 3 .

, q = 200 mW

m2

-----<305>
. This is about

a factor of 3 larger than the observed mean heat flow. The mean horizontal velocity
from eq <273> is u 0 = 84 mm

yr

. This is about twice the mean surface velocity

associate with plate tectonics.

Boundary layer theory can also be applied to a fluid layer that is heated from
within and cooled from above. The flow is again divided into counter-rotating, 2-D
cells with dimensions b and

. A cold TBL forms on the upper boundary of each

cell. When the two cold boundary layers from adjacent cells meet, they separate
from the boundary to form a cold descending thermal plume. However, for the fluid
layer heated from within. There is no heat flux across the lower boundary and hence
there is no hot TBL develops on the lower boundary, and no hot ascending plumes
between cells (see fig. 6-42).

Assume all fluid not in the cold TBL and plumes has the same temperature Tc . The
temperature in the upper cold TBL is given by eq <281> and the total rate at which
heat flows out of the top of each cell Q is given by eq <282>. In the layer there is a
uniform heat production H per unit mass. Thus, the total heat production in a cell is
0 Hb / 2 . Since we assume a steady state, Q must be equal to the rate of heat
generation in the cell,

0 Hb
2

u 2
= 2k (Tc T0 ) 0 .
2

-----<306>

79

The total buoyancy force on the descending cold plume is also the same as in the
previous problem; Fb given by eq <288>; however, there is only the single plume.
The rate of doing work on the boundaries is the same as in the previous problem;
however, the energy input comes from only the single plume. By equating the rate of
energy input to a cell to the rate of doing work on the boundaries, we find:

0 g v (Tc T0 )u 0
2u 0

2 4v0 2 u 0 2
=
+
.
2

-----<307>

Substitution of <291> 0 = u 0 b to eliminate v 0 and eq <306> to eliminate

2

Tc T0 yields

u0 =

2b

2b

1 +
4
16b
4

1
2

Ra1H2 ,

-----<308>

where the Ra # for a fluid layer heated from within is defined in eq <249>. We can
solve for the core temperature Tc by substituting eq <308> into <306>,
1

4 4
1
1
+
4
2
2 0 Hb 16b
Ra H1 4 ,
Tc T0 =
1
2
k

2

2b

-----<309>

where Tc T0 is the temperature rise across the fluid layer.

The efficiency with which convection cools the layer can be assessed by comparing
this temperature rise with that which would be obtained, T1c T0 if all the
internally generated heat were removed only by conduction. The dimensionless
temperature ratio ,
T T
c 0 ,
-----<310>
T1c T0
is thus a measure of convective efficiency for the internally heated fluid layer. The
smaller is, the more efficient convection is in removing the heat produced in the
fluid. Without convection, the temperature rise across the layer could be
T1c T0 =

Hb 2

.
(see section 4-6)
2k
By substituting <309> and <311> into <310>, we find

-----<311>

80

= ( 2 ) 2

4 4
1
+

4
16b Ra 1 4 .
H
1
2

2b

-----<312>

The dimensionless temperature is a function of the cell aspect ratio

2b

. The

aspect ratio that minimizes is found by setting

= 0.


2b

-----<313>

This gives

2b

= 1.

With this unity aspect ratio, the horizontal velocity and dimensionless temperature
are
u0 = 0.354

Ra1H2 ,

-----<314>

= 2.98RaH1 4 .

-----<315>

The non-dimensional temperature difference between the isothermal core and the
upper boundary decreases as convection becomes more vigorous with increasing Ra.
The boundary layer theory for thermal convection in a uniformly heated fluid
cooled from above can be applied to thermal convection in the whole mantle. Assume
H = 9 10 12 W

kg

, v = 3 10 5 0 K 1 , = 1 mm

, = 10 21 Pas, and we obtain

Ra H = 3 109 from <249>. Eqs <310>, <311> & <315> give Tc T0 = 550 0 K . This
is about a factor of 4 low. From eq <314>, we find that u 0 = 210 mm

yr

, which is

about a factor of 4 too high.

The parameterations of Nu and non-dimensional temperature obtained above
have been for free-slip surface bcs. For a fluid layer with no-slip surface bcs
(mostly occurs in the lab, experiments), if the fluid is heated from below:

Nu = 0.131Ra 0.3 .
For a layer heated from within and cooled from above,
5.95
= 0.23 .
Ra H

-----<316>

-----<317>

81

22. The forces that drive plate tectonics

The thermal convection of a fluid layer heated from within is very similar to mantle
convection. The TBL adjacent to the coded upper surface can be associated with the
oceanic lithosphere. The cold descending plume is associated with the subduction of
the lithosphere at an oceanic trench. As the gravitational body force on the cold plume
drives the convective flow, its also expected that the gravitational body force on the
descending lithosphere at a trench will be important in driving plate tectonics.
The gravitational body force Fb1 on the descending lithosphere due to its
temperature deficit relative to the adjacent mantle can be evaluated with eq <288>.
Because of the rigidity of the lithosphere, u 0 = v 0 . This also follows from eq <291>
for an aspect ratio

2b

= 1 . The equation for Fb1 is thus

Fb1 = 2 0 g v b(Tc T0 )
2u 0

-----<318>

In using this expression, the heating of the descending lithosphere by friction is

neglected. A primary uncertainty in evaluating the gravitational body force is the
depth of the convection cell b ; this is equivalent to the length of the descending
lithosphere beneath trenches. Based on the earthquake distribution that extends to a
depth of 700km , we take:
b = 700km, 0 = 3300 kg

m3

, g = 10 m

s2

, v = 3 10 5 0 K 1 , Tc T0 = 1200 0 K ,

= 1 mm s , g = 10 m 2 , and = 4000km , we obtain Fb1 = 3.3 1013 N m .

s
This is a force per unit length parallel to the trench.
There is also a force on the descending lithosphere due to the elevation of the
olivine-spinel phase change (see fig 6-43). The phase change occurs at a depth in the
surrounding mantle where the temperature is Tos . Because the descending lithosphere
is cooler than the mantle, the phase change occurs at lower pressure or shallow depth
in the slab (positive Claperon slope). The temperature of the descending slab Ts at
the depth where the phase change occurs depends on position Ts = Ts ( x ) , the phase
change boundary elevation hos also depends on position hos ( x ) . The downward
gravitational body force on the descending lithosphere due to the phase boundary
elevation Fb 2 is thus
Fb 2 = g os

x = xs

x = 0

hos ( x )dx ,

-----<319>

where os is the positive density difference between olivine and spinel phase. The

82

elevation of the phase boundary is given by

(Tos Ts )
dP hos 0 g
=
hos =
,
dP = hos 0 g , dT = Tos Ts =
0 g
dT Tos T
where is the slope of the Claperon curve (

dP
).
dT

Substitution of <320> into <319> yields

os x xs
Fb 2 =
(Tos Ts )dx .

-----<320>

------<321>

x = 0

The integral in <321> is the temperature deficit in the descending lithosphere at the
depth where the phase change occurs. This can be evaluated using the boundary layer
model discussed previously. In eq <287> evaluating f b , we found that the integrated
temperature deficit per unit depth of the descending plume is a constant. Its value,

fb
or 2(Tc T0 )
from <287> with u 0 = v 0 , is
0 g v
2u 0

2
.

Using this for the value of the integrated in eq <321>, we find:

2(Tc T0 ) os

Fb 2 =
0
2u 0
With os = 270 kg

m3

, = 4 MPa 0

obtain Fb 2 = 1.6 1013 N

2
.

-----<322>

, and other parameters given above, we

. The body force due to elevation of the olivine-spinel

phase change is about half the body force due to the thermal contraction. The total
body force on the descending lithosphere Fb = Fb1 + Fb 2 = 4.9 1013 N

. This force

is often refereed to as trench pull. If the force is transmitted to the surface plate as a

83

tensional stress in an elastic lithosphere with a thickness of 50km . The required

tensional stress is 1 GPa , clearly a very high stress.
The force is also exerted on the on the surface plates at oceanic ridges. The
elevation of the ridges establishes a pressure head that drives the flow horizontally
away from the ridge. This ridge push can also be thought of as gravitational sliding.
A component of the gravitational field causes the surface plate to slide downward
along the slope between the ridge crest and the deep oceanic basin. The force exerted
on a surface plate due to the elevation of a ridge can be evaluated from the force
balance in fig. 6-44.

The integrated horizontal force on the base of the lithosphere F1 can be determined
from equilibrium of section RCD of the mantle. The net horizontal pressure force on
RD , F5 must equal F1 . The force F5 is easily obtained by integrating the
lithostatic pressure beneath the ridge crest,
F5 = F1 =

w+ y L

m gydy ,

-----<323>

where m is the mantle density. This can be rewritten as:

w

yL

w + yL

F1 = g m ydy + g m w + y d y = g m ydy + g
0

m ydy ,

-----<324>

where y = y + w, and y = y w .
The integrated pressure force on the upper surface of the lithosphere F2 is
equal to F4 , the net pressure force on AB, because section of water RAB must in
equilibrium. Thus we can integrate the hydrostatic pressure in the water to obtain
w

F2 = F4 = w gydy ,
0

-----<325>

where w is water density. The horizontal force F3 acting on the section

lithosphere BC is the integral of the pressure in the lithosphere PL ,

84

yL

F3 = PL d y ,

-----<326>

where PL is given by

PL = w gw + L gd y ,
y

-----<327>

and L is the density of the lithosphere.

Substitution of <327> into <326> gives
yL
y

F3 = w gw + L gd y d y .
-----<328>
0
0

The net horizontal force on the lithosphere adjacent to an oceanic ridge FR is

obtained by combining <324>, <325>, <328>.

w
yL
y
FR = F1 F2 F3 = g ( m w ) ydy + g ( m w )w + m y L d y d y .
0
0
0

-----<299>
We substitute the isostatic relation from eq <4-204> and the identify

m (w y L ) = L y + w w w ( w m ) +
yL

yL

( m )dy = 0 ,

m y = m d y
y

-----<330>

to give:
FR = g ( m w )

yL
y
w2
+ g ( L m ) d y ( L m ) d y d y .
0
0
2
0

-----<331>

Substitution of eq <4-205> and <4-124> and

1

1
z=
2

u 2
y 0
x

-----<332>

yields
4x
w2
+ g m v (Tm T0 )
0 erfcz dz dz , -----<333>

2
u0
is the mantle temperature. The double internal of erfc( z ) has the value

FR = g ( m w )
where Tm
1
.
4

By substituting for w from eq <4-202>, we finally arrive at

2 m v (Tm T0 ) x
FR = g m v (Tm T0 )1 +

( m w ) u 0
2 m v (Tm T0 )
= g m v (Tm T0 )1 +
t ,
(
)

m
w

-----<334>

85

where t is the age of the seafloor.

The force due to the elevation of the ocean ridge is proportional to the age of the
lithosphere. Taking g = 10 m

, m = 3300 kg

, w = 1000 kg

, = 1 mm

Tm T0 = 1200 0 K and v = 3 10 5 0 K 1 . We find the total ridge push on 100 Myr

old oceanic lithosphere is 3.9 1012 N

This is a per unit length // the ridge.

Ridge push is an order of magnitude smaller than trench pull.

23. Heating by viscous dissipation

We have neglected the effect of viscous dissipation or frictional heating so far. In this
section, we calculate the temperature rise and the heat flux produced by viscous
dissipation in a simple Couette flow between plane-parallel walls (as shear in fig.
6-45). The velocity profile is

u = u 0 1 ,
h
Produced by the constant shear stress,

=
applied at the upper surface

( y = 0)

-----<335>

u 2
du
= 0 = 0 ,
d ( y)
h

-----<336>

of the channel. The rate at which shear force do

work on the entire fluid layer, per unit horizontal area, is given by the product of shear

0 and the velocity of the upper boundary u 0 , i.e.,

u02
h

u02

, the rate of shear heating

h2
per unit volume. This heating rate per unit volume is constant, because the shear stress
is constant and the velocity profile is linear. This volumetric heating rate can be

If we average this over the entire fluid layer, we get

86

identified with the internal volumetric heat production rate H in eq <4-12> to

obtain the equation for the temperature in the channel,

u02
d 2T
=

.
-----<336>
dy 2
h2
A straightforward integration of this equation with b.c.s T = T0 at y = 0 and
k

T = T1 at y = h gives
T = T0 +

u02 u02 y 2
y
.

+
T
T
1 0

2k 2k h 2
h

-----<337>

This can be written in the convenient dimensionless form

u02
u02
2

T T0 y
y
=
= 1 + 2k 2 2k .
T1 T0 h T1 T0 h T1 T0

-----<338>

The temperature in the channel is governed by the single dimensionless parameter:

u02
2k .
T
( 1 T0 )
This can be written as

1
times the product of the Pr and a dimensionless parameter
2

known as the Eckert number E

E

u02
,
c p (T1 T0 )

-----<339>

where c p is the specific heat at constant pressure. Thus we can write

u02
1
Pr E = 2k ,
2
(T1 T0 )
and

-----<340>

y Pr E y 2 Pr E
1 +

.
h
2 h2 2

-----<341>

The dimensionless temperature is plotted in fig. 6-45 for values of Pr E . The

conduction profile in the absence of frictional heating is the straight line for Pr E = 0 .
The temperature in excess of this linear profile are a consequence of viscous
dissipation. The excess temperature due to frictional heating e is obtained by
substracting the linear profile from eq <341>.

Pr E y

e =
1 .
2 h h

-----<342>

87

The maximum excess temperature is found by differentiating e w.r.t.

setting the result to zero. The maximum e occur at

y
and
h

y 1
=
and
h 2

Pr E
.
-----<343>
8
The upward heat flux at the upper boundary q 0 is found by using Fouriers law: (eq

e =

<4-1> and eq <337>).

u
T
k
= T1 T0 + 0 .
q 0 = k
2k
y y =0 h

-----<344>

The excess upward heat flux qe at y = 0 is

qe =

u02

.
-----<345>
2h
If all the frictionally-generated heat flowed out of the upper boundary, q e would
equal

u02

; half of the shear heating in the channel flows out through the lower
h
boundary. The ratio of the excess heat flowing through the upper boundary q e to the
heat flux without viscous dissipation q c =

k (T1 T0 )
is
h

qe 1
= Pr E .
qc 2

-----<346>

We can use this result to quantify the effect of frictional heating in an asthenospheric
shear flow. For example, take = 4 1019 Pa, u 0 = 50 mm

yr

, k = 4W

m 0K

and

88

T1 T0 = 300 0 K , we can find

Pr E
= 0.04 . Thus the maximum additional
2

temperature rise due to shear heating would be 1% of the temperature rise across the
asthenosphere or about 30 K in this example. The excess heat flux to the surface
would be 4% of the heat conducted across the asthenosphere in the absence of
dissipation. These results show that the trictional heating effects in mantle shear flow
are generally small.

24. Mantle recyling and mixing

The plate tectonic cycle is an inherent component of mantle convection. The
surface plate or lithosphere is the upper thermal boundary layer of convecting cells.
The oceanic lithosphere is created at mid-ocean ridges and recycled back to the
mantle at subduction zones.
Mid-ocean ridges migrate over the surface of the Earth in response to the
kinematic constraints of plate tectonics. Mantle rocks ascends passively beneath a
ridge in response to seafloor spreading and becomes partially molten due to
decompression on ascent. The magma percolates through the residual solid and then
solidifies to form the crust, with an average thickness of ~6 km. The result is
two-layer structure for the rigid oceanic lithosphere. The upper part of the
lithosphere is the solidified magma of the oceanic crust and the lower part is the
residual solid in the upper mantle. The residual solid also has a vertical stratification.
The uppermost mantle rock is highly depleted in the low-melting temperature basaltic
component and it grades into undepleted mantle over a depth range of ~50 km.
Isotopic and trace element studies of mid-ocean ridge basalts (MORB) show that
they are remarkably uniform and systematically depleted in incompatible trace
elements with respect to bulk Earth values. This indicates that, on average, the
mantle source of MORB is well-stirred depleted chemical reservoir on the scale at
which it is sampled by mid-ocean ridge processes. However, heterogeneities do
persist in this reservoir, as indicated by variation in MORB. Large-scale
heterogeneities are evident in variations between average Atlantic Ocean MORB and
average Indian Ocean MORB. In addition, small-scale heterogeneities deviated
from average MORB are evident, particularly when the mantle is locally sampled as it
is at young Pacific Ocean seamounts.
The depleted mantle source of MORB is complementary to the enriched
continental crust. Incompatible elements are partitioned into the continents by the
volcanic processes responsible for the formation of the continents; this occurs
primarily at island arcs. When the oceanic lithosphere is subducted, the oceanic
crust is partially melted; the resulting enriched magma ascends to the surface and

89

form island-arc volcanoes leaving a more strongly-depleted oceanic lithosphere.

The complementary nature of the continental crust and the MORB source reservoir
requires that this depleted oceanic lithosphere from which the continental crust has
been extracted, be mixed into the MORB source region.
Atomic diffusion plays a role in the homogenization of the mantle only on scales
of a meter or less because the solid-state diffusion coefficient is so small. Values of
the relevant diffusion coefficients are estimated to be in the range D = 10-18-10-20 m2/s.
Over the age of the Earth, 4.56x109 yr, the corresponding range of diffusion lengths is
0.3-0.03 m. We conclude that the subducted lithosphere is mixed back into the
mantle by convection, but that diffusive mixing is significant only on small scales.
This process of convective homogenization is known as kinematic mixing. The
mantle is composed of a matrix of discrete, elongated layers of subducted oceanic
lithosphere. Each layer has its own isotopic, chemical, and age identity. The older
the layer, the more it will have been elongated by mantle flow; on average, the older
layers will be thinner. The mantle thus has the appearance of a marble cake. The
marble cake comprises the enriched oceanic crust, which has been partially depleted
by subduction zone volcanism, and the complementary, highly depleted upper mantle.
Approximately the upper 60 km of the lithosphere is processed by the plate
tectonic cycle. We first ask the question: What fraction of the mantle has been
processed by the plate tectonic cycle since the Earth was formed? We consider the
two limiting cases of layered mantle convection and whole mantle convection.
To simply the analysis we assume that the rate M at which mass is processed
into a layered structure at ocean ridges is constant, and that the subducted rock is
uniformly distributed throughout the mantle (upper mantle).

We define M p to be

the primordial unprocessed mass in the mantle reservoir. The rate of loss of this
primordial mass by processing at ocean ridges is given by
dM p
Mp
=
M ,
-----<347>
dt
Mm
where M m is the mass of the mantle participating in the plate tectonic convection
cycle the whole mantle for whole mantle convection and the upper mantle for
layered mantle convection. The ratio M p (t ) M m is the fraction of the mantle
reservoir that has not been processed at an ocean ridge. Upon integration with the
initial condition M p = M m at t = 0 we obtain

M p = M me

t p

-----<348>

90

where

p =

Mm
M

-----<349>

is the characteristic time for processing the mantle in the plate tectonic cycle. The
processing rate M is given by

dS
M = m h p
,
dt

-----<350>

where m is the mantle density, h p is the thickness of the layered oceanic

lithosphere structure, and
(or subducted). Taking

dS
is the rate at which new surface plate area is created
dt

dS
= 0.0815 m 2 s 1 (see figure 4-26), h p = 60 km , and
dt

m = 3300 kg m 3 , we obtain M = 1.61 10 7 kg s 1 . For layered mantle convection

( M m = 1.05 10 24 kg ) the characteristic time for processing the mantle from eq
<349> is p = 2 Gyr ; for whole mantle convection p = 8 Gyr . The fraction of
primordial unprocessed mantle M p M m obtained for eq<348> is given as a
function of time t in Figure 6-47 for both layered and whole mantle convection. For
layered mantle convection 10.5% of the upper mantle is unprocessed at the present
time while for whole mantle convection 57% is unprocessed.

This analysis was carried out assuming a constant rate of recycling. As shown in
Section 4-5 the rate of radioactive heat generation in the Earth H was higher in the
past. To extract this heat from the Earths interior, the rate of plate tectonics was

91

probably also higher in the past. The time dependence of the radioactive heat
generation as given in Figure 4-4 can be approximated by the relation.

H = H 0 e (t e t ) ,

-----<351>

where H 0 is the present rate of heat production, t e is the present value of the time t,
and is the average decay constant for the mixture of radioactive isotopes in the
mantle. From the results given in Figure 4-4 we take = 2.77 10 10 yr 1 .
Assuming that the rate M at which mass is processed in to a layered structure at
ocean ridges is proportional to the rate of heat generation given in eq<351> we write
t t
M = M 0 e ( e )

-----<352>

where M 0 is the present rate of processing. Substitution of <352> into <347> gives

dM p
dt

Mp
Mm

M 0 e (te t ) .

-----<353>

Integration of eq<353> with the initial condition M p = M m at t = 0 gives

1 t e
M p = M m exp
e e (t e t )
p 0

where

p0 =

Mm
M 0

) ,

-----<354>

-----<355>

is again the characteristic time for processing the mantle in the plate tectonic cycle.
For layered mantle convection we again have p 0 = 2 Gyr and for whole mantle
convection we have p 0 = 8 Gyr .

With = 2.77 10 10 yr 1 , the fraction of

primordial unprocessed mantle M p M m obtained form eq<354> is given as a

function of time t in Figure 6-47 for both layered and whole mantle convection. For
layered mantle convection 1% of the upper mantle is unprocessed at the present time
while the whole mantle convection 33% is unprocessed. The time dependent
processing is more efficient, as expected. However, in all cases substantial fraction of
the mantle reservoir have been processed by the plate tectonic cycle.
The layered oceanic lithosphere is subducted back into the mantle at oceanic
trenches. The cold subducted lithosphere is heated by conduction from the
surrounding mantle on a time scale of 50 Myr . The heated and softened subducted
lithosphere is then entrained in the mantle convective flows and is subjected to the

92

fluid deformation. With the assumption that the subducted layered lithosphere
behaves passively, it is subject to kinematic mixing.

We next quantify the rate of kinematic mixing in the mantle. We consider the
problem layer stretching. As stated before, we hypothesize that the subducted oceanic
crust becomes entrained in the convecting mantle and is deformed by the strains
associated with thermal convection. Kinematic mixing can occur by both shear
strains and normal strains. We first consider the thinning of a passive layer in a
uniform shear flow. Initially we take the one dimensional channel flow (Couette flow,
see figure 6-2a) of width h as illustrated in Figure 6-48. The passive layer has an
initial width 0 and is assumed to be vertical with a length L. The linear velocity
profile from Equation <6-13> is
u =  (h y ) ,
-----<356>
where the strain rate = u 0 h . At a subsequent time t the top of the layer has moved
a distance u 0 t while the bottom boundary remains in place. The total length of the
strip is now

L = h 1 + (t )

2 12

-----<357>

However, to conserve the mass of material in the strip we require

h 0 = L .

-----<358>

Substitution of eq<357> into <358> gives

1
=
0 1 + (t )2

12

-----<359>

And for large strains, t >> 1 , this becomes

1
= (t ) .
0

-----<360>

Using eq<360> we can determine how long it takes to thin the subducted oceanic

93

crust ( 0 = 6 km ) to a thickness = 10 mm . For whole mantle convection we take

 = 50 mm yr -1 2886 km = 5.5 10 16 s 1 and find that t = 3.5 10 4 Gyr . For

layered mantle convection we take  = 50 mm yr -1 660 km = 2.4 10 15 s 1 and find
that t = 7.9 10 3 Gyr . Clearly this type of mixing is very inefficient.

In the relatively complex flows associated with mantle convection, normal

strains may also be important for mixing. An idealized flow that illustrates normal
strain is the two-dimensional stagnation point flow illustrated in figure 6-49. In this
flow
u = x
-----<361>
v = y ,
-----<362>
where u is the x-component of velocity and v is the y-component of velocity. The
strain rate  is independent of time. In the upper half-space ( y > 0 ) there is a
uniform downward flow and in the lower half space ( y < 0 ) there is a uniform
upward flow. These vertical flows converge on y = 0. There is a complementary
divergent horizontal flow. In the right half-space (x > 0) there is a uniform divergent
flow to the right. In the left half-space there is a uniform divergent flow to the left.
This steady stagnation flow satisfies the governing continuity equation <6-53> and
force balance equation <6-67> and <6-68>.
Problem 6-39 obtain the steam function corresponding to the two-dimensional
stagnation point flow giving in Equations <6-434> and <6-435>. Show that this
stream function satisfies the biharmonic Equation <6-74>.
We again consider the thinning of a passive layer by the stagnation point flow.
This passive layer initially occupies the region 0 2 y 0 2 . The deformation
of this layer is uniform in x direction and the change of the layer thickness with
time is given by

94

1 d
= v( ) =  .
2 dt
Integration with the initial condition = 0 at t = 0 gives

= 0 e 2t .

-----<363>

-----<364>

The stagnation point flow stretches and thins the passive layer. With normal strains
the passive layer thins exponentially with time. Normal strains are much more
effective in layer thinning than shear strains.
Based on <364>, we can determine how long it takes for normal strains to thin
the subducted oceanic crust ( 0 = 6 km) to a thickness of 10 mm. We again take

 = 5.5 10 16 s 1 and find that t = 380 Myr . For layered mantle convection we
take  = 2.4 10 15 s 1 and find that t = 88 Myr . Thus, normal strains can thin the
oceanic lithosphere to thickness that can be homogenized by diffusion in reasonable
lengths of geological times ( 108 yr).
The first question that arises is whether there is direct observational evidence of
an imperfectly mixed mantle. Allegra and Turcotte (1986) argued that the marble
cake structure associated with imperfect mixing can be seen in high-temperature
peridotitles (also called orogenic lherzolite massifs), which represent samples of the
Earths mantle. Typical locations include Beni Bousera in Morocco, Rhonda in
Spain and Lherzolite in France. These rocks consist primarily of depleted lherzolite.
Embedded in this matrix are bands of proxenite comprising a few percent of the
massif. Trace element studies of these bands indicate that they were originally
basaltic in composition.
These characteristics led to the postulate that the bands are former sample of
oceanic crust hat have been subducted and deformed by convective shear before
emplacement into their current locations. The bands range in thickness from a few
meters to a few centimeters and some have been extensively folded. Essentially, no
stripes with thickness of 1 cm or less are found, probably because stripes of this width
have been destroyed by diffusive processes. According to this hypothesis, they have
undergone 5 to 6 orders of magnitude of thinning from an initial thickness of 6 km
(Fig. 6-50).

95

Further evidence for the marble cake structure of the mantle comes from eclogitic
xenoliths associated with basaltic volcanism and kimberlites. In some kimberlites,
diamonds are found in the eclogite nodules. These eclogitic diamonds have been
found to have carbon isotope ratios characteristic of sediments. Some suggested that
subducted carbonates are one source of diamonds. Graphitized diamonds have been
found in the proxenite bands of the Beni Bousera high-temperature peridotite in
Morocco. These observations are completely consistent with the hypothesis that
sediments are entrained in the subducted oceanic crust. During convective mixing in
the deep interior some carbonate sediments are transformed to diamonds as the
oceanic crust in which they are embedded is stretched and thinned.