Geodynamics & Continental Deformation 2020 i

3rd-YEAR MODULE
GEODYNAMICS & CONTINENTAL DEFORMATION

Lecturers: Philip England and Richard Katz

Books (for geodynamics part) We will refer to various books throughout the course. There
is one book that you will be required to read: Mantle Convection for Geologists by Geoffrey
Davies, Cambridge University Press, 2011. This book is available from Amazon for about
£35. There are copies in the Earth Sciences, Radcliffe Science, EXT, STA, SEH, SPC,
and WORC libraries. You may also wish to look at Geodynamics by Donald Turcotte and
Gerald Schubert, Cambridge University Press, 3rd edition, 2014 (sometimes referred to as
“the bible of geodynamics”). There are copies in Earth Sciences, Radcliffe Science, EXT,
STA, SEH, SPC, and WORC libraries.
ii Geodynamics & Continental Deformation 2020

Contents
1 Refresher: Vectors and Tensors in Indical Notation 1
1.1 The Kronecker delta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Inner (dot) Product Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 The Permutation Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 The Vector (cross) Product Revisited . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.5 Vector Calculus: A Short Refresher . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.6 The Gradient and Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.7 The Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.8 Integral Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.8.1 The Generalised Divergence Theorem . . . . . . . . . . . . . . . . . . . . 4
1.8.2 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.9 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.9.1 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.10 Special Second Order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Fluid Mechanics 7
2.1 Continuum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Conservation of Mass and the Continuity Equation . . . . . . . . . . . . . . . . . 7
2.3 The Linear Momentum Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.1 Body Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.2 Surface Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.3 Back to the Linear Momentum Equation . . . . . . . . . . . . . . . . . . . 9
2.4 The Angular Momentum Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Relative Motion of Nearby Fluid Elements . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Stress in a Static (non moving) Fluid . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.7 Fluids in Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.8 Summary: Equations for Fluid Motion . . . . . . . . . . . . . . . . . . . . . . . . 12
2.9 The Reynolds and Rayleigh Numbers . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.10 1-Dimensional Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.10.1 1-D Channel Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.10.2 Asthenospheric Counter Flow . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.10.3 Viscous Flow Down an Incline . . . . . . . . . . . . . . . . . . . . . . . . 17
2.11 The Stream Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.12 Postglacial Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.13 Estimate of Mantle Viscosity from Glacial Isostatic Adjustment . . . . . . . . . . 23

3 Kinematics of Continental Deformation 25

3.1 Difference between Continental Tectonics and Plate Tectonics . . . . . . . . . . . 25
3.2 Scales of Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Models of Continental Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Measurements of Continental Deformation . . . . . . . . . . . . . . . . . . . . . . 31
3.4.1 Measuring Strain with Earthquakes . . . . . . . . . . . . . . . . . . . . . 31
3.4.2 Regional Strains from Earthquakes in the Aegean, Iran, and Asia . . . . . 36
3.4.3 Measurement of Strain with GPS . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.4 Velocities and Strain Rates . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.5 Advantages and Disadvantages of Different Methods of Measuring Strain 40
3.5 Derivation of Kostrov’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
 Geodynamics & Continental Deformation 2020 iii

4 Rheology of the Lithosphere 45

4.1 Strength Profile through the Lithosphere . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 Contribution of the Mantle Portion of the Lithosphere . . . . . . . . . . . 45
4.1.2 Contribution of the Lower Crust to Lithospheric Strength . . . . . . . . . 46
4.1.3 Contribution of Upper crust to Lithospheric Strength . . . . . . . . . . . 46
4.2 Vertically Averaged Rheology of the Lithosphere . . . . . . . . . . . . . . . . . . 47
4.3 Accommodation of Strain by the Upper Crust . . . . . . . . . . . . . . . . . . . . 49

5 Mechanics of Continental Deformation 51

5.1 Vertical Force Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.1.1 Isostasy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.1.2 Gravitational Potential Energy of the Lithosphere . . . . . . . . . . . . . 51
5.2 1D Thin-Sheet Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3 A Simple Solution, and Application to the Apennines of Italy . . . . . . . . . . . 57
5.4 2D Thin-Sheet Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.5 Length scales for Continental Deformation . . . . . . . . . . . . . . . . . . . . . . 60
5.6 Approximate solutions for a Power-Law Fluid . . . . . . . . . . . . . . . . . . . . 60

6 Estimates of Lithospheric Viscosity 63

6.1 Gravitational Potential Energy and deformation in Iran . . . . . . . . . . . . . . 63
6.2 Places in which it is possible only a bound on viscosity . . . . . . . . . . . . . . 65
6.2.1 The Tien Shan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2.2 The South Island of New Zealand . . . . . . . . . . . . . . . . . . . . . . . 67
6.3 Tectonics of the Aegean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.3.1 Flow of a rectangular Newtonian thin sheet . . . . . . . . . . . . . . . . . 71
6.4 Rheology of the lithosphere revisited . . . . . . . . . . . . . . . . . . . . . . . . . 73

7 Rayleigh-Benard Convection 75
7.1 Convection in experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.2 Density variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.3 The energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.4 The Rayleigh number and the Nusselt number . . . . . . . . . . . . . . . . . . . 77
7.5 The onset of convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.6 Convection at high Rayleigh number . . . . . . . . . . . . . . . . . . . . . . . . . 83

8 Mantle convection 87
8.1 The fluid mantle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.2 The plate mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.3 The plume mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.4 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.5 Energy-balance models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

9 Flow at Mid-Ocean Ridges and Subduction Zones 101

9.1 Setting up the corner-flow problem . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.2 Mid-ocean ridges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9.3 Subduction zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

10 Temperatures in Oceanic and Continental Lithosphere 107

10.1 Notes on the Solution of the 1D diffusion-advection equation . . . . . . . . . . . 107
10.1.1 The Péclet Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.1.2 Thermal diffusion in the solid bounded by two parallel planes (a slab) . . 108
10.2 Oceanic Heat Flux and Bathymetry . . . . . . . . . . . . . . . . . . . . . . . . . 110
10.2.1 Half-space solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
iv Geodynamics & Continental Deformation 2020

10.2.2 Depth of the Ocean Floor . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

10.3 The thermal plate model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
10.4 Onset of convection in a fluid cooled from the top . . . . . . . . . . . . . . . . . . 117
10.5 Convection and Dynamic Topography . . . . . . . . . . . . . . . . . . . . . . . . 119
10.6 Sedimentary Basins and Metamorphism . . . . . . . . . . . . . . . . . . . . . . . 123
10.6.1 Subsidence of sedimentary basins . . . . . . . . . . . . . . . . . . . . . . . 123
10.7 Regional metamorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
10.7.1 Exhumation and termination of metamorphism . . . . . . . . . . . . . . . 125
10.8 Temperatures at convergent plate boundaries . . . . . . . . . . . . . . . . . . . . 128
10.8.1 Temperatures in the subducting slab . . . . . . . . . . . . . . . . . . . . . 128
10.8.2 Flow at Subduction Zones and Temperature in the Mantle Wedge . . . . 130
Vectors and Tensors 1

1 Refresher: Vectors and Tensors in Indical Notation

In this chapter we will review of some useful vector calculus concepts and learn about tensors
and index notation. Tensors are extremely useful for dealing with fluid dynamics problems that
we shall address in this course.
A scalar quantity has magnitude. As an example, temperature is a scalar quantity. A vector
has both magnitude and direction. Examples of vector quantities would be force or position. The
quantitative description of a vector requires a choice of a coordinate system. We will be using
the Cartesian coordinate system for problems in this course. Consider the Cartesian coordinate
system with unit vectors ex , ey , and ez along the x, y, and z axes. A position vector can be
written as
x = xex + yey + zez , (1.1)

and an arbitrary vector a as

a = ax ex + ay ey + az ez , (1.2)

(recall that any unit vector has the property |ex | = 1.) Let us label the x, y, and z axes as 1,
2, and 3 instead. It is then convenient to introduce a compact notation that takes advantage of
the fact that the Cartesian coordinate system is right-handed and has mutually orthogonal axes.
Then we can write a as
3
X 3
X
a = a1 e1 + a2 e2 + a3 e3 = ai ei = aj ej . (1.3)
i=1 j=1

This is called index notation. Note that the summation index is arbitrary and thus any index
may be used. Since the presence of both the components (ai ’s) and the base vectors (ei ’s)
indicates that the summation occurs from 1 to 3 we now drop the summation symbol

3
X
a= ai ei = ai ei . (1.4)
i=1

A repeated index symbol in a product indicates a summation. This is the summation convention
and will reduce the complexity of manipulating vectors and tensors. As an example, consider
expanding ai bi
3
X
ai bi = ai bi = a1 b1 + a2 b2 + a3 b3 . (1.5)
i=1

As another example, expand kr wj er

3
X
kr wj er = wj kr er = wj kr er = wj (k1 e1 + k2 e2 + k3 e3 ) . (1.6)
i=1

In the above example only the r subscript appears twice and indicates a summation.

1.1 The Kronecker delta

In our Cartesian system we have e1 · e1 = 1, e1 · e2 = 0, e1 · e3 = 0, etc. (this is due to the base
vectors being mutually orthogonal). These relations can all be summarised using the Kronecker
delta operator, defined as (
1 f or i = j
δij = ei · ej = (1.7)
0 f or i 6= j
2 Geodynamics & Continental Deformation 2020

1.2 The Inner (dot) Product Revisited

Consider the vectors a = ai ei and b = bj ej . Then the dot product of the two is

a · b = ai ei · bj ej
= ai bj ei · ej
= ai bj δij
= ai bi or aj bj

This result should be familiar from your courses in vector calculus.

1.3 The Permutation Symbol

In our Cartesian coordinate system e1 × e1 = 0, e1 × e2 = e3 , e1 × e3 = −e2 , etc. A convenient
shorthand is
ei × ej = ijk ek (1.8)
where ijk is the permutation operator and i, j, and k are 1, 2, or 3. The permutation operator
is defined as 

1 if ijk = 123, 231, 312 (even permutation),
ijk = −1 if ijk = 132, 213, 321 (odd permutation), (1.9)


0 if (any two indices are the same).
Note the cyclic property ijk = kij = jki and that ijk = −jik .

1.4 The Vector (cross) Product Revisited

Consider again the vectors a and b. Let’s take their cross product and use the permutation
operator as a compact way of writing the result

c = a×b
= ai ei × bj ej
= ai bj ei × ej
= ai bj ijk ek . (1.10)

Consider the second component of the result (k = 2). Its magnitude will be

c2 = a1 b1 112 + a1 b2 122 + a1 b3 132

+a2 b1 212 + a2 b2 222 + a2 b3 232
+a3 b1 312 + a3 b2 322 + a3 b3 332
= a1 b1 (0) + a1 b2 (0) + a1 b3 (−1)
+a2 b1 (0) + a2 b2 (0) + a2 b3 (0)
+a3 b1 (1) + a3 b2 (0) + a3 b3 (0)
= a3 b1 − a1 b3 . (1.11)

This should look familiar from your past experience with cross product. The real benefit of the
permutation symbol is the ease of manipulation that it will provide when dealing with several
vector operations. We will see this shortly.
Another useful identity (you can easily show this using the definition of the permutation
operator) is
a × a = 0 → ai aj ijk = 0 for any k. (1.12)
Vectors and Tensors 3

Let’s consider another example. Using the definitions of δij and ijk , show that a · (b × a) = 0.

a · (b × a) = ai ei · (bj ej × ak ek )
= ai ei · (bj ak jkm em )
= ai bj ak jkm ei · em
= ai bj ak jkm δim
= ai bj ak jki
= ai bj ak ijk
= 0. (1.13)

Note that we used the cyclic property in the second last step and the identity in equation (1.12)
in the final step. This is a much cleaner way of solving the problem. Alternatively, you can
write out all the components in full and, if you don’t make an algebraic mistake along the way,
you’ll get the same answer with a lot more work.
As a final note, one more useful identity (you can show this) is

ijk klm = δil δjm − δim δjl . (1.14)

1.5 Vector Calculus: A Short Refresher

We know how to take a derivative of a scalar function, perhaps even a multivariable scalar
function. What about the derivative of a vector with respect to some scalar? To evaluate we
can use the definition of a derivative
df f (t + ∆t) − f (t)
= lim
dt ∆t→0 ∆t
f1 (t + ∆t)e1 + f2 (t + ∆t)e2 + f3 (t + ∆t)e3 − f1 (t)e1 − f2 (t)e2 − f3 (t)e3
= lim
∆t→0 ∆t
f1 (t + ∆t) − f1 (t) f2 (t + ∆t) − f2 (t) f3 (t + ∆t) − f3 (t)
= lim e1 + lim e2 + lim e3
∆t→0 ∆t ∆t→0 ∆t ∆t→0 ∆t
df1 df2 df3
= e1 + e2 + e3
dt dt dt
dfi
= ei (1.15)
dt
d da db
Using this result, let’s show that dt (a · b) = dt ·b+a· dt

d d
(a · b) = (ai ei · bj ej )
dt dt
d
= (ai bj δij )
dt
d
= (ai bi )
dt
dai dbi
= bi + ai
dt dt
dai dbj
= bj δij + ai δij
dt dt
dai dbj
= ei · bj ej + ai ei · ej
dt dt
da db
= ·b+a· (1.16)
dt dt
As an exercise, show that
d da db
(a × b) = ×b+a× (1.17)
dt dt dt
4 Geodynamics & Continental Deformation 2020

1.6 The Gradient and Divergence

The gradient, or ‘del’, operator is defined as
∂
∇ = ei . (1.18)
∂xi
Applying the del operator to a scalar field gives a vector result:
∂T
∇T = ei . (1.19)
∂xi
Dotting the gradient vector with another vector gives the divergence
∂
∇ · u = ei · uj ej
∂xi
∂
= uj ei · ej
∂xi
∂uj
= δij
∂xi
∂ui
= (1.20)
∂xi
Consider the result of d · ∇T where d is a (unit) direction vector and T is the temperature field.
Convince yourself that this quantity gives the scalar rate of change of the temperature field T
in the direction of the vector d. This is known as the directional derivative.

1.7 The Curl

Taking the cross product of the del operator with a vector gives the curl.
∂
∇×f = ei × fj ej
∂xi
∂fj
= ei × ej
∂xi
∂fj
= ijk ek (1.21)
∂xi

1.8 Integral Theorems

We will now consider several integral theorems that will be useful when considering conservation
laws.

1.8.1 The Generalised Divergence Theorem

Given a vector field f (x) defined in a domain (volume) V and bounded by the surface S with
unit normal n directed away (outwards from V ), then
Z Z
∇ · f dV = n · f dS, (1.22)
V S

See vector notes online for details and derivation. The following statements can also be shown
to be true Z Z
∇φ dV = nφ dS, (1.23)
V S
Z Z
∇ × g dV = n × g dS, (1.24)
V S
Vectors and Tensors 5

where φ is a scalar and g is a vector. The above three expressions can all be combined through
the statement Z Z
∇ ? Φ dV = n ? Φ dS, (1.25)
V S

where ? is any operation that ‘makes sense’ and Φ is the corresponding scalar or vector (or
tensor) function.

1.8.2 Stokes’ Theorem

Consider an open surface. There is a relationship between integrals over the surface S and
integrals around the contour C that bounds the opening
Z I
n · (∇ × f ) dS = t · f dS, (1.26)
S C

where t is a vector tangent to the curve C. The above relationship can be generalised as
Z I
n · (∇ × Φ) dS = t · Φ dS. (1.27)
S C

1.9 Tensors
Some situations require two or more vectors to characterise a physical phenomenon. As an
example, consider the stress tensor. The stress tensor specifies the force per unit area (magnitude
and direction) and the surface (unit normal) that the force is acting on. For this we introduce
a second-order tensor with components and two directions

T = Tij ei ej (1.28)

where i = 1, 2, 3, j = 1, 2, 3, and T therefore has nine components (all the permutations of ij).
We can perform operations on tensors in the same way as vectors.

1.9.1 Inner Product

The inner product of a vector a with a second order tensor T can be calculated as follows

a · T = ai ei · Tjk ej ek
= ai Tjk ei · ej ek
= ai Tjk δij ek
= ai Tik ek . (1.29)

Similarly,

T · a = Tjk ej ek · ai ei
= Tjk ai ej ek · ei
= Tjk ai ej δki
= Tji ai ej . (1.30)

As a final example, consider

b · (a · T) = bi ei · (aj ej · Tlm el em )
= bi ei · (aj Tjm em )
= bi aj Tji . (1.31)
6 Geodynamics & Continental Deformation 2020

1.10 Special Second Order Tensors

Let’s take a moment to consider a few special second order tensors that we will see from time
to time. The first is the identity tensor

I = δij ei ej . (1.32)

Try dotting the identity tensor with a vector a (you should recover the vector a!).
A symmetric tensor is any tensor for which Tij = Tji . Second order symmetric tensors have
6 independent components. An anti-symmetric second order tensor is any tensor for which
Tij = −Tji . An anti-symmetric tensor has only 3 independent components because the diagonal
components must be zero. An arbitrary second order tensor A can be written as the sum of a
symmetric and an anti-symmetric tensor
1
1

A= A + AT + A − AT . (1.33)
2 2
The first set of brackets contains a symmetric tensor while the second brackets contain an
anti-symmetric tensor.
Derivatives are taken by simply applying the same index and tensor algebra that we discussed
before. For example
∂
∇·T = ei · Tjk ej ek
∂xi
∂Tjk
= ei · ej ek
∂xi
∂Tjk
= δij ek
∂xi
∂Tjk
= ek . (1.34)
∂xj

As a final note on tensors, we will sometimes use a comma to denote a derivative. For
example
duj
= uj,i (1.35)
dxi
and
∂ 2 uj
= uj,ij . (1.36)
∂xi ∂xj
Fluid Mechanics 7

2 Fluid Mechanics
By the end of the course we would like to be able to treat several different fluid dynamics
problems.
Motivation Most of the earth is a fluid on the time scale of a few thousand years or greater.

The Continuum Hypothesis plus Newton’s Second Law lead to the Cauchy stress equation,
which describes the balance of forces in a continuous medium, and to the Navier-Stokes
equation for the flow of a fluid.

Pipe or Channel flow Flow in the asthenosphere, in the lower crust, of magma through vol-
canic conduits . . .

Continental Deformation Active tectonics, distributions of earthquakes and of seismic haz-

ard, causes of extensional, compressional, and strike-slip deformation. Evolution of moun-
tains, sedimentary basins, . . .

Mantle convection Plate motion, plumes, Earth’s thermal history.

To solve these problems we will need to spend a bit of time developing the appropriate
theory. We will begin by explaining the continuum hypothesis, then move on to consider the
important conservation principles.

2.1 Continuum Mechanics

Fluid mechanics is a branch of Continuum Mechanics,
which treats material as though it fills the space of inter-
est without gaps, cracks, or discontinuities. Obviously
all materials are composed of atoms, which are funda-
mentally discontinuous; we know that minerals deform
by the motions of dislocations, by cataclasis, by pres-
sure solution, etc.; the upper crust deforms primarily
by slip on faults – yet we shall treat all these phenom-
ena with the apparatus of continuum mechanics. Why
are we allowed to do this? The basic idea is illustrated Figure 2.1: The continuum approxima-
tion. The line indicates variation in some
in Figure 2.1: we suppose in all these cases that dis-
property of a material – for example
continuous, or unpredictable, fluctuations in physical density – as the scale of measurement
properties (e.g. density, cracks, dislocations) take place changes.
on scale, δV , that is small compared with the scale ∆V , http://pillars.che.pitt.edu/student/slide.cgi?
of the system that we investigate. course id=10&slide id=5.0

2.2 Conservation of Mass and the Continuity Equation

Consider a fixed volume V of arbitrary shape. In order for mass to be conserved the rate of
change of mass within the volume, with respect to time, must be equal to the net flow of mass
across the surface area S which defines the volume V (see Figure 2.2).
Z Z
d
ρ dV = − n · uρ dS,
dt V
Z ZS
dρ
dV = − ∇ · (uρ) dV,
V dt V
Z  
dρ
+ ∇ · (uρ) dV = 0 (2.1)
V dt
8 Geodynamics & Continental Deformation 2020

where in the second line we have used the divergence

theorem. Note that equation (2.1) must hold for any
volume V and thus the integrand must be true at every
point and
∂ρ
+ ∇ · (uρ) = 0. (2.2)
∂t
If the density is constant and uniform then we simply
have Figure 2.2: A volume V defined by the
∇ · u = 0. (2.3) closed surface S with unit normal n.

Equation (2.3) is the continuity equation for incompressible flows; we shall see this often in the
course.

2.3 The Linear Momentum Equation

P
The linear momentum equation is also sometimes referred to as Newton’s 2nd law ( F = M a).
It states that the time rate of change of linear momentum in a volume must be equal to the sum
of all forces acting on the volume. The forces acting on the volume are often separated into two
types: body forces and surface forces.

2.3.1 Body Forces

Body forces are long-range forces that act (perhaps unevenly) on all material points of a body.
Examples of body forces would be gravity, electric forces, and magnetic forces. Body forces arise
due to force fields. The body force we will be most concerned with is gravity. We will denote
the acceleration of gravity as g.

2.3.2 Surface Forces

Surface forces are short-range forces that represent two

types of molecular-scale interactions:

1. Short-range molecular forces between nearby fluid

molecules. Forces must be applied to slide
molecules past each other.

2. Momentum exchange due to random thermal mo-

tions (analogy: billiard balls).

Consider the small surface S shown in Figure 2.3

Figure 2.3: A force ∆F acting on a surface
that separates two regions of fluid. We will define the ∆S with unit normal n.
stress vector t as

force ∆F
stress vector = = lim = t(n). (2.4)
area ∆S→0 ∆S

The stress vector depends on the orientation of the surface S and thus on the normal vector n
which defines the surface. The stress tensor T contains all of the information on the forces on
surfaces in any direction at a given point. Taking the dot product of a normal vector n and the
stress tensor T gives the stress vector t for the surface S corresponding to n. In other words

t(n) = n · T. (2.5)
Fluid Mechanics 9

2.3.3 Back to the Linear Momentum Equation

We can now write out the linear momentum equation as
Z Z Z
d
ρu dV = fbody dV + t(n) dS. (2.6)
dt V V S

This statement says that the time rate of change of momentum in the volume V is balanced by
the sum of the body forces acting on the volume and the surface forces acting on the boundary
of the volume. At this point we need to make a little bit of a jump using something called the
Reynolds Transport Theorem. This can be found on page 15 of the online fluids notes. This
allows us to express the term on the left-hand-side of equation (2.6) in terms of the material
derivative Z Z Z
Du
ρ dV = fbody dV + t(n) dS, (2.7)
V Dt V S
where the material derivative is defined as
Dq ∂q
= + u · ∇q (2.8)
Dt ∂t
for any scalar (or vector) quantity q. We can now substitute equation (2.5) into equation (2.7)
and use the divergence theorem to obtain
Z Z Z
Du
ρ dV = fbody dV + n · T dS
V Dt V S
Z Z Z
Du
ρ dV = fbody dV + ∇ · T dV
V Dt V V
Z  
Du
ρ − fbody − ∇ · T dV = 0. (2.9)
V Dt

As the volume V is arbitrary the statement in equation (2.9) must be true at all points and we
therefore have
Du
ρ = fbody + ∇ · T. (2.10)
Dt
Equation (2.10) is the linear momentum equation and is also known as the Cauchy stress equation
of motion.
The stress tensor T was introduced in equation (2.5) as a relation between the stress vector
t and any given unit normal vector n for a surface. We still need to figure out what T is.

2.4 The Angular Momentum Balance

A similar conservation equation can be written for angular momentum. However, the details are
a bit more involved and we will simply state the result (details can be found in the fluids notes
posted on the course webpage). The angular momentum balance, in words, says that the time
rate of change of angular momentum (for a control volume) is equal to the sum of all torques
due to body and surface forces. The main result from the angular momentum balance is that it
reveals the relation

T = TT . (2.11)
This is important as it implies that the stress tensor is a symmetric tensor and thus has only 6
independent components.
10 Geodynamics & Continental Deformation 2020

2.5 Relative Motion of Nearby Fluid Elements

In order to relate the stress tensor to velocities in the
fluid we consider the deformation of a fluid body and
the relative motion of nearby fluid elements. Consider
the points P and Q shown in Figure 2.4 and specified
by the vectors x and x + δx. Let’s expand the velocity
at point Q in a Taylor series about the point P . Recall
that for a single variable x and function f (x) the Taylor Figure 2.4: Relative motion of nearby fluid
elements.
series about the point x0 is
 
df  1 2 d2 f 
f (x0 + δx) = f (x0 ) + δx + δx + O(δx3 ). (2.12)
dx x=x0 2 dx2 x=x0

For functions of multiple variables the Taylor series is

1
u(x + δx) = u(x) + δx · ∇u|x + δxδx : ∇∇u|x + O(|δx|3 ). (2.13)
2
The velocity at Q relative to P is (to first order)

δu(δx) = u(x + δx) − u(x)

= u(x) + δx · ∇u|x + O(|δx|2 ) − u(x)
≈ δx · ∇u|x . (2.14)

Equation (2.14) demonstrates that the relative motion of nearby fluid elements is determined
by the velocity gradients in the fluid. Earlier in this chapter we demonstrated that any second
order tensor may be written as the sum of a symmetric and anti-symmetric tensor. Let us define
the strain rate tensor as the symmetric part of the velocity gradient, E, and the vorticity tensor
as the anti-symmetric component, Ω:

1
1

∇u = ∇u + (∇u)T + ∇u − (∇u)T
2 2
= E+Ω (2.15)

The strain rate tensor E provides information on straining and stretching in the fluid while
the vorticity tensor Ω provides information on rotation. The angular momentum equation,
discussed in the previous section, told us that the stress tensor T is symmetric. For this reason
we expect T to depend on the symmetric strain rate tensor E and not on the vorticity Ω, which
is anti-symmetric (see the fluids notes for a more in depth discussion).

2.6 Stress in a Static (non moving) Fluid

Consider a fluid that is not moving and assume that the only body force we are interested in is
gravity, i.e. fbody = ρg. If the velocity is at rest then u and all its derivatives are equal to zero
and the linear momentum equation simplifies to

0 = ρg + ∇ · T. (2.16)

Fluids cannot support tangential (shear) stresses without motion. Tensile or compressive stresses
can be supported. Thus, for any surface we pick, the stress must lie normal to the surface and
we call this stress the static (or equilibrium) pressure pe . Then the stress vector to any surface
defined by the unit normal n is
t(n) = −pe n. (2.17)
Fluid Mechanics 11

The stress vector is negative as we consider pressure as compressive (pressing into the sur-
face). Recall that we defined the stress tensor as being related to the stress vector through the
relationship
t(n) = n · T. (2.18)
Then equations (2.17) and (2.18) imply that

n · T = −pe n. (2.19)

And this simplifies to

T = −pe I. (2.20)
Substituting this into equation (2.16) then gives

0 = ρg + ∇ · (−pe I) = ρg − ∇pe . (2.21)

As an exercise, show that the second equality in equation (2.21) is true and that ∇ · T =
−∇pe . Equation (2.21) is the governing equation for fluid statics. This equation allows for the
calculation of forces on dams, submerged objects, the shape of fluid interfaces, etc.

2.7 Fluids in Motion

Consider the simple shear flow in a layer of thickness H driven by a moving upper boundary
and shown in Figure 2.5. The upper surface moves at a velocity of u relative to the lower
boundary and a shear stress exists on the base of the upper surface equal to τyx (recall that this
represents the force per unit area in the x direction and with unit normal in the y direction). For
a Newtonian fluid and the case shown in Figure 2.5, the shear stress is related to the viscosity
of the fluid, η, through the relation

u du
τyx = η =η (2.22)
h dy

U
-

-
-
- y
u=U
y h
-
6 -
- x, u
Figure 2.5: Shear flow driven by a moving boundary.

For Newtonian fluids the stress depends linearly on the velocity gradients. For an incom-
pressible flow (∇ · u = 0) the stress tensor is

T = −pI + 2ηE = −pI + τ (2.23)

where p is now the dynamic pressure and the deviatoric stress tensor τ is equal to, again for a
Newtonian fluid
τ = 2ηE (2.24)
12 Geodynamics & Continental Deformation 2020

Recall that the Cauchy Stress Equation is

Du
ρ = ∇ · T + ρg (2.25)
Dt
Substituting equation (2.23) into equation (2.25) gives

Du
ρ = −∇p + η∇2 u + ρg (2.26)
Dt
where we have used the definition of the strain rate tensor E = (∇u + (∇u)T )/2 and the
continuity equation ∇ · u = 0 to express the pressure and deviatoric stress terms as (show as an
exercise) ∇ · (2ηE) = η∇2 u and ∇ · (−pI) = −∇p. Finally, we have arrived at the Navier-Stokes
equations for motion in a Newtonian incompressible flow
 
Du ∂u
ρ =ρ + u · ∇u = −∇p + η∇2 u + ρg (2.27)
Dt ∂t

∇·u=0 (2.28)
Equations (2.27) and (2.28) represent four equations (equation (2.27) represents 3 equations for
each component) for the four unknowns ux , uy , uz , and p. It is worth noting that the dynamic
pressure p is no longer equal to the equilibrium pressure pe that we discussed earlier (unless the
fluid is at rest). For more details on the pressure see the online fluids notes.

2.8 Summary: Equations for Fluid Motion

Let’s summarize everything that we have done so far in this chapter. We began by enforcing
conservation of mass and this led to the continuity equation
Dρ ∂ρ
= + u · ∇ρ = −ρ∇ · u (2.29)
Dt ∂t

For an incompressible and steady state flow ( Dρ

Dt = 0) the above simplified to

∇·u=0 (2.30)

We then used an argument based on conservation of linear momentum to arrive at the equation
Du
ρ = ∇ · T + ρg (2.31)
Dt
This equation is also often referred to as Newton’s Second Law since it is a statement of force
balance. The left-hand-side of equation (2.31) is the rate of change of momentum and it is
balanced by the sum of all surface forces (the first term on the right-hand-side) and the sum of
all body forces (the second term on the right-hand-side).
Conservation of Angular Momentum (in the fluids notes online - you do not need to know
the details here) led us to the conclusion that the stress tensor T must be symmetric

T = TT (2.32)

We considered the relative motion of nearby fluid elements and demonstrated that relative
velocities were controlled by the velocity gradient in the fluid. We decomposed the velocity
gradient into a symmetric component E and an anti-symmetric component Ω and called them
the strain rate and vorticity tensors, respectively.
1

E= ∇u + (∇u)T (2.33)
2
Fluid Mechanics 13

1

Ω= ∇u − (∇u)T (2.34)
2
∇u = E + Ω (2.35)
We argued, conceptually, that the stress in a fluid in motion should depend on the relative motion
of fluid elements and should therefore depend on the velocity gradient in the fluid. Knowing
that the stress tensor is symmetric then implied that the stress tensor should depend on the
symmetric part of the velocity gradient and thus on the strain rate tensor.
For a moving Newtonian and incompressible flow the stress tensor is

T = −pI + 2ηE = −pI + τ (2.36)

If the fluid is at rest then the stress is simply T = −pI and the pressure p = pe is equal to the
thermodynamic equilibrium pressure. Finally, the Navier-Stokes equations were found to be
 
Du ∂u
ρ =ρ + u · ∇u = −∇p + η∇2 u + ρg (2.37)
Dt ∂t

∇·u=0 (2.38)

2.9 The Reynolds and Rayleigh Numbers

Let’s now non-dimensionalize the equations using a characteristic velocity uc , a characteristic
length scale l, and a characteristic time tc = l/uc . Then we substitute into equation (2.37)

u = uc u0 (2.39)

xi = lx0i (2.40)
l 0
t= t (2.41)
uc
where u0 , x0i , and t0 are all dimensionless. This gives

uc Du0j 1 ∂p ηuc ∂ 2 u0j

ρ = − + + ρgj . (2.42)
l/uc Dt0 l ∂x0j l2 ∂x0i 2

Rearranging gives
ρluc Du0 0 2 0 ρgl2
= −∇p + ∇ u + (2.43)
η Dt0 uc η
where we have defined the characteristic pressure pc such that p = pc p0 = ηul c p0 . Also note that
∇ is now also dimensionless. If the characteristic velocity, length scale, and time scales are
close to the velocities, length scale over which velocity changes, and time scales involved in the
problem, then u0 , x0i , and t0 will all be of order O(1). This means that the terms Du0 /Dt0 , ∇p0 ,
and ∇2 u0 will all be of order O(1). We can then get some sense of how important each of the
different terms are by looking at the coefficients in front of them. Let us define the Reynolds
number as Re = ρluc /η. Then the above becomes

Du0 0 2 0 ρgl2
Re = −∇p + ∇ u + (2.44)
Dt0 uc η
The Reynolds number tells us how important the inertial terms are in comparison with the
surface force (stress) terms. For convection within the Earth’s mantle, mantle velocities are
of the order of uc ≈ 3 cm/year ≈ 10−9 m/s, velocities change over length scales of at least
l ≈ 100 km = 105 m, the mantle density is ρ ≈ 4000 kg/m3 , and the mantle viscosity is at least
η ≈ 1021 Pa s. This gives Re ≈ 4 × 10−22 . Thus, the inertial terms in equation (2.44) are
14 Geodynamics & Continental Deformation 2020

22 orders of magnitude smaller than the stress terms for mantle convection can be completely
ignored. Thus, for mantle convection, equation (2.44) can be simplified to
ρgl2
0 = −∇p0 + ∇2 u0 + (2.45)
uc η
Thermal convection drives flow in the Earth’s mantle. For such problems, the character-
istic velocity of interest is uc = κ/l, where κ is the thermal diffusivity of mantle rock. This
characteristic velocity is the velocity at which thermal diffusion and advection balance. The
density in the mantle is given by ρ(T ) = ρ0 (1 − α∆T ) where ρ0 is the references density, ∆T
is the difference between the mantle temperature and some reference temperature, and α is the
thermal expansion coefficient. Substituting this and g = gez into equation (2.45) then gives

ρ0 gl3
0 = −∇p0 + ∇2 u0 − Raez + ez (2.46)
κη
where Ra is the Rayleigh number and is defined as
αρ0 g∆T l3
Ra = (2.47)
κη
The last term in equation (2.46) results in a linear increase in pressure with depth (the hydro-
static component of pressure) while the third term gives the driving force due to changes in tem-
perature. For the Earth’s mantle α ≈ 10−5 K−1 , ρ0 ≈ 4000 kg/m3 , g ≈ 10 m/s2 , ∆T ≈ 1000 K,
l ≈ 3000 km = 3 × 106 m, κ ≈ 10−6 m2 /s, and η ≈ 1021 Pa s. Then Ra ≈ 107 . Thus, the term
associated with changes in temperature (buoyancy) is very important in driving mantle flow.

2.10 1-Dimensional Flows

Now that we have derived the equations for motion in a fluid and have discussed some of the
different terms in the equations we are ready to start looking at a few simple one dimensional
flows.

2.10.1 1-D Channel Flow

Consider the flow shown in Figure 2.6 and driven by a moving upper surface and a pressure
gradient applied from one end of the channel to the other. The boundary conditions are ux (y =
h) = u0 and ux (y = 0) = 0. The fluid viscosity η is constant and a constant pressure gradient
dP/dx exists across the channel. The viscosity is very high and you may assume that the
Reynolds number is very small and inertial terms can be ignored. You may also assume that
the force due to gravity is small and may also be neglected. Flow is in only one direction and
thus uy = uz = 0.
As with all fluids problems that we will solve, we begin by writing out the full Navier-Stokes
equations. For this problem, we then set the inertial and gravity terms to zero (as indicated
above). This gives
∇·u=0 (2.48)
0 = −∇p + η∇2 u (2.49)
Let’s start by learning as much as we can from equation (2.48) (it is often easiest to start with
the continuity equation due to its simplicity). Expanding this equation gives
∂ux ∂uy ∂uz ∂ux
∇·u= + + = +0+0=0 (2.50)
∂x ∂y ∂z ∂x
Then we are just left with
∂ux
=0 (2.51)
∂x
Fluid Mechanics 15

Figure 2.6: 1-dimensional channel flow driven both by moving boundaries and a pressure gradient.

This tells us that the horizontal velocity does not depend on the lateral position in the channel
and is independent of x. We might have expected this based on the fact that the boundary
conditions and pressure gradient are independent of x. This is all that we can get from the
continuity equation. Let’s now consider the x-component of equation (2.49)
 2 
∂p ∂ ux ∂ 2 ux ∂p ∂ 2 ux
0=− +η + = − + η (2.52)
∂x ∂x2 ∂y 2 ∂x ∂y 2

where ∂ 2 ux /∂x2 = 0 due to equation (2.51). We can’t really do anything else with this at the
moment as we do not know what either p or ux are. Let’s now consider the y-component of
equation (2.49)  2 
∂p ∂ uy ∂ 2 uy ∂p
0=− +η + =− (2.53)
∂y ∂x2 ∂y 2 ∂y
where ∂ 2 uy /∂x2 = ∂ 2 uy /∂y 2 = 0 since uy = 0. This is useful as it tells us that p is independent
of y. We can now return to equation (2.52) and integrate twice with respect to y

∂p ∂ 2 ux
0 = − +η
∂x ∂y 2
∂p ∂ux
0 = − y+η + c1
∂x ∂y
1 ∂p 2
0 = − y + ηux + c1 y + c2 (2.54)
2 ∂x
where c1 and c2 are constants of integration. Solving for ux from equation (2.54) gives

1 dp 2 c1 c2
ux = y − y− (2.55)
2η dx η η

Applying the boundary conditions ux (y = h) = u0 and ux (y = 0) = 0 gives

1 dp ηu0
c1 = h− (2.56)
2 dx h
c2 = 0 (2.57)
Substituting (2.56) and (2.57) into (2.55) then gives the final result

y dp y
ux = (y − h) + u0 (2.58)
2η dx h
This solution is useful and will be used in solving several other problems. Consider the two end
member scenarios that equation (2.58) can be used for: 1) Fixed boundaries (u0 = 0) and driven
16 Geodynamics & Continental Deformation 2020

Figure 2.7: 1-dimensional channel flow with fixed boundaries and driven by a pressure gradient.

by a pressure gradient or 2) zero pressure gradient with a moving boundary. For the first case
the solution simplifies to
y dp
ux = (y − h) (2.59)
2η dx
and the solution is a simple parabolic flow profile (shown in Figure 2.7). This type of flow is
often referred to as either a ‘pipe’ or ‘channel’ flow. For the second case, the solution simplifies
to
y
ux = u0 (2.60)
h
and the flow profile is linear (a simple shear profile). This is shown in Figure 2.5.

2.10.2 Asthenospheric Counter Flow

Consider the 1-D flow shown in Figure 2.8. In the early days of plate tectonics it was believed
that plates, in this example of thickness H and velocity u0 , moved along the surface of a relatively
thin, low-viscosity channel in the mantle layer as the asthenosphere. Plates would subduct into
the upper mantle and the material would return in the opposite direction by flowing within
this channel beneath the plate (the “return flow”). The velocity of the flow at the base of the
asthenosphere was assumed to be approximately zero (due to the higher viscosity of the mantle
beneath it). In this example the asthenosphere has a viscosity of η and a thickness of h. The
plate is assumed to be much stronger than the asthenosphere and thus moves as a rigid body
(we don’t have to consider fluid deformation within the plate). For such a case we may wish to
know how the pressure gradient in the asthenosphere is related to the thickness of the plate and
asthenosphere as well as the asthenospheric viscosity.

Plate

Asthenosphere

Figure 2.8: 1-dimensional asthenospheric counter-flow.

Fortunately, we don’t have to do very much work in determining the solution for the return
flow in the asthenosphere as it is identical to the 1-D flow that we considered in section 2.10.1.
The boundary conditions at the base and top of the asthenosphere are ux (y = 0) = 0 and
ux (y = h) = u0 and there is a pressure gradient, to be determined, of dp/dx. The solution for
this 1-D flow is given by equation (2.58) and is

1 dp y y
ux = (y − h) + u0 (2.61)
2 dx η h
Fluid Mechanics 17

for 0 ≤ y ≤ h and
ux = u0 (2.62)
for h ≤ y ≤ h + H. In order to determine dp/dx we require the return flow to balance the
forward flow. In other words, we require that
Z h+H
ux (y) dy = 0 (2.63)
0

Using equations (2.61) and (2.62) then gives

Z h+H Z h Z h+H
ux (y) dy = ux (y) dy + ux (y) dy
0 0 h
Z h  Z h+H
1 dp y y
= (y − h) + u0 dy + u0 dy
0 2 dx η h h
 
1 dp 1 h3 1 h3 1
= − + u0 h + u0 H (2.64)
2 dx 3 η 2 η 2
Simplifying and solving for dp/dx then gives
dp ηu0
= 6 3 (h + 2H) (2.65)
dx h
Equation (2.65) demonstrates that the pressure gradient is very sensitive to the thickness of the
asthenospheric channel. It’s always good to do a physical check to make sure that the units and
sign are what you expected. The units for the velocity, viscosity, and layer thicknesses will be
m/s, Pa s, and m, respectively. The answer therefore has units of Pa/m which is correct for a
pressure gradient. The sign of the pressure gradient, for a positive velocity u0 , is positive. This
implies that pressure increases with increasing position x and thus the pressure would force flow
in the −x direction, which is as expected.

2.10.3 Viscous Flow Down an Incline

We now consider the problem of a gravity driven viscous flow down a slope and shown in Figure
2.9. For simplicity the x direction is down-plane and y = 0 is at the surface of the fluid layer.
The fluid layer has a thickness of h and the plane is inclined at an angle α. We expect that
the base of the fluid layer will not move and enforce a ‘no-slip’ or ux (y = h) = 0 boundary
condition here. The surface of the fluid layer will be in contact with the air and we expect that
the air will not be able to exert any significant stress on the surface of the fluid. We then have a
zero-stress condition on the surface and require Tyy (y = 0) = 0 and Tyx (y = 0) = 0. Given that
it is a very viscous flow we can assume that the Reynolds number is sufficiently low that we do
not need to consider the inertial terms. The flow is also assumed to be 1-D and thus uy = 0.
Again, we start with the general Navier-Stokes equations and set the inertial term equal to
zero. The equations we are solving are:
∇·u=0 (2.66)
0 = −∇p + η∇2 u + ρg (2.67)
Again, let’s start with the continuity equation because it is the simplest. Expanding the conti-
nuity equation give
∂ux ∂uy ∂ux
∇·u= + = =0 (2.68)
∂x ∂y ∂x
This tells us that ux is independent of x. Next, let’s look at the x-component of equation (2.67)
 2 
∂p ∂ ux ∂ 2 ux ∂p ∂ 2 ux
0=− +η + + ρ sin(α)g = − + η + ρ sin(α)g (2.69)
∂x ∂x2 ∂y 2 ∂x ∂y 2
18 Geodynamics & Continental Deformation 2020

Figure 2.9: Viscous flow down an incline.

where we have used equation (2.68). We can’t really do anything else with equation (2.69) at
the moment. So let’s move on to the y-component of equation (2.67)
 2 
∂p ∂ uy ∂ 2 uy ∂p
0=− +η 2
+ 2
+ ρ cos(α)g = − + ρ cos(α)g (2.70)
∂y ∂x ∂y ∂y

Let’s take the derivative of equation (2.70) with respect to x

 
∂2p ∂2p ∂ ∂p
0=− +0=− =− (2.71)
∂x∂y ∂y∂x ∂y ∂x

Equation (2.71) is useful because it tells us that ∂p/∂x is independent of y. Let’s now integrate
equation (2.70) with respect to y. This gives

0 = −p + ρg cos(α)y + c1 (x) (2.72)

where c1 (x) is some function of x. At the surface we have Tyy (y = 0) = 0 as a boundary

condition. Recalling that T = −pI + 2ηE and E = (∇u + (∇u)T )/2, you can show that
Eyy = ∂uy /∂y = 0 and thus
Tyy = −p + 2ηEyy = −p (2.73)
Therefore at the surface p = 0. Then c1 (x) in equation (2.72) must be zero and p is then

p = ρg cos(α)y (2.74)

It is interesting to note that the pressure does not change with x as ∂p/∂x = 0. Thus, the flow
in the x direction is not driven by pressure gradients. The flow is driven by gravity. Substituting
equation (2.74) or simply noting that ∂p/∂x = 0 (as we just found) we return to equation (2.69)
and integrate with respect to y

∂p ∂ 2 ux
0 = − +η + ρ sin(α)g
∂x ∂y 2
∂ 2 ux
0 = η + ρ sin(α)g
∂y 2
∂ 2 ux
η = −ρ sin(α)g
∂y 2
Z Z
∂ 2 ux
η dy = − ρ sin(α)g dy
∂y 2
∂ux
η = −ρ sin(α)gy + c2 (2.75)
∂y
Fluid Mechanics 19

We now use the other boundary condition at the surface Tyx (y = 0) = 0. You can show,
using the definition of T and E as well as uy = 0 that Tyx = η∂ux /∂y. Then at the surface
Tyx = η∂ux /∂y = 0 and this implies that c2 in equation (2.75) must be zero. Then we can
integrate equation (2.75) once more with respect to y to obtain

ρg sin(α) 2
ux = − y + c3 (2.76)
2η

where c3 is a constant of integration. We now use the final boundary condition ux (y = h) = 0

to obtain
ρg sin(α) 2
c3 = h (2.77)
2η
and substituting this back into equation (2.76) gives us our final answer of

ρg sin(α) 2

ux = h − y2 (2.78)
2η

Again, it is important to check the sign of the solution to see if it makes physical sense. For
y ≤ h the solution has ux ≥ 0 and thus the fluid flows down the slope as we expect!

2.11 The Stream Function

We continue our discussion of fluid dynamics by now considering postglacial rebound and a
viscous (Re  1) incompressible two dimensional flow. To aid us with this we will first develop
an important tool called the stream function. Let us define a scalar function ψ that satisfies

∂ψ ∂ψ
ux = − and uy = (2.79)
∂y ∂x

With this definition the continuity equation is automatically satisfied

∂ux ∂uy ∂2ψ ∂2ψ

∇·u= + =− + =0 (2.80)
∂x ∂y ∂x∂y ∂y∂x

Now let’s take a look at the Stokes equation and assume that the body force is constant (this is
still reasonably general and can include gravity)

0 = −∇P + η∇2 u + ρg (2.81)

The x and y components are then

 2 
∂P ∂ ux ∂ 2 ux
0 = − +η + (2.82)
∂x ∂x2 ∂y 2
 2 
∂P ∂ uy ∂ 2 uy
0 = − +η + + ρg (2.83)
∂y ∂x2 ∂y 2

Substituting the relations in (2.79) then gives

 3 
∂P ∂ ψ ∂3ψ
0 = − −η + (2.84)
∂x ∂x2 ∂y ∂y 3
 3 
∂P ∂ ψ ∂3ψ
0 = − +η + 2 + ρg (2.85)
∂y ∂x3 ∂y ∂x

We can eliminate the pressure term by taking the partial derivative with respect to x of the y-
component equation and subtracting the partial derivative with respect to y of the x-component.
20 Geodynamics & Continental Deformation 2020

This gives
 4   4 
∂ ψ ∂4ψ ∂ ψ ∂4ψ
0 = η + + η +
∂x2 ∂y 2 ∂y 4 ∂x4 ∂y 2 ∂x2
 4 4 4

∂ ∂ ∂
0 = 4
+2 2 2 + 4 ψ
∂y ∂x ∂y ∂x
 2 2
 2 
∂ ∂ ∂ ∂2
0 = + + ψ
∂y 2 ∂x2 ∂y 2 ∂x2
0 = ∇2 ∇2 ψ
0 = ∇4 ψ (2.86)

The stream function ψ then satisfies the biharmonic equation given by (2.86). We will first solve
a problem to do with postglacial rebound and then plot the contours of ψ to discuss its physical
interpretation.

2.12 Postglacial Recovery

To determine the response of the Earth’s mantle to unloading following a glaciation, we consider
the flow in a semi-infinite and very viscous (Re  1) fluid half space (y > 0) subject to an initial
periodic vertical surface displacement of
 
2πx
w(x, 0) = w0 cos = w0 cos (kx) (2.87)
λ

where λ is the wavelength and k = 2π/λ is the wavenumber. In what follows, w0  λ, and this
will be useful for making a few approximations.
λ

w(x,t)
x

Figure 2.10: Definition sketch

We expect high pressures beneath areas with initial surface displacements that are high and
lower pressures beneath areas with initial surface displacements that are lower. These lateral
pressure gradients will drive a flow in the mantle and the rate at which this flow restores the
surface to its original shape (zero vertical displacement) will tell us about the viscosity of the
fluid in the mantle.
This problem meets all of the criteria discussed in Section 2.11 for using the stream function,
thus we are looking for a solution to the equation

0 = ∇4 ψ. (2.88)

Let’s start by making a simplifying assumption based on what we might physically expect. The
initial vertical displacement is periodic in x and we should therefore expect that our solution
for ψ is also periodic. We might try to find solutions that involve sines and cosines in the x
direction. But we can save ourselves a lot of work by noting that the initial surface displacement
Fluid Mechanics 21

involves the cosine function and we would expect the vertical velocity to therefore also depend on
cosine (we will have maximum vertical flow in areas that have maximum vertical displacement).
Noting that uy = ∂ψ/∂x we would then expect ψ to depend on the sine function. Applying the
method of separation of variables, we try to find a solution of the form

ψ = sin (kx) Y (y) (2.89)

where the function Y (y) is to be determined. Substituting this into the biharmonic equation
given by (2.88) and dividing by sin(kx) gives

d4 Y 2
2d Y
− 2k + k 4 Y = 0. (2.90)
dy 4 dy 2

This is a fourth-order linear ordinary differential equation with constant coefficients. You may
recall from your ODE’s class that equations like this have solutions of the form

Y ∝ y0 emy (2.91)

where m and y0 are to be determined. Substituting equation (2.91) into (2.90) gives the char-
acteristic equation for m
 2
m4 − 2k 2 m2 + k 4 = m2 − k 2 = 0. (2.92)
From this we see that solutions for m are m = ±k. This gives us two solutions for Y (y), but
we expect two more since equation (2.90) is fourth-order. The next step is to try solutions of
the form, and you can verify that this works, Y (y) ∝ y exp(±ky). Then the general solution for
Y (y) is
Y (y) = Ae−ky + Bye−ky + Ceky + Dyeky (2.93)
and thus the general solution for ψ is
h i
ψ = sin (kx) Ae−ky + Bye−ky + Ceky + Dyeky , (2.94)

where the four constants A, B, C, and D will be determined by the boundary conditions. We
expect the solution to be finite as y → ∞ and so C = D = 0. Using the relationships in (2.79)
we can calculate the components of the velocity from ψ

∂ψ
ux = − = sin (kx) e−ky [k(A + By) − B] (2.95)
∂y

and
∂ψ
uy = = k cos (kx) e−ky (A + By). (2.96)
∂x
For the Earth, the upper boundary of our half space is covered by a rigid lithosphere and we
require that the horizontal component of the velocity along the surface (y = w) to be equal to
zero (no slip condition). As w  λ, we can apply this boundary condition at y = 0 instead of
y = w. By setting ux = 0 at y = 0 we find

B = kA (2.97)

and ψ, ux , and uy then simplify to

ψ = A sin (kx) e−ky (1 + ky) , (2.98)

ux = A sin (kx) e−ky k 2 y, (2.99)

uy = Ak cos (kx) e−ky (1 + ky) . (2.100)
22 Geodynamics & Continental Deformation 2020

To determine the final constant A, we need to equate the lithostatic pressure ρgw(x, t) associated
with the topography w(x, t) to the vertical normal stress at the upper boundary of the fluid half-
space, y = 0. Then
∂uy
Tyy = −p + ηEyy = −p + 2η = ρgw. (2.101)
∂y
Let’s now work out what the pressure p is equal to at the surface (y = 0). To do this, we
substitute ux from equation (2.99) into the horizontal force balance (the x-component of the
Stokes equation) and we rearrange for ∂p/∂x. This gives

∂p
= −2ηAk 3 sin (kx) . (2.102)
∂x
We can integrate this with respect to x to obtain the pressure at y = 0

p(x, y = 0) = 2ηAk 2 cos (kx) (2.103)

This is the pressure associated with the deformed surface that is responsible for driving the flow.
Next, let’s evaluate ∂uy /∂y at the surface (y = 0). From equation (2.100) this gives
 
∂uy
=0 (2.104)
∂y y=0

Substituting (2.103) and (2.104) into equation (2.101) and solving for h gives

−2ηAk 2
w(x, t) = cos (kx) (2.105)
ρg

Equation (2.105) relates the surface displacement w to the integration constant A as well as
the material parameters ρ and η and the wavelength λ. The final step is to relate the rate
of change of the surface displacement w to the vertical velocity at the surface. This is simply
(again making use of the fact that w  λ)

∂w
= uy (y = 0). (2.106)
∂t
Using equation (2.100) for uy and evaluating at y = 0 then gives

∂w ρg
= Ak cos (kx) = −w , (2.107)
∂t 2kη

where we have used equation (2.105) to replace kA cos(kx). The solution to equation (2.107) is
simply
 
−ρg
w(x, t) = w(x, 0) exp t
2kη
 
ρg
= w0 cos (kx) exp − t . (2.108)
2kη

We can also write this as

w(x, t) = w0 cos (kx) e−t/τ0 (2.109)
where τ0 = 2kη/(ρg) = 4πη/(ρgλ) is the time constant for postglacial recovery. Note the way
that τ0 scales with the viscosity, density, and wavelength. Is this what you would expect?
Comparing equations (2.105) and (2.109) reveals that A is a function of time
ρgw0 −t/τ0 w0 −t/τ0
A(t) = − 2
e =− e . (2.110)
2ηk τ0 k
Fluid Mechanics 23

0.0

−0.5
0.5

−0.1
0.1
0.1

−0.4
0.4
0.2

.3
0.3

−0
0.3

0.4
2 .2
0. −0
y/λ

0.5

0.6 0.1 −0.1

0.7

0.8

0.9

1.0
−0.5 −0.4 −0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 0.4 0.5
x/λ
ψ
Figure 2.11: Contours of (Eq. (2.111)). Arrows show velocity vectors (Eqs. (2.112) and 2.113).
ψ0

Then our final set of equations for ψ, ux , and uy are

w0 e−t/τ0
ψ=− sin (kx) (1 + ky) e−ky = −ψ0 sin (kx) (1 + ky) e−ky , (2.111)
τ0 k
ux = −ψ0 k sin (kx) ky e−ky , (2.112)
−ky
uy = −ψ0 k cos (kx) (1 + ky) e . (2.113)

2.13 Estimate of Mantle Viscosity from Glacial Isostatic Adjustment

The ice sheets that covered Fennoscandia during the last ice age melted at about 11,000 years
before present. Since then, relative sea-level in those regions has been falling, indicating uplift
of the land surface relative to the geoid (Figure 8.1). This drop in sea-level is approximately
exponential, with a time-constant of 4.6 ka (Figure 8.1a). So we may use glacial isostatic
adjustment to obtain an approximate value for the viscosity of the mantle.
According to Eq. (2.109), the time constant for recovery of suface height is

τ0 = 4πη/(ρgλ)

whence
4.6 kyr × 3300 kg m−3 × 10 m s−2 × 2000 km
η∼ ∼ 1021 Pa s
4π
(Look at Figure 8.1 and make sure that you agree with the estimate of the wavelength, λ.)
24 Geodynamics & Continental Deformation 2020

Figure 2.12: Glacio-isostatic recovery of Fennoscandia. (Top) Rates of vertical surface motion, relative to
the geoid, of Fennoscandia, obtained with GPS by Johansson, J. et al., J. Geophys. Res., 107, B8,2157.
Bottom: (a) shows observational data of relative sea level versus time, with an exponential fit. (b) shows
a schematic diagram of the preglacial, synglacial, and postglacial state of the land surface.
 Fluid Mechanics

1. The following question parts will all assume an incompressible flow and a Newtonian fluid.
(a) Give two of the three conservation principles that were used in deriving the Navier-Stokes
equations.
(b) What physical information is contained in the ij th component of the stress tensor T? The
stress tensor is often written as Tij = −δij p + τij , where p is the pressure and τ is the
deviatoric stress tensor. What are the entries in τ if the fluid is at rest? Discuss how τ is
related to the velocity gradients in the flow.

Plate

Asthenosphere

Figure 1: (For question 2) Asthenospheric counter flow: A plate moves over a low viscosity as-
thenosphere and a return flow is driven by a lateral pressure gradient dp/dx. The return flow
balances the forward flow.

(c) The configuration of 1D asthenospheric counter flow is shown in Figure 1. In this problem
a plate of thickness H moves at a velocity uo to the right and over a low viscosity astheno-
sphere of thickness h and viscosity µ. The velocity at the base of the asthenosphere is
fixed at zero. The associated pressure gradient gives a return flow that is exactly balanced
by the forward flow. In this case the velocity, as a function of the vertical distance y from
the base of the asthenosphere, is given by:
(
uo for h ≤ y ≤ h + H
ux = 1 dp y y (1)
2 dx µ (y − h) + uo h for 0 ≤ y ≤ h

The pressure gradient that gives a balanced return flow is found to be:
dp uo
= 6µ 3 (h + 2H) (2)
dx h
(i) Calculate the yx component of the strain rate tensor E as a function of y in the
asthenosphere. (ii) Using this, calculate the drag stress τxy on the base of the plate at
y = h.
(d) For the asthenospheric counterflow problem, consider a plate of thickness H = 50 km
moving at a velocity of uo = 3 cm/year over an asthenosphere of thickness h = 100 km
and viscosity µ = 1021 Pa s. Calculate the pressure drop ∆P that would be required to
drive the return flow over a distance of L = 2000 km.
(e) Such a large pressure drop would also produce dynamic topography. (i) What change in
elevation might we expect over the distance of 2000 km to form this pressure drop? (Hint:
The increase in hydrostatic pressure from such a change in elevation can be calculated
as ∆P ≈ ρgl, where l is the difference in topography over the 2000 km. You may use
ρ = 3500 kg/m3 .) (ii) Is the magnitude of the dynamic topography reasonable? If not,
discuss which assumptions above might be in error?

Page 1 of 8
 Solution:
(a) Two of the following

1. Conservation of mass.
2. Conservation of linear momentum (also alright if you say Newton’s 2nd law here).
3. Conservation of angular momentum.

(b) Tij denotes the force per unit area in the j th direction on a surface with normal in
the ith direction. τ = 0 if the fluid is at rest. τ is related to the velocity gradients
in the fluid through the rate of strain tensor E. You could also say that τ is linearly
proportional to the velocity gradients (since we are talking about a Newtonian fluid).
 
1 dp 2y
(c) The yx component of E is Eyx = 12 du dy
x
= 4 dx µ − h uo
µ + 2h . Then τyx = 2µEyx =
 
µ dp 2y h µuo h dp µuo
2 dx µ − µ + h . Finally, τyx (y = h) = 2 dx + h .

dp dp
(d) Using the above values dx = 1.2 × 103 P a/m. Then ∆P = dx L = 2.4 × 109 P a.

(e) l = ∆P/ρg ≈ km. This is an absolutely huge change in topography and is unreason-
able. Possible reasons for this might be:

1. The estimate of viscosity in the asthenosphere is too high.

2. The thickness of the asthenosphere is larger than 100 km.
3. Complete asthenospheric counter flow does not occur in the Earth.
4. Any other reasonable comment here would be acceptable. The purpose of this
question is to get the students thinking beyond the simple problems that we did
in class.

2. (a) Discuss the fluid-dynamical meaning of the streamfunction ψ, indicate how it relates to
the two-dimensional velocity u = (ux , uy ), and then demonstrate that it automatically
satisifies the continuity equation for an incompressible fluid.
(b) The postglacial rebound of a surface with initial topography is given by wm = wm0 cos(2πx/λ),
where λ is the wavelength of the perturbation. The topography evolves as:
 
2πx −t/τr
w(t) = wm0 cos e
λ
(i) What does τr represent? (ii) Using physical reasoning and units, determine the defini-
tion of τr in terms of the parameters density ρ, gravitational acceleration g, viscosity µ,
and λ. Explain your logic.
(c) Beneath the topography at y > 0, the mantle is flowing in response to pressure gradients.
The streamfunction for flow of the mantle is:
   
ρgwm0 λ −t/τr 2πx −2πy/λ 2πy
ψ=− e sin e 1+
4πµ λ λ
(i) Compute the horizontal and vertical component of the mantle velocity. (ii) Discuss
how the resulting functions vary with x, y, t, and λ.
(d) Postglacial topographic perturbations decay with time because of mantle flow. Temper-
ature perturbations in a quiescent, bottom-heated fluid shows can decay or grow with
time. Explain this difference with physical reasoning.

Page 2 of 8
Solution:
(a) The streamfunction is a scalar potential for the fluid velocity. Contours of the stream-
function are instantaneous streamlines of the flow. Components of the velocity are
related to ψ as
∂ψ ∂ψ
ux = − , uy = .
∂y ∂x
We can see that
∂ux ∂uy ∂2ψ ∂2ψ
∇·v = + =− + = 0.
∂x ∂y ∂x∂y ∂x∂y

(b) τr represents the time-scale for the decay of the postglacial topography. Large viscosity
gives slower flow and hence a longer time-scale; density and gravity combine to give
larger pressure gradients in the mantle and hence faster flow and a shorter time-scale.
Then, to give τr the units of time, it must be of the form τr ∝ µ/(ρgλ). For full credit,
you can note that τr = 4πµ/(ρgλ).
Why does the PGR time constant decrease with increasing λ? Think of this as
a slightly complicated type of channel-flow problem. There is a pressure gradient
that looks like ρgwm0 /λ. This forces flow over a depth range in the mantle that is
proportional to λ; from the channel-flow problems, we know that the speed of the
flow is proportional to the pressure gradient multiplying the square of the channel
thickness (∼ λ2 ). The time constant will be proportional to the pressure gradient and
inversely proportional to the speed: hence, overall, inversely proportional to λ.

(c) The horizontal and vertical components of the mantle velocity are obtained using the
definition of the streamfunction from part (a) of this question. Calculating them gives
   
ρgwmo −t/τr 2πx −2πy/λ 2πy
ux = − e sin e ,
2µ λ λ
   
ρgwmo −t/τr 2πx −2πy/λ 2πy
uy = − e cos e 1+ .
2µ λ λ
(Note that there is a λ/2π embedded in the constant at the front of the equation in the
question, which is cancelled out from these expressions.) From this we can see that
the flow varies in x sinusoidally with the same wavelength as the surface topography.
In the y direction, the strength of flow goes to zero at y = 0 to match the boundary
condition. It increases with depth initially, but for y & λ/2π, the both components
decay with increasing y. The velocity also decays with time, as the surface topography
disappears. Large wavelength leads to larger penetration of flow in y but also more
rapid decay in time. The response to larger wavelength postglacial topography thus
samples the viscosity deeper into the mantle.

(d) In the case of PGR, flow is forced by lateral variation in the lithostatic pressure. This
flow reduces the topography. There is no force driving an increase in topography.
For the case of convection, there are two competing processes: diffusion of heat and
advection of heat. Diffusion of heat reduces thermal perturbations; advection of heat
amplifies them. Density differences associated with temperature causes hot fluid to
rise and cool fluid to sink. If advection dominates over diffusion, thermal perturbations
grow and the fluid convects; if diffusion dominates, thermal perturbations decay and
the fluid remains quiescent.

Page 3 of 8
3. Regarding the streamfunction, ψ:
(a) Define the streamfunction in terms of its relationship to the components of a two-dimensional
velocity vector u = (ux , uy ). Show that your choice of the streamfunction automatically
satisfies the continuity equation for an incompressible fluid.
(b) Discuss the conditions under which the streamfunction formulation of the Stokes equations
is a useful simplification.
(c) Consider a pure shear flow given by

v = A(xi − yj).

Find the corresponding streamfunction ψ, sketch its contours, and indicate the direction
of flow.
(d) Compute the strain-rate tensor E and the rotation-rate tensor Ω associated with v =
A(xi − yj). Describe in words the advection, deformation, and rotation of an infinitessimal
parcel of fluid embedded in the flow.

Solution:

Page 4 of 8
 (a) The streamfunction is defined as

∂ψ ∂ψ
ux = − , uy = .
∂y ∂x
The continuity equation for an incompressible fluid is ∇ · u = 0. Substitution gives

∂2ψ ∂2ψ
− + = 0,
∂x∂y ∂y∂x
which is true because the order of differentiation may be reversed.

(b) The streamfunction is useful for solving problems involving incompressible flow, es-
pecially if that flow is in two dimensions. In this case, the streamfunction can be
used to reduce the problem from solving for pressure and two velocity components to
solving for a single, scalar field, ψ. Once computed, the streamfunction is also useful
for visualising the flow: contours of the streamfunction are streamlines of the flow.

(c) The streamfunction for pure shear is

ψ = −Axy.

This function is contoured below. Flow is inward along the top and bottom of the
domain and outward along the left and right sides.
y=0

x=0

(d) The strain-rate tensor is

   
1 ∂ui ∂uj 1 0
E= + =A .
2 ∂xj ∂xi 0 −1

The rotation-rate tensor is

   
1 ∂ui ∂uj 0 0
Ω= − = .
2 ∂xj ∂xi 0 0

Infinitessimal parcels are stretched in the x-direction, compressed in the y-direction,

advected along contours of the streamfunction, and are unrotated by the flow.

4. Consider a channel of fluid of height h = 1 millimetre, aligned with the x-direction, with rigid
walls at y = ±h/2. A pressure gradient G ≡ dp/dx drives a flow in the +x direction with
velocity u = (u, 0). This flow must satisfy

∂ρ
+ ∇ · (ρu) = 0, (3a)
∂t
(3b)

Page 5 of 8
where g is the acceleration of gravity and points in the −y direction.
(a) The density of the fluid, ρ, is known to be constant and uniform with a value of 1000
kg/m3 . Use this knowledge to write a simplified version of the set of governing equations.
Explain your work.
(b) The fluid has a characteristic viscosity of approximately 1 Pa-sec and a characteristic
speed uc of about 10 cm/sec.
i. Compute a value for the Reynolds number using ρ, µ, uc , and h. Explain the physical
meaning of your result.
ii. Simplify the governing equations appropriately, given the value of the Reynolds num-
ber.
(c) Write the total stress tensor T in terms of the pressure p and the deviatoric stress tensor τ .
Use your result to derive the equation
dτxy
= G. (4)
dy

(d) The fluid is non-Newtonian. The relationship between stress and strain rate is
 1/n
du
τxy = C , (5)
dy

where n is and odd constant (n = 1, 3, 5...). Use equations (4) and (5) with boundary
conditions 
du 
= 0, and u(y = h/2) = 0,
dy y=0
to determine u(y). Calculate u, the mean flow speed.
(e) Refer to the plot of u(y)/u for n = 1, 3, 5. Describe how n affects the viscosity and the
profile of the flow. What would you expect to see as n → ∞? What type of rheology is
this?
0.5

0.4

0.3

0.2

0.1
y

-0.1

-0.2

-0.3

-0.4

-0.5
0 0.5 1 1.5

u=u

Solution:

Page 6 of 8
(a) If density is constant and uniform, the fluid is incompressible and hence we can write
(3a) as
∇ · u = 0.

Furthermore, since there are no gradients in density and the flow is driven by the
pressure gradient, we can write the Navier-Stokes equation as

Du
ρ =∇·T
Dt

(b) Using the values h = 10−3 mm, ρ = 1000 kg/m3 , uc = 10−1 m/sec, and µ ≈ 10 Pa-sec.
i. The Reynolds number is the ratio of inertial to viscous stresses. For this flow,

ρhuc 103 10−3 10−1

Re = = = 0.1
µ 1

which means that viscous stresses are much more important than inertia.
ii. Since inertia is small relative to viscous stresses, we can rewrite the Navier-Stokes
equation as
∇ · T = 0.

(c) We can expand the stress tensor as

T = −pI + τ ,

where τ is the deviatoric stress tensor. Since we know that u = (u, 0, 0), and by the
continuity equation we know that ∂u/∂x = 0, the only non-zero component of the
stress tensor is τxy . The x-component of the momentum equation is then

dp dτxy
0=− +
dx dy

or G = dτxy /dy.

(d) Combining (4) and (5) we have

" 1/n #
d du
G=C
dy dy

integrating once and using the boundary condition at y = 0 gives

du
(G/C)n y n = .
dy

Integrating again and using the boundary condition at y = h/2 gives

"  n+1 #
(G/C)n n+1 h
u= y −
n+1 2

Page 7 of 8
 The mean flow speed is given by
Z h/2
1
u= u(y)dy
h −h/2
Z "  n+1 #
h/2
2 (G/C)n h
n+1
= y − dy
h n+1 0 2
"  n+1 #h/2
2 (G/C)n y n+2 h
= −y
h n+1 n+2 2
0
 n+1
(G/C)n h
=−
n+2 2

(e) In the figure we see that as n increases, the viscosity becomes more sensitive to the
strain rate. The strain rate is largest near the walls, and so at higher n, the viscosity
is lower there. This causes the shear to localise along the wall of the channel, while the
middle of the flow becomes plug-like. As n → ∞, the viscosity goes to a plastic-type
rheology with a yield stress. All of the deformation is concentrated at the wall of the
pipe, with a plug of fluid flowing down the middle.

Page 8 of 8
Kinematics 25

3 Kinematics of Continental Deformation

3.1 Difference between Continental Tectonics and Plate Tectonics
Plate tectonics provides a useful description of the relative motions of the oceanic parts of the
Earth’s surface, and is an adequate framework in which to describe the current motions of those
parts of the continents that have not deformed since Mesozoic time. However, the boundary
between two plates should be a narrow band of faults (a few tens of kilometres wide at most)
that accommodates all the relative motion between two otherwise rigid plates (e.g. the narrow
bands of earthquakes at the oceanic ridges, and near subduction zones, shown in Figure 3.1); the
continents show broad regions of shallow seismicity (Figures 3.2 and 3.3) which are completely
inconsistent with this rule.
0˚ 30˚ 60˚ 90˚ 120˚
60˚ 60˚

45˚ 45˚

30˚ 30˚

15˚ 15˚

0˚ 0˚

−15˚ −15˚

0˚ 30˚ 60˚ 90˚ 120˚

Figure 3.1: Epicentres of earthquakes shallower than 35 km. Grey dots show earthquakes since 1964,
of magnitude 4 or greater; black dots show earthquakes since 1900 of magnitude 5 or greater. Note the
narrowness of the bands of seismicity along the ridges and trenches, in comparison with the distribution
of earthquakes within Eurasia. (The apparently broad zones of earthquakes along the trenches are, in
fact, the projections onto the surface of narrow, but dipping, bands of hypocentres.)

3.2 Scales of Deformation

The horizontal scale of deformation: Figures 3.2 and 3.3 show that the horizontal scale of
continental deformation is 100–1000-km. The locations of shallow earthquakes in the continental
interiors are closely related, spatially, with regions in which the surface lies either considerably
above or considerably below the elevation of undeforming continental lithosphere. This variation
in surface heights implies variation in crustal thickness, thus finite strain (of tens of percent) of
the crust and, by implication, of the lower lithosphere.
26 Kinematics and Dynamics of the Lithosphere 2020

Figure 3.2: The distribution of seismicity and the focal mechanisms of large earthquakes in central and
southern Asia. All the earthquakes shown here have hypocentres that are nominally shallower than 33 km;
most are probably shallower than 20 km. Focal mechanisms are of earthquakes Mw & 5.5 since 1977 from
the catalogue of Ekström et al. (2012). Double lines outline the Tarim Basin, a relatively aseismic region
between Tibet and the Tien Shan.
Kinematics 27

20˚ 30˚

40˚ 40˚

20˚ 30˚

Figure 3.3: The distribution of seismicity and the focal mechanisms of large earthquakes in the Aegean
(above) and Iran (below). Symbols as in Figure 3.2. Double lines show relatively aseismic regions whose
motions relative to Eurasia are shown in Figure 3.7.
28 Kinematics and Dynamics of the Lithosphere 2020

Seismogenic upper crust: Seismicity in continental crust usually shows an abrupt cut-off
in depth. That depth varies from place to place, but it is usually within the middle crust, in
the range 10–20 km.

Figure 3.4: Histograms of earthquake focal depths determined by modelling of long-period teleseismic
seismograms. Upper panels from Sloan et al. (2011), lower panels from Maggi et al. (2000).

The crust as a whole: Surface heights in regions of active continental deformation, when
averaged over ∼ 100 km, vary between about -2000 m and 6000 m. To a good approximation,
surface elevation contrasts whose horizontal scale exceeds about 100 km are compensated iso-
statically. Isostatically balanced columns may, however, differ in their gravitational potential
energy, and therefore have the potential to do work on each other. As we shall see, contrasts in
gravitational potential energy contribute importantly to the dynamics of the continents.

The lithosphere: We must draw a distinction between the lithosphere and the plates of Plate
Tectonics. By definition, a plate does not strain, except in a small elastic way. The continental
lithosphere obviously does deform, and by large strains. We shall use the term ‘lithosphere’ to
refer to the mechanical boundary layer at the top of the mantle. To deform this layer at geological
strain rates requires stresses that greatly exceed the stresses involved in mantle convection.
Kinematics 29

3.3 Models of Continental Deformation

Between the discovery of plate tectonics
and about the middle 1980s, the tech-
niques available to measure the deforma-
tion of the continents were not greatly
different from those used to identify plate
tectonics. Figures 3.2 and 3.3 show re-
gions in which the seismic activity is rela-
tively low, suggesting that they are plate-
like: For example, the Tarim Basin (40◦
N, 80–90◦ E, Figure 3.2), central Iran
(33◦ N, 55◦ E, Figure 3.3), the south-
ern Aegean and central Turkey (37◦ N,
26◦ E, and 40◦ N, 33◦ E, Figure 3.3),
and the Adriatic Sea (42◦ N, 15◦ E, Fig-
ure 3.7). It seemed a reasonable approx- Figure 3.5: A model of deformation in the Mediterranean
imation treat such areas as plates, and and Middle East, containing 3 major plates: Eurasia (1),
to suppose that continental deformation Africa (2), and Arabia (8), and 5 microplates, whose mo-
resembles plate tectonics, except with a tions relative to Eurasia are shown by arrows (Mc Kenzie,
larger number of smaller plates (e.g. Fig- 1972).
ures 3.5 and 3.6).

Figure 3.6: Interpretations of the tectonics of Asia in terms of horizontal relative motions. (Left panel)
Calculations of the orientations of slip lines in a plastic medium indented by rigid punches of different
shapes (Tapponnier and Molnar , 1976). The deformation in constrained to be in the plane of the image,
and the slip lines are taken to be equivalent to strike-slip faults. (Right panel) A model for the deformation
of Asia in which most of the convergence between India and Asia is accommodated by the relative motion
of rigid blocks of lithosphere separated by strike-slip faults (Tapponnier et al., 1986).
30 Kinematics and Dynamics of the Lithosphere 2020

Figure 3.7: An elaboration of Figure 3.5 (after Jackson and Mc Kenzie, 1988). The motions of Africa
and Arabia relative to Eurasia are shown by black arrows (scale at top left). Hatched regions represent
microplates; their motions relative to Eurasia, shown by white arrows, are estimated from the slip vectors
of earthquakes around their edges.

The questions soon arose, however, as to whether

this approach is useful – in the sense of helping us to
understand the deformation – and whether it stands
up to quantitative testing. One problem is that the
‘microplates’ are generally small compared with the de-
forming regions as a whole. They are 100–300-km scale
bodies embedded in deforming regions whose scale may
approach the size of the continents themselves. Even
when the relative motion of two block is known, the
style of deformation in the region between the bocks
remains unknown (see Figure 3.8).
Furthermore, the aseismic blocks do not remain Figure 3.8: Sketch of ways in which con-
rigid for times comparable with the duration of defor- vergence between two plates A and B may
mation in a region. For example, the southern Aegean, be accommodated by deformation in the
central Turkey, and the Central Iranian plateau all con- zone between them; any combination of
tain faults that appear to have been active in Quater- patterns (a) and (b) is permitted that
nary time. The most problematic observation is seen in matches the Eyy component of strain (see
the Tibetan plateau, whose surface is 5 km above sea Figure 3.16).
level, implying that its crust has been thickened during
the convergence of India with Asia. Yet the active deformation shows crustal extension (normal-
plus-strike-slip faulting, Figure 3.2). Not only is this region (which, incidentally, is the size of
the Cocos plate) deforming pervasively, but the nature of its deformation has changed over the
time during which India has been converging with Asia.
A theory for continental deformation needs to be able to account for the scales of actively
deforming regions, and for the time-dependence of the deformation. One approach has been
to treat the lithosphere as a fluid. This approach may seem wrong-headed, given that the
deformation of the upper crust is manifestly discontinuous: each of the earthquakes shown in
Figures 3.2 to 3.7 represents slip on a fault. Recall two points, however: (i) The upper crust
forms only a small part of the lithosphere (Section 3.2). (ii) Employing the continuum hypothesis
is not to assume that the lithosphere is free of discontinuity; rather, it is to recognise that there
are so many discontinuities that they cannot be analysed individually: it is simpler to analyse
the deformation in terms of the average properties of the medium (Figure 2.1).
Kinematics 31

Testing the notion that continental deformation is continuous introduces two significant
difficulties not present when testing plate tectonics. First, contrast the governing equations for
plates and continuous media:

Plate tectonics
v = ω×r
Testing plate tectonics requires only measurements of velocity. For continuous deformation:
∂Tji
= ρgi .
∂xj

Even in the simplest case of homogeneous Newtonian viscosity

Tji = τji − δij P

 
η ∂ui ∂uj
τji = +
2 ∂xj ∂xi

so testing the continuous approach requires measurements of the derivatives of strain rate

∂τxx ∂ 2 ux
(e.g.) =η .
∂x ∂x2
We now consider how such measurements can be made. In the following chapter we discuss
how to characterise the effective viscosity of continental lithosphere.

3.4 Measurements of Continental Deformation

The horizontal scale of continental deformation comfortably exceeds the thickness of the crust,
and in most places is several, to tens of, times the crustal thickness. Suppose we wish to
characterise the deformation of the Aegean (∼ 500 × 500 km, Figure 3.3). Twenty-five evenly
spaced measurements will give us the deformation at the scale of the lithospheric thickness; we
require 200 measurements if we are interested in deformation at the scale of the crust, or 2,500 if
we are interested in the scale of the seismogenic upper crust (which is about 10 km thick, here).
There are, currently, two ways of measuring strain on the appropriate scale: from the moment
tensors of earthquakes, and with the Global Positioning System of satellites (GPS). Each of these
methods has limitations. InSAR will eventually provide measurements with the necessary spatial
resolution, but those lie several years in the future.

3.4.1 Measuring Strain with Earthquakes

An early motivation for the use of earthquakes to measure distributed strain in the continents
was the observation (Molnar et al., 1973) that the P axes of earthquakes throughout Asia are
aligned, independently of whether the earthquakes have reverse-, strike-slip-, or normal-faulting
solutions. For example, compare the left-hand panel of Figure 3.9 with Figure 3.2: the reverse
faults beneath the Himalaya, and the strike-slip and normal faults in central Tibet all have
roughly north-south P axes, whereas the reverse faults have NS slip vectors, the normal faults
have EW slip vectors, and the strike-slip faults have NE/SW slip vectors. The NS alignment
is consistent with the shortening required to accommodate the convergence between India and
Asia. In eastern Tibet, the orientations of P axes become more easterly; here they are aligned
roughly with gradients of surface height.
There are two points to be made about these observations. First, if the rules of plate tectonics
applied to the deformation, the slip vectors of the earthquakes ought to align systematically; the
alignments of the P axes suggests, instead, a systematic pattern of strain in the region. Secondly,
the alignment of the P axes with surface slopes suggests that, somehow or other, the height of
32 Kinematics and Dynamics of the Lithosphere 2020

20˚ 25˚
75˚ 80˚ 85˚ 90˚ 95˚ 100˚ 105˚ 110˚ 115˚ 120˚
55˚ 55˚

50˚ 50˚
40˚ 40˚

45˚ 45˚

40˚ 40˚

35˚ 35˚

35˚ 35˚

30˚ 30˚

75˚ 80˚ 85˚ 90˚ 95˚ 100˚ 105˚ 110˚ 115˚ 120˚ 20˚ 25˚

Figure 3.9: (Left) P axes (contraction directions) of earthquakes in Asia. (Right) T axes (extension
directions) of earthquakes in the Aegean.

the topography influences deformation. This point is emphasised by the relation between fault
style and surface height; Figure 3.10 shows that all normal faults in Tibet occur at elevations
higher than 4500 m and, with a single exception, all reverse-faulting earthquakes are lower than
this height.

Figure 3.10: Relationship between surface height and fault type in Tibet (Elliott et al., 2010). Top
panel shows focal mechanisms of earthquakes superimposed on a topographic map; lower panel plots the
earthquake type agains the surface height at the epicentre of the earthquake.
Kinematics 33

A similar relationship is seen in the Aegean, where the extensional axes of earthquakes
within the Aegean systematically point towards the lower Mediterranean ocean floor to the SW,
S, and SE of the Hellenic plate boundary, irrespective of whether the earthquakes are strike-slip
or normal-faulting (compare the right-hand panel of Figure 3.9 with upper panel of Figure 3.3;
note that Figure 3.9 displays axes only of earthquakes within the Aegean continental lithosphere;
the reverse faulting on the plate boundary is not shown).

The Moment Tensor The measure of an earthquake is its Moment Tensor:

Mij = M0 (ûi n̂j + ûj n̂i ) (3.1)

M0 = ηAs (scalar moment)

A : area of fault plane
s : average slip on the plane
η : shear modulus of rocks in the neighbourhood of the fault
û : unit vector in direction of slip
n̂ : unit vector normal to fault plane

Moment magnitude:
2
Mw = log10 (M0 ) − 6.0 (3.2)
3
with M0 in newton metres (Hanks and Kanamori , 1979).

Kostrov’s Relation The strain rate of a volume, V , containing N earthquakes in time, t, is

related to the moment tensors of those earthquakes by:
N
1 X n
Eij = Mij (3.3)
2ηV t
n=1

where Eij is the ijth component of the strain rate and Mn is the moment tensor of earthquake
n contained in the volume. This relation assumes that there is no elastic strain stored within
the volume; any such strain caused by one earthquake must be released by other earthquakes
or relaxed aseismically. This result is given in a paper (Kostrov , 1974), that is not available
electronically at the moment. A derivation is given at the end of this chapter (Section 3.5).

Scaling of fault length and slip with moment The length and amount of slip in an
earthquake both scale with the moment (e.g. Wells and Coppersmith, 1994) (Figure 3.11). The
slip in an earthquake is about 3×10−5 times the fault length: a line of slope 1 through the
observations of slip versus length (Figure 3.11, right panel) gives a displacement of about 3 cm
for a 1-km fault, or 30 m for a 1000-km fault. But note the variability in the data: this ratio
varies by at least a factor of 3.

Large and small earthquakes An earthquake of magnitude Mw 5.5–6 will generally have a
fault break of 10–15 km (comparable with the thickness of the seismogenic upper crust) whereas
one of Mw 3.5–4 will involve a fault break of only 100 metres. Earthquakes whose magnitudes
exceed about Mw 6.5 generally involve fault slip with an along-strike length that exceeds the
thickness of the seismogenic layer. It is common to refer to those earthquakes whose fault breaks
are comparable in scale to the thickness of the seismogenic layer as ‘large’. ‘Small’ earthquakes,
in this context, deform a region smaller in size than the thickness of the seismogenic layer.
34 Kinematics and Dynamics of the Lithosphere 2020

Figure 3.11: [Left:] Observations of the length of fault rupture for earthquakes of different moment
magnitudes. [Right:] Relation between slip and length of rupture. All of these data are derived from
direct observation (Wells and Coppersmith, 1994).

For total cumulative offsets on faults the ratio γ of slip to length is more scattered than
for incremental slip in earthquakes, and lies in the range 10−1 to 10−3 (Figure 3.12). Note the
apparent change in γ near 10 km length, which corresponds to the break between ‘large’ and
‘small’ faults. The difference in scaling between incremental slip in earthquakes and the finite
displacement on faults shows that, in general, faults must increase their length as they increase
their cumulative offset.
Log10M0 (Nm)
15 16 17 18 19 20 21 22
5 5
Log10 Slip (m), Length (km)

4 4
Area
Log10 Area (km2)

3 3
th
Leng
2 2

1 Slip 1

0 0

-1 -1
4 5 6 7 8 9
MW
Figure 3.12: (Left:) Area, length, and Slip for earthquakes of different magnitudes. This sketch sum-
marises the observations in Figure 3.11. (Right:) Displacement-to-length ratios (γ) on faults. Shaded
bands with dots show total offsets on faults. Shaded band (without dots) to lower right indicates incre-
mental slip in earthquakes (Left). From Cowie and Scholz (1992). Breaks in slope corresponds to the
transition between ‘small’ and ‘large’ earthquakes.
Kinematics 35

Moment-frequency relation Local, regional,

and global catalogues of earthquakes show relations
between the size and the frequency of earthquakes
of the form:
N (M ) = aM −b (3.4)
where N is the number of events having moment
greater than or equal to M in a given time interval.
The value of b is found to be about 2/3.
In any given region, there will always be earth-
quakes that are too small to measure. The relation Figure 3.13: Moment-Frequency relationship
(3.4) allows one to estimate what fraction of the for global earthquakes (dots, number of earth-
seismic moment is released in unobserved earth- quakes on right-hand axis). Solid curve (scale
quakes. Suppose that in a particular region, the on right-hand axis) shows cumulative moment
maximum observed scalar moment is Mmax , and release in all earthquakes having moments up
the minimum moment for which a reliable moment to a given value of M0 .
tensor can be determined is Mmin . The total moment release over the time of observation is
Z 
Mmax 
dN 0
Mtot = M − 0
dM 0
0 dM
Z Mmax
= ab M 0(−b) dM 0
0
ab (1−b)
= M (3.5)
1 − b max
whereas the total observed moment release is:
Z Mmax
ab  (1−b) (1−b)

Mobs = M 0 n(M 0 ) dM 0 = Mmax − Mmin (3.6)
M 1−b
min
So:  (1−b)
Mobs Mmin
=1− (3.7)
Mtot Mmax
There are two important implications of Equation (3.7). First, it allows one to determine
what fraction of the deformation is not measured if earthquakes below a given magnitude are
ignored – or cannot be measured.

Suppose the maximum moment magnitude of earthquakes in a region is Mw 7.5, and the smallest
earthquake you choose to measure has Mw 6.5. Calculate the fraction of the seismic moment
released in the area that you do not observe.
Repeat your calculation for lower limits of Mw 6 and Mw 5. This exercise should convince you that,
although large earthquakes are much less frequent than small earthquakes, the large earthquakes
account for the majority of the seismic deformation of the crust (Figure 3.13).

Secondly, because the instrumental period of earthquake observation is short (about 100
years), whereas the repeat time for large earthquakes may be much longer than this, it is
always possible that Mmax in a particular period of observation is substantially smaller than
the moment of the largest earthquakes that ever occur in the region. Equation (3.7) allows
one to estimate the degree to which long-term seismic strain is underestimated in this case (see
Eq. (3.9, below).
36 Kinematics and Dynamics of the Lithosphere 2020

3.4.2 Regional Strains from Earthquakes in the Aegean, Iran, and Asia
w
Figure 3.14 sketches two (among an infinite
variety of) ways in which a zone of distributed
deformation can accommodate oblique con- a
vergence between two plates. Let the relative
velocities of the x- and y-edges of the deform-
ing zone be, respectively, y

x v
v = (vx , vy )
and
u = (ux , uy ) u

In Figure 3.14a, the velocities within the re-

gion are all parallel to the relative plate mo-
tion; v is the relative velocity of the plates,
v
and u = 0. In Figure 3.14b, some of the con-
vergence is accommodated by along-strike ex- Figure 3.14: (Sketch of two deforming regions, each
tension, which produces relative motion of the bounded by two plates whose relative velocity is shown
y-edges. The relative motion of the x-edges is by the large arrow (see text). Small arrows show the
still given by the relative plate velocity, v. velocities within the deforming zones.
If the zone between A and B accommodates v and u by homogeneous deformation, then:
ux 1 h vx uy i vy
Exx = ; Exy = + ; Eyy = (3.8)
w 2 a w a
The diagonal components of E have a simple physical meaning: Exx is the rate of extension
of material along strike of the zone; Eyy is the rate of horizontal shortening in the direction
perpendicular to the strike of the zone. The shear strain rate Exy contains information about
the relative motions of all the edges of the zone.

Eyy is predicted directly by the relative velocities of the two plates bounding the deforming zone.

One reasonable question to ask is: “Do strain rates calculated from earthquakes in continental
plate boundary zones agree with those expected from plate relative motions?’ ’

Jackson and Mc Kenzie (1988) approached this question by measuring the seismic strain
over the interval 1900 to 1981 in ten deforming region in the Mediterranean and Middle East
(Figure 3.15). They found, by comparing the value of Eyy determined seismically with that
predicted from the relative motions of the bounding plates (Eq. (3.8)), that, the fraction of the
total strain that is seismic lies between 10% and 100%.
Some of the differences between observed and expected seismic strain rates doubtless arise
from the short time interval of observation, but this explanation is unlikely to account for the
discrepancies in the Hellenic plate boundary, south of Greece, or in Zagros region of SW Iran.
Consider the moment-frequency relations discussed above (Eqs. (3.4) to (3.7)). Suppose that
over a short time interval the maximum moment of earthquakes observed in a region were M1 ,
whereas the actual maximum moment is M2 . Call the observed strain rate E1 , and let E2 be
the strain rate that would be obtained if the interval of observation were long enough for a
representative sample of the seismicity. From Eq. (3.5) and Eq. (3.3)
 1/(1−b)
ab (1−b) ab (1−b) M2 E2
E1 = M1 ; E2 = M2 ; = . (3.9)
2ηV (1 − b) 2ηV (1 − b) M1 E1
With b ∼ 2/3, the maximum moment required increases as the third power of the ratio of
expected to observed seismic strain rates.
Kinematics 37

Figure 3.15: From Jackson and Mc Kenzie (1988). Shading shows regions in which the moment release from
earthquakes 1900–1988 probably accounts for most of the deformation of the upper crust (50-100% – dark shading);
for a significant part (15–50% – hatching), or very little (less than 15% – circles). Regions marked with crosses
are those in which results are inconclusive.

The maximum observed magnitude of earthquakes in the Hellenic plate boundary, is about
Mw 7, and E2 /E1 ∼ 10 (Figure 3.15). To accommodate the relative motion across the Hellenic
plate boundary seismically would require a maximum earthquake size of Mw 9 (a 1000-fold in-
crease in moment is equal to an increase of 2 in the magnitude Eq. (3.2)). Such an earthquake
requires a fault length of ∼ 1000 km (Figure 3.12), which is larger than the whole plate bound-
ary and very much larger than any plausible fault segment along it. If we were to suppose that
the discrepancy were accounted for by more frequent smaller earthquakes, we should require
an Mw ∼8 earthquake approximately once per century and, although the historical record is
incomplete, it is inconceivable that the ∼ 20 great earthquakes that would be needed to remove
the deficit during the past 2,000 years could have occurred unremarked. A similar argument
applies to the Zagros region of SW Iran.

Another question is “What the style of strain rate is recorded by the seismicity? ” Table 1
and Figure 3.16 show strain rates calculated from the summed moment tensors of earthquakes
in Asia (Molnar and Deng, 1984) and Iran and the Aegean (Jackson and Mc Kenzie, 1988). The
strain rates show distinct differences from what is expected from the relative motions of the
bounding plates. For Asia and Iran, the predominant principal axis is of contraction roughly
aligned with the convergence direction between the plates, but there are significant components
of extension perpendicular to that direction. The Aegean, bounded by the Nubian and Eurasia
plates which converge NS, is extending in that direction.

Table 1: Strain rates from seismic moments (Jackson and Mc Kenzie, 1988; Molnar and Deng, 1984).

Region Exx Eyy Exy Ezz E1 E2 Azimuth

−8
10 /yr −8
10 /yr
Asia -5.6 -3.1 -11.6 8.7 7.3 -16.0 132
Iran -0.2 -4.0 -4.0 4.2 2.3 -6.5 148
Aegean .25 10.0 -1.0 -10.0 10. 0.1 6
38 Kinematics and Dynamics of the Lithosphere 2020

Figure 3.16: Seismic strain rates in Asia, Iran, and the Aegean (Ekström and England , 1989). Focal mechanisms
of large and great earthquakes in the interval 1900-1981 are plotted with their areas proportional the approximate
areas of their fault planes (Figure 3.12). Symbols on the lower right of each panel represent the summed moment
tensors of these earthquakes. Symbols on the lower left of each panel show the orientations and relative magnitudes
of the horizontal principal axes of seismic strain (Table 1). White arrows show motions of India and Arabia with
respect to Eurasia.

3.4.3 Measurement of Strain with GPS

The Global Positioning System (GPS) of satellites currently provides the most precise mea-
surements of displacements of the earth’s surface but is likely to be superseded, in the coming
years, by the far denser measurements available from satellite radar interferometry (InSAR).
The GPS system was designed for navigation (determination of the position of objects as they
move relative to the earth’s surface); the geophysics community has (as with other examples
of military hardware) turned the system to its own use. The geodetic use of GPS is for point
positioning: determining the postion of a static object. Because the object is static, its position
may be determined with precision by multiple repeated measurements. Displacements of such
GPS sites are interpreted tectonically by assuming that the motion of each site represents the
motion of a surrounding volume of rock that is several kilometres or tens of kilometres on a side.
There are two ways of acquiring GPS measurements of position. Campaign mode involves
repeated occupation for short intervals (8–72 hours) at separate occasions over many years.
In continuous mode a receiver is left in one location for years at a time. The distribution of
formal uncertainties for sites occupied in these two modes in the Aegean is shown in Figure 3.17.
These formal uncertainties have two fundamental components: first there are sources of noise
in the GPS signal (for example: atmospheric and ionospheric delays to the signal, and errors
in the satellite orbits). Secondly – and particularly for campaign-mode GPS – the estimates of
displacement between one measurement and the next may be contaminated by errors in setting
up the instruments, or by non-tectonic motions of the ground (Figure 3.18).
There is a third source of uncertainty (often swept under the carpet) attached to the question
of whether it is reasonable to assume that the motion of one small piece of rock on the surface
is related to the motions of the region that surrounds it. This is sometimes referred to as an
‘epistemic’ uncertainty: we are uncertain as to whether the measurement means what we think
it means. We defer discussion of epistemic uncertainties until Chapter 6.
Figure 3.17 shows that, after about 3 years of occupation, CGPS sites have formal uncer-
tainties in horizontal components of their velocities of a few tenths of millimetres per year,
but ten years or more are required to achieve such a level of precision in campaign mode. Un-
certainties in the vertical component of velocity are typically three times as great as in the
horizontal (see, for example, Figure 3.18). GPS determinations of vertical velocity are hard to
Kinematics 39

20˚ 22˚ 24˚ 26˚ 28˚

4 42˚ 42˚
Occupations
Standard Deviation mm/yr 10
9
3 8
40˚ 40˚
7
6

2 5
4 38˚ 38˚
3
2
1
36˚ 36˚

a b
0
0 2 4 6 8 10 12 14 16 18 20 20˚ 22˚ 24˚ 26˚ 28˚

Timespan years
Figure 3.17: Sites and data quality for GPS measurements in Greece and western Turkey Floyd et al. (2010). Left
Panel: The standard deviations of velocity components for the sites, plotted against the timespan of occupation
(time elapsed between first and last occupation of the site). Circles represent campaign sites, with the number
of occupations indicated by the shading; black squares represent continuous GPS (CGPS) sites. The ellipse
circumscribes campaign-mode sites that were disturbed by the 1995, Aigion, earthquake. (b) All GPS sites used
by Floyd et al. (2010), with shading as in (a); locations of sites whose velocities were determined by GPS (CGPS
sites as open squares).

Figure 3.18: Data from a continuous GPS site in western Crete. The panels show, from top to bottom,
displacements in the north, east, and up directions. The dotted lines joining large circles illustrate what
would have been the consequence of estimating the site velocity from two different pairs of campaign-mode
observations separated in time by about 1 year.
40 Kinematics and Dynamics of the Lithosphere 2020

interpret: they reflect the local upward motion of the rock to which the site is attached which, in
addition to tectonic motions, may be influenced by the much larger effects of isostatic response
to erosion, groundwater variations (anthropogenic and natural), or other sources of epistemic
uncertainty in this component of velocity.

3.4.4 Velocities and Strain Rates

Boring but important Strictly speaking, GPS measurements give a time series of displace-
ments of the sites with respect to a reference frame. Most of these measurements are, however,
reported as velocities under the assumption that velocity equals displacement divided by time.
Some CGPS measurements suggest that this assumption is reasonable (see, for example, Fig-
ure 3.18), but others, particularly near subduction zones, show significant variations in rates of
displacement over time. To date, such variations have not been reliably detected in the continen-
tal interiors – except in association with earthquakes. None of the data sets we discuss below is
known to be contaminated by transient variations in rate of displacement, so for convenience, we
shall refer to the average rate of displacement over time as the ‘velocity’.

The simplest interpretation of geodetically determined velocities in terms of strain rates

comes from dividing the region of interest into triangles, with each measurement site being a
vertex of one or more triangles (Figure 3.19).

Figure 3.19: A sketch of the relation between velocity differences across a triangle and the (assumed homogeneous)
velocity gradient.

The four horizontal components of the velocity gradient tensor may then be recovered from
the even-determined equations:
∂vx ∂vx ∂vy ∂vy
vx1 = ∂x ∆x1 + ∂y ∆y1 ; vy1 = ∂x ∆x1 + ∂y ∆y1
(3.10)
∂vx ∂vx ∂vy ∂vy
vx2 = ∂x ∆x2 + ∂y ∆y2 ; vy2 = ∂x ∆x2 + ∂y ∆y2

The even-determined problem for strain in a triangle (Equation 3.10) can be easily be gener-
alised to a least-squares problem for the strain of a region containing n ≥ 3 points. Strain-rate
fields such as those illustrated in the right-hand panel of Figures 3.20 and 3.21 are estimated by
moving a window over the observations, with some kind of weighting according to the distance
of the observation from the point of interest.

3.4.5 Advantages and Disadvantages of Different Methods of Measuring Strain

It is clear that the spatial resolution available from GPS data is far greater than that from seismic
strain rates, and we shall concentrate on GPS data when comparing theory with observation
Kinematics 41

20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚

20˚ 21˚ 22˚ 23˚ 24˚ 25˚ 26˚ 27˚

Geodetic Strains: 42˚ 42˚

40˚ Triangular Regions 40˚

10 ppm (dilatational)

40˚ 40˚
39˚ 39˚

38˚ 38˚ 38˚ 38˚

37˚ 37˚ 36˚ 36˚

200 nanostrain/yr
36˚ 36˚ 34˚ 34˚
20˚ 21˚ 22˚ 23˚ 24˚ 25˚ 26˚ 27˚ 20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚

Figure 3.20: (Left) Principal strain axes for triangular regions in Greece. Strains are determined from the
displacements of monuments whose relative positions were determined in 1892 by triangulation, and in 1992 by
GPS (Davies et al., 1997). Thick lines represent extension and thin lines show compression. Shaded polygons
cover groups of stations identified as co-moving. Active normal faults are identified by ticked lines. (Right)
Crosses show the orientations and magnitudes of the principal axes of strain rate obtained from about 250 GPS
velocities in the Aegean region (Floyd et al., 2010). Black bars correspond to extension, white bars to contraction.
Focal mechanisms of earthquakes since 1976 are from the GCMT catalogue (Ekström et al., 2012).

in the following chapters. It is important to remember, however, that there are two significant
reservations about the use of GPS: (a) the time scale of measurement is very short, and (b)
we have no guarantee that these measurements of points on the surface of the crust accurately
represent the deformation of the lithosphere as a whole. The observation that GPS measurements
agree with earthquakes and faulting gives us a measure of confidence.

60 0˚ ˚
˚ 60
˚
30˚ 120
45 60˚ 90˚ ˚
˚ 45

30 ˚
˚ 30

15 ˚
˚ 15

0˚
0˚
0˚

˚
30˚ 120

60˚ 90˚

1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 70 80 90 100

Strain rate 10−9/yr

Figure 3.21: Strain rates in Eurasia derived from GPS measurements (Kreemer et al., 2014).
42 Kinematics and Dynamics of the Lithosphere 2020

Earthquakes: Advantages
• Time scale: 100 years instrumental, 104 − 106 yr if earthquakes can be linked to faults.
• Scale of the observable: How long is the fault break of a magnitude 7 earthquake?
• Depth dimension: large earthquakes cut through the whole upper crust.

Earthquakes: Disadvantages
• Short time scale of observation, in comparison with recurrence time of earthquakes.
• Incomplete picture of deformation: cannot measure aseismic deformation – or rotations.

Space Geodesy: Advantages

• Greater precision and spatial resolution than seismic strain rates.
• Networks can be designed to address specific problems.
• Can be applied in places where InSAR lacks coherence.
• High precision and time resolution (days to seconds) with continuous GPS.

Space Geodesy: Disadvantages

• You have to wait between 3 years (CGPS) and 10 years (campaign) for reliable measurements.
• Short duration of strain: records a couple of decades long are rare.
• Displacement of points, rather than of large areas (contrast InSAR).
• Only a measure of superficial strain: How do we relate to this to the sub-surface?
Kinematics 43

3.5 Derivation of Kostrov’s Relation

Consider a simple example, in which a single fault cuts across an entire region of width w in
the y−direction, and length l in the x−direction (Figure 3.22); slip ∆u occurs on this fault.
Kostrov’s scheme calculates the average strain that this slip causes throughout the volume. In
the derivation that follows, ∆u can represent slip in a single earthquake or total displacement
on a fault over many earthquakes. To fix ideas, let ∆u represent slip in an earthquake: The
seismic moment resulting from the slip is
ηlh∆u
M0 = ηLh∆u = (3.11)
sin θ
where η is the shear modulus, and h is the thickness of the seismogenic layer.

Figure 3.22: Map view of a rectangular region that is cut by a vertical fault of strike θ (Molnar , 1983). The
dimensions of the region are l, w, and thickness, h. Note, however, that this sketch could equally well apply to a
cross-section of a region cut by a reverse fault or (if inverted) by a normal fault. The fault length, L = l/ sin θ,
and slip upon it, ∆u  l, w.

Recall the definition of the moment tensor

Mij = M0 (ûi n̂j + ûj n̂i ) (3.12)

where û and n̂ are unit vectors in the directions parallel to the slip and perpendicular to the
fault plane,
û = (sin θ, cos θ, 0) n̂ = (− cos θ, sin θ, 0). (3.13)
By inspection of Figure 3.22, the yy−component of strain due to the slip is
∆y ∆u cos θ
yy = = (3.14)
w w
sin θ cos θ
= M0 (3.15)
η(lwh)
sin θ cos θ
= M0 (3.16)
ηV
where V is the volume of the region.
44 Kinematics and Dynamics of the Lithosphere 2020

From Eqs. 3.12 and 3.13

Myy = 2 M0 sin θ cos θ, (3.17)
hence
Myy
yy = . (3.18)
2ηV
The other two components of the strain are slightly more complicated to calculate. The
mean displacement of the left and right sides of the body in Figure 3.22 are
∆u sin θ w3
right (3.19)
w
∆u sin θ w1
left ,
w
so the average change in x−direction length of body is

∆u sin θ(w3 − w1 )
∆l = (3.20)
w
and, because w1 − w3 = L cos θ,

∆l ∆uL sin θ cos θ sin θ cos θ

xx = =− = −M0 . (3.21)
l wl ηV
Similarly, the mean x−displacement of the top and bottom sides of the region is

M0 sin2 θ
, (3.22)
ηV
and the mean y−displacement of the right and left sides is

M0 cos2 θ
− , (3.23)
ηV
so

1 M0

xy = sin2 θ − cos2 θ (3.24)
2 ηV
Compare the different components of strain, above, with the elements of the moment tensor.
We can see that the increment of strain associated with the slip on this fault is
Mij
ij = . (3.25)
2ηV
If we were to sum the moment tensors in a region that took place in a time interval of length
t, we could estimate the strain rate as

Mij
E ij = . (3.26)
2ηV t

This is the form of Kostrov’s relation that we use here.

Measurement Problems

1. GPS measurements across a piece of continent yield the following values for horizontal com-
ponents of the velocity gradient tensor:
∂u1
= 3 × 10−7 yr−1
∂x1
∂u1
= 4 × 10−7 yr−1
∂x2
∂u2
= −1 × 10−7 yr−1
∂x1
∂u2
= −3 × 10−7 yr−1
∂x2
where the x1 direction is east, and x2 is north.
(i) Write down the horizontal components of the strain rate tensor.
(ii) Determine the orientation and magnitude of the greatest extensional strain rate.
(iii) The crustal thickness in the region is 40 km now. If these strain rates continue for another
10 million years, how thick will the crust be at the end of this time?

Remember for 2nd-year tectonics: The orientations of the principal axes can be
obtained as follows for the 2-D case is
q
E11 + E22 1 2
E1 = + (E11 − E22 )2 + 4E12
2 2
the angle θ1 between the E1 direction and the 1-axis is given by:
E1 − E11
tan θ1 =
E12
where cos θ1 = n1 and sin θ1 = n2

2. The total moment released in earthquakes in California since the 1906, Mw 7.9, San Francisco
earthquake has been 8×1020 N m. Is the rate of moment release consistent with the relative
plate motion across the region?
Treat the zone of active deformation in California as a rectangle, 1000 km long in the NW-
SE direction, and 400 km wide. You may assume that all earthquakes in the region occur by
right-lateral strike slip on vertical planes striking 135◦ . The relative motion of the Pacific plate
with respect to the North American plate in this region is 35 mm/yr in the direction 315◦ .

y, 45◦
6
n̂
6
û  - x, 135◦

Page 1 of 1
Rheology of the Lithosphere 45

4 Rheology of the Lithosphere

In discussing the rheology of the lithosphere we shall need to consider the physical properties of
three layers: the seismogenic upper crust, the lower crust, and the uppermost mantle.
• We shall use the term ‘strength’ to refer to the force per unit length required to make
the lithosphere, or a layer within it, deform at the strain rate(s) of interest. The strength
is the integral of deviatoric stress with respect to depth, and so has the units of newtons
per metre.
• It will also be convenient to refer to the average stress, which is the strength divided by
the thickness of the layer of interest.
• We shall use the term ‘effective viscosity’ to mean the vertical average of the stress
required to deform the lithosphere, or a layer within it, divided by strain rate.

4.1 Strength Profile through the Lithosphere

4.1.1 Contribution of the Mantle Portion of the Lithosphere
We may split the discussion of strength of the mantle portion of the lithosphere into two parts:
(i) strength arising from high-temperature creep, which is strongly temperature-dependent, and
(ii) strength arising from low-temperature plasticity, which is strongly dependent on deviatoric
stress and only weakly dependent on temperature.
Many laboratory determinations of the high-temperature creep of minerals, mineral ag-
gregates, and rocks take the form
 
n −Qc
Ec = A τ exp , (4.1)
RT
where Ec is strain rate, A is a constant, τ is differential stress, Qc is the activation energy, R
is the gas constant, and T is absolute temperature (e.g. Brace and Kohlstedt, 1980; Hirth and
Kohlstedt, 2003). (There is also a dependence on pressure, which we shall neglect.) We may
rewrite Eq. (4.1) as  
Qc
ηeff = η exp , (4.2)
nRT
where ηeff is the effective viscosity and η is a reference viscosity. Consider the change in viscosity
that occurs as a result of an increase in temperature from T0 to T0 + ∆T , where ∆T is small in
comparison with T0  
η1 −Qc ∆T
∼ exp . (4.3)
η0 nRT02
For olivine, Qc ∼ 530 kJ/mole, n = 3.5 (e.g. Hirth and Kohlstedt, 2003) and with T0 ∼ 900 −
1200 K, appropriate to the uppermost mantle, the strength decreases approximately five-fold
for each 100◦ C increase in temperature. Some crustal minerals, (e.g. pyroxenes and feldspars)
have activation energies comparable to, or higher than olivine; others, such as quartz, have
lower activation energies, in the range 200-400 kJ/mole, and are correspondingly less sensitive
to temperature.
To evaluate the strength of a layer characterised by the flow law of Eq. (4.1), we rearrange
that equation as
 1/n  
Ec Qc
τ= exp , (4.4)
A nRT
so that the strength of a layer described by such a constitutive relation is
 1/n Z T2  
Ec Qc
Σ= exp dT, (4.5)
A T1 nRT
46 Kinematics and Dynamics of the Lithosphere 2020

where T1 and T2 are the temperatures at top and bottom of the layer. In general, this expression
is messy to evaluate but if Q/nRT  1, which is the case for most rocks, and if the strength of
the bottom of the layer is much less than at the top (see Eq. (4.3) above), then the expression
reduces to
 
nRT12
Σ ∼ τ1 , (4.6)
γQc

where γ is the temperature gradient through the layer, and τ1 is the shear stress at the top of
the layer.
Laboratory measurements on olivine at low temperature and high differential stress, show a
flow law of the form (Mei et al., 2010):

  r 
2 −Qp τ
Ep = Ap τ exp 1− (4.7)
RT τp

with Ap = 1.4 × 10−7 s−1 MPa−2 , Qp = 320 kJ/mole and τp , = 5.9 GPa. This constitutive
relation is much less sensitive to temperature than the high-temperature flow laws
p (Eq. (4.1))
not only because of the lower activation energy but also because the term in τ /τp further
reduces temperature sensitivity. T (K)
We show the dependence of the 1600 1400 1200 1000 800
22
stress difference, τ , on tempera-
ture (Figure 4.1) for strain rates of city
re plasti
10−14 , 10−15 and 10−16 s−1 . mperatu
20 Low−te

Below a transition temperature

s −1

of about 1000 K, the stress differ-

4
−1
10

s −1

ences depend weakly upon tempera- 18

6
−1
Loge τ

ture, and are in the range of 0.1 to

10
p
ee
cr

1 GPa.
re
tu
ra

16
pe
em
−t
gh
Hi

4.1.2 Contribution of the Lower 14

Crust to Lithospheric Strength

Laboratory flow laws for quartz and 12

1.0 1.2 1.4 1.6 1.8 2.0
feldspar give stresses that vary by
1600/ T (K)
orders of magnitude for the same
conditions of temperature and strain Figure 4.1: Stress difference, τ , required to deform polycrys-
rate but suggest that the effective talline olivine at a range of temperatures and strain rates.
viscosity of the middle crust should The flow law for high-temperature creep is Eq. (4.1). Low-
temperature plasticity is described by Eq. (4.7).
be no higher than about 1020 Pa s.
This suggestion is supported by the
geodetic evidence, also summarised
by Bürgmann and Dresen (2008), from studies of post-seismic creep, for viscosities in the range
1018 –1020 Pa s for the mid-to-lower crust (see also Hilley et al., 2009; Wen et al., 2012; Yamasaki
and Houseman, 2012).
With a viscosity of 1020 Pa s, stresses required to deform the lower crust at a strain rate of
10−15 s−1 are of no higher than 1 MPa, much lower than the stresses required to deform the
mantle portion of the lithosphere.
Rheology of the Lithosphere 47

4.1.3 Contribution of Upper crust to Lithospheric Strength

Finite deformation of the upper crust involves frictional slip on faults which is usually, but not
always, seismic. Frictional resistance in rocks is usually described by ‘Byerlee’s Law’:

τ = 0.85 T n 0 < T n < 200MPa

τ = 60(±10) MPa + 0.6 T n T n > 200MPa (4.8)
T n = Tn − Pp = Tn (1 − λ) . (4.9)

where τ and T n are the shearing and effective normal stresses at which friction is overcome on
faults, Tn is the total normal stress and Pp is the pressure in the pore fluid, often expressed as
a fraction, λ, of Tn . (By convention, compressional stress is usually treated as positive in the
discussion of friction laws.) There are two major sets of uncertainty associated with Byerlee’s
law. One concerns the coefficient of friction, which is pretty much the same for all rocks, except
clays, for which the coefficient is low (∼ 0.2); clays form an important component of fault gouge.
Secondly the effective normal stress, T n is the total normal stress minus the pore pressure –
which is often unknown, and may be close to or exceed the normal stress.
One way to assess the stress in the upper crust is to measure the stress drops in earthquakes;
this gives estimates that concentrate between 1 and 10 MPa (Figure 4.2). Another is to estimate
the strain released by earthquakes from ratios of slip to rupture length; observations summarised
in Figure 4.2 show that these ratios vary between 10−4 and 10−5 . For a shear modulus of
3 × 1010 Pa, a reasonable value for the crust, this range corresponds to stresses of 3 to 0.3 MPa.

Figure 4.2: (Left:) Histogram of stress drop estimates for earthquakes observed by Allmann and Shearer
(2009). (Right) Ratios of slip to rupture length in continental earthquakes (Wells and Coppersmith,
1994).

4.2 Vertically Averaged Rheology of the Lithosphere

The discussion above shows that there is enormous uncertainty in the stresses acting at differ-
ent depths within the continental lithosphere. Fortunately, a simplification allows us to reduce
all this uncertainty to a rheological model that has only two parameters. Let us assume that
deformation of the lithosphere is governed by a non-linear rheological law in which strain-rate
varies as the nth power of deviatoric stress.

1
τ ij = BE n −1 Eij (4.10)
48 Kinematics and Dynamics of the Lithosphere 2020

where τ ij is the depth-averaged ij th component of the the deviatoric stress, Eij is the ij th
component of the strain-rate,  
1 ∂ui ∂uj
Eij = + (4.11)
2 ∂xj ∂xi
and 1
E = (Ekl Ekl ) 2 (4.12)
is the second invariant of the strain-rate tensor.
Sonder and England (1986) showed Equation 4.10 to be a close approximation to the vertical
average of the rheology of lithosphere failing by slip on faults in its upper crust and by failure
in the lower lithosphere, according to the strain rate and temperature there (Figure 4.3). The
success of this approximation derives from the ease with which one can obtain a straight line
by plotting the logarithm of one quantity against the logarithm of another, not from any single
fundamental physical process.
The exponent n in Eq. 4.10 does not
represent any individual deformation mech-
anism. Rather, it represents the ver-
tical averaging of deformation involving
slip on faults or low-temperature plastic-
ity (for which stress is essentially inde-
pendent of strain rate, and therefore n is
essentially infinite), with deformation by
creep (commonly with n being approxi-
mately 3).

We may use Eq. (4.10) to define, by

analogy with the definition of Newtonian
viscosity, an effective viscosity
1 1
ηeff = BE n −1 . (4.13)
2
Note that the effective viscosity increases as
the strain rate decreases. Continental litho-
sphere of homogeneous physical properties
(i.e. constant B) may nevertheless have
spatially varying effective viscosity; regions
that strain slowly will be stronger (have
higher viscosity) than regions that strain
more rapidly.

Figure 4.3: Plots of vertically-integrated strength (FL ) against

strain rate () for several combinations of the ratio of pore pres-
sure to lithostatic pressure in the crust (λ), depth to the crustal
brittle-ductile transition (zBD ), and Moho temperature (TM ).
The slope of each line corresponds to n, the effective stress ex-
ponent, and is indicated to the side of each line. The sub-Moho
thermal gradient is 10◦ C km−1 .
Rheology of the Lithosphere 49

4.3 Accommodation of Strain by the Upper Crust

In the following Chapter we develop the theory for deformation of a fluid lithosphere. An
obvious problem is that all the measurements we described in the previous chapter are made on
the upper crust, which deforms primarily by faulting. How do we compare observations with
theory?
Figure 4.4 suggests a way to think about the possible motions within a system consisting
of crustal blocks, strength B, faults strength F , and the ductile lower lithosphere, strength D.
(In each case “strength” is taken to mean the deviatoric stress required to make the part of the
system at a geologically interesting rate). More-or-less by definition, blocks are stronger than
the faults that separate them (B  F ), so there are three relevant cases:

• B  F  D: The strength of the ductile lower lithosphere is less than the strength of the
faults: this is the tectonics of faults and blocks – plate tectonics.

• B  F < D: The ductile lower lithosphere is stronger than the faults: the blocks of the
upper crust move in response to tractions applied to them by the lower lithosphere (lower
panel of Figure 4.4).

• D > B  F : The ductile lower lithosphere is stronger than the blocks, which still follow
the fluid, but may break up in response to forces applied by the fluid.

The discussion of this chapter suggests that the second or third of these conditions applies to
the deformation of the continents. In the following Chapters we develop the governing equations
for a fluid lithosphere, then apply them to the principal regions of active continental deformation.

Figure 4.4: Interseismic and long-term kinematics of crustal blocks within a zone of simple shear, after
Bourne et al. (1998). (Top panel) Between earthquakes, the faults remain locked and the crustal blocks
accumulate elastic strain through the action of basal stresses applied by the viscous layer beneath. The
elastic strain is greatly exaggerated here; shear strains in the interseismic period are usually of order
10−5 ). (Lower panel) The finite strain over many seismic cycles consists of ductile shear in the lower
lithosphere, and block-like motion of the upper crust. The long-term velocity of each block is the same
as the average velocity of the fluid beneath it. The transition between block motion at the surface and
distributed motion at depth takes place over a small depth interval, shown in lighter shading.
Thin-Sheet Equations 51

5 Mechanics of Continental Deformation

5.1 Vertical Force Balance
The stress balance equation with no acceleration term is:

0 = ∇ · T + ρg
∂σji
= −ρgi (5.1)
∂xj

Here the z−direction is vertically upwards. We proceed, as usual, by seeing what we can learn
from the three components of this equation, starting with the vertical.

5.1.1 Isostasy
The vertical component of the stress balance equation is:
∂σxz ∂σyz ∂σzz
+ + = ρg (5.2)
∂x ∂y ∂z
(Remember g = (0, 0, −g)). We know that, to a good approximation, over length scales compa-
rable with the thickness of the lithosphere (100s of kilometres) all lithospheric columns are close
to ‘local’ isostatic balance: i.e. the weight of any lithospheric column is supported by normal
tractions on its base, not by shear tractions on its edges:

σxz = σzx = σyz = σzy ' 0

Furthermore, we should expect horizontal gradients of these tractions to be much less than
vertical gradients of σzz . So equation (5.2) may be approximated by:
∂σzz
= ρg (5.3)
∂z

whence: Z z
σzz (z) = g ρ(z 0 ) dz 0 (5.4)
0
where ρ(z 0 ) is the density at z 0 , and σzz (0) is zero on the land surface, z 0 = 0. For example, in
the case of constant density σzz = −gρd at a depth d (z = −d).

5.1.2 Gravitational Potential Energy of the Lithosphere

The vertical traction on horizontal planes, σzz , obviously bears a relation to the forces required
to hold up the lithospheric column. If the density structure of the column changes then work is
done against, or by, gravity. In this section we shall see that the gravitational potential energy
per unit area of a lithospheric column is directly related to the vertically averaged value of σzz
in the column, σ zz . (From now on, the overbar denotes a vertical average).
Z Z z
g 0
σ zz = ρ(z 0 )dz 0 dz (5.5)
L −L 0

where L is the thickness of the lithosphere.

The difference, ∆σ zz , in σ zz between any two lithospheric columns is
Z Z z
g 0
∆σ zz = ∆ρ(z 0 )dz 0 dz (5.6)
L −L 0
52 Kinematics and Dynamics of the Lithosphere 2020

where ∆ρ(z 0 ) is the difference in density between the two columns at the depth z 0 .
In general ∆σ zz is not simply related to topography. It is, however, directly related to
differences, ∆Γ, in gravitational potential energy between columns. If the reference level for the
gravitational potential energy is taken to be z = 0, then the difference in gravitational potential
energy between any two columns of thickness L is:
Z 0
∆Γ = g z∆ρ(z)dz (5.7)
−L
Z 0 Z 0 Z z
= gL ∆ρ(z)dz − g ∆ρ(z 0 )dz 0 dz, (5.8)
−L −L 0

where we have integrated by parts. The first term on the right hand side is the difference in
weight per unit area between the two columns, multiplied by their length, L; this difference
is zero, because all isostatically balanced columns have the same weight. The second term is
L∆σ zz (Equation 5.6). Thus, equation (5.8) yields:

∆Γ = −L∆σ zz (5.9)

As an example, consider a pair of columns in Airy isostatic compensation. One column

contains crust of thickness, S0 , and density ρc , overlying mantle of density ρm . The other
column contains mantle alone.

Figure 5.1: Sketch of the buoyancy force arising from the difference in density structure between a crustal
column, thickness S0 and density ρc , and the column of mantle, density ρm .

By inspection of the figure:

gρc hS0
∆σ zz = − (5.10)
2L
where h is the difference in surface height between the two columns. But, by isostasy:

h = S0 (1 − ρc /ρm ) (5.11)
so
gρc (1 − ρc /ρm )S02
∆σ zz = − . (5.12)
2L
For two columns with crustal thicknesses S0 and S

gρc (1 − ρc /ρm )(S 2 − S02 )

∆Γ = . (5.13)
2
Thin-Sheet Equations 53

We can use the standard substitution for surface height in Airy isostasy to re-arrange
Eq. (5.13) as   
(S0 + S) ρm h
∆Γ = gρc h = gρc h S0 + , (5.14)
2 ρm − ρc 2
where h is surface height.
Figure 5.1 shows only one of many imaginable modes of compensation of surface height
variations. For example, the expression for difference in gravitational potential energy in the
case of Pratt compensation (density contrast spread throughout the lithosphere) is
gρ1 (L + h)2 gρ0 L2
∆Γ = −
2 2
gρ0 (L + h)L gρ0 L2
= −
2 2
gρ0 Lh
= , (5.15)
2
where ρ1 , ρ0 are the average densities of the lithospheric columns.
Table 2: Differences in gravitational potential energy of Airy-isostatically-compensated columns. Refer-
ence column is taken to have S0 = 30 km; ρc = 2800 kg m−3 , ρm = 3200 kg m−3 .

S (km) 5 10 20 30 40 50 60 70 80
h = (S − S0 ) (1 − ρc /ρm ) (km) -3 -2.5 -1.25 0 1.25 2.5 3.75 5 6.25

∆Γ (Airy TN m−1 ) -1.5 -1.4 -0.9 0 1.2 2.7 4.3 6.9 9.4

∆Γ (Pratt TN m−1 ) 0

5.2 1D Thin-Sheet Equation

Consider a long thin mountain belt resembling, for example, the Apennines of peninsular Italy
(Figure 5.2). For this configuration, we may neglect variations along strike, so let us align one
coordinate axis with that direction (say the y-direction), as shown below. We wish to determine
the state of stress in such a section of lithosphere.
In the previous section, we solved the vertical stress balance equation. Here, we are neglecting
variations in the y-direction, so it only remains to solve the x-equation:
∂σxx ∂σyx ∂σzx
+ + =0 (5.16)
∂x ∂y ∂z
We have chosen our coordinate system so that derivatives in the y-direction are negligible, so
we can drop the second term on the LHS.

∂σxx ∂σzx
+ =0 (5.17)
∂x ∂z
The second term in equation (5.17) represents vertical gradients of shear tractions in the
lithosphere. Recalling our discussion of lithospheric rheology, it is clearly impossible from first
principles to specify what those gradients might be. We can, however, write a simple expression
for the vertical average of those gradients: Let us integrate each term in equation (5.17) over
the thickness of the lithosphere, L, and divide by L to give that average:
Z Z
1 0 ∂σxx 1 0 ∂σzx
dz + dz = 0 (5.18)
L −L ∂x L −L ∂z
54 Kinematics and Dynamics of the Lithosphere 2020

˚
˚ ˚ ˚

44

44

44
10 10 10

˚
12

12

12
4 mm/yr 50 nanostrain/yr
˚

˚
42

42

42
˚

˚
D

14

14

14
˚ ˚ ˚
10 10 10
C

˚ ˚ ˚
44 44 44

˚
16

16

16
˚

˚
12

12

12
B

˚ ˚ ˚
40 40 40
A

˚ ˚ ˚
42 42 42
a b c
˚

˚
˚ ˚ ˚
14

14

14
40 40 40
˚

˚
16

16

16
Figure 5.2: Top: Continuous GPS observations in Italy and strain rates derived from them. Boxes
outline the region in which the orientations of both the velocity vectors and the principal extensional
strain rate axes remain constant, and which we investigate here; lines A–D show the profiles of Fig. 5.4.
(a) Arrows show velocities of CGPS sites in a frame of reference defined by sites on the west coast of
Italy. (b) Axes of principal horizontal strain rate, derived from the CGPS data. (c) Focal mechanisms
of earthquakes since 1976. Bottom: Definition sketch for a 2-D mountain belt (after Dalmayrac and
Molnar (1981)). Applying this sketch to the Apennines (panels a) and b)), the x−direction points across
the page – perpendicular to the strike of the mountain chain.

which gives:
∂σ xx σzx (0) − σzx (−L)
+ =0 (5.19)
∂x L
The first term on the left-hand side represents horizontal gradients of the vertically averaged
σxx . The second term on the LHS represents the difference between shear tractions on top and
bottom of the lithosphere. We may safely ignore shear tractions on the top surface (e.g. wind),
and here we neglect shear forces on the base of the lithosphere.

∂σ xx
'0 (5.20)
∂x
Thus we obtain the apparently uninteresting result that the vertically averaged horizontal
traction on vertical planes, σ xx , in the mountain belt is approximately constant. Recall, however,
that by the definition of deviatoric stress:

σzz = τzz + σm (5.21)

Thin-Sheet Equations 55

where σm is the mean stress

1
σm = (σxx + σyy + σzz ) (5.22)
3
Similarly,
σxx = τxx + σm (5.23)
Eliminating σm we find:
σxx = τxx − τzz + σzz (5.24)
Now recall, from the definition of deviatoric stress:

τxx + τyy + τzz = 0 (5.25)

And, with τyy = 0, τxx = −τzz , so:

σxx = 2τxx + σzz (5.26)
Taking vertical averages,
σ xx = 2τ xx + σ zz (5.27)
Substituting into equation (5.20) yields:

∂τ xx ∂σ zz 1 ∂Γ
2 =− = (5.28)
∂x ∂x L ∂x

Γ(x)
2τ xx = + constant (5.29)
L

In the sketch at the head of this section, we assumed Airy isostatic compensation, whence
from Eq. (5.12):

gρc (1 − ρc /ρm ) S 2 − S02
∆σ zz = − (5.30)
2L
So:
∂τ xx gρc (1 − ρc /ρm ) ∂S 2
2 = (5.31)
∂x 2L ∂x
or:
gρc (1 − ρc /ρm )S 2
2τ xx = + constant (5.32)
2L
This result shows that for some value of crustal thickness the horizontal deviatoric stress τ xx
will be zero. In parts of the mountain belt where crustal thicknesses are lower than this value
there will be compressional deviatoric stress. In any part where the crustal thickness is greater,
there will be extensional deviatoric stress (see Figure 5.3).
56 Kinematics and Dynamics of the Lithosphere 2020

Surface Height (m)

0 200 400 600 800 1000

−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
5 150 150
50 nanostrain/yr
1.50
4
100 100
3 1.25

Γ(x) (1012 N m−1)

2
50 50
ux (mm/yr)

1.00
1

y (km)
0 0 0
0.75
−1

0.50 −50 −50

−2

−3
0.25 −100 −100
−4
a b
−5 Γ0 0.00 −150 −150
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
x (km) x (km)
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
5 150 150
50 nanostrain/yr
1.50
4
100 100
3 1.25
Γ(x) (1012 N m−1)

2
50 50
ux (mm/yr)

1.00
1

y (km)
0 0 0
0.75
−1
−50 −50
−2 Γ0 0.50

−3
0.25 −100 −100
−4
c d
−5 0.00 −150 −150
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
x (km) x (km)
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
0 150 150
50 nanostrain/yr
−1
Γ0 1.50

100 100
−2 1.25
Γ(x) (1012 N m−1)

−3
50 50
ux (mm/yr)

1.00
−4
y (km)

−5 0 0
0.75
−6

0.50 −50 −50

−7

−8
0.25 −100 −100
−9
e f
−10 0.00 −150 −150
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
x (km) x (km)

Figure 5.3: Sketch illustrating solutions to the governing equation for a two-dimensional mountain belt
(Eq. (5.29)), following Dalmayrac and Molnar (1981). The surface heights are related to the distribution
of potential energy by Eq. (5.13), and are shown by the colour shading (scale at top right). Solid curves
in (a), (c), and (e) show the distribution of ux given by Eq. (5.35). Solutions are illustrated for three
different values of Γ0 , indicated by the horizontal bars in (a), (c), and (e). The right-hand set of panels
illustrate in map view the distributions of strain rate, Exx , corresponding to these solutions, superimposed
on the topography.
Thin-Sheet Equations 57

5.3 A Simple Solution, and Application to the Apennines of Italy

Let us first consider a Newtonian lithosphere of constant viscosity (τij = 2ηEij ). Eq. (5.29)
becomes
Γ(x)
2 (2ηExx ) = + constant
L
Γ(x) − Γ0
4ηExx =
L
∂ux 1
= (Γ(x) − Γ0 ) (5.33)
∂x 4ηL
The physical significance of the constant, Γ0 , is that any column of lithosphere having GPE
equal to Γ0 will have zero strain rate.

We refer to the quantity

∆Γ(x) = Γ(x) − Γ0 (5.34)
as the GPE excess. Integrating, we obtain
Z x 
0 0
ux = U0 + ∆Γ(x )dx /4ηL. (5.35)
x0

Here the lower limit of the range of integration, x0 , corresponds to the point at which ux = U0 .
Eq. (5.35) states that the velocity difference between one point and another on a profile across
the belt is directly proportional to the integral along that profile of the potential energy excess,
∆Γ(x). A similar solution holds for non-Newtonian lithosphere, but it is tedious to derive, and
we do not give it here. Figure 5.4 shows it to be comparable with the Newtonian solution.

Figure 5.4 shows the comparison between observed and theoretical distributions of velocity
across the Apennines of Italy.

• We can estimate the viscosity required to account for the observed velocity profiles from
Eq. (5.35).

• The integral with respect to distance of the GPE on the NE side of each profile is
40 to 60×1015 N (right-hand axes of Figure 5.4).

• The velocity difference is about 3 mm/yr.

From Eq. (5.35) R

∆Γ dx
η=
4L (ux − U0 )
5 × 1016 N
η∼ ∼ 1021 Pa s.
(4 × 105 m × 10−10 m/s)
58 Kinematics and Dynamics of the Lithosphere 2020

5 5
4 a 4 c
vx mm/yr

vx mm/yr
3 3
2 2
1 Profile A 1 Profile B
0 0
−1 −1

1000 100 1000 100

b 90
80
d 90
80
750 750

ΓnI (1015 N)

ΓnI (1015 N)
70 70
Topo (m)

Topo (m)
60 60
500 50 500 50
40 40
30 30
250 250
20 20
10 10
0 0 0 0
0 100 200 0 100 200
Distance (km) Distance (km)

5 5
4 e 4 g
vx mm/yr

vx mm/yr

3 3
2 2
1 Profile C 1 Profile D
0 0
−1 −1

1000 100 1000 100

f 90
80
h 90
80
750 750
ΓnI (1015 N)

ΓnI (1015 N)
70 70
Topo (m)

Topo (m)

60 60
500 50 500 50
40 40
30 30
250 250
20 20
10 10
0 0 0 0
0 100 200 0 100 200
Distance (km) Distance (km)

Figure 5.4: Smoothed surface height, integrated potential energy, and velocity profiles for Profiles A to D
(Fig. 5.2). (b), (d), (f), (h): Shading shows the topography, averaged over 50 km either side of the
profile. Double line shows the integral with respect to distance of the GPE (Eq. (5.13)); solid line shows
the equivalent quantity for the non-Newtonian case with n = 3. (a), (c), (e), (g): dots with error bars
show component of velocity parallel to the profile from sites within 50 km of the profile; uncertainties
are shown at the 2-standard-deviation level. Double line is fit to data with n = 1 (Eq. (5.35)); solid line
assumes n = 3.
Thin-Sheet Equations 59

5.4 2D Thin-Sheet Equations

In the three-dimensional world, we cannot neglect terms such as
∂σyx
∂y
because they represent horizontal gradients in stress, and so in deformation. By the arguments
of the previous section, however, we can still neglect terms such as
∂σzx ∂σzy
and
∂z ∂z
and the full equations for the stress in the sheet become:
∂τ xx ∂τ yy ∂τ xy 1 ∂Γ
2 + + =
∂x ∂x ∂y L ∂x
∂τ yy ∂τ xx ∂τ xy 1 ∂Γ
2 + + = (5.36)
∂y ∂y ∂x L ∂y

In any active region of continental deformation we can always chose a scale velocity, (e.g. the
magnitude of the relative velocity of the plates bounding the region); let us call this velocity U0 .
The combination of U0 with L, the thickness of the lithosphere, allows one to non-dimensionalize
distance and time, and thus the governing equations for the deformation of the sheet, which is
the model for the deforming lithosphere.

(x, S) u Eij
(x0 , S 0 ) = ; u0 = ; 0
Eij = ; (5.37)
L U0 (U0 /L)
Eij Γ
τij0 = 1/n
Γ0 = . (5.38)
B(U0 /L) Γ0
and dropping primes, we obtain dimensionless equations looking almost exactly like the dimen-
sional ones (equations 5.36):
∂τ xx ∂τ yy ∂τ xy ∂Γ
2 + + = Ar (5.39)
∂x ∂x ∂y ∂x
∂τ yy ∂τ xx ∂τ xy ∂Γ
2 + + = Ar (5.40)
∂y ∂y ∂x ∂y
where all τ s, xs, ys, and Γs are now dimensionless (Equation 5.38).
These equations (or their dimensional equivalents) express the fact that gradients in the
stress required to deform the lithosphere are balanced by stresses due to contrasts in gravita-
tional potential energy. The relative magnitudes of these stresses are given by a dimensionless
ratio, the Argand number, Ar:

Γ0 /L
Ar = . (5.41)
B(U0 /L)1/n
The choice of Γ0 depends upon the problem of interest; if we continue with the assumption that
GPE is described by Airy isostasy, a natural scaling is Γ0 = gρc (1 − ρc /ρm )L2 (the squared
crustal thicknesses in Eq. (5.13) being non-dimensionalised by L2 ) and

gρc (1 − ρc /ρm )L
Ar = (5.42)
B(U0 /L)1/n

where all quantities have been defined previously.

60 Kinematics and Dynamics of the Lithosphere 2020

5.5 Length scales for Continental Deformation

Although solutions to the governing equations (5.39 & 5.40) generally need to be obtained
numerically, we can find a couple of simple cases that cast light on the problem. First, we note
that for Newtonian fluids of constant viscosity (n = 1), the equations reduce to:
∂Exx ∂Eyy ∂Exy ∂Γ
2 + + = Ar (5.43)
∂x ∂x ∂y ∂x
∂Eyy ∂Exx ∂Exy ∂Γ
2 + + = Ar (5.44)
∂y ∂y ∂x ∂y
Differentiating the x−equation by x, and the y−equation by y and adding, we find
 2 
∂ ∂2
2 + (Exx + Eyy ) = −2∇2 (Ezz ) = Ar∇2 Γ. (5.45)
∂x2 ∂y 2

It is convenient, for later use, to consider a situation in which variation of velocity in one direction
(let’s say the x-direction) may be neglected. In this case, Eq. (5.44) reduces to

∂ 2 uy ∂ 2 uy ∂Γ
4 2
+ 2
= Ar (5.46)
∂y ∂x ∂y

Eq. (5.47) shows, for a constant compressional or extensional boundary condition (fixed Ezz )
applied over a portion of the edge of the continent, we should expect – from our knowledge of
solutions to Laplace’s equation – that its influence would extend over a region that is about as
broad, across strike, as the boundary is along strike (see Figure 5.5).

5.6 Approximate solutions for a Power-Law Fluid

Let us set the RHS of Equation 5.45 to zero (equivalent to
absence of crustal thickness variations):
∇2 (Ezz ) = 0. (5.47)
We do not expect the continental lithosphere to behave
as a Newtonian fluid, however. For n > 1, the problem is
slightly more complicated, but numerical solutions to the full
equations (illustrated in Figure 5.6) suggest a simplification.
Velocity vectors in the interior of the sheet are almost exactly
parallel to the orientation of the velocity applied to its edge,
which suggests that we can simplify the full equations by
neglecting the component of velocity that is perpendicular
Figure 5.5: Solution to Equation 5.47
to the imposed boundary condition.
with Ezz set to 1 on the grey part of
For example, in the case of a compressional or extensional the boundary, and zero on the rest of
boundary condition (illustrated in Figure 5.6a) the boundary.
1 ∂uy
τxx ∼ 0; τzz = −τyy ; Exy = (5.48)
2 ∂x
After some tedious the algebra (see England et al. (1985), if you wish, but you’ll probably prefer
to note the similarity to Eq. (5.46)) we obtain an equation for the velocity uy associated with a
stress applied perpendicular to the boundary
4 ∂ 2 uy ∂ 2 uy
+ = 0. (5.49)
n ∂y 2 ∂x2
Thin-Sheet Equations 61

Figure 5.6: Map views of velocity fields in thin viscous sheets with velocities imposed on their boundaries. Solid
lines indicate the boundaries on which the velocity is fixed to be zero. The lower boundary in each figure has a
constant velocity imposed over the sectors marked by arrows: in a) this velocity is perpendicular to the boundary,
in b), it is parallel to the boundary. Velocity vectors in the interiors of the sheets are aligned with the boundary
velocities.

In the case where the velocity condition is applied parallel to the boundary we find
∂ 2 ux 1 ∂ 2 ux
+ =0 (5.50)
∂x2 4n ∂y 2
Eqs. (5.49) and (5.50) are simply Laplace’s equations with one coordinate stretched relative
to the other, so we should expect the solutions to look like

U (y) = U0 exp (−y/λ)

where U is whichever component of velocity we are interested in, and λ is a scale length that
depends on which equation we are solving. If, instead of applying a constant velocity over
a distance D, we had applied a sinusoidal velocity with wavelength δ, then the solution to
Eq. (5.49) would have given
δ
λ= √ ,
π n
and Eq. (5.50) would have given
δ
λ= √ .
4π n
Obviously, the full solutions for constant velocity applied over a segment of length D require
summing a series of harmonic solutions, but a good approximation can be obtained by letting
D ∼ δ/π, which is the same as letting the integral of the velocity over the half-wavelength be
equal to the integral over D. We should therefore expect that the characteristic across-strike
dimension, λ⊥ , of a zone of convergent or extensional deformation, of along-strike width D, in
lithosphere whose vertically averaged rheology is given by equation 4.10, is:

D
λ⊥ ' √ , (5.51)
n

while the characteristic across-strike dimension, λk , of a zone of transcurrent deformation, of

along-strike width D is:

D
λk ' √ . (5.52)
4 n

The solution to Eq. (5.49), for n = 4 is shown in Figure 5.7a. The solution to Eq. (5.50), again
for n = 4 is shown in Figure 5.7b.
62 Kinematics and Dynamics of the Lithosphere 2020

a b

0.1
0.2

0.1

Figure 5.7: (a) Solution is to Equation 5.49, with n = 4, and with uy set to 1 on the grey part of
the boundary, and zero on the rest of the boundary. This is also the solution to Laplace’s equation
(Equation 5.47, Figure 5.5). (b) As (a), except that the solution is to Equation 5.50, with n = 4, with
the contours being contours of ux , the velocity perpendicular to the boundary.
 Thin-Sheet Problems

1. The figure below shows the results of a set of geodetic measurements across a plate boundary
zone between Plates A and B. Plate A is oceanic and Plate B is continental. Three faults, α, β,
and γ, whose locations are indicated by vertical lines, take up all the long-term deformation
of the upper crust within the plate boundary zone.
The horizontal velocities within the deforming zone are found to be approximately parallel to
the plate boundary, and their magnitudes, v, are shown by the thick line, whose form is given
by:

v = V0 exp (−x/λ) x>0

v = V0 x≤0

with λ = 30 km and V0 = 30 mm a−1 .

(a) Explain, briefly, the differences between the mechanical properties of oceanic and continen-
tal lithosphere, and how those differences might lead to an asymmetrical velocity distribution
of the form shown in the figure.
(b) Assume that the lithosphere of Plate B behaves as a power-law fluid in which strain rate,
E, is related to deviatoric stress, τ by:
1
τij = BE ( n −1) Eij
1
where E = (Eij Eij ) 2 and n = 3, estimate the length along strike (perpendicular to x-direction)
of the plate boundary zone.

2. Two sections of an isostatically balanced 2-dimensional mountain belt are in contact. One, hav-
ing crustal thickness 30 km and vertically averaged viscosity of 1022 Pa s, undergoes horizontal

Page 1 of 3
 compressional deformation at a strain rate of 10−15 s−1 . The crustal thickness of the second
section is 70 km and it is extending at 10−16 s−1 ; what is its vertically averaged viscosity?

3. To answer this question fully, you will need to make use of the following information: A moun-
tain range is approximately 500 km long in the N-S (y)-direction, and 100 km in the x-direction
(E-W). It is observed that the y-component of velocity, uy , is zero and the x-component, ux ,
increases by 10 mm/yr from one side of the range to the other (i.e. in the x-direction). You
may treat the mountains as having constant elevation, relative their surroundings, of 1000 m,
and you may assume that the thickness of the crust beneath them is 40 km. Density of
crust, ρc = 2800 kg m−3 ; density of mantle, ρm = 3300 kg m−3 ; shear modulus of crust,
µ = 3 × 1010 N m−2 .
a) The stress balance equation for creeping deformation is

∂σij
= −ρgi (1)
∂xj

where σij is the ij th component of the stress, ρ is density and gi is the ith component of
gravity. Adopt a coordinate system in which gravity acts in the −z-direction, and explain why
the z-component of the stress balance equation reduces to
∂σzz
= ρg
∂z
for a column of lithosphere in local isostatic balance.
b) Assume that differences in surface height are compensated by differences in crustal thickness
(Airy isostasy) and show that the difference in the average of σzz between two columns of
lithosphere may be written as

∆σ zz = −gρc (1 − ρc /ρm )(S12 − S22 )/2L

where S1 , S2 represent the crustal thicknesses of the two columns, and the overbar represents
the average through the lithosphere.
c) In the mountain range described here, the horizontal components of Equation (1) can be
simplified to
∂τxx gρc (1 − ρc /ρm ) ∂S 2
=
∂x 4L ∂x
Using this equation as your starting point, estimate the average viscosity of the lithosphere
of the region. You may assume that deformation arises entirely from the difference in poten-
tial energy between the mountains and their immediate surroundings; state clearly any other
assumptions that you make.
d) If all the strain in the mountains is released by earthquakes, what is the average rate of
seismic moment release? Approximately how many magnitude 6.5 earthquakes per century is
this rate equivalent to? [Recall: Units of seismic moment are N m.]

4. A rectangular region of continental deformation has east-west sides that are 1000 km long,
and north-south sides that are 500 km long. One thousand campaign-mode GPS sites are
distributed evenly over the region, and have been occupied once a year for the past decade.
The relative velocities of these stations show, to within measurement error, a homogenous
velocity-gradient field.

Page 2 of 3
  
0 0.3
× 10−8 /yr
0.1 3
where the 1-direction in east, and the 2-direction is north.

(a) The region is bounded, north and south, by rigid plates. The relative motion of these
plates is in the north-south direction. Interpret the observed velocity gradients in the light of
this statement. Calculate the magnitude of the relative velocity between the plates and, from
your knowledge of the precision of GPS measurements, indicate the uncertainty in that velocity.

(b) The present thickness of the crust in this region is 70 km. If the measured strain rate were
to continue unchanged, what would be the thickness of the crust after 10 Myr?

(c) Within the time interval AD 1900 to AD 2000, the following earthquakes took place

Number Scalar Moment

Nm
40 1019
8 1020
2 1021
1 1022
Each earthquake exhibited pure dip-slip normal faulting on planes striking East and dipping
at 45◦ . There has been no earthquake of magnitude greater than 6 since AD 2000.

Use Kostrov’s relation to estimate the rate at which strain in the region was released in earth-
quakes between AD 1900 and AD 2000. State, and justify, all your assumptions. Assume that
the shear modulus of rocks in the region is 3.3 × 1010 N m−2 .

(d) Is this strain rate is consistent with the strain rate measured by GPS? Provide quantitative
arguments to support your answer.

(e) This region has a constant surface height of 5000 m and is surrounded by continental
lithosphere whose surface is at sea level. Estimate the effective viscosity of the lithopshere of
the region.

Page 3 of 3
Analysis of velocity fields 63

6 Estimates of Lithospheric Viscosity

6.1 Gravitational Potential Energy and deformation in Iran
In the central part of Iran, between about 50◦ and 54◦ E, the velocities and principal horizontal
contractional strain rate axes are both aligned approximately north-south (Figure 6.1), which
allows us to use an approximate analytical solution to the thin-sheet equations, as we did for
the Apennines of Italy (Section 5.3).
0 200 400 600 800 1000 1200

X1 X2

Topography (m)
2000

40˚ 44˚ 48˚ 52˚ 56˚ 60˚

1000

44˚ 44˚

0
a
30mm/yr 0 200 400 600 800 1000 1200

Integrated GPE 1018 N

1.5

40˚ 40˚ 1.0

0.5
X2
0.0
36˚ 36˚

−0.5
b
0 200 400 600 800 1000 1200
30

32˚ 32˚ 25
X1 X2
vx mm/yr

20

15
28˚ 28˚
10
X1
5
c
0 200 400 600 800 1000 1200
24˚ 24˚
40˚ 44˚ 48˚ 52˚ 56˚ 60˚ Distance (km)

Figure 6.1: (Left Panel:) GPS velocities in Iran, relative to Eurasia. The box contains the sites whose
velocities are plotted in the right panel. (Right Panel:) (a) Shading shows the topography, averaged over
200 km either side of the profile X1 –X2 in left panel. Solid line shows the integral, from the origin to
the relevant distance, of the GPE, taking the GPE to be zero at the origin of the profile; this quantity is
calculated from the topography, assuming Airy isostasy (Eq. (5.13)). (b) The departure of the integral
of GPE (solid line in (a)) from a constant gradient (∆Γ, Eq. (6.5)). (c) Circles with error bars show
component of velocity parallel to the profile from sites within 200 km of the profile; uncertainties are
shown at the 2-standard-deviation level. Solid line shows the best fit to observed velocities (Eq. (6.2));
dashed line shows the linear part of this fit (first two terms in Eq. (6.2)).

We analyse the velocities within 200 km of profile X1 -X2 in Figure 6.1c. Neglecting the
component of velocity in the direction perpendicular to the profile, and integrating the thin-
sheet equation in the direction parallel to the profile allows us to use the same procedure as in
Section 5.3., except now we have to use velocity boundary conditions at both ends of the profile.
In the south, at X1 , the Arabian plate moves at about 23 mm/yr towards Eurasia (we’ll call
that rate U1 ), while at the north end the southern Caspian Sea moves at U2 ∼ 7 mm/yr with
respect to Eurasia. (Strictly speaking, that last clause should read “the component parallel to
X1 -X2 of the motion of the south Caspian Sea with respect to Eurasia is . . . ”.)
Starting with Eq. (5.33)
∂ux Γ(x)
= +A
∂x 4ηL
Z x
1
ux = Γ(x0 ) dx0 + A.x + B, (6.1)
4ηL 0
64 Kinematics and Dynamics of the Lithosphere 2020

where A and B are constants, to be determined. Call the length of the profile W ; the boundary
conditions give us

ux (0) = U1 −→ B = U1
Z W
1
ux (W ) = U2 = Γ(x0 )dx0 + AW + U1
4ηL 0
Z W
U2 − U1 1
A = − Γ(x0 ) dx0
W 4ηLW 0
Z x Z W 
(U2 − U1 ) x 1 0 0 x 0 0
ux = U1 + + Γ(x ) dx − Γ(x )dx
W 4ηL 0 W 0
Z x 
(U2 − U1 ) x 1

= U1 + + Γ(x0 ) − Γ0 dx0 , (6.2)
W 4ηL 0
Z W
1
where: Γ0 = Γ(x0 ) dx0 (6.3)
W 0
is the average GPE across the profile.
Equation (6.2) needs a bit more thought than the equivalent expression for the Apennines
(Eq. (5.33), Section 5.3).

• The first term on the RHS is OK: it’s the velocity at x = 0.

• The second term represents a constant velocity gradient, which is what would occur if
there were no potential energy in the problem at all. This corresponds to the state of
affairs illustrated in Figure 3.14a, in which all the relative motion between two plates is
taken up by a constant gradient of velocity.

• The way to think about the final term, in the square brackets, is to imagine what would
happen if the GPE across the region were constant (e.g. Γ(x) = Γ1 ). There’s no need to
specify the value of Γ1 : gradients of GPE would be zero, so they would not enter into the
solutions. The term in the square brackets represents the differences in velocity due to
differences between the GPE and the average GPE across the profile ( Γ0 , Eq. (6.3)).

• Note that the constraint on effective viscosity, or strength, of the lithosphere comes from
∆Γ, which is the integral of GPE contrasts over horizontal difference (Eq. (6.5)); in other
words, the constraint comes not only from the magnitude of the GPE contrast across the
central Iranian plateau, but also from the horizontal distance over which it is expressed.

Figures 6.1a shows the variation of topography,

Z x and of integrated GPE, along the profile X1 –


X2 ; Figure 6.1b shows the variation of the term Γ(x0 ) − Γ0 dx0 along the profile. Figure 6.1c
0
shows the variation of ux measured at GPS sites within 200 km of the profile.
We can estimate the viscosity of the Iranian lithosphere by eyeballing Figure 6.1. Call the
departure of the measured velocity from a straight line ∆U , then
∆Γ
∆U = (6.4)
4ηL
Z x 


where: ∆Γ = Γ(x0 ) − Γ0 dx0 (6.5)
0

The largest values of ∆U are co-located with the largest values of ∆Γ at about 400 km along
the profile, being respectively about -5 mm/yr (Figure 6.1c) and −2.5 × 1017 N (Figure 6.1b).
From Eq. (6.5)
∆Γ 2.5 × 1017
η∼ ∼ ∼ 5 × 1021 Pa s (6.6)
4L∆U 4 × 105 × 5/(3.16 × 1010 )
Analysis of velocity fields 65

The solid line in Figure 6.1 shows the least-squares fit of the velocities to Eq. (6.2), which has
an RMS misfit of 1.5 mm/yr. As you would hope, the values of the parameters to this fit agree
with the scaling analysis of the previous paragraph. They are: U1 , U2 = 23 and 7 mm/yr and
4ηeff L = 2 × 1027 N m−1 s which, for L ∼100 km, is equivalent to an effective viscosity for the
lithosphere of ∼ 5 × 1021 Pa s.

6.2 Places in which it is possible only a bound on viscosity

6.2.1 The Tien Shan

The distribution of GPS velocities observed across the Tien Shan is shown in Figure 6.2 (network
of about 400 sites, observed in campaign mode over about 16 years: uncertainties 1 to 3 mm/yr
Zubovich et al. (2010)). The gradients of velocity are remarkably constant with principal axes
perpendicular to the belt (Figure 6.2b), and velocities departing from a straight line by no more
than 1 to 2 mm/yr (Figure 6.3).

A note on epistemic uncertainty: This observation gives us an insight into the epistemic
uncertainties referred to in Section 3.4.3. It is extremely unlikely that these constant velocity
gradients are erroneous, so the misfits suggest that, in the case of this network, the velocities of
the GPS sites represent the velocities of the upper crust surrounding them to within one or two
millimetres per year.

70˚ 72˚ 80˚ 82˚ 70˚ 72˚ 80˚ 82˚

74˚ 76˚ 78˚ 74˚ 76˚ 78˚

C B A
46˚ 46˚ 46˚ 46˚
F D
20 mm/yr 50 nstrain/yr
E

44˚ 44˚ 44˚ 44˚

42˚ 42˚ 42˚ 42˚

40˚ 40˚ 40˚ 40˚

38˚ 38˚ 38˚ 38˚

km
a b 0 200 400

70˚ 72˚ 80˚ 82˚ 70˚ 72˚ 80˚ 82˚

74˚ 76˚ 78˚ 74˚ 76˚ 78˚

0 1000 1500 2000 2500 3000 3500 4000 5000

Figure 6.2: (a) Velocities of GPS sites relative to Eurasia (Zubovich et al., 2010); ellipses (mostly invisible)
show uncertainties at the 1-σ level. (b) Bars show orientation and relative magnitudes of the principal
horizontal axes of strain rate (red, contraction; white, barely visible, extension). Solid lines show locations
of profiles, chosen to lie roughly parallel to the principal contraction axes, that are displayed in Figure 6.3.
66 Kinematics and Dynamics of the Lithosphere 2020

Figure 6.3: Velocities along Profiles A–F (Figure 6.2). Lines show best-fitting straight lines through the
observations.

Figure 6.4: Profiles of GPE (Γ, Eq. 5.13) and ∆Γ (Eq. 6.5) across the Tien Shan, corresponding to the
profiles of Figure 6.3. Black lines show ∆Γ, grey shading shows the GPE derived from the topography.
Analysis of velocity fields 67

This is the same problem that we analysed for Iran, so we may use Equations 6.2 to 6.5. The
difference here is that the velocities don’t depart significantly from a straight line, so instead of
measuring the viscosity we can only place a lower bound on it.
The departures of the velocities along 6 profiles across the Tien Shan (Figure 6.2) are of the
order of one or two millimetres per year (Figure 6.3), while the magnitude of ∆Γ (Figure 6.4)
are about 1–2×1017 N. From Eq. (6.5)

∆Γ
η& . (6.7)
4 L ∆U
With ∆U ∼1–2 mm/yr (Figure 6.3), |∆Γ| ∼ 2 × 1017 N, and L ∼ 100 km, the effective viscosity
of the lithosphere of the Tien Shan must be greater than ∼ 1022 Pa s.

Exercise for the student: Check this bound independently by assuming that the force per unit
length required to deform the Tien Shan is equal to the difference in GPE between Tibet and
Central Asia.

6.2.2 The South Island of New Zealand

The convergence between the Pacific and Australian plates across the South Island of New
Zealand is highly oblique (Figure 6.5). As with the Tien Shan, we see no influence of the
topography (hence GPE) on strain rates. The velocities are aligned parallel to the Australia-
Pacific relative motions (see Figure 3.14a) and each component of velocity dies out exponentially
with distance from the plate boundary (Figure 6.5A-F).
So, again, we can place a bound on the viscosity. From Figure 6.5A-F, ∆U . 2 mm/yr
and from Figure 6.6 ∆Γ ∼ 5 × 1016 N. (I obtain this estimate by noting that the surface height
reaches a maximum of about 1 km, converting this to GPE using Eq. (5.13) then multiplying
by the width of the mountains, about 40 km).

Exercise for the student: Place abound on the viscosity of the lithosphere of New Zealand.
68 Kinematics and Dynamics of the Lithosphere 2020

168˚ 169˚ 170˚ 171˚ 172˚ 173˚ 174˚ 175˚

−40˚ −40˚

−41˚ −41˚

−42˚ −42˚

−43˚ −43˚

−44˚ −44˚
B

C
D

−45˚ E −45˚
F

168˚ 169˚ 170˚ 171˚ 172˚ 173˚ 174˚ 175˚

20 20 20
10 10 10

5 5 5

2 2 2

1 1 1
0.5 0.5 0.5
A B C
−100 −50 0 50 100 150 −100 −50 0 50 100 150 −100 −50 0 50 100 150

20 20 20
10 10 10

5 5 5

2 2 2

1 1 1
0.5 0.5 0.5
D E F
−100 −50 0 50 100 150 −100 −50 0 50 100 150 −100 −50 0 50 100 150

Figure 6.5: Profiles of velocity across the central South Island of New Zealand. Black dots show component of
velocity parallel to the plate boundary of GPS sites relative to the Pacific plate. Open symbls show the velocity
comonent perpendicular to the plate boundary. Grey lines show fits to each component of the form
V = V0 exp (−x/λ).
Analysis of velocity fields 69

−100 −50 0 50 100 150 −100 −50 0 50 100 150 −100 −50 0 50 100 150
4 2.0 4 2.0 4 2.0

Surface Height (km)

∆VPerp (mm/yr)

2 1.5 2 1.5 2 1.5

0
A 1.0 0 1.0 0 1.0

−2 0.5 −2 0.5 −2 0.5

B C
−4 0.0 −4 0.0 −4 0.0
−100 −50 0 50 100 150 −100 −50 0 50 100 150 −100 −50 0 50 100 150
4 2.0 4 4 2.0

Surface Height (km)

∆VPerp (mm/yr)

2 1.5 2 2 1.5

0 1.0 0 0 1.0

−2 0.5 −2 −2 0.5

D E F
−4 0.0 −4 −4 0.0
−100 −50 0 50 100 150 −100 −50 0 50 100 150 −100 −50 0 50 100 150
Distance (km) Distance (km) Distance (km)
Figure 6.6: Differences between strike-perpendicular components of velocity and the best-fitting line of
the form V = V0 exp (−x/λ) (Figure 6.5).
70 Kinematics and Dynamics of the Lithosphere 2020

6.3 Tectonics of the Aegean

20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

40 mm/yr
42˚ 42˚

40˚ 40˚

38˚ 38˚

36˚ 36˚

34˚ 34˚
a
20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

42˚ 42˚

40˚ 40˚

38˚ 38˚

36˚ 36˚

34˚ 34˚
b
20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚
.5
−1 −1.5

42˚ −1 42˚
1 0.5
0 −1 −0.
−0.50 5
1

1
y=w
0 5
0.5 0
0..5
1
2
1 1
40˚ 0.
5 40˚
1
0.

x=l
5

1
5
1.

38˚ 0.5 1.5 38˚

1
−0

0 0.5
.5

1
0
0.5
x=0 −0.5

y=−w
0
36˚ 36˚
−11.5

−0.5
−

−1
.5

−0.5
−1

−1

−2 −2 −1 −0.5
.5
0.
5
34˚ 34˚
c 1

20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

Figure 6.7: (a) Velocities, relative to Eurasia, of 346 GPS sites in the Aegean and Anatolia. (b) Focal
mechanisms of shallow earthquakes and locations of active faults. The covergent boundary between
Nubia and Eurasia is shown by the double line; red (blue) lines show active normal (strike-slip) faults.
(c) Double lines show the approximation to the geographical region of interest, in the form of a box of
length, l, and width 2w. Colours show GPE of the lithosphere, calculated from surface height using
Eq. (5.13), labels on contours are in units of TN m−1 .
Analysis of velocity fields 71

6.3.1 Flow of a rectangular Newtonian thin sheet

We treat the region of interest as a rectangular box (0 < x < l; −w < y < w, with the x-direction
being approximately eastward and the y-direction being approximately northward: Figure 6.7c).
We neglect the y-component of velocity, and allow the GPE of the sheet to vary only in the
x-direction, with a constant gradient ∆Γ/l, where ∆Γ represents the difference in GPE between
Eastern Anatolia and the Southern Aegean; this is an approximation to the GPE variation seen
in Figure 6.7c.
The x-equation of the thin sheet is
∂τ xx ∂τ xy ∆Γ
2 + = . (6.8)
∂x ∂y Ll
For a Newtonian fluid of constant viscosity η, Eq. (6.8) with uy = 0 becomes

∂ 2 ux 1 ∂ 2 ux ∆Γ
+ = , (6.9)
∂x2 4 ∂y 2 4ηLl
We set ux = 0 on y = ±w and
 
∂ux  ∂ux 
= =0 (6.10)
∂x x=0 ∂x x=l
With these boundary conditions, the solution to Eq. (6.9) is
∆Γ 1

ux = y 2 − w2 (6.11)
l 2ηL
You can see that this solution closely resembles the expression for flow in a channel (Eq. (2.59)).
∆Γ/l replaces the pressure gradient, dp/ dx, and there is the same parabolic dependence of ux
on y. You should also look at the Fluids problem for the channel flow of a power-law fluid and
note how the width of the shear zones near the edges of the channels become narrower as the
value of the power-law exponent, n, increases.
We can use Eq. (6.11) to estimate the effective viscosity of the lithosphere of Turkey and
the Aegean. The maximum magnitude of the velocity, U0 , is on the line y = 0; rearranging
Eq. (6.11)
∆Γ w2
η=− (6.12)
l 2U0 L
With L ∼ 100 km, U0 ∼ −20 mm/yr (it is in the negative-x-direction), ∆Γ ∼ 2 TN m−1 w ∼ 300 km
and l ∼ 1800 km (Figure 6.7), η ∼ 8 × 1020 Pa s.
This is, of course, a very crude approximation, but a full (numerical) solution, taking account
of the actual geometry of the deforming zone, and using a power-law fluid, gives a very similar
estimate. This solution is illustrated in Figure 6.8 for a lithosphere whose rheology is described
by:
τij = BE (1/n−1) Eij (6.13)
1 1 (1−n)
ηeff = BE (1/n−1) = B n T , (6.14)
2 2
where ηeff is the effective viscosity and T is the second invariant of the vertically-averaged
1
deviatoric stress tensor, (τ kl τ kl ) 2 . Note that for n 6= 1 the effective viscosity depends on the
deviatoric stress, or equivalently upon strain rate, so that lithosphere of spatially invariant
viscosity coefficient, B, will in general exhibit lateral variation in its effective viscosity.
The calculations show regions of slow strain rate (Figure 6.8b), hence high viscosity (Fig-
ure 6.8c), which correspond to the Southern Aegean and parts of Southern and Central Anatolia.
These regions are interpreted, on the basis of low strain rates seen in the GPS data, as being
rigid. In the calculation, they do have high effective viscosity, but that is entirely the consequence
of the low deviatoric stresses acting within them (see Eq. (6.14)).
72 Kinematics and Dynamics of the Lithosphere 2020

Second Invariant nanostrain/yr

1 2 5 10 20 50 100 200

20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

42˚ 42˚

40˚ 40˚

38˚ 38˚

36˚ 36˚
200 nanostrain/yr

34˚
a 34˚
20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

200 90
Observed Second Invariant nanostrain/yr

60
100
Observed Azimuth

30
50

20
−30

10
−60

5
b −90
c
5 10 20 50 100 200 −90 −60 −30 0 30 60 90
Model Second Invariant nanostrain/yr Model Azimuth

Effective Viscosity Pa s
21.00 21.25 21.50 21.75

20˚ 22˚ 24˚ 26˚ 28˚ 30˚ 32˚ 34˚ 36˚ 38˚ 40˚ 42˚

n=3
42˚ 42˚

40˚ 40˚

38˚ 38˚

36˚ 36˚
Analysis of velocity fields 73

6.4 Rheology of the lithosphere revisited

In Chapter 4, we argued that the strongest part of the continental lithosphere is probably the
part made up of mantle. If that is so, we may calculate the strength of the lithosphere using the
two flow laws for olivine, reproduced below. Consider a piece of lithosphere that is deforming
at a strain rate E in response to a vertically-averaged stress, τ . In the mantle portion of the
lithosphere, two processes will contribute to the strain rate:
 
n −Qc
High-temperature creep Ec = A τ exp , (6.15)
RT
  r 
2 −Qp τ
Low-temperature plasticity Ep = Ap τ exp 1− (6.16)
RT τp

where the Es are strain rates, τ is stress, and the other parameters are discussed in Chapter 4
(Eqs. (4.1) and (4.7)).

We can calculate the average stress in lithosphere deforming at strain rate E as follows.

• Specify the Moho temperature, thickness of mantle portion of the lithosphere and temper-
ature at its base (we’ll use the commonly assumed value of 1300◦ C).

• At each depth, z, and corresponding temperature T (z), find the stress τ (z) such that the
sum of Ec (Eq. (6.15)) and Ep (Eq. (6.16)) is equal to the strain rate of interest, E.

• Integrate stress with respect to depth from the Moho to the base of the lithosphere. Divide
by thickness of the layer.

This is done in Figure 6.9 for the flow-law parameters given in Chapter 4 for Eqs. (4.1) and (4.7).
There are several points to be made about these plots.

• There are large uncertainties in extrapolating flow laws determined in laboratories to

geological strain rates; the plot uses the low end of stresses extrapolated from lab studies.
One way of thinking of the uncertainties is that the temperature for a given strain rate is
uncertain by at least 100◦ C.

• As discussed in Chapter 4, you can get a straight line by making a log-log plot of pretty
much any function, particularly if you do so over a small range. Each of the lines on
Figure 6.9 may be approximated by a power law.

• Recall, that for a power-law fluid E ∝ τ n , or η ∝ E (1/n−1) . See grey lines on Figure 6.9a
equivalent to n = 3 and 6.

• The figure makes sense: The Indian Ocean plots in a place that implies a cold Moho.
We might expect Moho temperatures to be higher under regions that are extending, such
as the Apennines, Aegean, and Basin and Range, particularly as there are widespread
Neogene volcanics in the last two regions. Convergence in Tibet and Iran has been going
on for longer, allowing time for thermal relaxation (see Chapter 10), whereas the Tien
Shan and the South Island of New Zealand are young collisional range and, perhaps, have
not had enough time to warm up significantly.

• Although the previous bullet point may seem plausible, we don’t know enough, at present,
to say anything more definite about the rheology of the contiental lithosphere than that
the relationships between strain rate and GPE are consistent with the notion that the
strength of the lithosphere is controlled by low-temperature plasticity of the uppermost
mantle.
74 Kinematics and Dynamics of the Lithosphere 2020

8.5 24.0
IO 200

n=6 23.5 IO
300 SI
8.0
400 23.0
TS
Log10 η (Pa s)
Log10 τ (Pa)

500 22.5 TS
7.5 CA 20
0
30
600 22.0 TB 0
40
BR 0
IR
7.0 700 3 50
0
n= 21.5
AE 60
0
SI
800
IR AE 21.0 AP 700 CA
6.5 BR AP
800

TB 20.5 900

6.0
a 20.0
b
−16.5 −16.0 −15.5 −15.0 −14.5 −14.0 −13.5 −13.0 −16.5 −16.0 −15.5 −15.0 −14.5 −14.0 −13.5 −13.0

Log10 E (s−1) Log10 E (s−1)

Figure 6.9: Vertically averaged rheology for the continental lithosphere, assuming that its strength is
controlled by the mantle portion, with the stress-strain-rate relations of Eq. (4.1) and Eq. (4.7). Lines
relate average stress to strain rate (a) and effective viscosity to strain rate (b) for different values of Moho
temperature (labelled in ◦ C). Two-letter abbreviations refer to estimates of lithospheric viscosity that we
have made earlier in this Chapter, plus a couple of others: Indian Ocean (Gordon and Houseman, 2015);
South Island of New Zealand; Tien Shan (England and Molnar , 2015); TiBet (England and Molnar ,
1997); IRan (Walters et al., 2017); Aegean (England et al., 2016); APennines (D’Agostino et al., 2014);
Basin and Range, CAlifornia (Flesch et al., 2000).
Rayleigh-Benard Convection 75

7 Rayleigh-Benard Convection
We are interested in studying the fluid flows driven by temperature variations within the mantle.
We have learned that creeping flows, because of their low Reynolds number, satisfy Stokes
equation
∇p = η∇2 u + ρg. (7.1)
Note that in the absence of flow u = 0, the pressure gradient is lithostatic, p(z) = ρgz, where
z is depth into the Earth. Here ρg is the body force of gravity. Convection is driven by lateral
variations in the body force, as we shall see. Also note that the viscosity η has been assumed
constant and taken outside the divergence operator on the stress tensor.
To study the fundamental physics of convection, we will consider a simpler problem than full
mantle convection, developed in detail below, of a fluid that is heated from below and cooled
from above. Thermal expansion leads to a vertical density gradient, with cold, dense fluid above
hot, ligher fluid. As we shall see, this situation is unstable: small perturbations to the stillness
of the fluid grow exponentially, leading to convection.

7.1 Convection in experiments

See figure Figure 7.1 for an example of a laboratory experiment on convection. The liquid is a
viscous syrup containing temperature-sensitive liquid crystals. The crystals become visible at
certain temperatures, hence the green lines are isotherms. The tank is heated from below.

Figure 7.1: A tank of viscous fluid heated from below with a thermal plume forming in the boundary
layer. The green lines along the bottom of the tank and around the plume are isotherms. From the lab
of Anne Davaille.

7.2 Density variations

To a good approximation, the density of a fluid is linearly proportional to its temperature. The
constant of proportionality is typically called α. Hence we can write the density as

ρ = ρ0 [1 − α(T − T0 )], (7.2)

where ρ0 and T0 are reference values of density and temperature.

76 Kinematics and Dynamics of the Lithosphere 2020

Substitution of (7.2) into (7.1) gives

∇p = η∇2 u + ρ0 [1 − α(T − T0 )]g. (7.3)
We know that the background density ρ0 does not cause flow, but merely leads to the lithostatic
pressure gradient. This motivates us to write the pressure as p = P + ρ0 gz (where we have taken
z to be depth, and hence increasing downward). Substituting this into (7.3) gives
∇P = η∇2 u − ρ0 α(T − T0 )g. (7.4)
Variations in density arising from thermal expansion should appear in the conservation of
mass equation, which reads
∂ρ
+ ∇· (ρu) = 0. (7.5)
∂t
As an important and provably accurate approximation, we assume that the thermal density
variations can be neglected everywhere except in the body-force terms. This is because these
density changes are small relative to the fluid’s density. If we neglected them in the body-force
terms, however, we’d lose the driving force for convection, and hence would be unable to make
further progress. This is called the Boussinesq approximation. With it, the mass conservation
equation becomes
∇· u = 0. (7.6)

7.3 The energy equation

Since density variations are due to temperature, we require an equation to compute the tem-
perature variations through space and time. This can be derived from the first law of ther-
modynamics, which states that energy is conserved, and that heat and work are both forms of
energy:
dU = dQ + dW. (7.7)
In this case, we are only concerned with the thermal energy. For our idealised system, it can
be shown that the work done by convection is negligible, and hence we have dU = dQ. This
states that changes in internal energy are the result of addition or subtraction of heat.
The internal energy is related to the temperature by
dU = ρc dT, (7.8)
where c is the specific heat capacity. Substituting (7.8) into (7.7) and dividing by an infinitesimal
increment of time dt gives
dT dQ
ρc = , (7.9)
dt dt
the rate of change of temperature is proportional to the rate of addition of heat.
How is heat added? To answer this, consider a two-dimensional “volume” with sides of
length δx and δz. The mass within the box (per unit distance into the plane) is δxδzρ. Heat
can enter/leave the box in one of three ways: by diffusion through the sides, by advection through
the sides, or by radioactive decay within the box. We’ll neglect the latter in this model.
To handle the advection of heat, we’ll move into the reference frame that is moving with the
fluid. In this reference frame, we understand equation (7.9) as applying to a parcel of fluid of
volume (per distance into the plane) δx × δz that we are tracking as it moves. Hence we re-write
it as
DT
ρc δxδz = Q̇, (7.10)
Dt
where D/Dt is a Lagrangian derivative and Q̇ is the rate of addition of heat by diffusion. Recall,
from the definition of a Lagrangian derivative,
DT ∂T
≡ + u · ∇T. (7.11)
Dt ∂t
Rayleigh-Benard Convection 77

The first part of the right-hand side is the Eulerian derivative representing the rate-of-change of
temperature at a fixed point in space. The second part is the advective transport of temperature
by the flow.

Diffusion Recall Fourier’s law, which states that the heat flux q is proportional to the tem-
perature gradient:
q = −k∇T. (7.12)
Along the edges of the box in the x-direction, this leads to a net rate of heat loss of

−Q̇x = δz (qx (x + δx) − qx (x)) ,

  
∂qx
= δz qx (x) + δx − qx (x) ,
∂x
∂qx
= δzδx , (7.13)
∂x
∂2T
= −k 2 δzδx, (7.14)
∂x
where we have used a Taylor series, truncated at first order, to approximate the change in
diffusive heat flux across the box. The last line comes from substitution of (7.12), and we have
assumed that k is constant. The same applies in the z direction and hence the contribution from
diffusion of heat is  2 
∂ T ∂2T
− Q̇ = −k + δzδx. (7.15)
∂x2 ∂z 2
We can now combine the first law (8.4) applied to a moving parcel of fluid with our expression
for heat transport from diffusion (7.15) to obtain
 2 
DT ∂ T ∂2T
ρc δxδz = k + δzδx. (7.16)
Dt ∂x2 ∂z 2

Dividing through by ρc δxδz and using vector notation gives

DT
= κ∇2 T, (7.17)
Dt
where κ = k/(ρc) is the thermal conductivity. We can highlight the role of advection by
expanding the Lagrangian derivative using (7.11) to give

∂T
+ u · ∇T = κ∇2 T. (7.18)
∂t
This equation governs the evolution of temperature within the fluid. It says that changes in
temperature are due to advection by the flow and diffusion down thermal gradients.

7.4 The Rayleigh number and the Nusselt number

Consider a layer of fluid that is heated from below and cooled from above to create a vertical
temperature gradient. The equation for density variations tells us that hotter fluid below has
lower density than the colder fluid above. Hence the system has available potential energy that
could be used to drive fluid flow in which the hot fluid rises and the cold fluid sinks. This is
known as thermal convection.
We have learned, however, that viscous stresses resist fluid flow. So a primary question
is whether the driving force for convection is sufficiently large to overcome the resistance of
viscosity. Does flow occur?
78 Kinematics and Dynamics of the Lithosphere 2020

Two things could happen: the viscous resistance to flow could overcome the density inversion,
keeping the fluid static. In this case, heat would be conducted down the temperature gradient
from hot to cold. On the other hand, if the density variation is large enough, it can overcome
the viscous resistance to flow, and heat is transported by fluid motion. Even the latter case,
conduction still plays an important role, as we shall see.
We can understand and quantify this interplay of driving and resisting forces for convection
with a simple scaling analysis.

The Rayleigh number Consider a layer of fluid of height h. The boundary at the top of the
fluid is held at temperature T0 and the boundary at the bottom is held at the higher temperature
T0 + ∆T . Now consider a sphere of cold fluid with radius a that collects near the top boundary.
Because it is cold, it is denser and hence it sinks at a speed given by

g∆ρa2 gρ0 α∆T a2

W ∝ ∝ , (7.19)
η η

where we have used Stokes solution for the speed of settling of a dense spherical object, and
equation (7.2) to eliminate ∆ρ. Since this is a scaling analysis, we are not concerned with the
exact value of W , but rather how it scales with the problem parameters (such as ∆T ).
Since the layer has a thickness h, the timescale for sinking of the blob across the entire layer
at speed W is
h ηh
τsink = = . (7.20)
W gρ0 α∆T a2

As the sphere sinks, it is surrounded by hotter fluid. Heat diffuses inward and its temperature
goes up. The time-scale over which this heating occurs is given by the scaling relationship

a2
τwarm ∝ . (7.21)
κ

The question(s) then becomes: does the sphere sink to the bottom of the layer before it
heats up? Or does it heat up and lose its negative buoyancy before it sinks? To answer this we
consider the ratio of timescales
τwarm gρ0 α∆T a4
= (7.22)
τsink κηh

Note the a4 dependence on the radius of the sphere: this means that a small sphere can only
move a very short distance before it loses its relative coldness (and hence relative denseness)
to diffusion. In other words, small temperature perturbations get effectively damped out by
the system. The fastest-sinking and slowest warming sphere is the largest one possible. In this
context, the largest possible sphere has a size that scales as a ≈ h. Hence we have

τwarm ρ0 gα∆T h3
= ≡ Ra, (7.23)
τsink κη

where we have defined the constant to be called the Rayleigh number, Ra, after John William
Strutt, 3rd Baron Rayleigh. As we shall see in the next section, when this number exceeds
a critical value (dependent on the specifics of the experiment), it means that the time-scale
for sinking (or rising) is sufficiently small relative to the time-scale for warming (or cooling)
that convection occurs. When the Rayleigh number is smaller than the critical value, the fluid
remains static and heat is transported by conduction only.
Rayleigh-Benard Convection 79

The Nusselt number Whether by conduction, convection, or both, there is a flux of thermal
energy from the hot boundary to the cold boundary. This flux maintains the temperature
difference ∆T between the two sides. Let’s denote the actual heat flux qtotal . We can compare
qtotal to the heat flux that would be predicted if the fluid was still and heat transport was by
steady conduction only. This is given by Fourier’s law, |qconduct | = k∆T /h.
The ratio of the actual to the conductive heat fluxes is called the Nusselt number,
qtotal
Nu ≡ . (7.24)
k∆T /h
When qtotal > k∆T /h we have Nu > 1, which means that heat is being transported by conduction
and convection. We can obtain qtotal from an experiment (or numerical simulation) by direct
measurement.

7.5 The onset of convection

The Rayleigh number gives us a means of assessing the size of the driving force for convection
relative to the viscous resistance: when Ra is large enough, we expect convection to occur. But
how large is large enough? Determining this is our next goal. We seek to find the minimum
value of Ra when the system begins to convect. We can think of this as looking for the condition
when the static, conductive solution is unstable to perturbations. That is, a small disturbance
in temperature or flow will grow and lead to full-blown convection.
Amazingly, we can do this with a clever analysis of the governing equations. The trick is
that we know the static, conductive solution analytically, and the solution we seek is only a
very small perturbation of that known solution. Because of this, we can linearise the governing
equations around the known state, which is also called the base state. Once we have linearised
the equations, we will be able to solve them analytically for the convective disturbance.

The governing equations The equations that govern the pressure, temperature, and flow
within the fluid layer were given above but are collected here:
0 = −∇P + η∇2 u − ρ0 α(T − T0 )g, (7.25a)
0 = ∇· u, (7.25b)
∂T
0= + u · ∇T − κ∇2 T. (7.25c)
∂t
This is the system that we seek to understand in this analysis. Solving it analytically in this
form is not possible because it is non-linear: the term u · ∇T involves the product of two
unknowns. We therefore take a more clever approach that works when the non-linear term can
be approximated by a linear term.
In what follows we will use a coordinate system that goes from z = 0 at the top, cold
boundary, to z = h at the bottom, hot boundary.

The base state When the Rayleigh number is sufficiently small, we know that the viscous
resistance to flow will prevent fluid motion, and that conduction will transport heat across the
layer of fluid. In steady-state (∂T /∂t = 0), the equations are therefore solved by u(x) = 0 with
P (x) = P0 (z). Using this in the energy equation (7.25)c gives
∂2T
= 0, (7.26)
∂z 2
and hence for the case of no fluid motion, we can integrate this and apply boundary conditions
to obtain
∆T
Tc (x) = T0 + z, (7.27)
h
where the subscript c indicates that this is the conductive solution.
80 Kinematics and Dynamics of the Lithosphere 2020

The perturbed state If the Rayleigh number is just slightly above critical, we expect very
weak convection to occur, and we expect the solution to the equations to be different than, but
very close to, the base-state solution. Hence we can expand the variables in a series about the
base-state solution:
u = 0 + u1 + 2 u2 + ... (7.28a)
2
P = P0 + P1 +  P2 + ... (7.28b)
2
T = Tc + T1 +  T2 + ... (7.28c)
where  is a constant that is arbitrarily small and represents the size of the perturbations to
the system. Since   1 we have that 0  1  2 ..., at each order, the importance of the
solution at
that order is much smaller than the importance of the preceding order. The leading
order O 0 is the base state: it is the most important
but also boring (because it has no flow).
1
The perturbation to the base state
2

is first order O  : it is small but interesting to us. The
second order perturbation O  is extremely small and pertains to questions beyond what we
are trying to discover.
Hence we can truncate the series at order 1 and substitute into equations (7.25) to obtain


0 = −∇P0 − ∇P1 + η∇2 u1 − ρ0 α(Tc − T0 )g − ρ0 αT1 g + O 2 , (7.29a)
2


0 = ∇· u1 + O  , (7.29b)
∂T1

0= + u1 · ∇Tc − κ∇2 Tc − κ∇2 T1 + O 2 . (7.29c)
∂t
We can simplify these equations using the fact that


• the O 0 terms balance (i.e. they cancel each other),


• the O 2 terms are tiny and hence can be dropped,
• the base-state temperature gradient is non-zero only in the vertical direction: ∇Tc =
(∆T /h)k.
and we obtain equations for the first-order perturbations,
0 = −∇P1 + η∇2 u1 − ρ0 αT1 g, (7.30a)
0 = ∇· u1 , (7.30b)
∂T1 ∆T
0= + w1 − κ∇2 T1 . (7.30c)
∂t h
So now our problem is to solve equations (7.30) for the convective perturbation to the base state,
P1 , u1 = (u1 , w1 ), T1 . Fortunately, these equations are linear!

Boundary conditions We will assume that the boundaries are at constant temperature and
that no flow crosses them. Hence we have
w1 = T1 = 0 on z = 0, h. (7.31)
We need a boundary condition on the horizontal component of the perturbation velocity, u1 . One
possible choice is u1 (z = 0, h) = 0. However, it is mathematically convenient (and physically
realistic, in comparison with the mantle) to instead assume that there are no shear stresses on
the top and bottom boundaries. This is written
(1) ∂u1 ∂w1
τxz = + = 0 on z = 0, h. (7.32)
∂z ∂x
Since the perturbation to the vertical velocity vanishes on the boundaries (as given in (7.31)),
we can simplify this to
∂u1
= 0 on z = 0, h. (7.33)
∂z
Rayleigh-Benard Convection 81

The stream function As you have seen previously, the stream-function is useful for incom-
pressible flow, because it automatically satisfies the continuity equation. Recall that it is given
by
∂ψ1 ∂ψ1
u1 = , w1 = − . (7.34)
∂z ∂x
You should verify that this gives ∇· u1 = 0.
Introducing the stream-function into equations (7.30) gives
 3 
∂P1 ∂ ψ1 ∂ 3 ψ1
0=− +η + , (7.35a)
∂x ∂x2 ∂z ∂z 3
 3 
∂P1 ∂ ψ1 ∂ 3 ψ1
0=− −η + − ρ0 gαT1 , (7.35b)
∂z ∂z 2 ∂x ∂x3
∂T1 ∆T ∂ψ1
0= − − κ∇2 T1 . (7.35c)
∂t h ∂x
Eliminating the pressure from equations (7.35a) and (7.35b) gives the system
∂T1
0 = η∇4 ψ1 − ρ0 gα , (7.36)
∂x
∂T1 ∆T ∂ψ1
0= − κ∇2 T1 − , (7.37)
∂t h ∂x
where
∂4 ∂4 ∂4
∇4 = ∇2 ∇2 = + 2 +
∂x4 ∂x2 ∂z 2 ∂z 4
is the biharmonic operator (two-dimensional, in this case). The boundary conditions become

∂ 2 ψ1
= T1 = 0 on z = 0, h. (7.38)
∂z 2
The problem is now to solve for only two variables, ψ1 and T1 using two linear equations,
(7.36) and (7.37).

A trial solution Here’s what we know about the solution:

• it must consist of functions for ψ1 and T1 , and it must depend on x, z, and t;

• because of the boundary conditions, it must go to zero on z = 0, h;

• because the equation are linear, it must be separable: e.g. ψ1 (x, z, t) = F (x)G(z)H(t);

• because the equations are linear, each dependence on the independent variables must be
either sinusoidal or exponential.
To satisfy all of these conditions, we can propose the following ansatz1 :
 πz   
2πx σt
ψ1 = −ψ ∗ sin sin e , (7.39a)
h λ
 πz   
∗ 2πx σt
T1 = T sin cos e . (7.39b)
h λ
In this trial solution, ψ ∗ and T ∗ are unknown but constant coefficients; they are of no interest
to us. Of principal interest are λ, the horizontal wavelength of the perturbation, and σ, the
growth rate of the perturbation.
1
Ansatz is german for educated guess. It is the word typically used by physicists when they propose a trial
solution.
82 Kinematics and Dynamics of the Lithosphere 2020

x
y

Figure 7.2: Contours of T1 (colour) and ψ1 (black) within the layer. The contours of ψ1 are streamlines
of the convective flow. A wavelength λ has been chosen (with foresight) for plotting purposes. Solving
for λ as a function of other parameters remains part of the problem.

The solution specifies a periodic array of convective rolls, but doesn’t specify the aspect ratio
of those rolls (which is given, for a pair of rolls, by A = h/(λ/2), with the wavelength λ as yet
unknown). These rolls are shown in Figure 7.2. A value of the horizontal wavelength λ is chosen
for the purposes of plotting but it remains part of the problem to solve for λ.
If σ < 0 then we expect the perturbation to decay away, hence convection does not occur
and the system is termed as stable. If σ > 0 then we expect the arbitrarily small perturbations
to grow exponentially into a convective flow. In this case the system is called unstable. The
state where σ = 0 and the system is perched on the edge between stability and instability is
called the state of neutral stability.

Solving for λ and σ To obtain a solution we substitute (7.39) into (7.36) and (7.37)
 2 2
π 4π 2 2πρ0 gα ∗
η 2
+ 2 ψ∗ = − T , (7.40a)
h λ λ
 
κπ 2 κ4π 2 2π∆T ∗
σ+ 2 + 2 T∗ = − ψ . (7.40b)
h λ λh
These two algebraic equations can be combined to eliminate ψ ∗ and T ∗ giving the growth
rate,    
κ A2 2 2


σ = 2 Ra −π A +1 . (7.41)
h π 2 (A2 + 1)2
Recall that the aspect ratio of the solution and the Rayleigh number are given by
2h ρ0 gα∆T h3
A= and Ra = , (7.42)
λ ηκ
respectively.
We seek the values of the Rayleigh number Ra that, for any given value of A, is exactly
neutrally stable; i.e. we seek the critical Rayleigh number Racr below which the system is stable,
and above which it is unstable. This occurs precisely when the growth rate vanishes, so setting
σ = 0 in (7.41) and solving for Ra ≡ Racr gives

3
π 4 1 + A2
Racr = . (7.43)
A2
A plot of this curve is given in Figure 7.3.
The minimum value of the Rayleigh number at which convection (at any aspect ratio) can
occur is obtained by solving dRacr / dA = 0. The solution is
1
A= √ , (7.44)
2
Rayleigh-Benard Convection 83

2000

1500

1000

500

0
0 0.5 1 1.5 2

Figure 7.3: A plot of the critical Rayleigh number versus the aspect ratio of the convection solution

√
or λ = 2 2h. Substitution of this result back into (7.43) gives the minimum value of the
Rayleigh number for convection,

27π 4
min(Racr ) = = 657.5. (7.45)
4
For a stable, quiescent system where the Rayleigh number is slowly increased over time, it√is at
this value of the Rayleigh number that convection begins, and with an aspect ratio of π/ 2.

Applicability to the mantle Although a good illustration of the onset of convection, the
above analysis may not be applicable to the mantle for several reasons. The most important of
these is that the mantle has internal heating by radioactive decay, which was not included in
the analysis above.
A calculation along the same lines as the above, but with internal heating, a cold upper
surface, and a thermally insulating bottom boundary leads to a definition of the Rayleigh number
as
αρ20 gHh5
RaH = , (7.46)
kηκ
where H is the rate of internal heating and k is the thermal conductivity. The minimum critical
Rayleigh number for this system is 867.8, and the associated value of the aspect ratio is 1.79.
We can estimate the value of the Rayleigh number in the mantle. Taking η = 1021 Pa-
sec, k = 4 W/m/K, κ = 1 mm2 /sec, α = 3 × 10−5 K−1 , g = 10 m/sec2 , ρ0 = 4000 kg/m3 ,
H = 9 × 10−12 W/kg, and h = 2880 km, we find that RaH = 2 × 109 . This value is obviously
much higher than the minimum critical value. Convection is expected in the mantle.

7.6 Convection at high Rayleigh number

When Ra  Racr , thermal convection is rapid and disordered, with plumes rising and falling
from thermal boundary layers along the top and bottom of the fluid layer. Within the thermal
boundary layers, there is a sharp temperature gradient and little or no flow. Between the thermal
84 Kinematics and Dynamics of the Lithosphere 2020

boundary layers, within the interior of the fluid, convection is very vigorous, rapidly transporting
heat from the bottom to the top of the layer. The interior of the domain is therefore well-stirred,
with horizontally averaged temperature that is approximately independent of height, as shown
in Figure 7.4.

Figure 7.4: Average temperature profile through a convecting layer with Ra = 105 .

In this regime, it is impossible to obtain analytical solutions for the details of the flow. For
many purposes, however, the details of the flow are not important. What we care about is the
heat transported by convection and how it scales with the problem parameters. From looking at
Figure 7.4, we can see that heat enters and leaves the fluid by conduction through the boundaries
and the thermal boundary layers—hence we know that conduction is important. We also know
that in the interior, heat is transported by convection, and the vigor of convection scales with
the Rayleigh number. A dimensionally correct guess for the scaling of the heat flux is thus

Nu = Raβ (7.47)

where β an unknown exponent. Expanding the Nusselt number and rearranging gives
∆T β
qtotal ∝ k Ra . (7.48)
h
In experiments and numerical simulations of convection at high Rayleigh number, it is ob-
served that the heat flow qtotal does NOT depend on the height h of the fluid layer. Since the
Rayleigh number contains h3 , and the conductive scaling contains h−1 , this observation can be
accommodated if β ≈ 1/3.
Evidence from laboratory experiments (e.g. Figure 7.5) indicates that Nu ∝ Ra1/3 is very
accurate over many orders of magnitude in Ra > Racr . It is useful in analysing the thermal
history of the Earth.
Rayleigh-Benard Convection 85

Figure 7.5: Scaling of Nu versus Ra from experiments on Helium gas by Niemela et al, Nature 2000.
Mantle Convection 87

8 Mantle convection
With that basic introduction to thermal convection, we are ready to consider the mantle and how
it convects. The mantle can be thought of as having two modes of convection, the plate mode
and the plume mode. The plate mode is plate tectonics: a cold, stiff thermal boundary layer on
the outside of the planet founders and sinks into the mantle at subduction zones, pulling the
surface plate along behind it. Upwellings are passive (that is they are driven by plate spreading,
not by buoyancy). The plate mode arises from the strong cooling at the surface, and the weak
internal heating of the mantle. It is responsible for the majority of the heat transfer out of the
Earth.
The plume mode is associated with narrow, thermal upwellings in the mantle that may be
rooted at the core-mantle boundary. It arises because the mantle is heated from below by the
core. The plume mode accounts for only a small fraction of the heat transfer out of the Earth.
Mantle convection in both the plate and plume modes leads to cooling of the Earth over
time. Simple, globally integrated energy balances can allow us to formulate and test hypotheses
about the thermal history of the Earth.

8.1 The fluid mantle

We know the mantle is solid: seismic shear waves pass through it efficiently, traversing the
interior of the Earth multiple times. The stresses associated with these waves are sustained for
seconds, at most. When the mantle is subjected to longer-term stresses, it flows like a fluid: the
rate of deformation is proportional to the size of the stress. We know this from glacial isostatic
rebound.

Glacial isostatic rebound and mantle viscosity The ice sheets that covered Fennoscandia
during the last ice age melted at about 11,000 years before present. Since then, relative sea-level
in those regions has been falling, indicating a rebound of the land surface. This drop in sea-level
is approximately exponential, with a time-constant of 4.6 ka. We can use this information to
approximate the viscosity of the mantle. Our analysis will equate the isostatic force causing
rebound with the viscous resistance force of mantle flow back into the area.
Isostasy requires that pressure is constant at a depth D in the mantle. A column of rock
that was unaffected by the ice sheet has pressure ρm gD at its base. In the affected area, there
is a depression of depth d left by the ice after it melts. Suppose this depression is filled with
water. Then it has pressure g(ρw d + ρm (D − d)) at its base. The pressure difference between
these two columns of rock is thus

∆P = gρm D − g(ρw d + ρm (D − d)), (8.1)

= g(ρm − ρw )d. (8.2)

We can approximate the force arising from this pressure difference by multiplying by the area
over which it is applied. This area is roughly a circle with radius R and hence

Fd = πR2 g(ρm − ρw )d. (8.3)

This driving force is resisted by a viscous force in the mantle, as mantle rock flows into the
area that is rebounding. We can assume that the effect of the pressure gradient is felt over
a distance approximately equal to the radius of the ice sheet. If the flow has a speed that is
roughly v then a representative velocity gradient is v/R. The viscous resisting stress is then
approximated as v
τr = η = ηv/R, (8.4)
R
88 Kinematics and Dynamics of the Lithosphere 2020

where we have again used η for mantle viscosity (note that Davies uses η in his book). Again
we multiply by the area over which the stress is applied to get the total force as

Fr = πR2 ηv/R. (8.5)

Since the mantle has no appreciable momentum, the forces must balance: Fd + Fr = 0. This
gives
πR2 g(ρm − ρw )d − πR2 ηv/R = 0, (8.6)

where the negative sign comes from the fact that the forces are in opposite directions. We can
rearrange to give
v = g(ρm − ρw )Rd/η. (8.7)

The rate of flow v must be about equal to the rate of change of d, so we can write this as

dd gR∆ρ
=− d. (8.8)
dt η

This differential equation has the solution

d(t) = d0 e−t/τ0 , (8.9)

where τ0 = η/(gR∆ρ). Rearranging gives η = τ0 gR∆ρ. Using values g = 10 m/sec2 , R = 1000

km, ∆ρ = 2300 kg/m3 , and τ0 from Figure 8.1, this gives η ≈ 1021 Pa sec. This is a very large
number, which is why strain-rates in the mantle are very small.

Figure 8.1: Isostatic rebound of Fennoscandia. (a) shows observational data of relative sea level versus
time, with an exponential fit. (b) shows a schematic diagram of the preglacial, synglacial, and postglacial
state of the land surface.

So we have used glacial isostatic rebound to (a) demonstrate that the mantle flows over
geological time, and (b) obtain an approximate value for the viscosity of the mantle.
Mantle Convection 89

Temperature dependence of viscosity Our estimate of η actually pertains to the hot

interior of the mantle, rather than the cold lithosphere above. The viscosity of the whole mantle
over a range of temperatures can be described by an Arrhenius law
 ∗ 
E 1 1
η = ηr exp − , (8.10)
Rg T Tr

where ηr and Tr are reference values, E ∗ is the activation energy (J/mole), and Rg is the
universal gas constant. When T ≈ Tr , the exponential is approximately one, and the viscosity is
approximately equal to the reference viscosity. When T is much smaller than Tr , the argument
to the exponential is large, and hence the viscosity is much larger than ηr .
The temperature dependence of viscosity has an important effect on mantle convection.
Remember that Ra ∝ η −1 and Nu ∝ Ra1/3 . If the temperature of the mantle is high, the
viscosity is low, and the Rayleigh number is large. This means more vigorous convection, larger
Nusselt number, and higher heat flux. The heat flux cools the mantle and causes the viscosity
to increase.

8.2 The plate mode

We now know that the mantle is fluid with a large but finite viscosity. We know that plates move
and subduct. We know that the mantle convects, and that it is internally heated by radioactive
decay. Is plate tectonics the surface expression of mantle convection?

Simple force balance on oceanic plate To show that plate tectonics is mantle convection,
we can demonstrate the plate motions, to leading order, result from the balance of the negative
buoyancy of sinking plates and the viscous resistance of mantle flow. In simple mathematical
terms,
FB + Fr = 0, (8.11)
where FB is the buoyancy force and Fr is the resisting force.
We first turn our attention to FB . We’ll assume that buoyancy is entirely thermal in origin;
that the oceanic lithosphere and the asthenosphere have the same composition. We’ll consider a
cross-section of a subducting plate, as shown in Figure 8.2, and all of the equations that we write
will be per unit length into the plane of the cross-section. According to Archimedes principle,
the buoyancy force (per unit length) is then given by the product of the density difference and
the area of the plate in cross-section,

FB = gρ0 α∆T Dd, (8.12)

where Dd represents the thickness of the plate times its length in the mantle. The temperature
difference ∆T can be approximated if was assume that the plate has an average temperature
that is the mean of the mantle temperature and the surface temperature, so about T /2. Then
∆T = T − T /2 = T /2 and we have

FB = gρ0 αT Dd/2. (8.13)

Because the plate is colder than the mantle, it is negatively buoyant, and hence this force drives
it downward.
The resisting force comes from the viscous stress on the subducting plate and the ocean-floor
plate. Suppose they are moving with speed v through a mantle with viscosity η. This creates
a shear in the mantle between the plate and the conjugate mantle return flow −v. These are
separated by an approximate distance D and hence the velocity gradient is 2v/D and the shear
stress on the descending slab is
τr = 2ηv/D. (8.14)
90 Kinematics and Dynamics of the Lithosphere 2020

Figure 8.2: (a) Schematic diagram of the plate mode of convection. (b) Schematic diagram of a simple
convection cell driven by sinking of a cold plate into the mantle (from Davies’ Mantle Convection for
Geologists.)

The stress acts on an area D per unit length of the plate and the resisting force is given by
the product of the stress and the area D2ηv/D = 2ηv. This force acts on both sides of the
descending slab, and on the underside of the surface plate, to give a total resisting force of
Fr = 6ηv. Noting that the resisting and driving forces act in opposite directions and plugging
them into equation (8.11) gives

− gρ0 αT Dd/2 + 6ηv = 0, (8.15)

and solving for v gives

v = gρ0 αT Dd/12η. (8.16)
To compute the plate thickness d, we can use the conductive cooling of a plate. If the plate
moves a distance D with a speed v, the time taken is tp = D/v and the plate thickness is
approximately p p
d = 4κtp = 4κD/v. (8.17)
Substituting (8.17) into (8.16) and rearranging gives
 √ 2/3
gραT κ
v=D (8.18)
6η

Taking T = 1300◦ C, D = 3000 km (the height of the mantle), η = 1022 Pa sec (the viscosity
of the lower mantle) gives a velocity of about 10 cm/year, which is near the average observed
plate motion.
Mantle Convection 91

This analysis shows that by thinking of plate tectonics as a convective process driven by
the negative buoyancy of subducted plates, we can account for the approximate rate of plate
motions at the surface. This does not prove that plates move because of convective forces, nor is
this a complete theory of the forces involved in plate tectonics. It turns out that in much more
sophisticated models, however, this balance of forces plays a dominant role.

The yield stress and plate tectonics If mantle viscosity depended on temperature as
equation (8.10), there would be no plate tectonics because the lithosphere would be too stiff to
subduct. The rheology of rock is actually much more complicated, though, and one complication
is that is possesses a yield stress. This is a level of stress below which the material is rigid, and
above which it flows with little resistance. A yield stress allows plates to bend and break, and
hence allows for initiation of subduction or rifting.

Heat flow due to mantle convection We have seen that plates, and especially oceanic
plates, are part of mantle convection, not just a passive rider on top of it. They are the cold
surface boundary layer through which the Earth loses heat by conduction. How much heat does
the plate mode remove from the Earth, and what fraction of the total is due to the plate mode
of convection?
Let’s assume that just before subduction, the lithosphere is 100 Ma old and has a thickness
of d = 100 km. It has an average temperature that is about half the difference between the
mantle temperature (∼ 1300◦ C) and the surface temperature, so about 650◦ C. The amount of
heat lost from the lithosphere between when it left a mid-ocean ridge axis and when it arrives
at the subduction zone is
∆H = ρdcP ∆T, (8.19)
where cP is the specific heat capacity.
Sea-floor spreading creates about 3 km2 of new ocean floor each year; subduction removes a
similar amount. Hence the total rate of heat loss by cooling the oceanic lithosphere is approxi-
mately
Q = S∆H = SρdcP Tm /2, (8.20)
where S is the rate of subduction of sea-floor area. Plugging in typical numbers gives about
21 × 1012 Watts, or 21 TW. This estimate is in fact off by about 50%—the actual heat flow out
of the ocean basins is about 30 TW.
The total heat flow out of the Earth is 41 TW, so about 75% of this heat flow can be
attributed to the plate mode of convection. The oceanic heat flow is about 90% of the heat flow
coming out of the mantle.

8.3 The plume mode

For Rayleigh-Benard convection, we saw that there are both cold downwelling currents and hot
upwelling currents. The hot upwellings are called plumes. You are already aware that there are
plumes in the mantle. This is (probably) because the mantle is not only heated by radioactive
decay, but also has a basal heat flux from the hotter core. Secular cooling of the core produces
a thermal boundary layer at the base of the mantle (as shown in Figure 8.3), which gives rise to
thermal plumes (chemical buoyancy may also be important, but we neglect that here). In this
section, our goal is to understand the physics of an isolated plume.
Hawaii will be our focus because it is far from a mid-ocean ridge (unlike Iceland) and because
it has a relatively large flux and a long history.

Hotspot swells as a constraint on plume flux The Hawaiian volcanic chain is associated
with a broad swell in the sea floor. The swell is up to 1 km high and about 1000 km wide. There
is observational evidence that the crust in this swell is not anomalous; the swell is supported
92 Kinematics and Dynamics of the Lithosphere 2020

Figure 8.3: Schematic temperature profiles through the Earth. Profile (a) shows the temperature shortly
after accretion and segregation of the core. The mantle is still very hot, and there is no temperature
difference between the core and the mantle. Heat would then be lost from the mantle, which would cool
as a result. Later the profile would look like (b). Note the thermal boundary layer at the base of the
mantle. This could provide a site for formation of hot plumes.

by the buoyant material beneath the lithosphere. Because the Pacific plate is moving while
the hot-spot remains stationary, there must be a continuous flux of buoyant rock up the plume
conduit to maintain the swell.

Figure 8.4: Idealised section of a plume as a vertical cylinder.

To compute the buoyancy flux of a plume, we consider a simple model of a plume conduit
as a vertical cylinder with radius r and average upwelling speed u. We consider the short time
interval ∆t during which the rock in the plume moves, on average, a distance of u∆t. The
volume-rate of flow is then
V
φ≡ = πr2 u. (8.21)
∆t
And we recall that buoyancy is the gravitational force due to a deficit of density,

B = g∆ρV. (8.22)

This is the buoyancy of the material that has moved up the plume in a time ∆t. The buoyancy
flow rate is the rate at which buoyancy ascends the conduit,

b = g∆ρπr2 u. (8.23)
Mantle Convection 93

Figure 8.5: Sketch of a hotspot swell like that of Hawaii in map view (left) and two cross sections.

This buoyancy rate can be related to the size of the hotspot swell.
We again consider a balance of forces to obtain an equation. The swell has an excess weight
that is due to displacing ocean water with denser rock. Since the plate is moving over the plume,
new swell is constantly created as the buoyancy flux elevates the sea floor. If the Pacific plate
travels with a speed of v = 100 mm per year and the swell has a height of h = 1 km and width
of w = 1000 km, then the downward force of swell created over some time interval δt is

FW = g(ρm − ρw )wvhδt, (8.24)

where the density difference is the mantle density minus the density of water. This must balance
the buoyancy force of one year of plume flux, Fb = bδt, which acts in the upward direction.
Equating the two gives
b = g(ρm − ρw )wvh. (8.25)
Note that this simple equation must be used with care; it is invalid in the limit of zero v.1
Using values quoted above, we obtain b = 7 × 104 N/sec for Hawaii using eq. (8.25). This
is a somewhat abstract quantity, so let’s go further. Combine equation (8.23) with (8.21) and
solve for the volumetric flow rate to obtain
b
φ= . (8.26)
gρm α∆T

The petrology of Hawaiian lavas indicates a peak temperature of about 300 K above that of
normal mantle. Substituting this and other values gives φ = 7.4 km3 /yr.
The mean flow rate up the conduit is given as u = φ/πr2 . Taking a conduit diameter of 100
km for Hawaii (based on the region of active volcanism), we obtain an upwelling speed of nearly
1 m/yr, about 10 times faster than the fastest plate speed!

Heat transport by plumes We can relate the buoyancy rate to a heat flow rate by assuming
that the buoyancy is entirely thermal in origin. The plume has a temperature Tp , which is higher
than the mantle temperature Tm . The difference in density is

ρp − ρm = −ρm α∆T, (8.27)

and so the buoyancy rate can be rewritten as

b = πr2 ugρm α∆T. (8.28)

1
Equation (8.25) would seem to indicate that for a plate speed of zero with any finite value of w, the buoyancy
rate would go to zero! This is a false prediction, however, because the validity of the model (and hence of
equation (8.25)) breaks down if v is zero. If the plate doesn’t move (v = 0), the plume spreads symmetrically
beneath it; in this case the width of the plume trail, w, cannot be defined. For v sufficiently large, the product
wvh should remain constant, even if the plate speed changes.
94 Kinematics and Dynamics of the Lithosphere 2020

Recall that heat content per unit volume is given by H = ρcP ∆T , where cP is the specific heat
capacity. The rate at which heat flows up the conduit is then given by

Q = ρp cP ∆T πr2 u, (8.29)

where πr2 u is the volume flow rate of rock. Making the Boussinesq approximation ρp ≈ ρm , we
can use equation (8.28) to eliminate ρp ∆T πr2 u from (8.29) to obtain

cP b
Q= . (8.30)
gα

Hence we have related the heat flow rate up a plume conduit to the buoyancy flow rate with a
very simple formula, depending only on three known parameters. Note that we don’t need to
know ∆T , u, or r! Using standard values we obtain Q = 0.2 TW. This is only about 0.5% of
the global heat flow.
The total heat flow of all mantle plumes is only about 2.3 TW, which is about 6% of the
global total heat flow. The plate mode of convection accounts for 90% of the heat coming out
of the mantle, while the contribution from plumes is much less. This suggests that plumes
are a secondary form of convection in the mantle, which is fairly obvious if you look at global
topography and heat flow data.

Plume origins and dynamics We can think of plumes as originating with the instability
of a less dense layer underlying a more dense layer. A small perturbation to the interface can
lead to its exponential growth. This perturbation becomes a plume and rises through the dense
material above.
The structure of the plume has two parts: the head and the tail. The head is an approx-
imately spherical blob of buoyant fluid. Its size must be large enough for the buoyancy force
to overcome viscous resistance. At the same time, it is assimilating surrounding material (and
losing heat) by viscous entrainment.
The tail of the plume is an approximately cylindrical conduit of buoyant rock that connects
the plume to the boundary layer where it originated. Buoyancy drives a flow up the tail. The
excess temperature in the tail of the plume reduces the viscosity within it. Lower plume viscosity
means that the same flow rate is accommodated by a smaller radius.

Flow in a plume tail To model the flow of rock up the tail of a plume, we consider a
cylindrical conduit with fixed walls. Within the conduit there is a balance between buoyancy
forces (which we assume to be purely thermal) and viscous resistance force.
For a segment of the cylinder with radius a height δz, the buoyancy force within a radius
r ≤ a is
FB = −g∆ρπr2 δz. (8.31)
The minus sign cancels the negative in ∆ρ to give a positive buoyancy force.
To compute the viscous resistance, in this case, we can use calculus rather than scaling
analysis. The viscous resistance is proportional to the gradient in the speed of the flow, dv/ dr,
to give a shear stress that is τ = η dv/ dr. The stress acts over the internal area of the control
volume, 2πrδz. Thus the viscous resistance is

dv
FR = 2πrδzη . (8.32)
dr
Equating FB with FR , integrating once in r and using the boundary condition v = 0 at r = a
gives
g∆ρ 2
v= (r − a2 ). (8.33)
4η
Mantle Convection 95

Recall that ∆ρ < 0 and hence v ≥ 0 for r ≤ a.

We can integrate this equation to obtain the volumetric flow rate over the cross-section of
the cylinder. To do so, we must integrate infinitesimal rings of flow, each with an area given by
dA = 2πr dr. Hence we have
Z a
g∆ρ 2
φ= 2πr (r − a2 ) dr,
0 4η
 
πg∆ρ
=− a4 . (8.34)
8η

Equation (8.34) shows that by doubling the conduit radius a, the volumetric flow rate increases
by a factor of 24 = 16. This also means that large changes in the flow rate are accommodated
by only small changes in the conduit radius. The largest hotspot (Hawaii) has an estimated
flow rate of about 10 larger than the smallest; all other parameters being equal, this means the
difference in conduit radius is less than a factor of two. The seismic detectability of these two
plumes is about the same.
We can use our estimated volume flow rate from equation (8.26) in equation (8.34) with
∆ρ = ρα∆T to solve for the conduit radius a. Using a plume viscosity of 1018 Pa-sec gives a
radius a ≈ 50 km. Because a ∝ η −1/4 , a factor-of-ten uncertainty in η translates to a factor-of-
two uncertainty in a. Hence we conclude that plume conduits should be about 50–100 km in
radius in the upper mantle.

Ascent of plume heads Experiments and numerical simulations tell us that we can approx-
imate the head of a plume as a buoyant sphere. The rate of rise is then gives by Stokes law for
a spherical particle in a viscous fluid. This involves a balance between buoyancy

FB = 4πr3 g∆ρ/3 (upward) (8.35)

and Stokes drag

FR = 4πrηv (downward). (8.36)
Requiring these to balance yields
v = g∆ρr2 /3η. (8.37)
Using values appropriate to the lower mantle, a plume head with a radius of 500 km would
rise at a rate of about 4 cm/yr. In the upper mantle the rise speed with be larger. Figure 8.6
shows a time-series of results from an axisymmetric model of a plume ascending from a thermal
boundary layer. The black dots are passive markers that were initially in the thermal boundary
layer.
Note that the rate of flow that we calculated for the conduit (∼ 1 m/yr) is much larger than
the rate of rise of the head. We should therefore expect that the head of the plume is growing
with time, by this mechanism and by thermal entrainment.

Thermal entrainment The complex flow within the head of a plume leads to entrainment
of ambient mantle, as shown in Figure 8.6. Note how the material that was initially in the
boundary layer has been stirred into the head of the plume. Thermal diffusion out of the plume
head warms the surrounding mantle in another thermal boundary layer. Both entrainment and
thermal diffusion cause the plume head to grow with time.

8.4 Overview
Plumes and plates Mantle convection is driven, in part, by cooling from above, and in part
by heating from below. Why, then, are the downwellings (subducted plates) and the upwellings
(plumes) so different in shape and dynamics? What breaks the symmetry between the two? The
96 Kinematics and Dynamics of the Lithosphere 2020

Figure 8.6: Results from a axisymmetric numerical model in which a plume grows from a thermal
boundary layer.

difference is their viscosity. Plates are cold and behave as rigid units with deformation localised
into narrow shear zones. Plumes are hot and flow viscously, with a lower viscosity than their
surroundings.

Layered convection versus whole mantle convection There is a long-standing debate

about whether the mantle convects in two separate layers, or as a single whole. Early arguments
stated that the viscosity of the lower mantle (below 660 km) is too large for convection to occur
and that geochemical observations suggested the existence of a primitive reservoir of primordial
material that has been geochemically isolated over the age of the Earth (because it did not
participate in convection). Both of these lines of evidence have proven to be incorrect.

Figure 8.7: Subduction tomography showing slabs penetrating into the lower mantle. (From Stern, R.J.,
Subduction Zones, Rev. Geophys. 2002)

The most important evidence against layered convection is images from seismic tomography
(Figure 8.7) that show subducted slabs descending into the lower mantle. Despite their low
resolution, these images indicate that slabs penetrate the 660 km boundary. The mass flux
Mantle Convection 97

of slabs into the lower mantle precludes a persistant chemical isolation of the lower mantle.
Consider a simple box model: the lower mantle reservoir contains about 2.6 × 1024 kg of rock.
The subducted rate of addition of mass into this reservoir is about 3 km2 /yr subducted × 100
km plate thickness × 3300 kg/m3 density gives about 1015 kg/yr. The residence time of rock
in the lower mantle is therefore about 2.6 billion years. Since plate tectonics seems to have
operated over much of the history of the Earth, it is very unlikely that there exists a large,
pristine reservoir of primitive mantle rock—the lower mantle cannot be chemically isolated from
the upper mantle.

Figure 8.8: Heat budgets for three models of mantle convection. (a) Full-mantle convection. (b) With a
moderately thick lower mantle. (c) With a very thick lower mantle.

There is a second argument against a layered mantle with a lower mantle containing primitive
rocks. A deep layer that doesn’t participate in convection would have larger internal heating
than the upper mantle. This heating, plus the core heat flow would raise its temperature. A
thermal boundary layer would develop above it, in the upper mantle, and this would give rise
to large plumes, which would have a topographic signature on the sea floor. Such plumes are
not observed!

8.5 Energy-balance models

The discussion above has shown that we can consider the mantle to be a single, convecting
system that is driven by heating from the core, heating from radioactive decay, and cooling to
the atmosphere. We have approximately determined the balance of these rates at the present.
What was their balance in the past? Was convection more or less vigorous? We can investigate
this with energy-balance models that take current rates and extrapolate backward in time.
Fundamentally this means that we apply the first law of thermodynamics to the entire mantle:

rate of change of heat content = rate of heat input + rate of heat loss. (8.38)

If the rate of heat loss from the mantle exceeds the rate of heat input, then the mantle is losing
energy and cooling down. If the mantle is currently cooling, then we can run the equation
backward to determine the heat content (and temperature) of the mantle in the past.
To see this in more detail, we write equation (8.38) in terms of temperature and rates of
heat flow. Recall that changes in internal energy are related to changes in temperature by
dU = M C dT , where M is the mass of the system and C is its heat capacity. This can be
divided by dt and rearranged to give
dT 1 dU
= , (8.39)
dt M C dt
the rate of change of temperature is proportional to the rate of change of internal energy. By
the first law, the rate of change of internal energy is related to additions and subtractions of
heat, hence we can write
dT 1
= (QR + QC − QS ) . (8.40)
dt MC
98 Kinematics and Dynamics of the Lithosphere 2020

QR (radioactive heating), QC (heat from the core), and QS (heat lost through the surface) are
all rates of heat transfer. We need to quantify each of these.
Radioactive heating in the mantle is the mostly result of decay of 238 U, 235 U, 232 Th, and 40 K.
In total these have a time-dependent heating rate of HR in units of Watts per kilogram. We will
not worry about the details of this heating rate, but will express the total rate of radioactive
heating in terms of its ratio with the surface heat loss:

QR M HR
Ur = = , (8.41)
QS QS

were Ur is a nondimensional number called the Urey ratio.

The surface heat flux can be approximately modelled using the Nusselt number. Recall that
Nu = qtotal /qconductive ∝ Ra1/3 . Rearranging this and assuming the constant of proportionality
is about unity gives

qS = qtotal = qconductive Ra1/3 ,

 1/3
kT gρ0 αT h3
= , (8.42)
h κη(T )

where we have used the mean mantle temperature T as a proxy for ∆T , since the surface
temperature is constant. Temperature also appears implicitly in the viscosity η(T ). The rate of
surface heat loss is then the heat flux times the area of the Earth’s surface,
2
QS = 4πRE qS , (8.43)

and hence we can rewrite equation (8.40) as

"  1/3 #
dT 1 2 gρ0 αT
= (Ur − 1)4πRE kT + QC . (8.44)
dt MC κη(T )

We will not consider the core heat flow QC in detail; suffice to say that it can be reconstructed
similar in a manner similar to the surface heat flow.
In the end, we have an ordinary differential equation for the mean mantle temperature, T .
This equation can be solved numerically forward in time, given an initial condition, or it can be
solved backward in time from the present state. Figure Figure 8.9 shows the results of a slightly
more complicated model with independent temperatures for the upper and lower mantle. It was
solved forward in time from an initially hot Earth.
We can notice several things about Figure 8.9:

• There is a rapid cooling phase during the first half billion years of the model. This is the
phase when the initial heat of the mantle is lost through the surface, and the temperature
goes toward the temperature that can be maintained by radioactive heating.

• There is a slower cooling for the rest of the model time, where the surface heat flow QS
is approximately parallel to the radiogenic heating QR . (If QR were constant, QS would
approach it asymptotically).

• The heat flow from the core (by mantle plumes) remains very low through the entire run,
which keeps the core temperature very high.

• The overall picture is that primordial heat is lost quickly and for most of the evolution
of the model, the temperature is set by a balance between surface cooling and radiogenic
heating.
Mantle Convection 99

Figure 8.9: Results of a reference thermal evolution model of the mantle and core.

At the end of this run, the surface heat flow is about 36.4 TW and the core heat flow is
about 7.6 TW — both are close to the real present Earth value. The total radiogenic heating is
about 24.6 TW at the end of the model, so the final Urey ratio is about 24.6/36.4 ≈ 0.7. And
the heat flow corresponding to secular cooling is about 36.4-7.6-24.6 = 4.2 TW. Recall that
dT Qnet
= , (8.45)
dt MC
and so with M = 4 × 1024 kg and C = 1000 J/kg/K, the predicted cooling rate is about 30 K
per billion years.
So the model tells us that if the surface heat flow of the Earth is nearly balanced by radiogenic
heating, the Urey ratio must be near 0.7. Geochemists, in contrast, tell us that the “observed”
Urey ratio of the Earth is more like 0.3! This means that the present-day heat flow out of the
mantle is not supported by radiogenic heating! There are at least three possible explanations
1. The geochemists are wrong, and there are more heat-producing elements in the mantle
than they think.

2. The present rate of heat loss is anomalously high, representing a temporary fluctuation
in mantle convection. If we averaged over several fluctuations (say a few hundred Ma),
we would find that the heat flow is balanced by the observed radiogenic heating. In this
scenario, the present cooling rate would be about 140 K per billion years. Half a billion
years ago, the mantle would have been 70 K hotter, and there would have been a lot of
magma around! We don’t see evidence of this. So the fluctuations in this scenario would
have to be shorter term. This could come from something that has been called episodic
convection.

3. The present rate of heat flow is high, but not anomalously high. The rate of cooling of the
Earth is large, and the planet was much hotter in the past. Extrapolating back naively,
this gives unrealistically high temperatures in the past (magma ocean 1 Ga ago). This is
100 Kinematics and Dynamics of the Lithosphere 2020

known as the “thermal catastrophe.” Of course there was no thermal catastrophe in the
past! This just indicates that our understanding of mantle convection is flawed.

For case 3, there have been a number of possible scenarios advanced. One model from the
recent literature2 suggests that the mantle was molten at its base for much of the Earth’s history.
The heat flow that we see today that is unbalanced by radiogenic heating came not from secular
cooling of the mantle, but secular freezing of the basal magma ocean! Other possibilities are
discussed by Geoffrey Davies in his book.

2
Labrosse, Hernlund, and Coltice, A crystallizing dense magma ocean at the base of the Earth’s mantle, Nature,
2007.
Flow at Oceanic Ridges and Subduction Zones 101

9 Flow at Mid-Ocean Ridges and Subduction Zones

For isoviscous Rayleigh-Benard convection between parallel plates at low Rayleigh number,
we were able to obtain analytical solutions for a flow model. Then, in the context of the more
complicated flow associated with mantle convection, we could only use scaling analysis and highly
simplified models. But there are certain (interesting) areas where, with suitable assmptions, we
can obtain analytical results about the flow of mantle convection: at mid-ocean ridges and
subduction zones. This class of analytical solutions to the incompressible Stokes equations are
known as “corner-flow solutions.”1

9.1 Setting up the corner-flow problem

Recall that incompressible Stokes flow is by

∇· T + ρg = 0, (9.1a)
∇· u = 0, (9.1b)

where T is the viscous stress tensor, given by

T = −pI + 2ηE. (9.2)

Recall that E is the strain-rate tensor. Substituting equation (9.2) into (9.1a) and expanding
the strain-rate tensor gives, for the momentum equation,


∇p = ∇· η ∇u + ∇uT + ρg. (9.3)

Note that we have kept the viscosity within the divergence because, as we learned in the chapter
about plate tectonics, viscosity variations are of fundamental importance.
Equation (9.3), with η and ρ variable in space, could be used to model mantle convection,
with suitable constitutive laws to describe those variations. However, for most situations, this
would preclude analytical solutions. Hence we need to make simplifications.

Constant viscosity Plates are cold and hence have high viscosity, while the asthenosphere
is hot with low viscosity. We have previously interpreted this to mean that viscosity variations
are important. But the same statement points toward a simplification: let’s assume that the
“plate” has a uniform, very high viscosity and that the asthenosphere has a uniform, lower
viscosity. This tells us that the plate is too strong to deform and so it moves as a coherent unit;
in contrast, the asthenosphere can undergo shearing flow, which we’ll need to compute. We
have collapsed the temperature-dependent viscosity variation of the mantle to a simple, binary
distinction between plates and asthenosphere.

Constant density Plates are cold and hence have high density relative to the hot astheno-
sphere. We previously interpreted this to mean that density variations are important, but in
the context of the mid-ocean ridge (where the lithosphere doesn’t sink into the mantle), we
can simplify the problem. We shall assume that the motion of tectonic plates is fixed by the
large-scale, mantle convective flow, which is driven by density difference, but that locally 2 , we
can assume that the plate motions are given and treat them as boundary conditions on the
asthenosphere.
Treating the viscosity and density as constants and using incompressibility, we can define
the dynamic pressure as
∇P = ∇p + ρg (9.4)
1
For more details see: Spiegelman & McKenzie (1987) Simple 2-D models for melt extraction at mid-ocean
ridges and island arcs, EPSL, 83, 137–152.
2
Here, “local” means nearby a plate boundary. So say within a few hundred kilometres of the boundary.
102 Kinematics and Dynamics of the Lithosphere 2020

and simplify (9.3) to obtain the system,

∇P = η∇2 u, (9.5a)
∇· u = 0, (9.5b)

which will apply only within the asthenosphere, not within the plates. Since the flow is considered
to be incompressible, we can use a streamfunction ψ, which we define by
∂ψ ∂ψ
u = ∇× (ψẑ) = x̂ − ŷ. (9.6)
∂y ∂x
Using manipulations introduced in earlier chapters, the system (9.5) reduces to


∇2 ∇2 ψ = 0. (9.7)

This is the equation that must be solved for ψ, subject to the boundary conditions representing
the plate motions.

9.2 Mid-ocean ridges

We first consider the case of a mid-ocean ridge. We define a domain as in Figure 9.1(a) to be a
semi-infinite half space (all x but only y < 0) and define a cylindrical coordinate system with θ
the angle measured counter-clockwise from −ŷ. Clearly there is a mirror symmetry across the
y axis, and hence we can focus attention on the flow for θ > 0.

3p
u3 (r; :=2 ! 3p )
U0 U0
3
e re
h o sph e re
U0
li t sph 3p
o
r

e n
as t h
(a) (b) ur (r; :=2 ! 3p )

Figure 9.1: Mid-ocean ridge set-up for the corner-flow problem with θp = 25◦ . (a) Schematic of domain
and cylindrical coordinate system. Dashed line shows the symmetry plane. (b) Plate motion vector
decomposed into its components in the r̂ and θ̂ directions on the base of the lithosphere, where θ =
π/2 − θp .

The lithosphere can be approximated as a rigid plate with a base that makes an angle θp to
the surface. In our cylindrical coordinate system, the plate motion can be conveniently assigned
as a boundary condition at constant θ = π/2 − θp for all r. Resolving the vector into cylindrical
coordinates (Fig. 9.1(b)),
 
u(θ = π/2 − θp ) = U0 cos θp r̂ + sin θp θ̂ . (9.8)

Symmetry across the y axis provides two more boundary conditions as

∂ur
uθ = 0, = 0 at θ = 0. (9.9)
∂θ
We now require a general solution to (9.7) in cylindrical coordinates where
 
2 1∂ ∂ 1 ∂2
∇ = r + 2 2. (9.10)
r ∂r ∂r r ∂θ
Fortunately, this is available in terms of the separable solution

ψ = U0 rf (θ), (9.11)
Flow at Oceanic Ridges and Subduction Zones 103

where f (θ) is a function of theta only that is given by

f (θ) = C1 sin θ + C2 θ sin θ + C3 cos θ + C4 θ cos θ, (9.12)

where Ci are unknown constants, to be deterimined using the boundary conditions. In cylindrical
coordinates, the relationship between streamfunction and velocity is given by
1 ∂ψ ∂ψ
u= r̂ − θ̂. (9.13)
r ∂θ ∂r
Substituting (9.11) into (9.13) gives
h i
u = U0 f 0 (θ)r̂ − f (θ)θ̂ . (9.14)

Using equation (9.14) we can re-write the boundary conditions as

f (0) = 0, (9.15a)
00
f (0) = 0, (9.15b)
f (π/2 − θp ) = − sin θp , (9.15c)
0
f (π/2 − θp ) = cos θp . (9.15d)

The first of these tells us that C3 = 0; the other three can be used to obtain C1 , C2 , C4 with
some algebra. The result is

2 sin2 θp −2
C1 = , C2 = C3 = 0, C4 = . (9.16)
π − 2θp − sin 2θp π − 2θp − sin 2θp

Thus we obtain the required solution,

2U0 r  2 
ψ= sin θp sin θ − θ cos θ . (9.17)
π − 2θp − sin 2θp

Contours of this function are plotted in Figure 9.2 along with contours of pressure. Dynamic
pressure P is computed by substituting (9.17) and (9.13) into (9.5) and integrating. The pressure
goes to −∞ at r = 0 and is sharply negative in a region around this point. This is because the
imposed plate divergence creates an infinite stress at the origin.

U0 U0

Figure 9.2: Results of the mid-ocean ridge corner-flow model with θp = 25◦ . Black lines are contours
of the streamfunction ψ from equation (9.17). Grey lines are contours of pressure. Pressure becomes
increasing negative toward the ridge axis. At r = 0 there is a negative singularity in the pressure.

Examine the pattern of flow in Figure 9.2. Note the shape of the region of upwelling: it
is roughly triangular. Also note that the sides of that triangle slope downward more steeply
than the lithosphere itself. Where is melt produced? How might this picture be different if the
lithosphere had a boundary that was proportional to x1/2 ?
104 Kinematics and Dynamics of the Lithosphere 2020

9.3 Subduction zones

The domain for the subduction zone corner-flow model is shown in Figure 9.3(a). The overlying
lithosphere is prescribed to have zero velocity while the downgoing slab has a dip θp and a speed
U0 . No-slip and no-penetration boundary conditions at the bottom of the lithospheric lid (θ = 0)
and along the top of the slab (θ = θp ) are given by

uθ = 0 at θ = 0, θp , (9.18a)
ur = 0 at θ = 0, (9.18b)
ur = U0 at θ = θp . (9.18c)

We seek a solution to (9.7) with boundary conditions (9.18) within the domain 0 ≤ r < ∞, 0 ≤
θ ≤ θp .

lithosphere
3
asthenosphere
U0 U0
r
re
he
sp
ho
lit

(a) (b)

Figure 9.3: Subduction zone domain and results with θp = 45◦ . (a) Schematic diagram of the subduction
zone domain, showing lithospheric and asthenospheric regions and motion of the downgoing slab. A plane
polar coordinate system is shown with θ measured counter-clockwise from the −x̂ direction. (b) Black
lines are linearly spaced contours of the streamfunction ψ from eqn. (9.21). Grey lines are logarithmically
spaced contours of the dynamic pressure, which is increasingly negative for decreasing r and has a negative
singularity at r = 0.

The general solution (9.11) again applies, with f given by (9.12). Hence the problem becomes
determination of the constants Ci using the boundary conditions. Expressing the boundary
conditions in terms of f we have,

f (0) = 0, (9.19a)
0
f (0) = 0, (9.19b)
f (θp ) = 0, (9.19c)
0
f (θp ) = 1. (9.19d)

Combining these conditions with (9.12) gives

θp sin θp θp cos θp − sin θp
C1 = , C2 = < 0, C3 = 0, C4 = −C1 . (9.20)
θp2 − sin2 θp θp2 − sin2 θp

Hence the streamfunction solution is written as

U0 r
ψ= [θp sin θp (sin θ − θ cos θ) + (θp cos θp − sin θp )θ sin θ] . (9.21)
θp2 − sin2 θp

Contours of this function are plotted in Figure 9.3(b). We can again determine the dynamic
pressure by substituting (9.21) and (9.13) into (9.5) and integrating. The pressure goes to −∞
Flow at Oceanic Ridges and Subduction Zones 105

at r = 0 and is sharply negative in a region around this point. This is because the discontinuity
in the boundary conditions creates an infinite stress at the origin.
Examine the pattern of flow in Figure 9.3(b). Where is there upwelling? Is it strong relative
to that of the mid-ocean ridge model? Where are the fastest velocities? Are they upward or
downward?
Mantle Convection and the Lithosphere 107

10 Temperatures in Oceanic and Continental Lithosphere

10.1 Notes on the Solution of the 1D diffusion-advection equation
Our motivation is to interpret observations of the depth of the ocean floor, and of oceanic heat
flux in order to estimate the thickness of plates. The tools we develop here will also be used to
investigate sedimentary basins and regional metamorphism.
The energy equation is:
∂T
+ u · ∇T = κ∇2 T. (10.1)
∂t
We non-dimensionalize this equation using a characteristic length scale l, a characteristic
temperature, Tc , and a characteristic time l2 /κ:
T = Tc T 0 (10.2)
xi = l x0i (10.3)
l2
t = t0 (10.4)
κ
κ 0
u = u (10.5)
l
Eq. (10.1) becomes
∂T
+ u · ∇T = ∇2 T, (10.6)
∂t
where u, xi , T , and t are all dimensionless, and we have dropped the primes.

10.1.1 The Péclet Number

Although the adoption of a time scale and a length scale in the non-dimensionalisation that leads
to Eq. (10.6) is sufficient to define a velocity scale (κ/l) it is more natural, for many problems,
to recognise that the velocity scale is set independently of thermal parameters. For example,
we shall consider problems in which the velocity is that of plate motion, or of erosion. If we
introduce a velocity scale uc Eq. (10.6) becomes
∂T
+ Pe u · ∇T = ∇2 T, (10.7)
∂t
where Pe is the Péclet number
uc l
Pe = .
κ
Pe, the Péclet number expresses the relative importance of advection and diffusion. It can
be thought of as the ratio of two time scales:
l2 uc
Pe = . (10.8)
κ l
Convince yourself that you understand what those time scales are.
It is instructive to examine the relative importances of diffusion and advection of heat by
considering a simple situation in which the temperature varies in only one dimension. We imagine
layer 0 < z < l, moving at a constant speed −U towards the boundary z = 0. We further suppose
that this motion has occurred for long enough that the temperature is independent of time. The
dimensional diffusion-advection equation (10.1) becomes
d2 T dT
κ 2
= −U . (10.9)
dz dz
which has the solution  
Uz
T (z) = A exp − +B (10.10)
κ
108 Kinematics and Dynamics of the Lithosphere 2020

Let us impose the conditions T (0) = 0, T (l) = T/T0

T0 , then 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0.0
 
1 − exp (−U z/κ) 0.1 10.0
T (z) = T0 (10.11)
1 − exp (−U l/κ)
0.2
or  
1 − exp (−Pe z 0 ) 0.3
T 0 (z) = (10.12) 3.0
1 − exp (−Pe)
0.4
where we have written

x/l
0.5 0 1.0
Ul 0z
Pe = ; z = . (10.13) 0.6 0.3
κ l
0.7
This solution is illustrated in Figure 10.1. Ex-
amine the structure of this solution as Pe → 0 0.8
and for large Pe. You can see that as Pe → 0,
0.9
the temperature profile approaches the purely con-
ductive solution. For large Pe the temperature is 1.0
close to T0 everywhere except in a layer near the
Figure 10.1: Solution to Equation (10.12). La-
surface z = 0, whose dimensionless thickness is bels on the curves show the value of Pe and,
∼ 1/Pe or, equivalently, whose dimensional thick- with the exception of the label for Pe= 0, are
ness is ∼ κ/U . centred on the places where T 0 = 1 − 1/e.

10.1.2 Thermal diffusion in the solid bounded by two parallel planes (a slab)
The thermal definition of the lithosphere is the region
that transfers heat solely by conduction. This defini-
tion implies that it is appropriate to use the diffusion
equation only down to some depth (see for example Sec-
tion 10.2). Here we give the means of solving the 1-D
diffusion equation in a solid bounded by two parallel
planes (representing the top and bottom of the plate).
These solutions, and many others relevant to the Earth
Sciences, can be found in Carslaw and Jaeger (1959).
Suppose that the temperature at the base of the
plate, of thickness l, is held at a constant value, Tm .
The steady state solution to this problem (neglecting
heat production) is obtained by setting the LHS of the
1D energy equation (10.1) to zero and integrating twice:

d2 Ts
= 0 (10.14) Figure 10.2: Sketch to define steady and
dz 2 unsteady temperatures in a slab
Ts = az + b (10.15)
z
Ts = T0 + (Tm − T0 ) , (10.16)
l
where T0 is the temperature on z = 0 which, hereafter,
we set to zero.
The trick with most thermal problems concerning the lithosphere is to recognise that you
can always write the temperature as the sum of two temperatures: the first is the steady-state
temperature Ts , and the second is the unsteady temperature Tu , which is the difference between
the actual temperature at any time, and the steady-state temperature (Figure 10.2). Tu must
obey the 1D diffusion equation
∂T ∂2T
=κ 2 (10.17)
∂t ∂z
Mantle Convection and the Lithosphere 109

which we solve by separation of variables.

Let
Tu (z, t) = Θ(t)Z(z), (10.18)
then substitution into (10.17) gives

dΘ(t) d2 Z(z)
Z(z) = −c2 = κΘ(t) ,
dt dz 2
1 dΘ(t) 1 d2 Z(z)
= −c2 = κ , (10.19)
Θ(t) dt Z(z) dz 2
to give the solutions:

Θ(t) = exp (−c2 t), (10.20)

   
cz cz
Z(z) = A sin √ + B cos √ . (10.21)
κ κ
(10.22)

The boundary conditions on Tu are that it be equal to zero both at z = 0 and z = l (Figure 10.2).
The condition on z = 0 requires that B = 0. The condition on z = l requires that
√
nπ κ
c= (10.23)
l
Thus
 nπz 
Z(z) = sin (10.24)
l 2 2 
−n π κt
Θ(t) = exp (10.25)
l2
The general solution for the unsteady temperature Tu is, therefore:
∞
X  nπz   2 2 
−n π κt
Tu (z, t) = An sin exp (10.26)
l l2
n=1

Where Z  
l
2 0

nπz 0
An = Tu z , 0 sin dz 0 (10.27)
l 0 l
For example, in the oceanic ridge problem (below),

Tu (z, 0) = Tm (1 − z/l) (10.28)

so
2Tm
An = (10.29)
nπ
The key things to note about solutions of this sort are:
• Each term looks like a sine wave in z multiplied by a term that decays exponentially with
time.

• The argument of the exponential decay term varies as n2 , thus the terms with smaller
wavelengths (higher n) decay MUCH faster than the terms with longer wavelength.

• In fact, the decay time for the term that has wavelength l/(nπ) is tn = l2 /(n2 π 2 κ). As
you can see from Figure 10.3, after 10 million years, only the first two terms matter, and
after 30 million years, only the first term matters.
110 Kinematics and Dynamics of the Lithosphere 2020

1600 1600

1400 1400

1200 1200
Temperature oC

Temperature oC
1000 1000

800 800

600 600

400 400

200
1 Myr 3 Myr 200

0 0
0 25 50 75 100 125 0 25 50 75 100 125
Depth (km) Depth (km)

1600 1600

1400 1400

1200 1200
Temperature oC

Temperature oC
1000 1000

800 800

600 600

400 400

200
10 Myr 30 Myr 200

0 0
0 25 50 75 100 125 0 25 50 75 100 125
Depth (km) Depth (km)

Figure 10.3: Successive approximations to the full solution given by (10.27) with An = 2/nπ. (This is
the oceanic ridge problem.) Curves show the influence of evaluating the series with increasing numbers
of terms. The double line corresponds to evaluating only the first term in the series; the line with one
maximum and one minimum evaluates only the first two terms, and so on. The black lines correspond
to the full solutions.

10.2 Oceanic Heat Flux and Bathymetry

For the cooling of the ocean floor, we treat the mantle as being at a uniformly high temperature
at the ridges, subsequently having its surface held at a constant temperature (through contact
with the water column). We choose a coordinate system with x parallel to the direction of plate
motion and z vertical. The velocity is (uc , 0, 0), and we may neglect variation in the y-direction
so Eq. (10.7) becomes
∂T ∂2T ∂T ∂2T
= − Pe + (10.30)
∂t ∂x2 ∂x ∂z 2
If Pe is large, then the horizontal heat transfer is dominated by the advection term, and we
may choose a reference frame fixed to the moving plate. Eq. (10.30) then reduces to the one-
dimensional diffusion equation
∂T ∂2T
=κ 2 (10.31)
∂t ∂z

Exercise for the student: Calculate the Péclet number for a typical oceanic plate, hence
satisfy yourself that the horizontal diffusion of heat is negligible in comparison with horizontal
advection.
Mantle Convection and the Lithosphere 111

10.2.1 Half-space solution

Initially, we consider the half-space z > 0, described by Eq. (10.31). The initial condition is

T (z, 0) = Tm (10.32)

where Tm is the mantle temperature, and the boundary conditions are

T (0, t) = 0 (10.33)
T (∞, t) = Tm (10.34)

Because we are considering a half-space (semi-infinite medium) there is no natural choice of

length scale. We know, however, that whatever solution we obtain must be expressible in
dimensionless form. In other words, there must be some quantity having the dimensions of
length
√ that will appear in the solutions as the divisor of z. The only possibility for that length
is κt, so we seek a solution that depends on
z
z0 = √ . (10.35)
2 κt

Because z 0 contains both dimensions of Eq. (10.31) (time and the one space dimension) we might
expect that this non-dimensionalisation will lead to an ordinary differential equation. And it
does:
 2 2
∂2T 1 d T
= √
∂z 2
2 κt dz 02
∂T z 0 dT
= −
∂t 2t dz 0
2
d T dT
02
= −2z 0 0 (10.36)
dz dz

   
1 d dT
= −2z 0
dT / dz 0 dz 0 dz 0
dT 02
0
= Ae−z
dz
Z z0
0 2
T (z ) = A e−v dv + B (10.37)
0

The first boundary condition given by 10.33 is equivalent to

T (z 0 ) = 0, z 0 = 0

which requires that B = 0. The second boundary condition (Eq. 10.34) is equivalent to

T (z 0 ) → Tm , z 0 → ∞

Note that the initial condition also requires T (z 0 ) → Tm , z 0 → ∞, so is automatically satisfied.

These conditions require that Z ∞
2
Tm = A e−v dv
0

Making use of the result that Z √

∞
−v 2 π
e dv =
0 2
112 Kinematics and Dynamics of the Lithosphere 2020

the constant A is given by

2Tm
A= √
π
The solution is therefore given by
Z z
√
2Tm 2 κt 2
T (z, t) = √ e−v dv (10.38)
π 0

The error function erf(x) is defined by

Z x
2 2
erf(x) = √ e−v dv
π 0

so that the solution 10.38 of the heat diffusion equation can be expressed as
 
z
T (z, t) = Tm erf √ , (10.39)
2 κt
and the surface heat flux is
kTm
q=√ . (10.40)
πκt
It is observed that heat flux through the ocean floor decreases as the square root of its
age (Figure 10.4) and that depth of the ocean floor increases as the square root of its age
(Figure 10.7).

Figure 10.4: (Left) Variation of oceanic heat flux with age (Parsons and Sclater , 1977); solid line shows
the square-root of age dependence predicted by Eq. (10.40). (Right) Blow-up of reliable heat flow mea-
surements in ocean floor older than 100 Myr B (Jaupart and Mareschal , 2007); solid line shows the
square-root of age dependence predicted by Eq. (10.40).

Although the observations of heat flux agree well with this model, there are small discrepan-
cies for ocean floor older than 100 Myr (Figure 10.4, right panel). The discrepancies are much
more clear, however, for the depth of the ocean floor.
Mantle Convection and the Lithosphere 113

10.2.2 Depth of the Ocean Floor

The condition of isostatic balance gives us an expression for the
subsidence w of the ocean floor relative to the depth at the mid-
ocean ridge (Figure 10.5). Set the level z = 0 at the top of the
old ocean floor and define the density of the mantle as

ρ(z) = ρ0 (1 − α T (z)) (10.41)

Then

M1 = M2
Z ∞ Z ∞
ρm dz = wρw + ρ(z) dz Figure 10.5: Definition sketch
−w 0
Z ∞ for isostatic balance of oceanic
(ρm − ρ(z)) dz = w (ρw − ρm ) ridge and ocean floor.
0
Z ∞
ρ0 α
w = (Tm − T (z)) dz (10.42)
(ρm − ρw ) 0
Z ∞  
ρ0 α Tm z
= 1 − erf √ dz (10.43)
(ρm − ρw ) 0 2 κt
The expression in parentheses within the integral defines the complementary error function
erfc(x) given by

erfc(x) = 1 − erf(x)
Z ∞
2 2
= √ e−v dv
π x

Using this result in the expression 10.43 for the subsidence w gives
Z ∞ Z ∞
2ρ0 αTm 2
w= √ dz e−v dv (10.44)
(ρm − ρw ) π 0 z
√
2 κt

By changing the order of integration as shown in the

figure, the integration over z can be carried out first in
a straightforward manner,
Z ∞ Z (2√κt)v
2ρ0 αTm 2
w = √ e−v dv dz
(ρm − ρw ) π 0 0
r Z ∞
4ρ0 αTm κt 2
= ve−v dv
(ρm − ρw ) π 0
r " #
2 ∞
4ρ0 αTm κt e−v
= − Figure 10.6: Domain of integration of the
(ρm − ρw ) π 2
0 integral in expression 10.44.
leading to the result
r
2ρ0 αTm κt
w= (10.45)
(ρm − ρw ) π

As for the heat flux, the bathymetry agrees with the half-space model out to about 80 Myr
(Figure 10.7), but the departures from the square-root of age dependence beyond this age are
clear (Figure 10.14).
114 Kinematics and Dynamics of the Lithosphere 2020

Figure 10.7: Median sediment-corrected depth of the ocean floor as a function of the square root of
median age within bins of 1 Ma1/2 for different oceans. The grey bars show the interquartile range. The
best-fitting straight lines were calculated by minimizing the weighted rms misfit to the data, where the
weight is the inverse of the interquartile range.(Crosby and Mc Kenzie, 2009)
Mantle Convection and the Lithosphere 115

Figure 10.8: Reference depth versus the square root of age in the Pacific and Atlantic Oceans. (Crosby
et al., 2006)
116 Kinematics and Dynamics of the Lithosphere 2020

10.3 The thermal plate model 1400

We have so far solved the 1-D heat equation imag- 1200

ining that the bottom of the region of interest lies
so far away from the surface that it effectively has 1000
no control on thermal structure near the surface.

Temperature oC
10 30
100
We have now seen, from data on the subsidence 800 150
of the ocean floor, that the cooling of the oceanic
lithosphere appears to slow once it passes the age 600
of approximately 80 million years. One explana-
tion for this behaviour is that some process – most 400
plausibly convection in the mantle – is supplying
heat to the base of the oceanic plates. 200
Once we invoke the idea of a base to the sys-
tem we are looking at, then we must apply a second 0
boundary condition to the system – corresponding 0 25 50 75 100 125

to whatever is going on at its base. We will investi- Depth (km)

gate the simplest such condition: that the temper- Figure 10.9: Comparison between the half-
ature at the base of the plates is held at a constant space solution for the cooling ocean floor and
value, Tm , and we shall suppose that the plate has the plate-model solution. Double lines are the
a thickness l. half-space solution; black lines are the plate
The solution to this problem is given by model (only distinguishable at 100 Myr). The
straight line corresponds to the steady state so-
Eq. (10.27) (Section 10.1.2), with
lution Ts .
Z l
2 z  nπz  2 Tm
An = 1− sin dz = (10.46)
l 0 l l nπ
∞  nπz   2 2 
2 Tm X 1 −n π κt
Tu (z, t) = sin exp (10.47)
π n l l2
n=1

which is compared with the half-space solution in Figure 10.9. Note that the two solutions for
temperature differ little until the age of the ocean floor exceeds about 100 Myr. For this reason,
and because measurements of heat flux depend on measuring temperature gradients in the upper
few tens of metres of the ocean floor, heat flux does not provide a sensitive test of the plate
model.
On the other hand, the depth of the ocean floor, which is a measure of the integral of
temperature difference through the lithosphere, is a much more sensitive discriminant between
the plate and half-space models (Figure 10.10).
In the same way that we calculated the depth of the ocean floor as a function of its age for
the half-space model (Eq. 10.43), we may calculate the difference in height W between ocean
floor of age T and ocean floor whose temperature is in steady state. We consider ocean floor
older than a few tens of million years, for which we may take only the first term in the series in
Eq. (10.47)
Z l
ρ0 α
W = (T (z, t) − Ts ) dz (10.48)
(ρm − ρw ) 0
Z l
ρ0 α
= Tu (z, t) dz (10.49)
(ρm − ρw ) 0
 2 Z l  πz 
2 ρ0 α T m −π κt
= exp sin dz (10.50)
π(ρm − ρw ) l2 0 l
 2 
4 ρ0 α Tm l −π κt
= 2
exp (10.51)
π (ρm − ρw ) l2
Mantle Convection and the Lithosphere 117

Exercise for the student: Suppose that the “flattening” observed in Figure 10.14 results
not from heating of the lithosphere, but from the pressure gradient required to drive the return
flow that we discussed in the Fluid Mechanics section of the course. What combinations of
asthenospheric viscosity and channel thickness are permitted by the observations?

Exercise for the student: Use Figure 10.10 to estimate Tm and the thickness of the plates,
l. Assume α = 4 × 10−5 K−1 ; κ = 8 × 10−7 m2 s−1 .

Figure 10.10: Sketch of the way in which plate parameters are estimated from the depth-age curve. The
√
slope of the bathymetry on the depth- age plot is proportional to the temperature at the base of the
√ √
plate. The bathymetry departs measurably from the age-dependence when 3 κt ∼ l (Parsons and
Sclater , 1977).

10.4 Onset of convection in a fluid cooled from the top

We now see (from data on the subsidence of the ocean floor) that the oceanic lithosphere appears
to stop cooling once it passes the age of approximately 80 million years. Parsons and Mc Kenzie
(1978) pointed out that the strong temperature-dependence of the viscosity of mantle rocks
means that we should consider the cooling ocean floor as consisting of a mechanical boundary
layer, cooler than a temperature T1 , and a thermal boundary layer, in the temperature range
T1 < T < Tm . The mechanical boundary layer is too strong to convect, but the thermal
boundary layer can become involved in convection if the conditions are favourable.
Let us consider the Rayleigh number of this boundary layer:
gρα (Tm − T1 ) β 3
Ra = (10.52)
κη
118 Kinematics and Dynamics of the Lithosphere 2020

√
where β ∝ κt (by Eq. (10.39).
√ T1 is ∼ 3/4Tm (temperatures measured with respect to a surface at
A reasonable choice for
T = 0), which gives β ∼ 2κt, hence
p
gραTm t κt/2
Ra ∼ (10.53)
η

Exercise for the student: Test the plausibility of this idea by using the time t =80 Myr for
the breakdown of the square-root-of-age dependence in the bathymetry to estimate the viscosity
of the thermal boundary layer required for it to reach a critical Rayleigh number.

Figure 10.11: (Top) Sketch of a longitudinal section of a convection experiment. (Bottom) Photograph
showing the flow in a transverse section through line B in the upper panel (from Parsons and Mc Kenzie
(1978)).

This discussion introduces the idea of another scale of convection in the mantle: small-scale
flow due to convective instability near the base of the lithosphere, where both temperature and
viscosity exhibit significant vertical variations.
Mantle Convection and the Lithosphere 119

10.5 Convection and Dynamic Topography

We consider a simple model for convection in a fluid half-space (z ≤ 0); this consists of setting
up a temperature field and determining how it drives flow, but the non-linearity we alluded to
when discussing the onset of convection is circumvented by neglecting the influence of the flow
on the temperature field. Models such as this have a lengthy tradition in the study of convection
in the Earth; the one we discuss here was presented by (Mc Kenzie, 1977, and see Figure 10.12).
Consider a half-space with a temperature T (x, 0) = T0 cos (kx) imposed on its upper surface.
The internal temperature is
T (x, 0) = T0 cos (kx) exp (kz) . (10.54)

Recall that
ρ0 gα ∂T
∇4 ψ = − (10.55)
η ∂x
(sign correct for z downwards). After application of the usual means, which we have seen in
earlier Chapters, the solution when τxz = 0 is

ρ0 gαT0  2 z 
ψ= z − exp (kz) sin (kx) (10.56)
8ηk k

Figure 10.12: (a) Dimensionless temperature and (b) stream function as a function of depth. Dimension-
less temperatures are calculated from Equation 10.54. Note that upwelling flow occurs over the centre,
where temperature is high, and downwelling where it is low.
120 Kinematics and Dynamics of the Lithosphere 2020

We are interested in the deflection of the surface due to the convection. Recall, from our
discussion of post-glacial recovery, that we equate the hydrostatic pressure ρgh associated with
the topography h to the vertical normal stress at the upper boundary of the fluid half-space,
y = 0. Then
∂uz
ρg∆h = Tzz = −p + ηEzz = −p + 2η . (10.57)
∂z
from Equation 10.56
∂uz ∂2ψ ρ0 gαT0
2η = −2η = cos (kx) (10.58)
∂z ∂x∂z 4k
and the dynamic pressure on z = 0 is
ρ0 gαT0
p= cos (kx) (10.59)
2k
Hence
3αT0
∆h = cos (kx) (10.60)
4k
We state the expression for the resulting gravity anomaly (Mc Kenzie, 1977) without deriving
it:
π G ∆ρ α T0
∆g(x) = cos (kx) , (10.61)
2k
where ∆ρ is the difference in density between the convecting fluid and what overlies it (water
or air).
In the limit of long wavelength (kh  1), we may approximate the part of the gravity
anomaly due to the variation of the surface height, alone, by the attraction of a slab of thickness
∆h and density ∆ρ is(the Bouguer formula):

∆g(x) = 2π ∆ρ G ∆h(x), (10.62)

3π G ∆ρ α T0
= cos (kx) . (10.63)
2k
This anomaly is exactly three times the anomaly given by Eq. (10.61). This results arises because
the gravity anomaly due to surface height variations is counteracted by larger (positive)/negative
contributions to gravity from the (colder, sinking)/hotter rising fluid.
The ratio, Z, of gravity anomaly to surface height is independent of the wavelength of the
temperature field and depends only on ∆ρ, the difference in density between the convecting fluid
and what overlies it.
∆g(x) 2 π G ∆ρ
Z= = . (10.64)
∆h(x) 3
If air overlies, then Z evaluates to about 50 mGal per kilometre of surface height; if water, then
Z evaluates to about 30 mGal/km.
Figure 10.13 shows a free-air anomaly map of the world; note that, in almost all places on
Earth, except near subduction zones, the free-air anomaly is smaller than 15 mGal, suggesting
that support of topography by convection is no more than about 500 m (Figure 10.13).
A detailed analysis of the gravity anomalies and topography in the Pacific (Crosby et al.,
2006) reveals a correlation between surface height and gravity which suggests that some of the
variations in surface height are supported by convection (Figure 10.14). Again, note that the
majority of observations suggest that the support of topography by convection is limited to
± ∼ 500 metres.

Exercise for the student If the gravity anomalies thought to be due to convection are
generally no greater than 15 mGal, estimate the associated temperature variations in the upper
mantle.
Mantle Convection and the Lithosphere 121

210˚ 240˚ 270˚ 300˚ 330˚ 0˚ 30˚ 60˚ 90˚ 120˚ 150˚ 180˚ 210˚
75˚ 75˚
60˚ 60˚
45˚ 45˚
30˚ 30˚
15˚ 15˚
0˚ 0˚
−15˚ −15˚
−30˚ −30˚
−45˚ −45˚
−60˚ −60˚
−75˚ −75˚
210˚ 240˚ 270˚ 300˚ 330˚ 0˚ 30˚ 60˚ 90˚ 120˚ 150˚ 180˚ 210˚

−45 −30 −15 0 15 30 45

Free−air Anomaly mGal
Figure 10.13: Free-air gravity anomalies band-passed in the wavelengths between 500 and 6000 km.
Triangles show the locations of proposed mantle plumes.
122 Kinematics and Dynamics of the Lithosphere 2020

Figure 10.14: (Top) Gravity anomalies in the North Pacific ocean. Black polygons show regions in which
the topography was excluded (seamounts, plateaus, subduction zones, certain fracture zones and flexural
moats). (Bottom) Plots of observed gravity versus sediment-corrected topography from the north and
equatorial Pacific (regions excluded as in top panel). (Crosby et al., 2006).
Mantle Convection and the Lithosphere 123

10.6 Sedimentary Basins and Metamorphism

10.6.1 Subsidence of sedimentary basins
We consider a simple configuration in which the
continental lithosphere is instantaneously stretched
horizontally by a factor, β and, therefore, thinned
by the same factor (Figure 10.15). We shall con-
sider only the case in which the strain is uniform;
other models, in which the strain is assumed to
vary with depth, can be treated by simple exten-
sion of the methods described here.
We neglect radiogenic heating and assume
that, immediately before time T = 0, the
temperature in the lithosphere is Tm (1 − z/l)
where Tm is the temperature of the up-
per mantle at the base of the lithosphere,
which has thickness l, and z is measured upwards
from the base of the lithosphere. At time t = 0
the lithosphere is extended horizontally by a fac-
tor β. The density structure is changed, and iso-
static compensation causes upwelling of hot as-
thenosphere (Figure 10.15). This causes an instan-
taneous change in surface height:
h i Figure 10.15: Sketch to show the principal
  t (ρ − ρ ) 1 − tc αTm
− aρ0 αTm features of the subsidence model (Mc Kenzie,
1 c 0 c 2l 2
Si = 1 − 1978).
β ρm − ρfill
(10.65)
where ρc , ρ0 , and ρm are respectively the density of
crust and mantle at 0◦ C and the density of mantle
at temperature Tm , and tc is the initial thickness of the crust. ρfill is the density of what is
deposited on top of the crust during stretching (air, water, or sediment).
Because the temperature of the rocks remains unchanged during the extension, the new
temperature structure is
 
z 1
T = Tm 0< < 1−
l β
   
z 1 z
T = Tm β 1 − , 1− < <0
l β l
 
z z 1
Tu = Tm 0< < 1− (10.66)
l l β
  
z 1 z
Tu = Tm (β − 1) 1 − , 1− < <0 (10.67)
l β l
As the temperature perturbation, Tu , decays back towards its steady-state condition, given by
Equation 10.66, the density of the stretched crust and upper mantle increases, and the original
land surface subsides. With a small algebraic difference, this problem is the same as the cooling
of the oceanic plates. We find the solution for the decay of the thermal perturbation by the
same techniques as are described in Section 10.1.2, obtaining an infinite series that we hope
(justifiably as it turns out) will be adequately described by its first term alone:
∞
X  nπz   2 2 
−n π κt
Tu (z, t) = An sin exp (10.68)
l l2
n=1
124 Kinematics and Dynamics of the Lithosphere 2020

Where Z    
a
2 0

nπz 0 0 (−1)n+1 β nπ
An = Tu z , 0 sin dz = sin (10.69)
l 0 l n nπ β
Again, as in the oceanic case, one obtains the subsidence by integrating the temperature per-
turbation through the thickness of the lithosphere. Mc Kenzie (1978) gives the full expression,
which is another infinite series, but it is of interest to us to consider only the first term in the
series, which gives an excellent approximation to the full solution. It is useful to express the
subsidence in terms of the quantity e(t), which is the elevation of the surface of the pre-stretching
crust above the final depth to which it sinks:
     2 
4lρ0 αTm β π π κt
e(t) ∼ 2 sin exp − 2 (10.70)
π (ρ0 − ρfill ) π β l

Note that, while the magnitude of the subsidence depends on β, the time scale is always l2 /π 2 κ ∼
60 Myr (Figure 10.16).
A useful rule of thumb can be obtained by noting that with l = 125 km, and with the fill
being water (ρfill = ρw = 1000 kg/m3 ):

4lρ0 αTm
2
∼ 3.2km
π (ρ0 − ρw )

So for your favourite basin, filled with sediment of density ρs , on a lithosphere of thickness L,
you will expect the total subsidence for a stretching factor of β to be
    
(ρ0 − ρw ) β π L
E = 3.2 km sin (10.71)
(ρ0 − ρs ) π β l

where ρw is the density of water. This expression is a useful guide as to how uncertainties in
model parameters feed into uncertainties in subsidence (see Figure 10.16).
5.0
1.2

4.5
3.6

4.0
3.2

3.5
2.8

τ (Myr)
β

3.0
2.4

2.5
2
1.
6

2.0
1.2

1.5 0.8

0.4

1.0
50 75 100 125 150 175 200 225 250
L (km)

Figure 10.16: (Left:) Log10 e(t) where e(t) is in metres (Mc Kenzie, 1978). The lines for different values
of β are obtained from the full solution; the arrows show the straight lines fitted to the full solutions for
T > 20 Myr intersect the t = 0 axis. (Right:) Contours show total subsidence in kilometres as a function
of lithospheric thickness, L, and β for a basin filled with water (Eq. (10.71)). Red line shows the time
constant, τ = L2 /(π 2 κ), right-hand scale, as a function of L.
Mantle Convection and the Lithosphere 125

10.7 Regional metamorphism

We are motivated by the observation that the pressure-temperature paths followed by rocks that
are involved in regional metamorphism often have a characteristic form (Figure 10.17) with a
phase of heating at high pressure, which is followed by approximately isothermal decompression
then cooling plus decompression.

Figure 10.17: (Left:) Pressure-temperature-time (PTt) trajectories for metamorphic realms in New Eng-
land (M. Brown, J. Geol. Soc., London,. 150, 1993, pp. 227-241 ). (Right:) P-T-t history for kyanite-
bearing pelites from the migmatite core of Naxos. 1 and 2 represent the beginning and end of the
deformational event that is inferred to represent crustal extension active from (at least) 25 to 12 Ma.
I. Buick and T. Holland, 1989; Geol. Soc. Lond. Spec. Pub. No. 43, pp 365–369. Pressures may be
converted to burial depth in kilometres by dividing by ρc g to give a conversion factor of about 3.6 km
per kilobar.

Figure 10.18 illustrates several different modes of crustal thickening and the temperature
perturbations associated with them. As we discussed in see Section 10.1.2, the evolution of
the temperature profiles may be assessed by identifying a steady-state temperature distribution
Ts (z), then calculating the decay of the unsteady temperature perturbation Tu , which is shown
by shading in Figure 10.18.
Because the crust is thickened, the geothermal gradient is lower initially than it would be
in steady state, Tu is negative, so temperatures everywhere increase with time. If nothing else
were to happen, then temperatures in the thickened crust would approach Ts on a time scale of
∼ l2 /(π 2 κ), or about 60 Myr.

10.7.1 Exhumation and termination of metamorphism

Thickening of the crust raises the surface height and two consequences, at least, of this process
can lead to exhumation and cooling of the buried rocks.

Erosion In general, regions of high elevation erode at rates of order a millimetre per year.
Recall the discussion of the Péclet number (Section 10.1.1) and consider the influence of diffusion
and advection on a rock buried initially at a depth of 30 km. The time scale for the exhumation
of that rock at an erosion rate of 1 mm/yr is 30 Myr, and for erosion at 3 mm/yr, the time
scale would be 10 Myr. These time scales are shorter than the time for the decay of the initial
temperature perturbation, so the heating is terminated as the rocks are brought towards the
cool land surface.
Eq. (10.11) suggests that a rock that is being brought towards the surface at speed U will
cool significantly once it is shallower than a depth ∼ κ/U . With κ ∼ 10−6 m2 s−2 , this depth is
30(10) km for U = 1(3) mm/yr.
126 Kinematics and Dynamics of the Lithosphere 2020

Figure 10.18: Sketches of different modes of crustal thickening and their temperature perturbations Tu
(England and Thompson, 1984). a), b) correspond to thickening by thrusting; e), f) correspond to
thickening by homogeneous strain; two cases are illustrated: when the whole lithosphere is thickened
(A) and when only the crust is thickened (C). All temperature scales are purely illustrative. Cross-
hatching shows the difference (Tu ) between the initial temperature distribution (Ti ) and the steady-state
temperature distribution that would be obtained if diffusion were allowed to continue indefinitely (Ts ).

Figure 10.19: Geotherms (dashed) in continental terrain at times of 20, 40, 60 and 80 Ma after over-
thrusting. Erosion of the pile above rock A lowers the pressure whilst heating is in progress (arrows) so
that maximum temperature is reached at 33 Ma. The array of Tmax − PT max points achieved by rocks
which subsequently reach the surface is shown as the metamorphic field array (solid line) which is labelled
with the times at which the rocks appear on surface (England and Richardson, 1977).
Mantle Convection and the Lithosphere 127

Extension of lithosphere We now turn to exhumation as a result of extension, supposing

that the extensional strain of the lithosphere causes uniform vertical thinning at a rate Ė, so
that the velocity of rocks, relative to the land surface, is −Ėz. The (steady-state) 1D diffusion-
advection equation becomes
dT d2 T
− Ėz =κ 2. (10.72)
dz dz
which has the solution
!
dT Ėz 2
= A exp −
dz 2κ
 s 
Ė 
T (z) = Tm erfz (10.73)
2κ

q
Hence, the rocks will cool significantly once they are shallower than a depth ∼ 2κ/Ė. Fig-
ure 10.17 (right panel) illustrates a PTt path from the island of Naxos, in the Aegean, which
underwent extension in middle-Miocene time. We do not know the extensional strain rate at
the time, but if we were to suppose that it were at a rate typical of today’s extension – about
3 × 10−15 s−1 – then the depth would be about 25 km, which is consistent with the observed
PTt path.
It therefore seems likely that erosion and/or exhumation by horizontal extension (vertical
thinning) of the crust during thermal relaxation of thickened crust accounts for the principal
features of many metamorphic PT paths.
128 Kinematics and Dynamics of the Lithosphere 2020

10.8 Temperatures at convergent plate boundaries

10.8.1 Temperatures in the subducting slab
We consider a model for temperatures in the
subducting slab (Figure 10.20). We take the
x-coordinate to be along the slab, and the z-coordinate to be perpendicular to
the slab, with z = 0 at the base of the slab, thickness l.
We assume that the subduction has been going on long
enough that the temperature is in steady state. The
non-dimensional advection-diffusion equation is

∂2T 0 ∂2T 0 ∂T 0
+ − 2Pe = 0, (10.74)
∂x0 2 ∂z 0 2 ∂x0
where the Pe is here defined (for algebraic convenience) Figure 10.20: Sketch of the geometry of
as V l/(2κ) and V is the convergence rate at the trench. the slab calculation. Positive x is along
the slab, in the direction of motion. Posi-
We first obtain an approximate solution to this prob-
tive z is upwards from base of slab.
lem, as we did for the cooling ocean floor, by neglecting
diffusion in the x-direction, and fixing our coordinate system to the descending slab. The initial
temperature profile is  z
Ti = Tm 1 − , (10.75)
l
while the steady-state temperature is Tm , so th temperature perturbation at time t = 0 is
z
Tu (z, 0) = − ,
l
so
The general solution to this problem is (Mc Kenzie, 1969)
∞
X h
1/2  0 i
T 0 = A + Bz 0 + Cn exp Pe − Pe2 + n2 π 2 x sin nπz 0 (10.76)
n=1

With T 0 = 1 on bottom and top of the slab (z 0 = 0, 1), and at infinite x0 , the temperature
everywhere in the slab must equal the mantle temperature (T 0 = 1), so

A = 1; B = 0.

and using the boundary condition

0 0 2 (−1)n
T = 1 − z −→ Cn =
nπ
∞
X (−1)n h
1/2  0 i
0
T =1+2 exp Pe − Pe2 + n2 π 2 x sin nπz 0 (10.77)
nπ
n=1
Recall that we are probably going to get a good approximation to the temperature by considering
only the first term in the sum.
2 h 
1/2  0 i
T 0 ∼ 1 + exp Pe 1 − 1 + π 2 /Pe2 x sin πz 0 (10.78)
π
Convince yourself that the Péclet number for a slab subducting at about 100 mm/yr is very
much greater than π. These simplifications give us an approximate solution:
 2 0
0 2 π x
T = 1 − exp − sin πz 0 (10.79)
π Pe
Mantle Convection and the Lithosphere 129

Putting this expression back into dimensional terms, we can say that the maximum temperature
inside the slab is
  2    2 
2 π κx 2 π κt
Tmax = Tm 1 − exp − = Tm 1 − exp − 2 (10.80)
π V l2 π l

where t is the length of time that it has taken the piece of slab to move the distance x from the
trench.
Alternatively, Equation 10.79 can be rearranged to give the greatest distance that a given
isotherm, T , can penetrate:  
V l2 2
xmax = 2 ln (10.81)
π κ π(1 − T /Tm )
Figure 10.21 shows the distribution of depths of earthquakes in the Tonga subduction zone
plotted against isotherms, calculated from Equation 10.78. The convergence rate increases from
north to south along the subduction zone. It makes sense to suppose that seismicity in the
slab cuts off at some given temperature; Mc Kenzie (1969) made this calculation assuming a
thickness of 50 km for the slab; calculations with modern parameters give a cut-off temperature
of 600-700◦ C.

Figure 10.21: Projection of hypocentres of earthquakes onto a vertical plane along the profile AA’ (inset,
left: the Tonga subduction zone). Lines represent isotherms calculated from Equation 10.78. The right-
hand inset sketches the geometry of the calculation, and isotherms. (Mc Kenzie, 1969).

Exercise for the student: Compare the solution of Equation 10.80 with that for the n = 1
term of the solution for the thermal plate model (Equation 10.47). Make sure that you under-
stand why they are similar.
130 Kinematics and Dynamics of the Lithosphere 2020

10.8.2 Flow at Subduction Zones and Temperature in the Mantle Wedge

We are motivated by the observation of a systematic

variation in depth to the slab beneath volcanic arcs.
When the depth, D, to the top of the slab immediately
beneath the volcanic arcs is plotted against the product
of convergence rate, V , and slab dip, δ, with a small
number of exceptions they lie on an array for which D
decreases with V δ (England and Katz , 2010, see Fig-
ure 10.22).
At first sight, this result might seem surprising,
given that we have shown in the previous section that
the faster the slab is descending, the lower is its tem-
perature at a given depth. However, we need also to
consider the advection of heat in the wedge of man-
tle above the slab. Flow in the mantle wedge brings
hot material towards the top of the slab, forming an
advective boundary layer (Figure 10.23). The key to
understanding the observations in Figure 10.22 lies in Figure 10.22: Depth to the top of the slab
understanding how closely this boundary layer can ap- beneath volcanic arcs, plotted against V δ,
where V is convergence rate perpendicular
proach to the wedge corner. Within the boundary layer,
to the arc and δ is the dip of the slab (after
temperatures are too cool for melting to occur: outside England and Katz (2010)).
the boundary layer, temperatures are close to the upper
mantle temperatures, and therefore high enough for melting to occur.

0.00

0.25
θ/δ

0.50

0.75

1.00
0.00 0.25 0.50 0.75 1.00
2 π uθ / Vδ

Figure 10.23: (Left:) Sketch of a subduction zone. Two plates converge at a speed V; the slab dips at
an angle, δ, and relative motion between the slab and overriding plate generates circulation in the wedge
of mantle between them, which follows stream lines, shown as curved lines; illustrative arrows show the
relative magnitudes of the velocity in different parts of the wedge. Shading depicts the portion of the
wedge that is cooler than the background temperature of the mantle, Tm , while still being hot enough
to participate in the flow. This layer forms an insulating boundary on the top of the subducting slab.
(Right:) uθ (Equation 10.83), divided by V δ/2π and plotted against θ/δ for δ 10◦ to 90◦ in steps of 10◦ .

We calculate flow in the wedge using the ‘corner flow’ solution (see Chapter 6). The velocity
Mantle Convection and the Lithosphere 131

in the wedge, u(ur , uθ ), can be expressed in terms of a stream function, ψ:

1 ∂ψ
ur = (10.82)
r ∂θ
∂ψ
uθ = − (10.83)
∂r
where r, θ are polar coordinates whose origin is at the corner of the flow (Figure 10.23).
With the overriding plate having zero velocity, the stream function, ψ, for the flow in the
wedge above the slab is given by:
rV [(δ − θ) sin θ sin δ − δ θ sin (δ − θ)]
ψ= , (10.84)
δ 2 − sin2 δ
where θ is measured counter-clockwise from the base of the overriding plate (Figure 10.23). This
is the same expression as Eq. (313) in Chapter 6, with δ replacing θp ; the apparent difference
arises from the use of different forms of trigonometric identies.
The tangential component of velocity in the wedge (uθ = ψ/r, Eqs. 10.83, 10.84) scales with
the dip of the slab (see Figure 10.23, right-hand panel). In the bottom one-quarter of the wedge
(θ/δ > 3/4) uθ varies almost linearly from its peak value of V δ/(2π) to zero over the angular
distance of δ/4. So
V (δ − θ)
uθ ∼ (10.85)
2π δ/4
We recast this expression into a more familiar form, involving distances, rather than angular
distances. We use z = r (δ − θ) to describe the distance in the tangential direction away from
the top of the slab, and uz = −uθ .
2V z
uz ∼ − (10.86)
πr
Neglecting diffusion of heat parallel to the top of the slab, and fixing our coordinate axes to
the top of the descending slab, the temperature equation for this boundary layer becomes

d2 T dT
κ = uz ,
dz 2 dz
d2 T 2V z dT
= − ,
dz 2 πκr dz
(10.87)

which has the solution, as we saw in Section 10.7.1,

r !
V
T (z) = Aerf z +B
πκ r
r !
V
T (z) = TS + (Tm − TS )erf z (10.88)
πκ r

where TS is the temperature on the top of the slab. We do not need to know TS ; all we are
interested in here is the thickness of the advective
p boundary layer which this solution tells us,
from the properties of the error function, is πκ r/V . When this boundary layer thickness
becomes comparable with the thickness of the wedge, hot material can no longer be dragged
toward the wedge corner, so melting cannot take place. This condition is met when
p
πκ r/V ∼ rδ (10.89)
πκ
r ∼ (10.90)
V δ2
πκ
D = r sin δ ∼ (10.91)
Vδ
132 Kinematics and Dynamics of the Lithosphere 2020

where we have used the small-angle approximation for sin δ. Thus, the observations of Fig-
ure 10.22 may be explained if we assume that the volcanic arcs form above the place where a
critical isotherm in the mantle wedge comes closest to the wedge corner.
Mantle Convection and the Lithosphere 133

References
Allmann, B. P., and P. M. Shearer (2009), Global variations of stress drop for moderate to large earth-
quakes, Journal of Geophysical Research, 114, 22.

Bourne, S., P. England, and B. Parsons (1998), The motion of crustal blocks driven by flow of the lower
lithosphere and implications for slip rates of continental strike-slip faults, Nature, 391, 655–659.

Brace, W., and D. Kohlstedt (1980), Limits on lithostatic stress imposed by laboratory experiments,
J. Geophys. Res., 85, 6,248–6,252.

Bürgmann, R., and G. Dresen (2008), Rheology of the lower crust and upper mantle: Evidence from rock
mechanics, geodesy, and field observations, Ann. Rev. Earth Planet. Sci., 36, 531–567.

Carslaw, H. S., and J. C. Jaeger (1959), Conduction of Heat in Solids, second ed., Oxford University
Press.

Cowie, P., and C. Scholz (1992), Displacement-length scaling relationship for faults: data synthesis and
discussion, J Struct Geol, 14, 1149–1156.

Crosby, A. G., and D. Mc Kenzie (2009), An analysis of young ocean depth, gravity and global residual
topography, Geophys. J. Int., 178, 1198–1219.

Crosby, A. G., D. Mc Kenzie, and J. G. Sclater (2006), The relationship between depth, age, and gravity
in the oceans, Geophys. J. Int., 166, 553–573.

D’Agostino, N., P. England, I. Hunstad, and G. Selvaggi (2014), Gravitational potential energy and
deformation of the Apennines, Earth Planetary Science Lett., 397, 121–132.

Dalmayrac, B., and P. Molnar (1981), Parallel thrust and normal faulting in Peru and constraints on the
state of stress, Earth Planet. Sci. Lett., 55, 473–481.

Davies, R., P. England, B. Parsons, H. Billiris, D. Paradissis, and G. Veis (1997), Geodetic strain of
Greece in the interval 1892–1992, J. Geophys. Res., 102, 24,571–24,588.

Ekström, G., and P. C. England (1989), Seismic strain rates in regions of continental deformation,
J.Geophys. Res., 94, 10,231– 10,245.

Ekström, G., M. Nettles, and A. Dziewonski (2012), The Global CMT Project 2004–2010: centroid-
moment tensors for 13,017 earthquakes, Phys. Earth Planet Inter., 200-201, 1–9.

Elliott, J. R., R. J. Walters, P. C. England, J. A. Jackson, Z. Li, and B. Parsons (2010), Extension on the
Tibetan plateau: recent normal faulting measured by InSAR and body wave seismology, Geophysical
Journal International, 183, 505–535.

England, P., and P. Molnar (1997), Active deformation of Asia: From kinematics to dynamics, Science,
278, 647–650.

England, P., and P. Molnar (2015), Rheology of the lithosphere beneath the central and western Tien
Shan, Journal of Geophysical Research-Solid Earth, 120, 3803–3823.

England, P., G. Houseman, and L. Sonder (1985), Length scales for continental deformation in convergent,
divergent and strike-slip environments: analytical and approximate solutions for a thin viscous sheet
model, J. Geophys. Res., 90, 3,551–3,557.

England, P., G. Houseman, and J.-M. Nocquet (2016), Constraints from GPS measurements on the
dynamics of deformation in Anatolia and the Aegean, Journal Of Geophysical Research-Solid Earth,
121, 8888–8916.

England, P. C., and R. F. Katz (2010), Melting above the anhydrous solidus controls the location of
volcanic arcs, Nature, 467, 700–703.

England, P. C., and S. W. Richardson (1977), The influence of erosion upon the mineral facies of rocks
from different metamorphic environments, J. Geol. Soc., 134, 201–213.
134 Kinematics and Dynamics of the Lithosphere 2020

England, P. C., and A. B. Thompson (1984), Pressure–Temperature–time paths of regional metamor-

phism I. Heat transfer during the evolution of regions of thickened crust, J. Petrol., 25, 894–928.

Flesch, L. M., W. E. Holt, A. J. Haines, and T. B. Shen (2000), Dynamics of the Pacific-North American
plate boundary in the Western United States, Science, 287, 834–836.

Floyd, M. A., H. Billiris, D. Paradissis, G. Veis, A. Avallone, P. Briole, S. McClusky, J.-M. Nocquet,
B. Parsons, and P. C. England (2010), A new velocity field for Greece: Implications for the kinematics
and dynamics of the Aegean, J. Geophys. Res., 115, B10,403.

Gordon, R. G., and G. A. Houseman (2015), Deformation of Indian Ocean lithosphere: Evidence for a
highly nonlinear rheological law, J. Geophys. Res. Solid Earth, 120, 4434–4449.

Hanks, T., and H. Kanamori (1979), A moment magnitude scale, J. Geophys. Res, 84, 2348–2350.

Hilley, G. E., K. Johnson, M. Wang, Z.-K. Shen, and R. Burgmann (2009), Earthquake-cycle deformation
and fault slip rates in northern Tibet, Geology, 37, 31–34.

Hirth, G., and D. Kohlstedt (2003), Rheology of the upper mantle and the mantle wedge: A view from
the experimentalists, Inside the subduction Factory, pp. 83–105.

Jackson, J., and D. P. Mc Kenzie (1988), The relationship between plate motions and seismic moment
tensors and the rate of active deformation in the Mediterranean and Middle East, Geophys. J. R. Astr.
Soc., 93, 45–73.

Jaupart, C., and J.-C. Mareschal (2007), Heat flow and thermal structure of the lithosphere, in Treatise
on Geophysics, chap. 6.05, Elsevier, Amsterdam.

Kostrov, B. (1974), Seismic moment and energy of earthquakes, and seismic flow of rock, Izv. Acad. Sci.
USSR Phys. Solid Earth, 97, 23–44.

Kreemer, C., G. Blewitt, and E. C. Klein (2014), A geodetic plate motion and Global Strain Rate Model,
Geochemistry, Geophysics, Geosystems, 15, 3849–3889.

Maggi, A., J. A. Jackson, D. Mc Kenzie, and K. Priestley (2000), Earthquake focal depths, effective elastic
thickness, and the strength of the continental lithosphere, Geology (Boulder), 28, 495–498.

Mc Kenzie, D. (1969), Speculations on the consequences and causes of plate motion, Geophys. J. Royal
Astron. Soc., 18, 1–32.

Mc Kenzie, D. (1972), Active tectonics of the Mediterranean region, Geophys. J. Royal Astron. Soc., 30,
109–185.

Mc Kenzie, D. (1977), Surface deformation, gravity anomalies and convection, Geophysical Journal of the
Royal Astronomical Society, 48, 211–238.

Mc Kenzie, D. (1978), Some remarks on the development of sedimentary basins, Earth Planet. Sci. Lett.,
40, 25–32.

Mei, S., A. M. Suzuki, D. L. Kohlstedt, N. A. Dixon, and W. B. Durham (2010), Experimental constraints
on the strength of the lithospheric mantle, Journal of geophysical research, 115, B08,204.

Molnar, P. (1983), Average regional strain due to slip on numerous faults of different orientations, J.
Geophys. Res., 88, 6,430–6,432.

Molnar, P., and Q. Deng (1984), Faulting associated with large earthquakes and the average rate of
deformation in central and eastern Asia, J. Geophys. Res., 89, 6203–6227.

Molnar, P., T. Fitch, and F. Wu (1973), Fault plane solutions of shallow earthquakes and contemporary
tectonics in Asia, Earth Planet. Sci. Lett., 19, 101–112.

Parsons, B., and Mc Kenzie (1978), Mantle convection and the thermal structure of the plates, J. Geophys.
Res., 83, 4485–4496.
Mantle Convection and the Lithosphere 135

Parsons, B., and J. G. Sclater (1977), An analysis of the variation of ocean floor bathymetry and heat
flow with age, J. Geophys. Res., 82, 802–827.

Sloan, R. A., J. A. Jackson, D. Mc Kenzie, and K. Priestley (2011), Earthquake depth distributions in
central Asia, and their relations with lithosphere thickness, shortening and extension, Geophysical
Journal International, pp. 1–29.

Sonder, L., and P. England (1986), Vertical averages of rheology of the continental lithosphere: relation
to thin sheet parameters, Earth Planet. Sci. Lett., 77, 81–90.

Tapponnier, P., and P. Molnar (1976), Slip-line field theory and large-scale continental tectonics, Nature,
264, 319.

Tapponnier, P., G. Peltzer, and R. Armijo (1986), On the mechanics of the collision between India and
Asia, in Collision Tectonics, Spec. Publ. 19, edited by M. Coward and A. Ries, pp. 115–157, Geological
Society London, Special Publication No. 19.

Walters, R. J., P. C. England, and G. A. Houseman (2017), Constraints from GPS measurements on the
dynamics of the zone of convergence between Arabia and Eurasia, Journal of Geophysical Research-
Solid Earth, 67, 40–26.

Wells, D., and K. Coppersmith (1994), New empirical relationships among magnitude, rupture length,
rupture width, rupture area, and surface displacement, Bull. Seismol. Soc. Am., 84, 974–1002.

Wen, Y., Z. Li, C. Xu, I. Ryder, and R. Bürgmann (2012), Postseismic motion after the 2001 Mw 7.8
Kokoxili earthquake in Tibet observed by InSAR time series, Journal of Geophysical Research, 117,
B08,405.

Yamasaki, T., and G. A. Houseman (2012), The crustal viscosity gradient measured from post-seismic
deformation a case study of the 1997 Manyi (Tibet) earthquake, Earth Planet. Sci. Lett., 351-352,
105–114.

Zubovich, A. V., X.-Q. Wang, Y. G. Scherba, G. G. Schelochkov, R. Reilinger, C. Reigber, O. I. Mosienko,

P. Molnar, W. Michajljow, V. I. Makarov, J. Li, S. I. Kuzikov, T. A. Herring, M. W. Hamburger, B. H.
Hager, Y.-M. Dang, V. D. Bragin, and R. T. Beisenbaev (2010), GPS velocity field for the Tien Shan
and surrounding regions, Tectonics, 29, TC6014.